Research On Short-term Load Forecasting Of Microgrid Based On Improved LS-SVR

Short-term load forecasting of distribution networks is an important prerequisite for grid stability and economic operation.With the increase of distributed photovoltaic power generation capacity on the distribution side and the promotion of microgrid structure on load side,more and more photovoltaic microgrids are connected to the distribution network.Since solar energy is affected by a variety of random factors,the output power of photovoltaic power generation varies in a variety of ways,with intermittent,volatility and unpredictability.Directly forecasting the microgrid containing distributed photovoltaic power on the distribution network perspective,according to which the appropriate adjustment of the operation mode and control strategy of the distribution network is essential for the stable economic operation of the power grid.From the perspective of distribution network,this paper regards the microgrid load with distributed photovoltaic power supply as a whole to predict.Construct a direct prediction model of PV microgrid net load to reduce the number of models of the prediction system and to effectively avoid indirect Predict the cumulative prediction error.Then improve the model to achieve the improvement of prediction accuracy and operational efficiency,realizing more accurate and faster one-time prediction of the net load.The specific research content mainly includes:(1)The basic principles of statistical learning theory and support vector machine are briefly described.By analyzing the improvement process of support vector machine,it is found that the solution of least squares support vector machine lacks the sparseness and robust line.The angle is analyzed to explain the reasons for the lack of robustness of the least squares support vector machine.(2)The characteristics of microgrid load and microgrid PV output and its influencing factors are introduced.The concept of net load is put forward from the perspective of global control of distribution network.The shortcomings of indirect forecasting method of net load are analyzed.Methods based on least squares support vector machine verify the effectiveness of the direct load forecasting method.(3)Due to the great uncertainty of distributed PV power,the sparse training sample set is realized by finding the approximate base of the input sample in the feature space,then the IGGIII weighting function is selected.Improves the robustness of the previous Net load directly prediction model based on least squares support vector machine and achieves more accurate and stable prediction accuracy.(4)For the problem that the model running time is greatly increased after the robustness and the sparseness are improved,the fine particle size parallel particle swarm optimization(PSO)is applied on the graphics processing unit(GPU).This optimization of the sparse robust least squares support vector machine penalty factor C and the kernel parameter ? can reduce the optimization time of the parameters and improves the overall running speed of the model.