Research On Short-Term Power Load Forecasting Based On LSTM Combination Model

People’s life and production are inseparable from the use of electric energy.All over the world,the power industry is a major basic industry.To develop the economy,we must first ensure power.Power load forecasting is a key part of ensuring the safe operation of power grids.The power department can reasonably dispatch the generator sets according to the results of the power load forecast,so as to ensure the normal power supply.The research of relevant scholars in the field of power load forecasting has been very mature,and the technology has also developed from the ancient empirical forecasting method and traditional forecasting method to the modern forecasting method using high-tech means.Machine learning is widely used in power load forecasting because of its ability to fit nonlinear data well.As an extension of machine learning,deep learning is the most popular research direction,and is widely used in power load forecasting.Due to the large amount of data and strong nonlinearity of power load data,the error is often very large when using experience and traditional forecasting methods.The longterm memory functional unit of the long short-term memory neural network(LSTM)can better screen data and information,and in the field of short-term power load forecasting,it can well handle the long sequence problems generated in the forecasting.This paper focuses on the data set and processing methods of power load forecasting.First,the original data is processed,and the redundant data is eliminated by scientific means to fill in the missing data.Then,the LSTM combined model is used to forecast the power load.In order to improve the prediction accuracy of long short-term memory neural network,a combined PSO-CNN-LSTM model is constructed.The model fully combines the optimization ability of particle swarm optimization algorithm(PSO)and the advantages of local feature extraction of convolutional neural network(CNN),making up for the shortcomings of CNN-LSTM and PSO-LSTM models.The prediction effect of the PSO-CNN-LSTM model is compared and analyzed with the prediction effect of the LSTM,PSO-LSTM,and CNN-LSTM models.Through the analysis of the evaluation error index parameters of the model,the results show that the prediction effect of the PSO-CNN-LSTM model is the best,and the prediction accuracy is as high as 98.2%.Due to the periodic nature of the power load,the peak-valley time-sharing forecasting method is used,and the time is divided into peaks and troughs,and the PSO-CNN-LSTM model is used to predict and stack.After several simulation experiments and analysis of error evaluation indicators,it is shown that the time-sharing prediction method has a small improvement in the prediction accuracy,which further verifies the effectiveness of the combined model.In order to facilitate users to observe and summarize the power load data and facilitate the rational dispatching of the power department,a visual platform for power load prediction is built.Firstly,the overall design scheme of the system is analyzed.The system is developed based on Java Script,adopts browser/server architecture,and uses My Sql database.The power load prediction results are obtained by the combined model.Then the functions and achievements of the system are briefly introduced.Figure [46] Table [26] Reference [80]...