Short-term Power Load Forecasting Based On EMD And SVM

Power system load forecasting level as a measure of one of the important ways of power system operation and management,the forecasting result is affected by multiple factors,the current power system short-term load forecasting is to realize the power grid and the important measures for power supply enterprises to participate in market competition,its purpose is to promote the economic and social benefits for the power industry,at the same time in order to ensure the security and stability on the basis of reduce the cost of power grid,you need to load forecasting accuracy reaches the highest,but the traditional load forecasting methods are often on the precision of prediction is difficult to meet the requirements,therefore,the study of the method for predicting become the focus of research scholars.In this paper,a short-term power load forecasting model based on EMD(empirical mode decomposition method)and SVM(support vector machine)is adopted to solve the problem of load forecasting accuracy.Firstly,this paper briefly describes the research background and significance of short-term power load prediction,expounds the research status at home and abroad,and analyzes the characteristics and flow of power load.For the characteristics of power load uncertainty,conditionality,timeliness and multi-scheme,this paper adopts the support vector machine model for prediction.Support vector machine can effectively accommodate the complexity and self-learning ability of the model,and compared with other methods,it has better generalization ability.Moreover,this method has strong convergence,relatively few adjustable parameters,and does not need too much experimental data.In this paper,based on the traditional vector machine,the least squares support vector machine model is adopted,which has a better performance in power load forecasting.Secondly,PSO(particle swarm optimization algorithm)is adopted to optimize the parameters of SVM kernel function,and PSO is improved from two aspects,namely topology structure and algorithm parameters.In terms of topological structure,the standard deviation of particle distance is proposed for the selection of initial population of particle swarm,the mean value of optimal reference value of particle swarm around the optimal reference value is taken for reference,and the fitness variance is set for judging premature convergence.In terms of algorithm parameters,a dynamic weight inertia value is proposed.On the basis of the above improvements,this paper established a SVM model for improved PSO optimization,and verified it with an example.Through comparative analysis,it was concluded that this model had better performance in short-term power load forecasting.For the instability of power load data,periodic not obvious characteristics,this paper improved particle swarm algorithm based on support vector machine(SVM)introduced the EMD decomposition method,by the complex of the original signal feature extraction separation,can be complex and unstable load signal is decomposed into relatively stable signal,is more advantageous to forecast load data processing.Combined with the improved PSO-SVM model proposed previously,a model combining EMD and improved PSO-SVM was established,and an example was used to verify the comparison between the results and the improved PSO-SVM model without EMD.The prediction results proved that the prediction accuracy was greatly improved after the introduction of EMD decomposition method.In order to better reflect its application,the combined method is compared with other mainstream forecasting methods,and it is concluded that this method is more accurate and feasible.