Load Forecasting Of Microgrid In Smart Grid Environment

Load forecasting provides reliable theoretical support for the operation of smart grid.It is a qualitative and reasonable prediction of future data based on the rule summary and expansion of historical data.However,with the development of more complex data,the traditional load forecasting methods can no longer meet the requirements.With self-learning artificial neural and combination algorithm for the lack of algorithms in the power system.However,the prediction accuracy will be different with different theoretical basis of the algorithm.Massive data also increases the computational complexity,which slows down the computational speed of the algorithm to some extent.Therefore,how to improve the computational speed and have better prediction accuracy has become a hot topic of The Times.Based on the multiple factors affecting the load and the characteristics of the load,this paper presents an improved RBF????neural network prediction model for LSSVM training.The main work in this paper is as follows:(1)Based on the characteristic analysis of load data and the judgment of load influencing factors,the grey relational degree algorithm was used to calculate the sexual correlation of real-time data and determine the input variables of the prediction model.This method can well screen the factors that can affect the load data and reduce the complexity of the data to some extent.At the same time,the method is also used to divide the similarity of the input data,and the training sample set of the prediction model is obtained.(2)Proposed chaos ant colony algorithm to optimize LSSVM.LSSVM algorithm has good generalization ability and optimization ability,but its parameters are usually set by human experience input,too much subjectivity.In this case,combined with the better combinatorial search ability of chaos ant colony algorithm,it is applied to the parameter optimization of LSSVM.Firstly,chaos theory is introduced into ant colony algorithm based on the chaos of ant behavior.Then,the combination parameters of LSSVM are used as the positions of ants to construct the objective function,and the communication ability between groups is used to exchange the information of the optimal position until the optimal value is reached or the set number of times is reached.So that it is not easy to fall into the local optimization,improve the performance of the prediction model.(3)Based on the similarity between RBF neural network and LSSVM,the LSSVM output regression machine was used to determine the training model of RBF.The performance of RBF neural network is determined by parameter setting and central function.The regression machine will be output after LSSVM training.The parameter design and mechanism design of RBF neural network are determined by the optimal regression machine.Finally,MATLAB is used to simulate the real-time data of the optimization algorithm proposed in this paper.The simulation results show that compared with the traditional LSSVM or RBF neural network,the improved LSSVM in this paper has higher load accuracy and faster calculation speed in training the RBF????neural network.