Search

Atmospheric extreme events cause severe damage to human societies and ecosystems. The frequency and intensity of extremes and other associated events are continuously increasing due to climate change and global warming. The accurate prediction, characterization, and attribution of atmospheric extreme events is, therefore, a key research field in which many groups are currently working by applying different methodologies and computational tools. Machine learning and deep learning methods have arisen in the last years as powerful techniques to tackle many of the problems related to atmospheric extreme events. This paper reviews machine learning and deep learning approaches applied to the analysis, characterization, prediction, and attribution of the most important atmospheric extremes. A summary of the most used machine learning and deep learning techniques in this area, and a comprehensive critical review of literature related to ML in EEs, are provided. The critical literature review has been extended to extreme events related to rainfall and floods, heatwaves and extreme temperatures, droughts, severe weather events and fog, and low-visibility episodes. A case study focused on the analysis of extreme atmospheric temperature prediction with ML and DL techniques is also presented in the paper. Conclusions, perspectives, and outlooks on the field are finally drawn. [ABSTRACT FROM AUTHOR]

Colloidal quantum dots (CQDs) have been proposed to obtain intermediate band (IB) materials. The IB solar cell can absorb sub-band-gap photons via an isolated IB within the gap, generating extra electron-hole pairs that increase the current without degrading the voltage, as has been demonstrated experimentally for real cells. In this paper, we model the electron hopping transport (HT) as a network embedded in space and energy so that a node represents the first excited electron state localized in a CQD while a link encodes the Miller–Abrahams (MA) hopping rate for the electron to hop from one node (=state) to another, forming an "electron-HT network". Similarly, we model the hole-HT system as a network so that a node encodes the first hole state localized in a CQD while a link represents the MA hopping rate for the hole to hop between nodes, leading to a "hole-HT network". The associated network Laplacian matrices allow for studying carrier dynamics in both networks. Our simulations suggest that reducing both the carrier effective mass in the ligand and the inter-dot distance increases HT efficiency. We have found a design constraint: It is necessary for the average barrier height to be larger than the energetic disorder to not degrade intra-band absorption. [ABSTRACT FROM AUTHOR]

Organic disordered semiconductors have a growing importance because of their low cost, mechanical flexibility, and multiple applications in thermoelectric devices, biosensors, and optoelectronic devices. Carrier transport consists of variable-range hopping between localized quantum states, which are disordered in both space and energy within the Gaussian disorder model. In this paper, we model an organic disordered semiconductor system as a network embedded in both space and energy so that a node represents a localized state while a link encodes the probability (or, equivalently, the Miller–Abrahams hopping rate) for carriers to hop between nodes. The associated network Laplacian matrix allows for the study of carrier dynamics using edge-centric random walks, in which links are activated by the corresponding carrier hopping rates. Our simulation work suggests that at room temperature the network exhibits a strong propensity for small-network nature, a beneficial property that in network science is related to the ease of exchanging information, particles, or energy in many different systems. However, this is not the case at low temperature. Our analysis suggests that there could be a parallelism between the well-known dependence of carrier mobility on temperature and the potential emergence of the small-world property with increasing temperature. [ABSTRACT FROM AUTHOR]

Numerical association rule mining (NARM) is a popular method under the umbrella of data mining, focused on finding relationships between attributes in transaction databases. Numerical association rules for time series are a new paradigm that extends the applicability of NARM to the domain of time series. Association rule mining algorithms result in numerous rules, the interpretation of which is sometimes not easy for human experts. Therefore, various visualization methods have been developed to improve the explanation results of the rule mining process. This article is a novel contribution to the development of a new visualization method capable of presenting the association rules for time series developed according to the principles of explainable artificial intelligence. The experiments are conducted in the context of smart agriculture (i.e., agricultural time series data), and show the great potential of the proposed visualization method for the future. [ABSTRACT FROM AUTHOR]

This research paper aims to develop a hybrid neuro-evolutionary wrapper-based model for daily global solar radiation estimation in the solar-rich Sunshine State of Queensland, Australia. To design a robust hybrid modelling mechanism, the Interim-ERA European Centre for Medium Range Weather Forecasting (ECMWF) Reanalysis data are employed to train and cross-validate the estimation model that is formulated by an evolutionary-type algorithm: the Coral Reefs Optimization (CRO) integrated with an Extreme Learning Machine (ELM) model. The hybrid CRO-(ELM) algorithm is applied in two stages: first for the feature selection process guided by an ELM algorithm (a class of fast training neural network tool) as the model’s fitness function to screen an optimal set of predictor variables and second, for the estimation of the solar radiation using the optimally screened variables by the final hybrid CRO-(ELM)-ELM system. To benchmark the performance of the hybrid CRO-ELM algorithm for this estimation problem we apply an alternative, yet a similar feature screening approach: the Grouping Genetic Algorithm encoded into the ELM-based model (GGA-(ELM) used as the predictor mechanism). After the feature selection process is performed by the CRO algorithm that extracts the data patterns for accurate estimation the alternative objective algorithm is applied (in this case the ELM again) to formulate the hybrid CRO-(ELM)-ELM modelling system. To provide a rigorous evaluation of the CRO-(ELM)-ELM hybrid system, alternative estimation approaches are considered: the Multivariate Adaptive Regression Splines (MARS), Multiple Linear Regression (MLR) and the Support Vector Regression (SVR). The hybrid CRO-(ELM)-ELM system is tested in a real problem where the results are evaluated by means of several statistical score metrics and diagnostic plots of the observed and the estimated daily global solar radiation in the testing phase. We show that the hybrid CRO-(ELM)-ELM model is able to yield promising results; thus improving those attained by the 7 alternative models (i.e., hybrid CRO-(ELM)-MARS, CRO-(ELM)-MLR and CRO-(ELM)-SVR and the GGA equivalent models). The study ascertains that the CRO-based hybrid system where a large pool of predictor data are carefully screened through a wrapper-based modelling system and the ELM model is applied as a objective estimation tool can be adopted as a qualified stratagem in solar radiation estimation problems. The proposed hybrid CRO-(ELM)-ELM system exhibits clear advantages over the alternative machine learning approaches tested and also over the other machine learning algorithms used without the feature selection tool; thus advocating its scientific utility in renewable energy applications. [ABSTRACT FROM AUTHOR]

