An Improved Extreme Learning Machine Approach to Real-Time Electricity Price Forecasting in Electricity Markets

Responding to the problem that the high volatility of electricity price data and the complexity of influencing factors in the electricity market led to the low stability and prediction accuracy of the existing prediction models, an improved extreme learning machine method for real-time electricity price predication in the electricity market is proposed. An empirical wavelet transform-based decompose of the electricity price data is designed to decompose the electricity price sequence into low-frequency signals that are closely aligned with the price and high-frequency signals that are influenced by noise. An enhanced random forest-based feature extraction price is proposed to identify the optimal feature combinations of electricity price influencing factors based on prediction importance. The kernel function replaces the hidden layer of the extreme leaning machine to construct the electricity price prediction model. The result, based on the real-time electricity price data of an electricity market, demonstrates the accuracy and stability of the proposed prediction method.