Search

The complexity in stock index futures markets, influenced by the intricate interplay of human behavior, is characterized as nonlinearity and dynamism, contributing to significant uncertainty in long-term price forecasting. While machine learning models have demonstrated their efficacy in stock price forecasting, they rely solely on historical price data, which, given the inherent volatility and dynamic nature of financial markets, are insufficient to address the complexity and uncertainty in long-term forecasting due to the limited connection between historical and forecasting prices. This paper introduces a pioneering approach that integrates financial theory with advanced deep learning methods to enhance predictive accuracy and risk management in China's stock index futures market. The SF-Transformer model, combining spot-forward parity and the Transformer model, is proposed to improve forecasting accuracy across short and long-term horizons. Formulated upon the arbitrage-free futures pricing model, the spot-forward parity model offers variables such as stock index price, risk-free rate, and stock index dividend yield for forecasting. Our insight is that the mutual information generated by these variables has the potential to significantly reduce uncertainty in long-term forecasting. A case study on predicting major stock index futures prices in China demonstrates the superiority of the SF-Transformer model over models based on LSTM, MLP, and the stock index futures arbitrage-free pricing model, covering both short and long-term forecasting up to 28 days. Unlike existing machine learning models, the Transformer processes entire time series concurrently, leveraging its attention mechanism to discern intricate dependencies and capture long-range relationships, thereby offering a holistic understanding of time series data. An enhancement of mutual information is observed after introducing spot-forward parity in the forecasting. The variation of mutual information and ablation study results highlights the significant contributions of spot-forward parity, particularly to the long-term forecasting. Overall, these findings highlight the SF-Transformer model's efficacy in leveraging spot-forward parity for reducing uncertainty and advancing robust and comprehensive approaches in long-term stock index futures price forecasting. [ABSTRACT FROM AUTHOR]

The existing prediction methods have complex model application, high requirements for data parameters, and are limited to the prediction of a single observation point. To address this problem, this paper proposes a deep learning-based surface subsidence prediction method. Taking Hefei City of China as the research area, the time-series surface deformation results of this area are obtained by using SBAS-InSAR, and then the SFLA intelligent algorithm and Elman neural network model are combined to predict the surface deformation of key urban areas, and the prediction results are compared and analyzed.The experimental results show that the prediction model proposed in this paper can not only accurately predict a single deformation point, but also predict regional land subsidence, and can be used for auxiliary decision-making of urban spatial planning, early warning of geological hazards and hazard mitigation. [ABSTRACT FROM AUTHOR]

Going beyond previous work, this paper presents a systematic literature review that explores the deployment of satellites, drones, and ground-based sensors for yield prediction in agriculture. It covers multiple aspects of the topic, including crop types, key sensor platforms, data analysis techniques, and performance in estimating yield. To this end, datasets from Scopus and Web of Science were analyzed, resulting in the full review of 269 out of 1429 retrieved publications. Our study revealed that China (93 articles, >1800 citations) and the USA (58 articles, >1600 citations) are prominent contributors in this field; while satellites were the primary remote sensing platform (62%), followed by airborne (30%) and proximal sensors (27%). Additionally, statistical methods were used in 157 articles, and model-based approaches were utilized in 60 articles, while machine learning and deep learning were employed in 142 articles and 62 articles, respectively. When comparing methods, machine learning and deep learning methods exhibited high accuracy in crop yield prediction, while other techniques also demonstrated success, contingent on the specific crop platform and method employed. The findings of this study serve as a comprehensive roadmap for researchers and farmers, enabling them to make data-driven decisions and optimize agricultural practices, paving the way towards a fully digitized yield prediction. [ABSTRACT FROM AUTHOR]

Day by day pollution in cities is increasing due to urbanization. One of the biggest challenges posed by the rapid migration of inhabitants into cities is increased air pollution. Sustainable Development Goal 11 indicates that 99 percent of the world's urban population breathes polluted air. In such a trend of urbanization, predicting the concentrations of pollutants in advance is very important. Predictions of pollutants would help city administrations to take timely measures for ensuring Sustainable Development Goal 11. In data engineering, imputation and the removal of outliers are very important steps prior to forecasting the concentration of air pollutants. For pollution and meteorological data, missing values and outliers are critical problems that need to be addressed. This paper proposes a novel method called multiple iterative imputation using autoencoder-based long short-term memory (MIA-LSTM) which uses iterative imputation using an extra tree regressor as an estimator for the missing values in multivariate data followed by an LSTM autoencoder for the detection and removal of outliers present in the dataset. The preprocessed data were given to a multivariate LSTM for forecasting PM2.5 concentration. This paper also presents the effect of removing outliers and missing values from the dataset as well as the effect of imputing missing values in the process of forecasting the concentrations of air pollutants. The proposed method provides better results for forecasting with a root mean square error (RMSE) value of 9.8883. The obtained results were compared with the traditional gated recurrent unit (GRU), 1D convolutional neural network (CNN), and long short-term memory (LSTM) approaches for a dataset of the Aotizhonhxin area of Beijing in China. Similar results were observed for another two locations in China and one location in India. The results obtained show that imputation and outlier/anomaly removal improve the accuracy of air pollution forecasting. [ABSTRACT FROM AUTHOR]