This paper proposes a novel optimization algorithm for grouping problems, the Grouping Coral Reefs Optimization algorithm, and describes its application to a Mobile Network Deployment Problem (MNDP) under four optimization criteria. These criteria include economical cost and coverage, and also electromagnetic pollution control and capacity constraints imposed at the base stations controllers, which are novel in this study. The Coral Reefs Optimization algorithm (CRO) is a recently-proposed bio-inspired approach for optimization, based on the simulation of the processes that occur in coral reefs, including reproduction, fight for space or depredation. This paper presents a grouping version of the CRO, which has not previously evaluated before. Grouping meta-heuristics are characterized by variable-length encoding solutions, and have been successfully applied to a number of different optimization and assignment problems. The GCRO proposed is a novel contribution to the intelligent systems field, which is able to improve results obtained by two alternative grouping algorithms such as grouping genetic algorithms and grouping Harmony Search. The performance of the proposed GCRO and the algorithms for comparison has been tested with real data in a case study of a MNDP in Alcalá de Henares, Madrid, Spain. [ABSTRACT FROM AUTHOR]

This paper reviews the most relevant works that have investigated robustness in power grids using Complex Networks (CN) concepts. In this broad field there are two different approaches. The first one is based solely on topological concepts, and uses metrics such as mean path length, clustering coefficient, efficiency and betweenness centrality, among many others. The second, hybrid approach consists of introducing (into the CN framework) some concepts from Electrical Engineering (EE) in the effort of enhancing the topological approach, and uses novel, more efficient electrical metrics such as electrical betweenness, net-ability, and others. There is however a controversy about whether these approaches are able to provide insights into all aspects of real power grids. The CN community argues that the topological approach does not aim to focus on the detailed operation, but to discover the unexpected emergence of collective behavior, while part of the EE community asserts that this leads to an excessive simplification. Beyond this open debate it seems to be no predominant structure (scale-free, small-world) in high-voltage transmission power grids, the vast majority of power grids studied so far. Most of them have in common that they are vulnerable to targeted attacks on the most connected nodes and robust to random failure. In this respect there are only a few works that propose strategies to improve robustness such as intentional islanding, restricted link addition, microgrids and smart grids, for which novel studies suggest that small-world networks seem to be the best topology. [ABSTRACT FROM AUTHOR]

An iterative neural network framework is proposed in this paper for the human-induced Ground Reaction Forces (GRF) replication with an inertial electrodynamic mass actuator (APS 400). This is the first approach to the systematization of dynamic load tests on structures on a purely objective, repeatable and pedestrian-independent basis. Therefore, an inversion-free offline algorithm based on Machine Learning techniques has been applied for the first time on an electrodynamic shaker, without requiring its inverse model to tackle the inverse problem of successful force reconstruction. The proposed approach aims to obtain the optimal drive signal to minimize the error between the experimental shaker output and the reference force signal, measured with a pair of instrumented insoles (Loadsol©) for human bouncing at different frequencies and amplitudes. The optimal performance, stability and convergence of the system are verified through experimental tests, achieving excellent results in both time and frequency domain. [ABSTRACT FROM AUTHOR]

Abstract: Hyper-parameters estimation in regression Support Vector Machines (SVMr) is one of the main problems in the application of this type of algorithms to learning problems. This is a hot topic in which very recent approaches have shown very good results in different applications in fields such as bio-medicine, manufacturing, control, etc. Different evolutionary approaches have been tested to be hybridized with SVMr, though the most used are evolutionary approaches for continuous problems, such as evolutionary strategies or particle swarm optimization algorithms. In this paper we discuss the application of two different evolutionary computation techniques to tackle the hyper-parameters estimation problem in SVMrs. Specifically we test an Evolutionary Programming algorithm (EP) and a Particle Swarm Optimization approach (PSO). We focus the paper on the discussion of the application of the complete evolutionary-SVMr algorithm to a real problem of wind speed prediction in wind turbines of a Spanish wind farm. [ABSTRACT FROM AUTHOR]

Color reduction in images is an important problem in image processing, since it is a pre-processing step in applications such as image segmentation or compression. Different methods have been proposed in the literature, several of them involving nature-inspired algorithms such as neural networks. However, not many works involving evolutionary computation techniques have been applied to this problem. This paper proposes a novel evolutionary algorithm to tackle the color reduction of RGB images. The proposed evolutionary algorithm incorporates a procedure called incremental-encoding, consisting in starting the image quantization with a small number of colors, and including additional colors in a gradual form, until reaching the final number of quantization colors. In the experiments carried out we show that the incremental-encoding evolutionary algorithm improves the performance of the standard evolutionary algorithm in this problem. Also we show that it obtains better results than several existing color reduction techniques for color quantization problems. [ABSTRACT FROM AUTHOR]

Abstract: The selection of hyper-parameters in support vector machines (SVM) is a key point in the training process of these models when applied to regression problems. Unfortunately, an exact method to obtain the optimal set of SVM hyper-parameters is unknown, and search algorithms are usually applied to obtain the best possible set of hyper-parameters. In general these search algorithms are implemented as grid searches, which are time consuming, so the computational cost of the SVM training process increases considerably. This paper presents a novel study of the effect of including reductions in the range of SVM hyper-parameters, in order to reduce the SVM training time, but with the minimum possible impact in its performance. The paper presents reduction in parameter C, by considering its relation with the rest of SVM hyper-parameters ( and ), through an approximation of the SVM model. On the other hand, we use some characteristics of the Gaussian kernel function and a previous result in the literature to obtain novel bounds for and hyper-parameters. The search space reductions proposed are evaluated in different regression problems from UCI and StatLib databases. All the experiments carried out applying the popular LIBSVM solver have shown that our approach reduces the SVM training time, maintaining the SVM performance similar to when the complete range in SVM parameters is considered. [Copyright &y& Elsevier]

This paper describes an educational simulation tool, GSMSIM, which has been developed to teach GSM mobile communications in laboratory lectures. The tool has been programmed in OMNET++, a flexible environment which allows its extension to different aspects of GSM technology, such as the simulation of calls, traffic, handoffs, etc. This tool has been used in the courses 'Communication Systems' and 'Telecommunication Networks I', at the University of Alcalá, Madrid, Spain. We report some of the results obtained in three different scenarios simulating a small, medium and large city, with different characteristics of network coverage. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents the hybridization of the fifth generation mesoscale model (MM5) with neural networks in order to tackle a problem of short-term wind speed prediction. The mean hourly wind speed forecast at wind turbines in a wind park is an important parameter used to predict the total power production of the park. Our model for short-term wind speed forecast integrates a global numerical weather prediction model and observations at different heights (using atmospheric soundings) as initial and boundary conditions for the MM5 model. Then, the outputs of this model are processed using a neural network to obtain the wind speed forecast in specific points of the wind park. In the experiments carried out, we present some results of wind speed forecasting in a wind park located at the south-east of Spain. The results are encouraging, and show that our hybrid MM5-neural network approach is able to obtain good short-term predictions of wind speed at specific points. [Copyright &y& Elsevier]

