Short-term electric load forecasting based on improved Extreme Learning Machine Mode

Abstract

Aiming at the nonlinear characteristics of short-term electric load series, a short-term electric load forecasting model with improved Extreme Learning Machine (ELM) is proposed. This model is based on the data feature analysis of Ensemble Empirical Mode Decomposition (EEMD) and Long Short-Term Memory (LSTM). EEMD is used to decompose the load sequence into several regular modal components to reduce the error caused by the randomness of the sequence. LSTM and ELM are used to predict the high-frequency and low-frequency components respectively. After using the improved Extreme Learning Machine (ELM), the prediction results of each component are solved. Then the optimal results of power load forecasting are obtained. The results show that the proposed model can judge the overall change trend of the electric load and predict the details of the local change. The average absolute percentage error is 2.24% comparing to 4.78% of the best experiment result of single model, and it has better generality.

Keywords

• Electric load
• Extreme learning
• High frequency component sequence
• Adaptive decomposition
• Data feature mining