The correction of model forecast is an important step in evaluating weather forecast results. In recent years, post-processing models based on deep learning have become prominent. In this paper, a deep learning model named EDConvLSTM based on encoder-decoder structure and ConvLSTM is developed, which appears to be able to effectively correct numerical weather forecasts. Compared with traditional post-processing methods and convolutional neural networks, ED-ConvLSTM has strong collaborative extraction ability to effectively extract the temporal and spatial features of numerical weather forecasts and fit the complex nonlinear relationship between forecast field and observation field. In this paper, the post-processing method of ED-ConvLSTM for 2 m temperature prediction is tested using The International Grand Global Ensemble dataset and ERA5-Land data from the European Centre for Medium-Range Weather Forecasts (ECMWF). Root mean square error and temperature prediction accuracy are used as evaluation indexes to compare ED-ConvLSTM with the method of model output statistics, convolutional neural network postprocessing methods, and the original prediction by the ECMWF. The results show that the correction effect of EDConvLSTM is better than that of the other two postprocessing methods in terms of the two indexes, especially in the long forecast time. [ABSTRACT FROM AUTHOR]

In this paper, we forecasted the ionospheric total electron content (TEC) over the region of China using the bidirectional convolutional gated recurrent unit (BiConvGRU) model. We first generated the China Regional Ionospheric Maps (CRIMs) using GNSS observations provide by the Crustal Movement Observation Network of China (CMONOC). We then used gridded TEC maps from 2015 to 2018 with a 1 h interval from the CRIMs as the dataset, including quiet periods and storm periods of ionospheric TEC. The BiConvGRU model was then utilized to forecast the ionospheric TEC across China for the year 2018. The forecasted TEC was compared with the TEC from the International Reference Ionosphere (IRI-2016), Convolutional Long Short-Term Memory (ConvLSTM), Convolutional Gated Recurrent Unit (ConvGRU), Bidirectional Convolutional Long Short-Term Memory (BiConvLSTM), and the 1-day Predicted Global Ionospheric Map (C1PG) provided by the Center for Orbit Determination in Europe (CODE). In addition, indices including Kp, ap, Dst and F10.7 were added to the training dataset to improve the forecasting accuracy of the model (-A indicates no indices, while -B indicates with indices). The results verified that the prediction accuracies of the models integrated with these indices were significantly improved, especially during geomagnetic storms. The BiConvGRU-B model presented a decrease of 41.5%, 22.3%, and 13.2% in the root mean square error (RMSE) compared to the IRI-2016, ConvGRU, and BiConvLSTM-B models during geomagnetic storm days. Furthermore, at a specific grid point, the BiConvGRU-B model showed a decrease of 42.6%, 49.1%, and 31.9% in RMSE during geomagnetic quiet days and 30.6%, 34.1%, and 15.1% during geomagnetic storm days compared to the IRI-2016, C1PG, and BiConvLSTM-B models, respectively. In the cumulative percentage analysis, the BiConvGRU-B model had a significantly higher percentage of mean absolute error (MAE) within the range of 0–1 TECU in all seasons compared to the BiConvLSTM-B model. Meanwhile, the BiConvGRU-B model outperformed the BiConvLSTM-B model with lower RMSE for each month of 2018. [ABSTRACT FROM AUTHOR]

The accurate prediction of photovoltaic (PV) power is essential for planning power systems and constructing intelligent grids. However, this has become difficult due to the intermittency and instability of PV power data. This paper introduces a deep learning framework based on 7.5 min-ahead and 15 min-ahead approaches to predict short-term PV power. Specifically, we propose a hybrid model based on singular spectrum analysis (SSA) and bidirectional long short-term memory (BiLSTM) networks with the Bayesian optimization (BO) algorithm. To begin, the SSA decomposes the PV power series into several sub-signals. Then, the BO algorithm automatically adjusts hyperparameters for the deep neural network architecture. Following that, parallel BiLSTM networks predict the value of each component. Finally, the prediction of the sub-signals is summed to generate the final prediction results. The performance of the proposed model is investigated using two datasets collected from real-world rooftop stations in eastern China. The 7.5 min-ahead predictions generated by the proposed model can reduce up to 380.51% error, and the 15 min-ahead predictions decrease by up to 296.01% error. The experimental results demonstrate the superiority of the proposed model in comparison to other forecasting methods. [ABSTRACT FROM AUTHOR]

This paper is concerned with the unsolved issue of how to accurately predict the financial market volatility. We propose a novel volatility prediction method for stock index futures prediction based on LSTM, PCA, stock indices and relevant futures. Inspired by the recent advancement of deep learning methodology, six models that combine a variety of artificial intelligence techniques are compared, including ANN, ANN(PCA), ANN(AE), LSTM, LSTM(PCA), and LSTM(AE). That is, in the design and comparison of the proposed AI models, we consider the combination of two dimensionality reduction methods (PCA and AE) and two typical neural networks (ANN and LSTM) in processing time series data. Besides, to further assess the prediction performance of the proposed models, two widely-applied statistical models (i.e. AR and EGARCH) on volatility prediction are used as benchmarks. In the empirical study, we collect financial trading data in both China and the US, and compare the performances of different models in predicting 5 days and 10 days ahead volatilities of stock index futures. In all, our analysis supports the use of LSTM(PCA) model to tackle those irregular and complex datasets. [ABSTRACT FROM AUTHOR]

Accurate short-term forecasting of intensive rainfall has high practical value but remains difficult to achieve. Based on deep learning and spatial–temporal sequence predictions, this paper proposes a hierarchical dynamic graph network. To fully model the correlations among data, the model uses a dynamically constructed graph convolution operator to model the spatial correlation, a recurrent structure to model the time correlation, and a hierarchical architecture built with graph pooling to extract and fuse multi-level feature spaces. Experiments on two datasets, based on the measured cumulative rainfall data at a ground station in Fujian Province, China, and the corresponding numerical weather grid product, show that this method can model various correlations among data more effectively than the baseline methods, achieving further improvements owing to reversed sequence enhancement and low-rainfall sequence removal. [ABSTRACT FROM AUTHOR]