Abstract: This paper presents a novel application of the hybrid grouping genetic algorithm in a problem related to university timetabling. Specifically, the assignment of students to laboratory groups is tackled. This problem includes an important constraint of capacity, due to laboratories usually have a maximum number of equips or computers available, so the number of total students in a group is constrained to be equal or less than the capacity of the laboratory. In addition, our approach considers the case in which the students provide a sorted list of preferred laboratory groups, so the objective of the assignment must take this point into account. A variation of the problem in which a balanced number of students per group is required (lecturer preferences) is also studied in this paper. The performance of the approach is shown in different test problems and in a real application in a Spanish University. [Copyright &y& Elsevier]

In this paper a novel scheme to solve multiclass classification problems using pure evolutionary techniques is presented. Our proposal consists of the evolution of several geometric structures such as hypercubes, hyperspheres or hyperoctahedrons, to obtain a first division of the space, in which the training samples are assigned to one or several structures. In in a second step, the samples belonging to two or more structures are re-evolved in order to obtain a single classifier. We call this approach the Evolution of Geometric Structures (EGS) algorithm. Its main characteristics are described in the paper. An application of the EGS to a well known multiclass classification problem, the automatic classification of high range resolution (HRR) radar targets, is also discussed. [ABSTRACT FROM AUTHOR]

Abstract: Wind speed prediction is a very important part of wind parks management. Currently, hybrid physical-statistical wind speed forecasting models are used to this end, some of them using neural networks as the final step to obtain accurate wind speed predictions. In this paper we propose a method to improve the performance of one of these hybrid systems, by exploiting diversity in the input data of the neural network part of the system. The diversity in the data is produced by the physical models of the system, applied with different parameterizations. Two structures of neural network banks are used to exploit the input data diversity. We will show that our method is able to improve the performance of the system, obtaining accurate wind speed predictions better than the one obtained by the system using single neural networks. [Copyright &y& Elsevier]

Abstract: This paper presents three memetic algorithms to solve the spread spectrum radar polyphase code design problem, based on Evolutionary Programming, Particle Swarm Optimization and Differential Evolution, respectively. These global search heuristics are hybridized with a gradient-based local search procedure which includes a dynamic step adaptation procedure to perform accurate and efficient local search for better solutions. We have compared the different memetic algorithms proposed in several numerical examples, and we have also demonstrated the performance of our approaches against existing approaches for this problem. [Copyright &y& Elsevier]

Abstract: In this paper we present a novel approach to the problem of speaker segmentation, which is an unavoidable previous step to audio indexing. Mutual information is used for evaluating the accuracy of the segmentation, as a function to be maximized by a simulated annealing (SA) algorithm. We introduce a novel mutation operator for the SA, the Consecutive Bits Mutation operator, which improves the performance of the SA in this problem. We also use the so-called Compaction Factor, which allows the SA to operate in a reduced search space. Our algorithm has been tested in the segmentation of real audio databases, and it has been compared to several existing algorithms for speaker segmentation, obtaining very good results in the test problems considered. [Copyright &y& Elsevier]

Abstract: This paper proposes the use of a bank of Hopfield networks to solve a class of constraints which appear in combinatorial optimization problems. Specifically, we deal with problems which constraints’ structure can be represented by a binary matrix C, and can be separated in independent substructures. We show that a bank of Hopfield networks can solve these constraints, and also can be easily hybridized with a global search algorithm, such as simulated annealing, to obtain a final solution to the problem. We apply our approach to the solution of a famous logic-type puzzle, the light-up puzzle, where we report improvements over a branch and bound algorithm, and to an important problem which arises in electronic control: the so-called Crossbar Switch Problem. [Copyright &y& Elsevier]

Abstract: A picture-logic puzzle is a game that takes the form of an grid, with numbers situated on the left of its rows and on the top of its columns, which give the clues for solving the puzzle. These puzzles have gained popularity in the last years all over the world, and there are companies involved in the commercialization of products related to them, mainly magazines, on-line puzzles via the web and games for mobile phones. The main problem with selling picture-logic puzzles is the need to generate a large number of picture-logic puzzles per month, or even per week, so automating the generation process of such puzzles is an important task. In this paper we propose algorithms for the automated generation of picture-logic puzzles (both black and white and color puzzles), starting from any digital RGB color image. We also present a visualization interface which provides several useful tools for finishing the puzzle generation process, like the possibility of changing colors and displaying the process of puzzle solution. We show the performance of the algorithms for generating picture-logic puzzles with several examples. [Copyright &y& Elsevier]

Abstract: In this paper a sequential binary Hopfield network architecture to solve crossbar switch problems (CSP) is proposed. It is shown that the sequential Hopfield algorithm''s performance on the CSP depends on the neurons’ updating ordering, and also an optimal ordering for the CSP is proposed. Our sequential binary Hopfield network has been applied to the resolution of large and very large CSP instances, obtaining optimal solutions within less steps than previous approaches to this problem. [Copyright &y& Elsevier]

This paper discusses new results about the effect of linear transformations (LTs) over the representation, the spectrum and the performance of standard genetic algorithms (GAs). Unitary LTs are used to set the basic outcomes and, although they can be used for theoretic GA-hard problems, their applicability is shown to be quite limited. Nevertheless, the redundancy of problems can be exploited by means of non-unitary LTs. We also propose an heuristic for construction of non-unitary LTs. Several experiments are performed in order to discuss the theoretical results of the paper. [ABSTRACT FROM AUTHOR]

In this paper, a method to teach advanced features of evolutionary algorithms (EAs), using a famous game known as Japanese puzzles is presented. The authors show that Japanese puzzles are constrained combinatorial optimization problems, that can be solved using EAs with different encodings, and are challenging problems for EAs. Other features, such as special operators and local search heuristics and its hybridization with genetic algorithms, can also be taught using these puzzles. The authors report an experience using this method in a course taught at the Universidad de Alcalá, Madrid, Spain. [ABSTRACT FROM AUTHOR]

