Search

Differential evolution (DE) is a popular evolutionary algorithm inspired by Darwin’s theory of evolution and has been studied extensively to solve different areas of optimisation and engineering applications since its introduction by Storn in 1997. This study aims to review the massive progress of DE in the research community by analysing the 192 articles published on this subject from 1997 to 2021, particularly studies in the past five years. The methodology used to search for relevant DE papers and an overview of the original DE are firstly explained. Recent advances in the modifications proposed to enhance the effectiveness and efficiency of the original DE are reviewed by analysing the strengths and weaknesses of each published work, followed by the potential applications of these DE variants in solving different real-world engineering problems. In contrast to most existing DE review papers, additional analyses are performed in this survey by investigating the impacts of various parameter settings on given DE variants to identify their optimal values required for solving certain problem classes. The qualities of modifications incorporated into selected DE variants are also evaluated by measuring the performance gains achieved in terms of search accuracy and/or efficiency against the original DE. The additional surveys conducted in this study are anticipated to provide more insightful perspectives for both beginners and experts of DE research, enabling their better understanding about current research trends and new motivations to outline appropriate strategic planning for future development works.

Artificial neural networks (ANNs) have achieved great success in performing machine learning tasks, including classification, regression, prediction, image processing, image recognition, etc., due to their outstanding training, learning, and organizing of data. Conventionally, a gradient-based algorithm known as backpropagation (BP) is frequently used to train the parameters’ value of ANN. However, this method has inherent drawbacks of slow convergence speed, sensitivity to initial solutions, and high tendency to be trapped into local optima. This paper proposes a modified particle swarm optimization (PSO) variant with two-level learning phases to train ANN for image classification. A multi-swarm approach and a social learning scheme are designed into the primary learning phase to enhance the population diversity and the solution quality, respectively. Two modified search operators with different search characteristics are incorporated into the secondary learning phase to improve the algorithm’s robustness in handling various optimization problems. Finally, the proposed algorithm is formulated as a training algorithm of ANN to optimize its neuron weights, biases, and selection of activation function based on the given classification dataset. The ANN model trained by the proposed algorithm is reported to outperform those trained by existing PSO variants in terms of classification accuracy when solving the majority of selected datasets, suggesting its potential applications in challenging real-world problems, such as intelligent condition monitoring of complex industrial systems.

Feedforward neural network (FNN) emerges as a promising machine learning algorithm widely used for crucial tasks such as prediction, classification, clustering, regression, and etc. due to its excellent performance in training, learning and organizing data. Conventional approaches such as gradient-based algorithm is commonly used to train FNNs, but it has drawbacks of being trapped into local optima easily, slow convergence rate and high sensitivity to initial solutions generated. Metaheuristic search algorithm such as teaching-learning-based optimization (TLBO) is envisioned as a potential solution to train FNNs due to their stochastic nature and excellent global search ability. In this paper, a new TLBO variant named as teaching-learning-based optimization with modified learning phases (TLBO-MLPs) is designed as the training algorithm of FNN, aiming to optimize its neuron weights, biases and selection of activation functions by referring to the datasets of given classification problems. The classification performances of FNN trained by TLBO-MLPs are evaluated by using the real-life classification datasets from UCI Machine Learning Repository and compared with those optimized by the other well-established TLBO variants. Rigorous simulation studies show that the FNN trained by TLBO-MLPs outperforms most of its peer algorithms significantly in solving different classification problems. Particularly, the FNN classifiers optimized by proposed TLBO-MLPs are reported to outperform its competitors up to 15.72% and 29.07% in terms of classification accuracy for both training and testing datasets, respectively.

Although different modified versions of particle swarm optimization (PSO) were proposed in past decades to solve global optimization problems, the appropriate mechanism used to attain proper balancing of algorithm’s exploration and exploitation searches remains as an open-ended challenges. A modified PSO with unique self-cognitive learning (MPSO-USCL) is proposed in this paper to address this issue. For each particle, a unique exemplar can be generated by the proposed USCL module to replace the self-cognitive component of each particle and guide its search process towards the promising regions of search space with different levels of exploration and exploitation strengths. Extensive simulation studies are performed to compare the optimization performances of MPSO-USCL with six existing PSO variants using 12 benchmark functions. The proposed MPSO-USCL is reported to outperform its peer algorithms for all benchmark functions.

An improved multi-objective particle swarm optimization (MOPSO) variant known as the MOPSO with alternate learning strategies (MOPSOALS) is proposed to overcome the drawbacks of most existing MOPSO variants that can only solve the selected categories of optimization problems with good performance due to the limited directional information brought by search operators. Particularly, both of the current and memory swarm evolution are incorporated into MOPSOALS as the more robust mechanisms in handling different types of problems. Two search operators are introduced in current swarm evolution to determine the particle’s new velocity, while three operators are proposed to fine tune the particle’s personal best position. These five proposed search operators are anticipated to guide all MOPSOALS particles to perform thorough searching in the solution search spaces with various exploration and exploitation strengths by fully utilizing all useful information contained in the non-dominated solution set. The proposed MOPSOALS is reported to have better performance in solving all selected test functions than the five peer algorithms.

Metaheuristic search algorithms (MSAs) such as particle swarm optimization (PSO) and differential evolution (DE) have gained significant attentions from practitioners to solve different engineering applications. Nevertheless, these MSAs are only effective to solve certain problem categories due to No Free Lunch Theorem. A modified hybrid algorithm namely chaotic oppositional based hybridized DE with PSO (CO-HDEPSO) is proposed to address the aforementioned drawbacks in order to solve different global optimization problems more effectively. The chaotic oppositional based initialization scheme is first designed into CO-HDEPSO, aiming to produce initial population with better quality. A hybridization framework is introduced in CO-HDEPSO to combine the strengths of PSO and DE simultaneously for further optimization performance enhancement. Comprehensive analyses are performed to compare the performances of CO-HDEPSO with five MS algorithms using CEC 2014 benchmark functions. The proposed CO-HDEPSO is reported to demonstrate more competitive search accuracy than its peer algorithms in most benchmark functions and able to solve 26 out of 30 CEC 2014 test functions with best Fmean values.

In this paper, wearable sensors based remote patient monitoring system using IoT analytic platform is proposed. This system is targeted for cardiovascular diseases as it can be used to monitor and check the current and historical health status of patient through IoT data analytic platform although the patient is not at the hospital. The low-cost and compact remote patient monitoring is designed and constructed with implementation of various wearable sensors, Arduino UNO, Wi-Fi shield, pump and valve. ThingSpeak is used as the data analytics platform for data monitoring which features an automatic measuring process for all age groups.