Application of least squares support vector machines in medium-term load forecasting

Power load forecasting is essential in the task scheduling of every electricity production and distribution facility. In this project, we study the applications of modern artificial intelligence techniques in power load forecasting. We first investigate the application of principal component analysis (PCA) to least squares support vector machines (LS-SVM) in a week-ahead load forecasting problem. Then, we study a variety of tuning techniques for optimizing the least squares support vector machines' (LS-SVM) hyper-parameters. The construction of any effective and accurate LS-SVM model depends on carefully setting the associated hyper-parameters. Poplular optimization techniques including Genetic Algorithm (GA), Simulated Annealing (SA), Bayesian Evidence Framework and Cross Validation (CV) are applied to the target application and then compared for performance time, accuracy and computational cost. Analysis of the experimental results proves that LS-SVM by feature extraction using PCA can achieve greater accuracy and faster speed than other models including LS-SVM without feature extraction and the popular feed forward neural network (FFNN). Also, it is observed that optimized LS-SVM by Bayesian Evidence Framework can achieve greater accuracy and faster speed than other techniques including LS-SVM tuned with genetic algorithm, simulated annealing and 10-fold cross validation.

Application of least squares support vector machines in medium-term load forecasting

Power load forecasting is essential in the task scheduling of every electricity production and distribution facility. In this project, we study the applications of modern artificial intelligence techniques in power load forecasting. We first investigate the application of principal component analysis (PCA) to least squares support vector machines (LS-SVM) in a week-ahead load forecasting problem. Then, we study a variety of tuning techniques for optimizing the least squares support vector machines' (LS-SVM) hyper-parameters. The construction of any effective and accurate LS-SVM model depends on carefully setting the associated hyper-parameters. Poplular optimization techniques including Genetic Algorithm (GA), Simulated Annealing (SA), Bayesian Evidence Framework and Cross Validation (CV) are applied to the target application and then compared for performance time, accuracy and computational cost. Analysis of the experimental results proves that LS-SVM by feature extraction using PCA can achieve greater accuracy and faster speed than other models including LS-SVM without feature extraction and the popular feed forward neural network (FFNN). Also, it is observed that optimized LS-SVM by Bayesian Evidence Framework can achieve greater accuracy and faster speed than other techniques including LS-SVM tuned with genetic algorithm, simulated annealing and 10-fold cross validation.

Feature Extraction from Spectral Data Using the Bayesian Evidence Framework

The optimal selection of discriminatory features from large datasets remains a pressing problem in damage identification. In this paper, a Bayesian approach to classification and feature selection is introduced and applied to a challenging experimental problem.