In this paper, two evolutionary programming (EP) algorithms (classical EP and fast EP) are applied to design prototype lowpass Finite Impulse Response filters for use in a modulated filter bank. The chosen filter design technique is based on frequency-sampling (where the Fourier transform magnitudes of the filter are the objective variables). Design is simplified by constraining most of these values, leaving only a small number of values in the filter transition band to be optimized. The EP algorithms were used to determine the optimum values for this subset of values. Since there is an additional monotomc constraint on the transition band values, a modification to the EP algorithms was developed called variable limits evolutionary programming. Results indicate that a) both EP algorithms were insensitive to initial conditions, and reliably found the minimum values of the chosen objective functions, and b) the designed prototype filters are suitable to obtain near-perfect reconstruction filter banks, offering quality parameters comparable or better than systems obtained using other techniques. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents an extension of the terminal assignment problem (TA) in the case that groups of terminals must be assigned together. We analyze this situation by means of an equivalent problem: the wedding banquet problem (WBP). We provide a description of the problem and its mathematical definition. We also describe an application of the WBP to mobile communications network design. Two hybrid metaheuristics algorithms for the WBP are presented in the paper. We test their performance in several computational experiments, including synthetic instances of the WBP, and a mobile network design application. [Copyright &y& Elsevier]

No abstract received. [ABSTRACT FROM AUTHOR]

The widespread deployment of WDM optical networks posts lots of new challenges for network designers. Traffic grooming is one of the most common problems. Efficient grooming of traffic can effectively reduce the overall cost of the network. But unfortunately, traffic grooming problems have been shown to be NP-hard. Therefore, new heuristics must be devised to tackle them. Among these approaches, metaheuristics are among the most promising ones. In this paper, we present a thorough and comprehensive survey on various metaheuristic approaches to the grooming of traffic in both static and dynamic patterns in WDM optical networks. Some future challenges and research directions are also discussed in this paper. [ABSTRACT FROM AUTHOR]

This paper presents a hybrid Hopfield network-genetic algorithm (GA) approach to tackle the terminal assignment (TA) problem. TA involves determining minimum cost links to form a communications network, by connecting a given set of terminals to a given collection of concentrators. Some previous approaches provide very good results if the cost associated with assigning a single terminal to a given concentrator is known. However, there are situations in which the cost of a single assignment is not known in advance, and only the cost associated with feasible solutions can be calculated. In these situations, previous algorithms for TA based on greedy heuristics are no longer valid, or fail to get feasible solutions. Our approach involves a Hopfield neural network (HNN) which manages the problem's constraints, whereas a GA searches for high quality solutions with the minimum possible cost. We show that our algorithm is able to achieve feasible solutions to the TA in instances where the cost of a single assignment in not known in advance, improving the results obtained by previous approaches. We also show the applicability of our approach to other problems related to the TA. [ABSTRACT FROM AUTHOR]

This article focuses on a twofold approach to improve the performance of genetic algorithms (GAs) in the feature selection problem (FSP). The feature selection problem (FSP) is an open issue in machine learning, which basically consists of finding a subset of input features that describes the underlying system structure as well or better than all available features. There is a more general problem, known as feature extraction problem, in which one has to construct a few features that encode all of the information contained in the available original ones; the FSP is, in fact a particularization of this problem.

In trainable wireless communications systems, the use of deep learning for over-the-air training aims to address the discontinuity in backpropagation learning caused by the channel environment. The primary methods supporting this learning procedure either directly approximate the backpropagation gradients using techniques derived from reinforcement learning, or explicitly model the channel environment by training a generative channel model. In both cases, over-the-air training of transmitter and receiver requires a feedback channel to sound the channel environment and obtain measurements of the learning objective. The use of continuous feedback not only demands extra system resources but also makes the training process more susceptible to adversarial attacks. Conversely, opting for a feedback-free approach to train the models over the forward link, exclusively on the receiver side, could pose challenges to reliably end the training process without intermittent testing over the actual channel environment. In this article, we propose a novel method for the over-the-air training of wireless communication systems that does not require a feedback channel to train the transmitter and receiver. Random samples are transmitted through the channel environment to train a mixture density network to approximate the channel distribution on the receiver side of the network. The transmitter and receiver models are trained with the resulting channel model, and the transmitter can be deployed after training. We show that the block error rate measurements obtained with the simulated channel are suitable for monitoring as a stopping criterion during the training process. The resulting method is demonstrated to have equivalent performance to the end-to-end autoencoder training on small message sequences. [ABSTRACT FROM AUTHOR]

In recent years, reinforcement learning (RL) techniques have achieved great success in many different applications. However, their heavy reliance on complex deep neural networks makes most RL models uninterpretable, limiting their application in domains where trust and security are important. To address this challenge, we propose MENS-DT-RL, an algorithm capable of constructing interpretable models for RL via the evolution of decision tree (DT) models. MENS-DT-RL uses a multi-method ensemble algorithm to evolve univariate DTs, guiding the process with a fitness metric that prioritizes interpretability and consistent high performance. Three different initializations for the MENS-DT-RL are proposed, including the use of Imitation Learning (IL) techniques, and a novel pruning approach that reduces solution size without compromising performance. To evaluate the proposed approach, we compare it with other models from the literature on three benchmark tasks from the OpenAI Gym library, as well as on a fertilization problem inspired by real-world crop management. To the best of our knowledge, the proposed scheme is the first to solve the Lunar Lander benchmark with both interpretability and a high confidence rate (90% of episodes are successful), as well as the first to solve the Mountain Car environment with a tree of only 7 nodes. On the real-world task, the proposed MENS-DT-RL is able to produce solutions with the same quality as deep RL policies, with the added bonus of interpretability. We also analyze the best solutions found by the algorithm and show that they are not only interpretable but also diverse in their behavior, empowering the end user with the choice of which model to apply. Overall, the findings show that the proposed approach is capable of producing high-quality transparent models for RL, achieving interpretability without losing performance. • A multi-method ensemble evolutionary algorithm is proposed for Interpretable RL. • Interpretability is achieved by using Decision Trees as RL agents. • Imitation Learning is leveraged to jumpstart the evolutionary algorithm. • A novel method is proposed to prune DTs in RL environments. • State-of-the-art results are obtained for several RL benchmarks. [ABSTRACT FROM AUTHOR]