Short-term load forecasting plays a significant role in the operation of power systems. Recently, deep learning has been generally employed in short-term load forecasting, primarily in the extraction of the characteristics of digital information in a single dimension without taking into account of the impact of external variables, particularly non-digital elements on load characteristics. In this paper, we propose a joint MFF-SAM-GCN to realize short-term load forecasting. First, we utilize a Bi-directional Long Short-Term Memory (Bi-LSTM) network and One-Dimensional Convolutional Neural Network (1D-CNN) in parallel connection to form a multi-feature fusion (MFF) framework, which can extract spatiotemporal correlation features of the load data. In addition, we introduce a Self-Attention Mechanism (SAM) to further enhance the feature extraction capability of the 1D-CNN network. Then with the deployment of a Graph Convolutional Network (GCN), the external non-digital features such as weather, strength, and direction of wind, etc., are extracted. Moreover, the generated weight matrices are incorporated into the load features to enhance feature recognition ability. Finally, we exploit Bayesian Optimization (BO) to find the optimal hyperparameters of the model to further improve the prediction accuracy. The simulation is taken from our proposed model and six benchmark schemes by using the bus load dataset of the Shandong Open Data Network, China. The results show that the RMSE of our proposed MFF-SAM-GCN model is 0.0284, while the SMAPE is 9.453%,the MBE is 0.025, and R-squared is 0.989, which is better than the selected three traditional machine learning methods and the three deep learning models. [ABSTRACT FROM AUTHOR]

This paper presents a multi-timescale receding horizon framework for the load forecast of large power customers. The future load pattern of individual users could be very difficult to predict because of its chronological and high volatile properties. Also, the sampling of nonaggregated load data may suffer from severe information missing issues. To address these challenges, we first develop an online singular value thresholding (SVT) algorithm, which utilizes the approximate low-rank property of load data matrices to efficiently recover the missing information. Then, a combinatorial deep learning method is developed, which applies the multi-layer perception (MLP) neural network and the long short-term memory (LSTM) neural network with gated recurrent unit (GRU) to deal with the short-term and ultra-short-term load forecast, respectively. Specifically, an early stopping strategy is designed and implemented to avoid the over-fitting of model training. Moreover, the receding time window is imposed to dynamically update the data recovery and load forecast outcomes, which supports the online computing on a Spark platform. Numerical experiments on real-world load data from North China confirms the effectiveness of the proposed methodology, which can support more complex applications in embedded systems and cyber physical systems. [ABSTRACT FROM AUTHOR]

Different from traditional algorithms and model, machine learning is a systematic and comprehensive application of computer algorithms and statistical models, and it has been widely used in many fields. In the field of finance, machine learning is mainly used to study the future trend of capital market price. In this paper, to predict the time-series data of stock, we applied the traditional models and machine learning models for forecasting the linear and non-linear problem, respectively. First, stock samples that occurred from year 2010 to 2019 at the New York Stock Exchange are collected. Next, the ARIMA (autoregressive integrated moving average model) model and LSTM (long short-term memory) neural network model are applied to train and predict stock price and stock price subcorrelation. Finally, we evaluate the proposed model by several indicators, and the experiment results show that: (1) Stock price and stock price correlation are accurately predicted by the ARIMA model and LSTM model; (2) compared with ARIMA, the LSTM model performance better in prediction; and (3) the ensemble model of ARIMA-LSTM significantly outperforms other benchmark methods. Therefore, our proposed method provides theoretical support and method reference for investors about stock trading in China stock market. [ABSTRACT FROM AUTHOR]

Many calculation methods have been developed to forecast carbon emissions (CE). These methods can be classified into two categories: time series prediction and variables regression. However, time series prediction cannot fully use all CE‐related information and variables regression can only predict CE by knowing other variables in advance. To solve these problems, KLS is proposed in this paper which is a fusion of Kalman filter (KF), long short‐term memory (LSTM), and support vector machine (SVM). LSTM is used to predict CE as a time series, and 10 of 14 variables are selected by ridge regression (RR) to regress CE by SVM. Finally, KF is used to integrate the results of LSTM and SVM based on their variances. Overall, statistical filtering and deep learning algorithms are used in this paper, which achieved the fusion of time series prediction and variables regression. To confirm the effectiveness of KLS, we trained KLS model by CE from 1965 to 2005 and tested the model by the data from 2006 to 2014. The results showed KLS was more accurate than other four existing methods. Finally, we predicted the CE of China in next decade. [ABSTRACT FROM AUTHOR]

Accurate photovoltaic (PV) power forecast is crucial for carbon neutrality. Current researches on PV power forecast mainly focus on using temporal information from single PV station, and the spatial information in multiple PV power stations are often neglected. To address this problem, this paper introduces Moran index to verify the spatial autocorrelation of PV power for the first time, uses Granger causality test and transfer entropy to reveal the spatial information gain in PV power forecast for the first time, and proposes a novel spatial–temporal probabilistic PV forecast method using deep Graph Gated Recurrent Unit (GraphGRU) network-based spatial–temporal multi-task learning and Kernel Density Estimation (KDE). Deep GraphGRU combines the advantages of Graph Convolutional Network (GCN) in spatial feature extraction and the advantages of Gated Recurrent Unit (GRU) network in temporal feature extraction, and thus has strong ability to extract spatial–temporal information in historical data of multiple different PV power stations. Through GraphGRU, temporal dependency information extracted from historical data of multiple PV stations can promote each other to improve the forecast accuracy of each PV stations. KDE is used for estimating the joint probabilistic density function and giving the spatial–temporal probabilistic confidence interval of PV power. Experiments were performed in the five-year actual PV power data from 11 provinces of Belgium and the three-year solar irradiation data from 12 places in China to verify the superiorities of the proposed method. Comparison with conventional spatial–temporal and temporal forecast methods show that the proposed GraphGRU-based spatial–temporal forecast method can extract the spatial–temporal information from multiple PV power stations well and significantly outperform conventional temporal forecast methods. [ABSTRACT FROM AUTHOR]

