Power Material Demand Forecasting Based on LSTM and Random Forecast

Abstract

This paper takes an electric power company as a case study. Through actual investigation and analysis of the data provided by the company, we found that the material demand forecast of the company is not accurate enough, so that the company cannot effectively arrange the material purchase, resulting in a large backlog of inventory. To improve the reliability and accuracy of the power material demand forecast of the company, this paper proposes a combined forecasting model which combines Long Short-Term Memory Network (LSTM) model and Random Forest (RF) algorithm to predict power material demand. First, the LSTM model is used for preliminary prediction, and then RF algorithm is used to further predict and optimize the residual of LSTM prediction, so as to improve the overall prediction effect. Finally, the combined prediction model proposed in this paper is compared with the single LSTM model and RF algorithm, and it is verified that the combined prediction model has higher prediction accuracy. The research on the demand forecast provides a valuable reference for the company to make material purchase plan and solve the problem of high inventory, and affords a certain reference for the material demand forecast of many similar industries.