A method for short-term electric load forecasting based on the FMLP-iTransformer model

Abstract

The massive integration of intermittent renewable energy sources into the power grid poses a formidable challenge to the accuracy of power load forecasting. To address this challenge, this paper proposes the FMLP-iTransformer model, which integrates a novel feature capture module with the iTransformer to enhance the model’s capability in extracting features from time-series data. Consequently, the model achieves flexible and precise predictions of short-term power loads, accommodating periodicity with greater agility. Through experiments conducted on actual datasets and evaluation criteria including the Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE), the model has demonstrated more superior in the both standards compared to others. This result verifies that the means proposed by this paper could predict more accurate short-term power load, providing precise information for operation and dispatch of the grid. • FMLP-iTransformer enhances feature and temporal learning for load forecasting. • Model shows robust accuracy across short-term forecast lengths. • Simplified Transformer with single encoder boosts training efficiency. • CNN and FMLP improve feature extraction and prediction accuracy. • FMLP-iTransformer outperforms iTransformer and others in accuracy and efficiency.