Accurately predicting carbon trading prices using deep learning models can help enterprises understand the operational mechanisms and regulations of the carbon market. This is crucial for expanding the industries covered by the carbon market and ensuring its stable and healthy development. To ensure the accuracy and reliability of the predictions in practical applications, it is important to evaluate the model's robustness. In this paper, we built models with different parameters to predict carbon trading prices, and proposed models with high accuracy and robustness. The accuracy of the models was assessed using traditional survey indicators. The robustness of the CNN-LSTM model was compared to that of the LSTM model using Z-scores. The CNN-LSTM model with the best prediction performance was compared to a single LSTM model, resulting in a 9% reduction in MSE and a 0.0133 shortening of the Z-score range. Furthermore, the CNN-LSTM model achieved a level of accuracy comparable to other popular models such as CEEMDAN, Boosting, and GRU. It also demonstrated a training speed improvement of at least 40% compared to the aforementioned methods. These results suggest that the CNN-LSTM enhances model resilience. Moreover, the practicality of using Z-score to evaluate model robustness is confirmed. • Deep learning by combined CNN-LSTM networks for carbon price forecasting. • Hyperparameter tuning of deep learning models for improving forecast accuracy. • CNN-LSTM models outperform single CNN models and single LSTM models. • Performance metric of deep learning for abrupt changes in carbon price variation. [ABSTRACT FROM AUTHOR]

The inflow forecasting is one of the most important technologies for modern hydropower station. Under the joint influence of soil, upstream inflow, and precipitation, the inflow is often characterized by time lag, nonlinearity, and uncertainty and then results in the difficulty of accurate multistep prediction of inflow. To address the coupling relationship between inflow and the related factors, this paper proposes a long short-term memory deep learning model based on the Bagging algorithm (Bagging-LSTM) to predict the inflows of future 3 h, 12 h, and 24 h, respectively. To validate the proposed model, the inflow and related weather data come from a hydropower station in southern China. Compared with the classical time series models, the results show that the proposed model outperforms them on different accuracy metrics, especially in the scenario of multistep prediction. [ABSTRACT FROM AUTHOR]

Accurate PM 2.5 forecasting provides a possibility for establishing an early warning system to notify the public and take precautionary measures to prevent negative effects on ambient air quality and public health. Considering strong seasonal variation in meteorological conditions, in this paper, a seasonal stacked autoencoder model combining seasonal analysis and deep feature learning is proposed for forecasting the hourly PM 2.5 concentration, named DL-SSAE model. The original data are firstly decomposed into four seasonal subseries according to the Chinese calendar, and then the Kendall correlation coefficient method is employed to search inherent relationships between PM 2.5 concentrations and meteorological parameters within 1-h ahead for each seasonal time series. The inherent relationships of each seasonal subseries are finally extracted, learned, and modeled by different deep neural networks (stacked autoencoders for regression), and the hourly PM 2.5 forecasts are yielded. The addressed model is tested by the dataset collected from three environmental monitoring stations in Beijing, China. The results demonstrate that the proposed model outperforms all other considered models with/without seasonality consideration in this paper. [ABSTRACT FROM AUTHOR]

Probabilistic load forecasting (PLF) are of critical importance to reliable and economical power system operations. To further improve the PLF performance, this paper extends the load forecasting model with high accuracy to PLF by using Monte Carlo (MC) Dropout and soft thresholding (ST) for uncertainty modeling and noise reduction, respectively. Specifically, the method first builds a Dual-stage Attention-based Long and Short Time Pattern Network (DA-LSTPNet) by weighting the temporal information extracted from Gate Recurrent Unit (GRU) and convolutional neural networks (CNN) through the time attention mechanism(TAM). Then, an adaptive ST structure is used to reduce the impact of meteorological data noise on model prediction performance. The term ST-DA-LSTPNet refers to DA-LSTPNet with the addition of the ST structure. Finally, an appropriate Dropout setting is inserted in the neural network layers during the model's testing phase to achieve the model parameters' stochastic process. After multiple sampling and calculations, the final PLF result is obtained. To demonstrate the effectiveness of the proposed method, this work construct a case study using the actual data from a prefecture-level city in southern China. Compared with several benchmark methods, the proposed method significantly improves the performance of PLF and has good anti-noise generalization ability. [Display omitted] • The high-accuracy load forecasting model extends to probabilistic load forecasting. • The adaptive soft thresholding can mitigate noise impact on prediction. • The model performs well on actual grid data, with over 92% prediction coverage. • This work provides effective support for economic scheduling and management. [ABSTRACT FROM AUTHOR]

