Research On Grid Monthly Load Forecasting Based On Relevance Vector Machine

The accuracy of load forecasting is an important guarantee for the safety and economic operation of power system.The development of power market also increases the prediction accurate request.Monthly load has a double trend of growth and fluctuation,which presents complex nonlinear characteristic and increases the difficulty of prediction.With the development of artificial intelligence(AI),it appears strong non-linear processing ability and extensive adaptability in regression forecasting problems,which provides a new idea for load forecasting.Therefore applying AI to monthly load forecasting can improve the accuracy and adaptability of the prediction model and is of great significance to power system operation.In the beginning this thesis introduces the research status of monthly load forecasting and the application of AI in load forecasting.Then from the view of machine learning problem,statistical learning theory and relevance vector machine(RVM)are expounded.Aiming at the defects of RVM’s sensitive to the kernel function and model parameters,combined kernel function and improved particle swarm optimization are used to combine the advantages of different kernel functions and optimize parameters.Then the load forecasting model based on PSO-RVM is established.The thesis also analyzes the external and internal influence factors of monthly load.The external factors include economic,industrial structure and temperature.By K-L information method,high correlation factor is extracted as model’s input to establish the load forecasting model based on K-L information and PSO-RVM.Business expansion capacity is considered as the internal factor.Electricity trend after business expansion is studied by growth curve fitting and k-means clustering algorithm,which is used to obtain monthly effect ratio and calculate business expansion increment that having practical impact on the load.Then the load forecasting method considering business expansion based on K-L information and PSO-RVM is proposed.Finally,the case study indicates that all three models proposed in this thesis are of high precision and have guiding significance for monthly load forecasting.