Exposure to solar ultraviolet (UV) radiation can cause malignant keratinocyte cancer and eye disease. Developing a user-friendly, portable, real-time solar UV alert system especially or wearable electronic mobile devices can help reduce the exposure to UV as a key measure for personal and occupational management of the UV risks. This research aims to design artificial intelligence-inspired early warning tool tailored for short-term forecasting of UV index (UVI) integrating satellite-derived and ground-based predictors for Australian hotspots receiving high UV exposures. The study further improves the trustworthiness of the newly designed tool using an explainable artificial intelligence approach. An enhanced joint hybrid explainable deep neural network model (called EJH-X-DNN) is constructed involving two phases of feature selection and hyperparameter tuning using Bayesian optimization. A comprehensive assessment of EJH-X- DNN is conducted with six other competing benchmarked models. The proposed model is explained locally and globally using robust model-agnostic explainable artificial intelligence frameworks such as Local Interpretable Model-Agnostic Explanations (LIME), Shapley additive explanations (SHAP), and permutation feature importance (PFI). The newly proposed model outperformed all benchmarked models for forecasting hourly horizons UVI , with correlation coefficients of 0.900, 0.960, 0.897, and 0.913, respectively, for Darwin, Alice Springs, Townsville, and Emerald hotspots. According to the combined local and global explainable model outcomes, the site-based results indicate that antecedent lagged memory of UVI and solar zenith angle are influential features. Predictions made by EJH-X-DNN model are strongly influenced by factors such as ozone effect, cloud conditions, and precipitation. With its superiority and skillful interpretation, the UVI prediction system reaffirms its benefits for providing real-time UV alerts to mitigate risks of skin and eye health complications, reducing healthcare costs and contributing to outdoor exposure policy. • Explainable artificial intelligence xAI for 1-hr ahead solar ultraviolet UV index prediction. • xAI model is optimized with the Bayesian feature selection algorithm. • xAI methods provide physical understandings of model prediction outcomes. • xAI model is significantly accurate for solar UV index predictions. • xAI model can help in sun exposure risk mitigation, skin and eye health protection. [ABSTRACT FROM AUTHOR]

Wireless communications systems are traditionally designed by independently optimising signal processing functions based on a mathematical model. Deep learning-enabled communications have demonstrated end-to-end design by jointly optimising all components with respect to the communications environment. In the end-to-end approach, an assumed channel model is necessary to support training of the transmitter and receiver. This limitation has motivated recent work on over-the-air training to explore disjoint training for the transmitter and receiver without an assumed channel. These methods approximate the channel through a generative adversarial model or perform gradient approximation through reinforcement learning or similar methods. However, the generative adversarial model adds complexity by requiring an additional discriminator during training, while reinforcement learning methods require multiple forward passes to approximate the gradient and are sensitive to high variance in the error signal. A third, collaborative agent-based approach relies on an echo protocol to conduct training without channel assumptions. However, the coordination between agents increases the complexity and channel usage during training. In this article, we propose a simpler approach for disjoint training in which a local receiver model approximates the remote receiver model and is used to train the local transmitter. This simplified approach performs well under several different channel conditions, has equivalent performance to end-to-end training, and is well suited to adaptation to changing channel environments. [ABSTRACT FROM AUTHOR]

In this paper, new probabilistic and dynamic (adaptive) strategies for creating multi-method ensembles based on the coral reef optimization with substrate layers (CRO-SL) algorithm are proposed. CRO-SL is an evolutionary-based ensemble approach that is able to combine different search procedures for a single population. In this work, two different probabilistic strategies to improve the algorithm are analyzed. First, the probabilistic CRO-SL (PCRO-SL) is presented, which substitutes the substrates in the CRO-SL population with tags associated with each individual. Each tag represents a different operator which will modify the individual in the reproduction phase. In each generation of the algorithm, the tags are randomly assigned to the individuals with similar probabilities, obtaining this way an ensemble that sees more intense changes with the application of different operators to a given individual than CRO-SL. Second, the dynamic probabilistic CRO-SL (DPCRO-SL) is presented, in which the probability of tag assignment is modified during the evolution of the algorithm, depending on the quality of the solutions generated in each substrate. Thus, the best substrates in the search process will be assigned higher probabilities than those which showed worse performance during the search. The performances of the proposed probabilistic and dynamic ensembles were tested for different optimization problems, including benchmark functions and a real application of wind-turbine-layout optimization, comparing the results obtained with those of existing algorithms in the literature. [ABSTRACT FROM AUTHOR]

Wind is an important source of renewable energy, often used to provide clean electricity to remote areas. For optimal extraction of this energy source, there is a need for an accurate and robust wind speed forecasting. The intermittent nature of wind makes this goal quite challenging. This research proposes a novel hybrid bidirectional LSTM (BiLSTM) model for near real-time wind speed forecasting. The hybrid model is developed using wind speed and selected climate indices from a group of neighbouring reference stations as predictors to forecast wind speed of a target station. A 3-stage feature selection is applied on the predictors to robustly extract highly significant input features. Stage 1 employs partial auto-correlation and cross-correlation, stage 2 uses the RReliefF filter algorithm, and Boruta-RF wrapper method is implemented in the final stage to improve the BiLSTM model with an efficient Bayesian optimization used for hyperparameter tuning. The proposed model has been benchmarked with comparative models including standalone and hybrid LSTM, RNN, MLP and RF. The proposed hybrid BiLSTM algorithm is found to be superior in wind speed prediction for all tested sites with ≈ 76. 6 − 84. 8 % of errors being ≤ | 0. 5 | ms − 1 . The hybrid BiLSTM model also registered the lowest Relative Root Mean Square Error (9. 6 − 23. 8 %) and Mean Absolute Percentage Error (8. 8 − 21. 5 %) among all the tested algorithms. This research ascertains that the proposed model can accurately predict wind speed and capacitate wind energy availability to be regularly monitored at a near real-time level. [Display omitted] • A novel deep learning model is proposed for near real-time wind forecasting. • BiLSTM integrated with 3-stage feature selection and Bayesian optimization is used. • Wind speed at six target sites is predicted using reference stations data. • Proposed hybrid model outperforms all comparative models. • The study has prospective implications for intermittent wind energy forecast. [ABSTRACT FROM AUTHOR]

