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

The power industry is the basic guarantee for social production and people’s daily life.Highprecision short-term load forecasting of power loads is of great significance to maintaining the reliable operation of the power system.The research on related technologies has also received more and more attention from the industry.With the proposal of the "dual carbon" goal and the improvement of the working mode of the power grid,the power system is constantly developing and becoming more complex,and the difficulty of short-term load forecasting is greatly increased.Traditional forecasting methods are difficult to meet the requirements of short-term load forecasting in the current environment,so this paper improves and optimizes based on traditional neural network,aiming to explore new methods to improve the accuracy of load forecasting.This paper fully investigated and analyzed the power load prediction characteristics and load influence.Based on Recurrent Neural Networks(RNN),Long Short Term Memory(LSTM)and Gate Recurrent Unit(GRU)Neural network model,Finally,the optimization effect of the prediction model is verified by experimental simulation.The main work is as follows:(1)Firstly,the characteristics of power load are analyzed,and the rules are mined to lay a theoretical foundation for load prediction;Secondly,the external factors affecting power load and the causes of prediction errors are analyzed.For the collected historical sample data,several common pre-processing methods are analyzed in detail,including the elimination of abnormal data,the supplement of missing data,data normalization processing,etc.Finally,several evaluation indexes of power load forecasting and basic realization steps of load forecasting are analyzed.(2)Based on the RNN,LSTM and GRU neural network models as the basic models,the structure and characteristics of the neural network are analyzed,and the advantages and disadvantages of each neural network in the application of load prediction are compared and analyzed.In order to improve the accuracy of short-term load forecasting,a short-term load forecasting model based on gradient descent GRU neural network is proposed.(3)This paper analyzes in detail the modeling process and steps of the neural network forecasting model,as well as the flow of load forecasting;makes a detailed comparative analysis on the selection of key parameters affecting the forecasting results,and selects actual data sample sets of different seasons and different date types.Simulation experiments are carried out,and the comparison and analysis of the experimental results show that the GRU neural network model has high prediction accuracy and prediction stability,and has good general adaptability.(4)According to the influencing factors in actual production,the prediction accuracy is further improved and the error is reduced.Combined with the Attention mechanism,the climate characteristics and date type characteristics are normalized and processed together with the historical load data as the input information of the neural network prediction model.The experimental results of the data analyze the influence of climatic factors and date type factors on the load retest,and verify the optimization effect of the Attention mechanism on the prediction accuracy and stability.