The air pollutant diffusion trend prediction plays an important role in the environment protection. The approaches in the existing studies rarely consider spatial and temporal information sequentially or cannot use spatiotemporal structure information effectively. In this paper, we propose a new deep learning-based air pollutant diffusion trend prediction model, called GATBL-Learning. It is an end-to-end neural network model consisting of Graph Attention Network (GAT) and Encoder-Decoder model. Since each city in the region will set as a node, GAT can extract spatial information from them by aggregation operation. Then Encoder-Decoder model, consisting of Bi-LSTM as Encoder and LSTM as Decoder, mines time series features and outputs the data containing spatiotemporal features. In addition, we also observe that the changes of air pollutants show seasonality discipline and training the seasonal model can improve the model adaptability. The results demonstrate the proposed prediction model is effective and superior. [ABSTRACT FROM AUTHOR]

This paper develops a hierarchical deep learning machine (HDLM) to efficiently achieve both quantitative and qualitative online transient stability prediction (TSP). For the sake of improving its online efficiency, multiple generators’ fault-on trajectories as well as the two closest data-points in pre-/post-fault stages are acquired by PMUs to form its raw inputs. An anti-noise graphical transient characterization technique is tactfully designed to transform multiplex trajectories into 2-D images, within which system-wide transients are concisely described. Then, following the divide-and-conquer philosophy, the HDLM trains a two-level convolutional neural network (CNN) based regression model. With stability margin regressions hierarchically refined, it manages to perform reliable and adaptive online TSP almost immediately after fault clearance. Test results on the IEEE 39-bus test system and the real-world Guangdong Power Grid in South China demonstrate the HDLM's superior performances on both stability status and stability margin predictions. [ABSTRACT FROM AUTHOR]

Accurate and reliable low-voltage load forecasting is critical to optimal operation and control of distribution network and smart grid. However, compared to traditional regional load forecasting at high-voltage level, it faces tough challenges due to the inherent high uncertainty of the low-capacity load and distributed renewable energy integrated in the demand side. This paper proposes a novel hybrid ensemble deep learning (HEDL) approach for deterministic and probabilistic low-voltage load forecasting. The deep belief network (DBN) is applied to low-voltage load point prediction with the strong ability of approximating nonlinear mapping. A series of ensemble learning methods including bagging and boosting variants are introduced to improve the regression ability of DBN. In addition, the differencing transformation technique is utilized to ensure the stationarity of load time series for the application bagging and boosting methods. On the basis of the integrated thought of ensemble learning, a new hybrid ensemble algorithm is developed via integrating multiple separate ensemble methods. Considering the diversity in various ensemble algorithms, an effective K nearest neighbor classification method is utilized to adaptively determine the weights of sub-models. Furthermore, HEDL based probabilistic forecasting is proposed by taking advantage of the inherent resample idea in bagging and boosting. The effectiveness of the HEDL method for both deterministic and probabilistic forecasting has been systematically verified based on realistic load data from East China and Australia, indicating its promising prospective for practical applications in distribution networks. [ABSTRACT FROM AUTHOR]

[Display omitted] • In the case of a small number of known deposits and mineral occurrences, the prospecting prediction model based on the Siamese network can extract effective features without relying on prior knowledge to realize supervised prospecting prediction. • On the basis of sliding window data enhancement, this paper generates paired sample data sets by randomly combining deposit samples and non-deposit samples, which can further enhance the data and improve the generalization ability and prediction accuracy of the model to a certain extent. • Compared with the prediction results of the weight of evidence method and the convolutional neural network, the prediction results of the Siamese network are better, and 7 new gold prospecting areas have been discovered in Zhangjiakou, Hebei Province, China. Supervised neural networks constitute an important research area for the intelligent prediction of locations for prospecting mineral deposits. Accurate training of a supervised neural network requires many training samples, something that is often difficult to obtain. This paper reports the use of geological, geochemical, and geophysical data by a Siamese network to overcome the problem of insufficient training samples, and implements a supervised deep learning prospecting prediction method based on the Siamese network. Intelligent prediction for gold deposits prospecting is carried out for the Chongli mineral concentrate area in Zhangjiakou, Hebei Province, China, and compared with the weight of evidence method and convolutional neural network model. The results show that:(a) the performance of the Siamese network model is no less than that of the convolutional neural network (CNN) model and better than that of the weight of evidence (WOE) method; (b) the gold prospective areas differentiated by the established models are strongly consistent with geological and metallogenic characteristics in the study area. This study suggests Siamese network model as an effective mineral prospectivity modeling tool. This method is also suitable for prospecting prediction using geoscience data in other areas. [ABSTRACT FROM AUTHOR]

• A hybrid deep learning model within decomposition-ensemble strategy named DEASeq2Seq is proposed for multistep metro passenger flow prediction. • DEASeq2Seq employs CEEMDAN and RQA to capture intrinsic dynamics. • An attention-based Seq2Seq model is developed to characterize temporal dependencies. • Real-world datasets to verify the proposed model and further model interpretation. Short-term passenger flow prediction has practical significance for metro management and operation. However, the complex nonlinear and non-stationary characteristics make it challenging to detect evolution characteristics of passenger flow. To address this problem, a hybrid short-term metro passenger flow prediction model named decomposition ensemble attention sequence to sequence (DEASeq2Seq) is proposed in this paper. The proposed DEASeq2Seq includes three phases: decomposition, ensemble, and prediction. First, complete empirical mode decomposition with adaptive noise (CEEMDAN) is utilized to decompose the original passenger flow data into several intrinsic mode functions (IMFs) and a residue. Second, recurrence quantification analysis (RQA) is performed to reconstruct the decomposed modes into a stochastic part, a deterministic part, and a trend part via determinism evaluation. Third, a Seq2Seq model with the attention mechanism is proposed to execute multistep prediction for short-term passenger flow and explore the influence mechanism of the reconstructed components on the prediction targets. The real dataset from Chengdu metro, China, is used to verify the proposed model. The experiment results show that the proposed DEASeq2Seq model outperforms the benchmark models. Further model interpretations are conducted to analyze the impacts of decomposition strategy, ensemble strategy, and attention mechanism. [ABSTRACT FROM AUTHOR]

