Research On Medium And Long-term Power Load Forecasting For User Side Of Power Market

With the development of our country’s society and economy,the reform of power marketization has been fully implemented,the development pattern of "opening the two ends and controlling the middle" has become increasingly clear,the national power market trading system has basically been formed,and the opening of the power market has continued to expand.User-side medium-and long-term power load forecasting is an important basis for power market-oriented operation.Improving the accuracy of user-side medium-and longterm power load forecasting can help enhance the bargaining power of power purchases and reduce operating costs.In this paper,the research on the above situation is carried out,and the main work includes:(1)In response to the fact that there are many outliers and missing values in the corporate electricity load datasets publicly available on the Internet,which are not conducive to the establishment of accurate medium-and long-term electricity load forecasting models,this paper introduces Time-series Generative Adversarial Networks(TimeGAN)to enhance the electricity load This paper introduces Time-series Generative Adversarial Networks(TimeGAN)to enhance the power load data set.The algorithmic principles of the classical Generative Adversarial Networks(GAN)are firstly explored,and their shortcomings and shortcomings are explained;then the algorithmic principles of TimeGAN are investigated,and its advantages in generating time-series data are pointed out;finally,experiments are designed to compare the data generated by TimeGAN and GAN,respectively Finally,we design experiments to compare the data generated by TimeGAN and GAN,and visualize the feature distributions of the generated data and the original data using Principal Component Analysis(PCA)and t-Distributed Stochastic Neighbor Embedding(t-SNE),and compare the original data and GAN data by Prophet algorithm.The original data,GAN-enhanced data and TimeGAN-enhanced data are modeled and predicted respectively,proving the effectiveness of TimeGAN-generated data to achieve the replacement of outliers and filling of missing values in the dataset for the purpose of enhancing the electricity load dataset.(2)In response to the current demand for medium and long-term transactions in the electricity market,and the need for electricity users to submit typical load curves to the electricity dispatching department,this paper introduces the NeuralProphet algorithm to build a medium-and long-term electricity load forecasting model based on the enhanced electricity load dataset.NeuralProphet embeds the Auto-Regressive Neural Network(ARNet)into the Prophet algorithm,which enhances the non-linear fitting capability of the Prophet algorithm while retaining the component interpretability of the Prophet algorithm.NeuralProphet decomposes the electricity load time series into a Trend term,a Seasonal term and an Autoregressive term.Seasonality and Auto-Regression modules are visualised and presented,with the seasonality term giving typical annual,weekly and daily load curves.Finally,experiments were designed to evaluate four forecasting models,NeuralProphet,Prophet,LSTM and ARIMA,using Root Mean Square Error(RMSE)and Mean Absolute Percentage Error(MAPE)to demonstrate NeuralProphet’s superior performance in medium and long-term electricity load forecasting,with its RMSE and MAPE of 532.59 k W and 6.49%respectively,both of which are the lowest values,achieves the purpose of improving the accuracy of medium and long-term forecasting.