Neural Network Short-term Load Forecasting Based On WOA And Clustering Algorithm

The electric power industry occupies a self-evident position in the national economy,society and other fields,and is one of the most important basic industries in the country.Power system load forecasting is related to the safe,reliable and economic operation of power system,and its accuracy is the basis of the rationality of power system dispatching.In view of this,on the basis of load data analysis and processing,the neural network is combined with Whale Optimization Algorithm(WOA)and clustering algorithm respectively,and an improved WOA-LSTM neural network prediction model and piecewise back propagation(BP)neural network prediction model are established,the former is used to solve the problem that the accuracy of Long Short-Term Memory(LSTM)neural network prediction model is too low due to the initialization of weights and bias values,and the latter is used to solve the problem that the accuracy of neural network prediction model declines seriously in a slightly long period of time.The main contents and research results are as follows:(1)The main factors affecting the accuracy of load forecasting are studied,the characteristics of load are analyzed,and the data processing method of load historical data is introduced in this paper.On this basis,the general process of power load forecasting is summarized.(2)The theory and method of load forecasting are expounded,and the main factors affecting the load change are analyzed in this paper.The date constant is introduced into the historical load data,and the method of feature contribution is used to prove its effectiveness.(3)The WOA algorithm is used to optimize the weight and bias of LSTM neural network,which reduces the difficulty of back propagation through time(BPTT)algorithm.In addition,the mutation algorithm is used to improve the WOA.Finally,the improved WOA-LSTM neural network is established,and the load data in the next 48 hours is predicted through simulation experiments.At the same time,the forecasting results of LSTM forecasting model,WOA-LSTM forecasting model and improved WOA-LSTM forecasting model are compared.It is proved that the improved WOA has obvious effect on the application of LSTM forecasting model in short-term load forecasting,and can effectively reduce the forecasting error of the model.(4)The working principle and process of K-means algorithm in clustering analysis are studied,BP neural network is combined,and K-means algorithm is used to segment historical data.Then BP neural network is used to predict the load in each data segment,and the piecewise BP neural prediction model is got by stitching all the in and out of BP neural network finally.After the piecewise BP neural network prediction model is established,the influence of K value selection on the prediction accuracy of the model in detail is studied,and K=2,K=3,K=4 are taken as examples to carry out the simulation test on the model.It is concluded that the selection of K value has a great influence on the prediction accuracy of the piecewise BP neural network prediction model,and its larger or smaller value will increase the prediction error of the model.In addition,the prediction output from different K values is compared with the prediction results of the improved WOA-LSTM prediction model,which proves that the prediction accuracy of the piecewise BP neural network prediction model is better than the improved WOA-LSTM prediction model in a slightly long period of time.