It will be beneficial to devise an effective approach for short-term macroeconomic forecasting. Existing traditional statistics-based macroeconomic forecasting mainly focuses on exploring feasible methods for improving the accuracy of long-term predictions. However, the performance of short-term predictions was far less impressive under the impact of unexpected incidents. Furthermore, some deep learning-based approaches can achieve fine-grained variable frequency forecasting and preliminarily demonstrate effective results, but the interpretability is still controversial. Therefore, how to consider both the performance and the interpretability has already become a universal concern and an urgent unsolved problem. In this paper, we identified the above issue and proposed an interpretable data-driven approach, named EcoForecast 1 1 The code and data are available at https://github.com/navfour/ecoforecast. , for short-term macroeconomic forecasting based on the N-BEATS (neural basis expansion analysis for interpretable time series forecasting) neural network. To the best of our knowledge, EcoForecast is the first interpretable purely data-driven unified normative scheme for macroeconomic forecasting that achieves variable forecast frequencies and prediction domains, surpassing traditional statistics-based and deep learning-based approaches in performance or interpretability. EcoForecast used a three-level hierarchical signal encoding, including the fully connected neural network (FCNN) level, the block level, and the stack level. The FCNN level implemented both forecast and backcast information extraction for the temporal prediction and parameter learning of context. Block levels were connected by residuals so that the block's backcast could be sequentially filtered on the input. The stack level was used to form the top-level system of EcoForecast, where each stack was constrained to specialize in different inductive functions. To some extent, EcoForecast can balance effectiveness, efficiency, generalizability and interpretability while conditions such as forecast frequency and time window change, even when unexpected incidents occur. Based on the actual macroeconomic data for China from 1992 to 2022, the data-driven EcoForecast demonstrated high stability in different sequence learning scenarios and the accompanying high-accuracy performance. This stability was reflected in smaller prediction error expectation and variance, tolerance of fewer input samples, and robustness across prediction domains. The experimental results indicated that EcoForecast improved the accuracy up to 3.94 times compared with the traditional BVAR. In the robustness test, EcoForecast required only a quarter of the data to achieve 2.51 times smaller forecast errors than the BVAR while also improving the accuracy of varied macroeconomic indicators such as the Purchasing Managers' Index (PMI) and national electricity generation (ELEC)forecasting by 2.38 and 1.45 times. EcoForecast had a high sensitivity to the emergence of economic inflection points and adapts quickly when the economic environment changes, thus demonstrating a performance that exceeds traditional solutions in GDP forecasting during epidemics and PMI forecasting during economic turmoil. Interfacing with traditional economics research, the interpretable EcoForecast can uncover the trends and cycles of economic change, from which the conclusions are validated with the actual economics practice in China. Our findings can provide a new possible research direction for short-term macroeconomic forecasting. [ABSTRACT FROM AUTHOR]

Integrating return prediction of traditional time series models in portfolio formation can improve the performance of original portfolio optimization model. Since machine learning and deep learning models have shown overwhelming superiority than time series models, this paper combines return prediction in portfolio formation with two machine learning models, i.e., random forest (RF) and support vector regression (SVR), and three deep learning models, i.e., LSTM neural network, deep multilayer perceptron (DMLP) and convolutional neural network. To be specific, this paper first applies these prediction models for stock preselection before portfolio formation. Then, this paper incorporates their predictive results in advancing mean–variance (MV) and omega portfolio optimization models. In order to present the superiority of these models, portfolio models with autoregressive integrated moving average's return prediction are used as benchmarks. Evaluation is based on historical data of 9 years from 2007 to 2015 of component stocks of China securities 100 index. Experimental results show that MV and omega models with RF return prediction, i.e., RF+MVF and RF+OF, outperform the other models. Further, RF+MVF is superior to RF+OF. Due to the high turnover of these two models, this paper discusses their performance after deducting the transaction fee cased by turnover. Experiments present that RF+MVF still performs the best among MVF models and omega model with SVR prediction (SVR+OF) performs the best among OF models. Moreover, RF+MVF performs better than SVR+OF and high turnover erodes nearly half of their total returns especially for RF+OF and RF+MVF. Therefore, this paper recommends investors to build MVF with RF return prediction for daily trading investment. • Compares the performance of machine learning and deep learning in stock preselection. • Combining return prediction of machine learning and deep learning in portfolio formation. • Emphasis on advancing portfolio optimization with return prediction. • Advanced mean–variance model with random forest forecasts performs the best. [ABSTRACT FROM AUTHOR]

