Medium and Short-Term Prediction of Power System Load Based on Improved LSTM Algorithm

Abstract

Electricity production has the characteristics of “production and sale”, and the entire production and consumption are completed in an instant. Therefore, predicting the load changes of the power system in the future and selecting the appropriate operation mode in advance is an important task to maintain the power balance and promote economic operation. In power system load forecasting, the influence of historical system load social conditions and natural conditions should be fully considered, and the system load will be predicted for a period of time in the future. Load forecasting is an important part of power system planning and dispatching. The medium and short-term forecast analysis of regional load includes the load forecast in units of 15 minutes in the next 10 days, as well as the maximum and minimum load forecasts and their time in daily units in the next 3 months, while ensuring the accuracy of the two forecast results. This paper adopts the univariate ARIMA model and the LSTM model based on the neural network method. After comparing the results, the model with higher accuracy is selected as the best result. By comparing the ARIMA and LSTM models, we found that LSTM has higher prediction accuracy in terms of electricity load, and use the improved LSTM model based on grid search to predict the maximum and minimum daily load and accuracy of each industry in the next 3 months. We propose the idea of predicting the future through a time series with a certain period of time, and establish an adaptive time window LSTM model through the combination of algorithms.