Research On Short-term Load Forecasting Of Power System Based On Improved GRU Neural Network

Electric energy is an important energy related to the development of national economy,the power system responsible for the production,transmission,distribution and use of electric energy is a very complex system,high precision short term load forecasting provides an important basis for maintaining the stable operation of power system.With the continuous development of the power system,lots of distributed generation access to the power grid,greatly increasing the difficulty of short-term load forecasting.Reasonable and effective forecasting model has become the focus of current research.The load of power system is affected by many factors.The traditional load forecasting models is difficult to meet the requirements of forecasting accuracy in the current environment.In order to effectively improve the prediction accuracy and reduce the cost of power production,this dissertation studies the artificial neural networks which has good effect in the field of prediction,and mainly analyzes the gated recurrent unit(GRU)neural networks in the neural networks.The traditional GRU neural networks is improved by different ways,and a variety of prediction models are proposed.The input information of the previous time of the prediction point is considered in the prediction process of GRU neural networks,which can be used more effectively than other neural networks.Through particle swarm optimization(PSO),which can effectively achieve global optimization,to optimize GRU neural networks can improve the effect.The results show that the GRU neural networks optimized by PSO algorithm has higher effect and stability.In order to make full use of the input information,a prediction model based on PSO algorithm to optimize the bidirectional weighted GRU neural networks is proposed.The prediction process of GRU neural networks is divided into forward and reverse directions,and the hidden layer information of the two directions is weighted summation.PSO algorithm is used to train the weights of GRU neural networks in two directions and the coefficients of weighted sum,which improves the prediction accuracy of the prediction model.In order to further improve the stability and generalization ability,a model based on Bagging and bidirectional weighted GRU ensemble neural networks is proposed.The bidirectional GRU neural networks is used as the base learner in Bagging.The PSO is used to optimize the base learner.The prediction results of multiple base learners are averaged to get the prediction results.which further improves the prediction accuracy of the prediction model.Taking the ridgelet function as the excitation function of bidirectional GRU neural networks,the bidirectional ridgelet GRU neural networks is obtained,which effectively improves the expression ability of nonlinear function.The bidirectional ridgelet GRU neural networks is used as the base learner,and a model based on Bagging and bidirectional ridgelet GRU ensemble neural networks is obtained.Through the example simulation,it can be concluded that the effect of the proposed model is better.