Owing to the importance of coalbed methane (CBM) as a source of energy, it is necessary to predict its future production. However, the production process of CBM is the result of the interaction of many factors, making it difficult to perform accurate simulations through mathematical models. We must therefore rely on the historical data of CBM production to understand its inherent features and predict its future performance. The objective of this paper is to establish a deep learning prediction method for coalbed methane production without considering complex geological factors. In this paper, we propose a multivariate long short-term memory neural network (M-LSTM NN) model to predict CBM production. We tested the performance of this model using the production data of CBM wells in the Panhe Demonstration Area in the Qinshui Basin of China. The production of different CBM wells has similar characteristics in time. We can use the symmetric similarity of the data to transfer the model to the production forecasting of different CBM wells. Our results demonstrate that the M-LSTM NN model, utilizing the historical yield data of CBM as well as other auxiliary information such as casing pressures, water production levels, and bottom hole temperatures (including the highest and lowest temperatures), can predict CBM production successfully while obtaining a mean absolute percentage error (MAPE) of 0.91%. This is an improvement when compared with the traditional LSTM NN model, which has an MAPE of 1.14%. In addition to this, we conducted multi-step predictions at a daily and monthly scale and obtained similar results. It should be noted that with an increase in time lag, the prediction performance became less accurate. At the daily level, the MAPE value increased from 0.24% to 2.09% over 10 successive days. The predictions on the monthly scale also saw an increase in the MAPE value from 2.68% to 5.95% over three months. This tendency suggests that long-term forecasts are more difficult than short-term ones, and more historical data are required to produce more accurate results. [ABSTRACT FROM AUTHOR]

• A linear components extraction method is proposed to identify the daily fluctuating patterns of the solar power series. • A scenarios generation algorithm is proposed to provide forecasting linear trend data. • Temporal distributed deep learning models are developed based on the detected daily fluctuating patterns of the PV power generation. • The proposed algorithm can provide accurate day-ahead solar power forecasts for distributed PV farms without local weather information. In recent years, the integration of distributed photovoltaic has witnessed explosive growth because of its great economic and environmental benefits. However, the intermittence and randomicity of PV power generation are challenging the stable and economic operation of active distribution system, especially when the local historical and forecasting meteorological data are not available. Thus, this paper presents a novel Gated Recurrent Unit based hybrid model for forecasting distributed PV power generations. The proposed method uses locally historical PV power generation data to forecast one-day-ahead solar power time series with five-minute intervals. First, a linear components extraction method is proposed to identify the daily fluctuating patterns of the solar power series. Then, temporal distributed Gated Recurrent Unit models are developed based on the detected daily fluctuating patterns of the PV power generation. A scenarios generation algorithm is proposed to provide forecasting linear trend data for each Gated Recurrent Unit model. Finally, the daily linear series scenarios, the linear and nonlinear parts of the historical data are fed into the Gated Recurrent Unit to forecast the PV generation. Data collected from an actual distributed PV farm in southeastern China are utilized to validate the effectiveness of the proposed model. The results show that the proposed model can provide a nominalized root mean square error of 6.83% and a nominalized mean absolute error of 4.12%. Compared to a single Gated Recurrent Unit forecasting model, the developed model has improved the forecasting accuracy by 62.7%. [ABSTRACT FROM AUTHOR]

• A satellite image-based photovoltaic power forecasting model is proposed. • A nonlinear cloud movement forecasting method based on deep learning is proposed. • An algorithm for intra-hour cloud condition estimation is established. • An algorithm for recognizing the cloud region that blocks sunlight is designed. The rapid variation of clouds is the main factor that causes the fluctuation of photovoltaic power. 1 1 This work was supported by the National Key Research and Development Program of China under Grant 2019YFE0118400. The satellite images contain plenty of information about clouds, applicable for photovoltaic power forecast. However, in practice, two main factors obstruct the application of the satellite images: 1) the relatively low update frequency of the satellite images mismatches the photovoltaic power forecasting frequency, and 2) the cloud region that blocks the sunlight changes significantly with time. In this paper, a novel satellite image-based approach for photovoltaic power forecast is proposed to overcome these obstacles and achieve accurate forecasting results. Firstly, concerning the hourly updated satellite images, a nonlinear cloud movement forecasting model, considering the thickness and shape changes of the cloud, is presented to forecast the hourly variation of the images. Secondly, an active cloud region selection rule is derived based on the changing solar position to dynamically select the cloud region that blocks the concerned photovoltaic power station in a satellite image. Thirdly, a sequential cloud region selection algorithm is provided to estimate the intra-hour variation of the cloud to match the photovoltaic power forecasting frequency. Finally, the photovoltaic power is predicted using the XGBoost algorithm concerning the effects of the cloud and other influencing factors. Testing results show that the proposed method can achieve more accurate photovoltaic power forecasts using the low update frequency satellite images. Meanwhile, the superior performance compared with other benchmarks also verifies the effectiveness of considering cloud information obtained by the proposed method for photovoltaic power forecast. [ABSTRACT FROM AUTHOR]

In this paper, a deep learning long-short-term memory (LSTM) method is applied to the forecasting of the critical frequency of the ionosphere F2 layer (foF2). Hourly values of foF2 from 10 ionospheric stations in China and Australia (based on availability) from 2006 to 2019 are used for training and verifying. While 2015 and 2019 are exclusive for verifying the forecasting accuracy. The inputs of the LSTM model are sequential data for the previous values, which include local time (LT), day number, solar zenith angle, the sunspot number (SSN), the daily F10.7 solar flux, geomagnetic the Ap and Kp indices, geographic coordinates, neutral winds, and the observed value of foF2 at the previous moment. To evaluate the forecasting ability of the deep learning LSTM model, two different neural network forecasting models: a back-propagation neural network (BPNN) and a genetic algorithm optimized backpropagation neural network (GABP) were established for comparative analysis. The foF2 parameters were forecasted under geomagnetic quiet and geomagnetic disturbed conditions during solar activity maximum (2015) and minimum (2019), respectively. The forecasting results of these models are compared with those of the international reference ionosphere model (IRI2016) and the measurements. The diurnal and seasonal variations of foF2 for the 4 models were compared and analyzed from 8 selected verification stations. The forecasting results reveal that the deep learning LSTM model presents the optimal performance of all models in forecasting the time series of foF2, while the IRI2016 model has the poorest forecasting performance, and the BPNN model and GABP model are between two of them. [ABSTRACT FROM AUTHOR]

