Particle Swarm Optimization With Extended Memory And Its Application In Power System Short-term Load Forecasting

Along with the recent privatization and deregulation of the electricity market, the accuracy of future electricity demand forecasting has received growing attention, particularly in the areas of electricity load planning, energy expenditure and secure operation fields, in regional and national systems.During recent decades, numerous investigations have been proposed to improve the accuracy of electricity load forecasting. Prediction models based on classical mathematical statistics methods don't achieve satisfactory prediction accuracy for those models difficult to manifest complex and non-linear relationships between the load and corresponding factors.Support vector machine (SVM) has been one of hot research topics of the power system load forecasting,which implements the structural risk minimization(SRM) principle rather than the empirical risk minimization principle implemented by most of the traditional neural network models. The most important concept of SRM is the application of minimizing an upper bound to the generalization error instead of minimizing the training error. Based on this principle, SVM achieves an optimum networks structure. In addition, SVM will be equivalent to solving a linear constrained quadratic programming problem so that the solution of SVM is always unique and globally optimal. It is SVM that wins these prominent advantages over solving nonlinear regression problems. Therefore, it's feasible to introduce SVM to short-term load forecasting.Particle swarm optimization algorithm with extended memory (PSOEM) is presented for the problem that particles often lost their way when applying the standard particle swarm optimization (PSO) algorithm to find the best solution. This paper combines SVM with PSOEM and then builds PSOEM-SVM forecasting model. The PSOEM searches the solution space intelligently and finds out the best one. Parameters in SVM are optimized by PSOEM, which implements automation of the parameter optimization avoiding the blindness of selecting parameters. Not only does it utilize the generalization feature of SVM, but enhance the global search ability of PSO (Particle Swarm Optimization). Thus, both accuracy and speed are increased at the same time.However, the improvement of load forecast accuracy needs to fully understand and strong grasp the influences of environmental factors. EMD (Empirical Mode Decomposition) is employed to decompose the nonlinear and non-stationary power load series into IMFs (Intrinsic Mode Functions) and the residue during the parse of data preprocessing. Analysis of IMFs and the residue disclose the character of load variation and the effect of environment factors. Then choose appropriate kernel functions, build prediction models respectively. All components'prediction results are reconstructed and the final results are gained.In view of the reduction of computing effort and the increase of computing efficiency, characteristics of IMFs and the residue are summarized and flow of prediction is simplified. The simulation example of the EUNITE (Uropean Network on Intelligent TEchnologies for Smart Adaptive Systems) power load prediction competition verifies the effectiveness of the proposed prediction model.