In this paper, we propose different explicable forecasting approaches, based on inductive and evolutionary decision rules, for extreme low-visibility events prediction. Explicability of the processes given by the rules is in the core of the proposal. We propose two different methodologies: first, we apply the PRIM algorithm and evolution to obtain induced and evolved rules, and subsequently these rules and boxes of rules are used as a possible simpler alternative to ML/DL classifiers. Second, we propose to integrate the information provided by the induced/evolved rules in the ML/DL techniques, as extra inputs, in order to enrich the complex ML/DL models. Experiments in the prediction of extreme low-visibility events in Northern Spain due to orographic fog show the good performance of the proposed approaches. [ABSTRACT FROM AUTHOR]

Prediction of Total Cloud Cover (TCDC) from numerical weather simulation models, such as Global Forecast System (GFS), can aid renewable energy engineers in monitoring and forecasting solar photovoltaic power generation. A major challenge is the systematic bias in TCDC simulations induced by the errors in the numerical model parameterization stages. Correction of GFS-derived cloud forecasts at multiple time steps can improve energy forecasts in electricity grids to bring better grid stability or certainty in the supply of solar energy. We propose a new kernel ridge regression (KRR) model to reduce bias in TCDC simulations for medium-term prediction at the inter-daily, e.g., 2–8 day-ahead predicted TCDC values. The proposed KRR model is evaluated against multivariate recursive nesting bias correction (MRNBC), a conventional approach and eight machine learning (ML) methods. In terms of the mean absolute error (MAE), the proposed KRR model outperforms MRNBC and ML models at 2–8 day ahead forecasts, with MAE ≈ 20–27%. A notable reduction in the simulated cloud cover mean bias error of 20–50% is achieved against the MRNBC and reference accuracy values generated using proxy-observed and non-corrected GFS-predicted TCDC in the model's testing phase. The study ascertains that the proposed KRR model can be explored further to operationalize its capabilities, reduce uncertainties in weather simulation models, and its possible consideration for practical use in improving solar monitoring and forecasting systems that utilize cloud cover simulations from numerical weather predictions. [Display omitted] • KRR model for bias correction of Total Cloud Cover forecasts is proposed. • KRR reduces bias in two-to-eight-day ahead forecasts at 0, 3, 6 UTC. • Cloud cover forecast error is reduced between 20-50% against conventional error reduction methods. • KRR is validated with multivariate recursive nested bias correction and other models. • KRR reduces bias in Total Cloud Cover forecasts for solar energy forecast systems. [ABSTRACT FROM AUTHOR]

Concept-bottleneck models (CBMs) are a new paradigm to construct interpretable classifiers. The CBM architecture can be regarded as a neural network with a single hidden layer whose neurons are replaced by binary classifiers. Each of these binary classifiers implements a concept, that is, it attempts to answer a meaningful yes/no question: the presence or absence of a concept in the observation presented to the network. The output layer is usually implemented as a linear stage that combines the concepts into final decisions. This paper presents an application of the CBM paradigm to a problem of Extreme Wind Speed prediction, such that the forecasting properties of the system can be interpreted in terms of different concepts considered for this problem. A significant contribution of this paper is the proposal of an automated concept generation method, with controlled sparsity, in which the concepts are automatically extracted from the branches of decision trees fitted with informative features that capture the dynamics of the wind speed time series. The final forecasting system is able to predict extreme wind speed values in wind farms with good accuracy and interpretable results, as experiments over real wind speed data from a wind farm in Spain show. • Application of Concept Bottleneck Models for Extreme Wind Speed prediction. • Features capturing the dynamics of the wind speed borrowed from finance analysis. • Concepts are automatically extracted from the branches of decision trees. • Results in real data from a Spanish wind farm are discussed. [ABSTRACT FROM AUTHOR]

Electricity demand prediction is crucial to ensure the operational safety and cost-efficient operation of the power system. Electricity demand has predominantly been predicted deterministically, while uncertainty analysis has been usually overlooked. To address this research gap, an integrated Neural Facebook Prophet (NFBP) model and Gaussian Kernel Density Estimation (KDE) model is proposed in this paper, as a way to obtain point and interval predictions of electricity demand, quantifying this way the uncertainty in the predictions. First, historical lagged data, created by utilizing the Partial Auto-correlation Function and Mutual Information Test, is applied to train a prediction model based on NFBP, Deep Learning (DL) as well as Statistical Models. Second, the model Prediction Errors (PE) are derived from the difference between actual and predicted values. A splitting strategy based on the mean and standard deviation of PE is proposed. Finally, electricity demand prediction intervals are obtained by applying Gaussian KDE on split PE. To verify the effectiveness of the proposed model, simulation studies are carried out for three prediction horizons on freely available datasets for the Bulimba sub-station in Southeast Queensland, Australia. Compared with DL models (Long-Short Term Memory Network and Deep Neural Network), the Root Mean Square Error of the NFBP model was reduced by 6.1% and 11.3% for 0.5-hr ahead, 22.7% and 26.3% for 6-hr ahead, and 31.8% and 29.9% for daily prediction. In addition, the Prediction Interval normalized Interval width is smaller in magnitude for the proposed NFBP-KDE model compared to other DL and Statistical models. [Display omitted] • A novel method is proposed for interval prediction of electricity demand. • Kernel density estimation with Gaussian function is applied to obtain interval prediction based on point prediction error. • Parameter of KDE distribution is optimized by minimizing Mean Integrated Square Error. • Deterministic and probabilistic metrics verify the superiority of the proposed prediction model. [ABSTRACT FROM AUTHOR]

Prediction problems in timed datasets related to human activities are especially difficult to solve, because of the specific characteristics and the scarce number of predictive (input) variables available to tackle these problems. In this paper, we try to find out whether Machine Learning (ML) approaches can be successfully applied to these problems. We deal with timed datasets with human activity patterns, in which the input variables are exclusively related to the day or type of day when the prediction is carried out and, usually, to the meteorology of those days. These problems with a marked human activity pattern frequently appear in mobility and traffic-related problems, delivery prediction (packets, food), and many other activities, usually in cities. We evaluate the performance in these problems of different ML methods such as artificial neural networks (multi-layer perceptrons, extreme learning machines) and support vector regression algorithms, together with an Analogue-type (KNN) approach, which serves as a baseline algorithm and provides information about when it is expected that ML approaches will fail, by looking for similar situations in the past. The considered ML algorithms are evaluated in four real prediction problems with human activity patterns, such as school absences, bike-sharing demand, parking occupation, and packets delivered in a post office. The results obtained show the good performance of the ML algorithms, revealing that they can deal with scarce information in all the problems considered. The results obtained have also revealed the importance of including meteorology as the input variables, showing that meteorology is frequently behind demand peaks or valleys in this kind of problem. Finally, we show that having a number of similar situations in the past (training set) prevents ML algorithms from making important mistakes in the prediction obtained. [ABSTRACT FROM AUTHOR]