This paper proposes a multimodal deep learning method for forecasting the daily power generation of small hydropower stations that considers the temporal and spatial distribution of precipitation, which compensates for the shortcomings of traditional forecasting methods that do not consider differences in the spatial distribution of precipitation. First, the actual precipitation values measured by ground weather stations and the spatial distribution of precipitation observed by meteorological satellite remote sensing are used to complete the missing precipitation data through linear interpolation, and the gridded precipitation data covering a group of small hydropower stations are constructed. Then, considering the time lag between changes in the daily power generation of the group of small hydropower stations and precipitation, the partial mutual information method is used to estimate the "time difference" between the two, and combined with the precipitation grid data, a data set of the temporal and spatial distribution of precipitation is generated. Finally, using only the temporal and spatial distribution of precipitation and historical power generation data, a multimodal deep learning network based on a convolutional neural network (CNN) and multilayer perceptron (MLP) is constructed, and a highly accurate prediction model for the daily power generation of small hydropower stations is obtained. Taking the real power generation data of a group of small hydropower stations in southern China as an example, after considering the temporal and spatial distribution of precipitation, the prediction accuracy of the proposed method is as high as 93%, which is approximately 5.8% higher than before considering the temporal and spatial distribution of precipitation. In addition, compared with mainstream methods such as support vector regression (SVR) and the long–short-term memory network (LSTM) (the average accuracy is about 87%), and the average accuracy improvement of the proposed method is approximately 6%. [ABSTRACT FROM AUTHOR]

• Deep learning models were employed to predict train delays. • Train operations were modeled as sequences. • Interactions were captured from train groups in the prediction model. • The proposed model shows satisfactory performance on different railway lines. This paper presents a carefully designed train delay prediction model, called FCLL-Net, which combines a fully-connected neural network (FCNN) and two long short-term memory (LSTM) components, to capture operational interactions. The performance of FCLL-Net is tested using data from two high speed railway lines in China. The results show that FCLL-Net has significantly improved prediction performance, over 9.4% on both lines, in terms of the selected absolute and relative metrics compared to the commonly used state-of-the-art models. Additionally, the sensitivity analysis demonstrates that interactions of train operations and weather-related features are of great significance to consider in delay prediction models. [ABSTRACT FROM AUTHOR]

• An end-to-end model is proposed for rice yield prediction using deep learning fusion. • Deep spatial-temporal features significantly reduced prediction errors. • The model converges quickly to low errors and provides stable prediction performance. • The model achieves the optimal prediction performance when the number of layers is 6. Rice yield is essential for more than half of the world's population, and thus, accurate predictions of rice yield are of great importance for trade, development policies, humanitarian assistance, decision-makers, etc. However, traditional mechanistic models and statistical machine learning models need to identify features, making the research on and application of these models laborious and time-consuming. In this paper, a novel end-to-end prediction model that fuses two back-propagation neural networks (BPNNs) with an independently recurrent neural network (IndRNN), named BBI-model, is proposed to address these challenges. In stage one, BBI-model preprocesses the original area and meteorology data. In stage two, one BPNN and the IndRNN are used to learn deep spatial and temporal features in parallel. In stage three, another BPNN combines two kinds of deep features and learns the relationships between these deep features and rice yields to make predictions for summer and winter rice yields. The experimental results indicate that BBI-model achieved the lowest mean absolute error (MAE) and root mean square error (RMSE) for the summer rice prediction (0.0044 and 0.0057, respectively) and corresponding values of 0.0074 and 0.0192 for the winter rice prediction when the number of layers in the network was set to six. Moreover, the errors of the model using the combination of deep spatial-temporal features were significantly lower than when simply using deep temporal features. Furthermore, the model converged quickly with 100 iterations and then remained stable. These findings confirm that the model can make accurate predictions for summer and winter rice yields of 81 counties in the Guangxi Zhuang Autonomous Region, China. [ABSTRACT FROM AUTHOR]

Oil production forecasting is an important means of understanding and effectively developing reservoirs. Reservoir numerical simulation is the most mature and effective method for production forecasting, but its accuracy mostly depends on high-quality history matching and accurate geological models. In order to achieve fast and accurate production predicting, an ensemble empirical mode decomposition (EEMD) based Long Short-Term Memory (LSTM) learning paradigm is proposed for oil production forecasting. In this paper, the original oil production series are first split into training set and test set. The data of test set is gradually added to the training set and decomposed by EEMD to obtain multiple intrinsic mode functions (IMFs). The stability of IMFs is analyzed by its Means and curve similarity computed by Dynamic time warping (DTW). Then proper number of stable IMFs are selected as predictor variables for machine learning. Considering the variation trend and context information of production series, LSTM is utilized to establish predictive model for production forecasting. The optimal hyper-parameters of LSTM are determined by Genetic algorithm (GA). For evaluation and verification purpose, the proposed model is applied to two actual oilfields from China. Empirical results demonstrated that the proposed approach is capable of giving almost perfect production forecasting. • An EEMD based LSTM model is developed for time series forecasting of oil production. • A method to select appropriate stable IMFs decomposed by EEMD is developed based on DTW. • GA is introduced for optimizing hyper-parameters and architecture of LSTM. • It is demonstrated that the EEMD-LSTM can make perfect prediction of oil production using selected IMFs. • Empirical results show that the proposed model outperform other models, such as EEMD-ANN and EEMD-SVM. [ABSTRACT FROM AUTHOR]