A day-ahead industrial load forecasting model using load change rate features and combining FA-ELM and the AdaBoost algorithm

Industrial customers consume a large part of the total electricity demand. In the operation of industrial energy systems, accurate prediction of electric loads is a prerequisite to help industrial users adjust their electric load dispatch and improve energy efficiency. Therefore, this paper proposes a day-ahead industrial load forecasting model employing load change rate features and combining the firefly algorithm to optimize the extreme learning machine and adaptive boosting algorithm (LCR-AdaBoost-FA-ELM). The industrial load is mainly influenced by the power users’ production schedules, making its forecasting laws mainly analyzed by the changing laws of the load data itself. Given this, the rate of load change feature is introduced to form a candidate feature set with variables such as the date and lag load. In order to decrease the number of parameters required to train the model, the Spearman correlation coefficient is used to select high-quality input features and eliminate variables that are weakly associated with electricity consumption. The basic algorithm of the prediction model is the ELM, based on which the FA is used to optimize its weights and biases. Finally, the ensemble learning concept is introduced to learn to combine multiple FA-ELM weak predictors by AdaBoost to correct the prediction errors. In this paper, the proposed model is validated using a typical industrial industry, a furniture factory, as a research case. The results show that the proposed LCR features can capture the nonlinear characteristics of the load sequence, resulting in more precise prediction outcomes. Additionally, based on the FA boosting ELM prediction accuracy, AdaBoost can lower the day-ahead load prediction error once more. Using the mean absolute percentage error (MAPE) as an example, AdaBoost-FA-ELM declines by 76.85% compared to ELM and decreases by 23.90% before and after the LCR features are applied. The proposed forecasting model framework in this paper provides a new research strategy for the field of energy forecasting.