Global Solar Radiation (GSR) prediction models are critical to improve the dispatch, control, and stabilization of solar renewable power, and to integrate the solar energy into the electrical grid system. GSR, especially on a short-term scale, can have important fluctuations, which may affect the total energy expected to be supplied to the grid. To overcome this issue, prediction models with a high forecasting performance are needed. In this paper a novel framework based on Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) and Deep Residual Network with Bidirectional long short-term memory, i.e., DRESNET model, is proposed for obtaining accurate multi-step ahead GSR predictions. To train the proposed ICEEMDAN-DRESNET hybrid model, minute-level daylight data from Energy Sector Management Assistance Program in Nepalgunj (mid-western Nepal) are used. The results demonstrate ICEEMDAN-DRESNET model is an excellent tool for short-term solar energy monitoring, yielding excellent predictions, in all metrics such as MAE 9.769 W/m2, MAPE 5.657 % , TIC 1.143, CPI 4.739, and TIC 0.023 for 5-min time-horizon predictions, improving the results from the benchmark models. As the forecasting time-horizon is increased, the ICEEMDAN-DRESNET model accuracy drops, with MAE 33.672 W/m2; MAPE 31.749 % for 1-hr, MAE 22.625 W/m2; MAPE 18.312 % (30-min) and MAE 14.897 W/m2; MAPE 10.358 % (15-min), also better than the benchmark models. The results confirm the competitive merit of ICEEMDAN and DRESNET integration to improve deep learning and the potential of proposed model for the monitoring of solar or other renewable (e.g., wind or solar) energies. [Display omitted] • A deep learning algorithm for short-term solar prediction is proposed. • The system considers deep residual and bidirectional long short-term memory network models. • Nepal (a solar-rich region) is selected to evaluate the deep learning algorithm. • Tests are carried out at multi-step ahead horizons of 5-min, 10-min, 30-min and 1-h. • Excellent performance of the proposed deep learning system is reported. [ABSTRACT FROM AUTHOR]

Digital technologies with predictive modelling capabilities are revolutionizing electricity markets, especially in demand-side management. Accurate electricity price prediction is essential in deregulated markets; however, developing effective models is challenging due to high-frequency fluctuations and price volatility. This study introduces a hybrid prediction system that addresses these challenges through a comprehensive data processing and modelling framework for half-hourly electricity price predictions. The preprocessing stage employs the Maximum Overlap Discrete Wavelet Transform (MoDWT) to enhance input quality by reducing overlap and revealing underlying price patterns. The prediction model integrates Convolutional Neural Networks with Random Vector Functional Link (CRVFL) in a deep learning hybrid approach. Bayesian Optimization fine-tunes the MoDWT-CRVFL model for optimal performance. Validation of the model is conducted using half-hourly electricity prices from New South Wales. The results highlight the efficacy of the MoDWT-CRVFL model, achieving high accuracy with superior Global Performance Indicator (G P I) values of approximately 1.61, 1.33, 1.85, 1.30, and 0.78 for Summer, Autumn, Winter, Spring, and Annual (Year 2022), respectively, outperforming alternative models. Similarly, the Kling–Gupta Efficiency (K G E) metrics for the proposed model consistently surpassed those of both decomposition-based and standalone models. For instance, the K G E value for MoDWT-CRVFL was approximately 0.972, significantly higher than values of approximately 0.958, 0.899, 0.963, 0.943, 0.930, 0.661, 0.708, 0.696, 0.739, and 0.738 for MoDWT-LSTM, MoDWT-DNN, MoDWT-XGB, MoDWT-RF, MoDWT-MLP, Bi-LSTM, LSTM, DNN, RF, XGB, and MLP, respectively. The methodologies proposed in this study optimize energy resource allocation, market prices, and network management, empowering market operators to make informed decisions for a resilient and efficient electricity market. • Designs deep hybrid Convolutional Neural-Random Vector Functional Link-Wavelet data decomposition model. • Performs half-hourly electricity price predictions over seasonal and annual periods. • Performs detailed appraisal of proposed deep hybrid decomposition model with visual and statistical predictive analytics. • Proposed deep hybrid decomposition model outperforms data decomposition and standalone predictive models. • Demonstrates a practical method for seasonal/annual electricity market insight analysis. [ABSTRACT FROM AUTHOR]

Implementing key engineering solutions to optimise the operation of energy industries requires daily electricity demand forecasting and including uncertainty, to promote markets insight analysis as part of their strategic planning, regulating and supplying electricity to consumers. This paper proposes hybrid artificial intelligence models combining convolutional neural networks (CNN) as a feature extraction algorithm with extreme learning machines (ELM) as a framework to predict electricity demand with confidence intervals generated by Kernel Density Estimation (KDE) approaches. In order to develop CELM-KDE model, time-lagged series of daily electricity demand with local climate variables based on the air temperature, atmospheric vapour pressure, evaporation, solar radiation, humidity and sea level pressures are used to train the proposed CELM-KDE hybrid model. In order to fully evaluate the newly developed model from a point-based, as well as a probabilistic prediction strategy, the observed and predicted electricity demand as well as the probability distribution of errors are analysed using KDE method that operates without prior data distribution assumptions. Based on observed and predicted electricity demand and the relevant probabilistic confidence intervals generated by the CELM-KDE model, the final results show that the proposed method attains significantly better probability interval predictions than traditionally-used point-based models. The proposed CELM-KDE model is demonstrated to be highly effective in providing a comprehensive coverage of predicted errors, as well as providing greater insights into the average bandwidth and detailed predicted electricity demand in the testing stage. The results also indicate that the proposed hybrid model is a reliable decision support tool to develop engineering solutions in area of energy modelling, monitoring and forecasting, which could potentially be useful to the industry policymakers. We show that the point-and probabilistic-based electricity demand predictive models can be employed as an effective tool to improve accuracy of forecasting and provision of insights for national electricity markets and key energy industry stakeholder application tools. [Display omitted] • A hybrid deep learning model for daily electricity demand prediction is proposed. • A Kernel Density Estimation is applied to derive probabilistic prediction intervals. • The algorithm quantifies uncertainties in point-based electricity demand predictions. • The proposed model outperforms deep learning models using statistical uncertainty indicators. [ABSTRACT FROM AUTHOR]

Wind energy is an environment friendly, low-carbon, and cost-effective renewable energy source. It is, however, difficult to integrate wind energy into a mixed energy grid due to its high volatility and intermittency. For wind energy conversion systems to be reliable and efficient, accurate wind speed (WS) forecasting is fundamental. This study cascades a convolutional neural network (CNN) with a bidirectional long short-term memory (BiLSTM) in order to obtain a model for hourly WS forecasting by utilizing several meteorological variables as model inputs to study their effects on predicted WS. For input selection, the mutation grey wolf optimizer (TMGWO) is used. For efficient optimization of CBiLSTM hyperparameters, a hybrid Bayesian Optimization and HyperBand (BOHB) algorithm is used. The combined usage of TMGWO, BOHB, and CBiLSTM leads to a three-phase hybrid model (i.e., 3P-CBiLSTM). The performance of 3P-CBiLSTM is benchmarked against the standalone and hybrid BiLSTMs, LSTMs, gradient boosting (GBRs), random forest (RFRs), and decision tree regressors (DTRs). The statistical analysis of forecasted WS reveals that the 3P-CBiLSTM is highly effective over the other benchmark forecasting methods. This objective model also registers the highest percentage of forecasted errors (≈ 53.4 – 81.8%) within the smallest error range ≤ | 0. 25 | ms − 1 amongst all tested study sites. Despite the remarkable results achieved, the CBiLSTM model cannot be generally understood, so the eXplainable Artificial Intelligence (xAI) technique was used for explaining local and global model outputs, based on Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP). Both of the xAI methods determined that the antecedent WS is the most significant predictor of the short-term WS forecasting. Therefore, we aver that the proposed model can be employed to help wind farm operators in making quality decisions in maximizing wind power integration into the grid with reduced intermittency. • Explainable hybrid deep learning model proposed for hourly wind speed forecasting. • Grey wolf optimizer with a two-phase mutation designed for feature selection. • Bayesian Optimization and HyperBand used for efficient hyperparameter tuning. • Local Interpretable Model-Agnostic Explanations offer local model interpretation. • SHapley Additive exPlanations provide global model interpretation. [Display omitted] [ABSTRACT FROM AUTHOR]

Abstract: In this paper we present the application of a hybrid harmony search (HS) algorithm to the Spread-Spectrum Radar Polyphase (SSRP) codes design. Such a design can be formulated as a non-linear max–min optimization problem, hard to be solved using classical numerical techniques. Soft-computing approaches have then been successfully applied to solve the SSRP in the past, such as evolutionary computation techniques, variable neighborhood approaches or tabu search algorithms. In this paper we elaborate on the proposed hybrid HS approach, which consists of a naive implementation of the HS algorithm along with an adaptive-step gradient-guided local search procedure. Intensive computer simulations show that the proposed hybrid HS algorithm is able to outperform existing algorithms for the SSRP design problem (including the best reported so far), with significant differences in large-size SSRP instances. [Copyright &y& Elsevier]

This paper presents a project-based learning experience to teach mobile communications courses, carried out at Alcalá University, Madrid, Spain. The experience consists of a competition among teams formed by students, who act as consultant companies working for imaginary operators. The objective of the teams is to obtain the best network design and the most economical budget for an imaginary operator in the city of Alcalá de Henares. The proposed learning scheme uses existing concepts of different fields, such as the project-based or competitive games methodologies, and applies them to the teaching of telecommunications subjects. Details of the practical application of the methodology and results obtained at Alcalá University are discussed in this paper. [ABSTRACT FROM AUTHOR]

In this paper the performance of different Machine Learning and Deep Learning approaches is evaluated in problems related to green mobility in big cities. Specifically, the forecasting of bike sharing demand in Madrid and Barcelona (Spain) is approached, for different prediction time-horizons, and also a problem of cable car demand forecasting in Madrid city. An important number of predictive variables are considered, which are grouped into four different sets (categorical/calendrical, persistence-based, meteorological and, as a novelty of the paper, information about analogue past instances), whose relevance is studied for all cases. A feature selection mechanism is also incorporated in order to improve the prediction accuracy of the proposed algorithms. A total of 12 different multivariate regression techniques are implemented, covering from Machine Learning methods to time-series Deep Learning approaches. Excellent results in all the prediction problems approached are reported. Finally, the consequences of obtaining accurate prediction in these three problem of green mobility in big cities are discussed. In addition, it is studied how the results could be exported to other similar cases in more general urban mobility studies. Novelties of the work include: (1) Addressing the forecast problem of passenger flow on a cable car using ML and DL multivariate techniques; (2) using the demand of analogous past instances as an additional feature to solve the demand prediction problems; and (3) the extraction of global conclusions about feature relevance when addressing a demand forecasting problem in green mobility. [ABSTRACT FROM AUTHOR]

Reliable forecast of electricity demand is crucial to stability, supply, and management of electricity grids. Short-term hourly and sub-hourly demand forecasts are difficult due to the stochastic nature of electricity demand. To improve the electricity demand forecasts, artificial neural networks are combined with attention-based bi-directional neural network as well as Error Correction methods, and Bayesian hyperparameter optimizations. As part of the hybrid VMD-CABLSTM-ANN-EC model, the intrinsic mode functions, representing distinct predictive features, are separated from electricity demand data using variational mode decomposition algorithm. For predicting each intrinsic mode function and respective error time series, convolutional neural networks-bi-directional long short-term memory algorithms are applied. In a second stage, error time series are split, followed by a model to predict the error series' intrinsic mode functions, and a final integration stage to predict the error. The results of the proposed VMD-CABLSTM-ANN-EC model are benchmarked against decomposition-based and standalone models at substations in Queensland, Australia. The proposed VMD-CABLSTM-ANN-EC model outperformed all benchmark models. By splitting the predictor and target variables, it will be shown that the variational model decomposition method improves error corrections by revealing key features of historical demand data. Using error correction method in short-term electricity demand modelling can provide greater confidence in the prediction of electricity demand. The models can be used by energy providers for market analysis and insight research, to manage power failure risk, and make financial decisions. [Display omitted] • Proposes double decomposition method for electricity demand error correction. • Decomposition method generate significant important predictive features. • Integrates convolutional neural, attention long short term memory and artificial neural networks. • Hybrid model adopts Bayesian optimization, outperforms benchmark models. • Our model is useful for energy market insights/electricity demand forecasting. [ABSTRACT FROM AUTHOR]