A Two-stage Forecasting Approach for Day-ahead Electricity Price Based on Improved Wavelet Neural Network with ELM Initialization

Abstract

In deregulated electricity markets, reliable electricity price forecasting (EPF) is the basis for developing bidding strategies, operating dispatch controls, and hedging volatility risks. However, electricity prices are highly volatile, non-stationary and multi-seasonal, making it challenging to estimate future trends, so the accuracy of most existing forecasting models falls short of the practical requirements. To this end, a hybrid model combining feature extraction, pattern recognition, neural network models and machine learning is proposed for day-ahead EPF. The model is divided into two main steps: first, feature extraction is performed with Lasso. And then, k-means is used to cluster all historical daily electricity price curves into different patterns, and the SVM model is proposed to recognize the price patterns. Second, a novel improved wavelet neural network (IWNN) model supported by extreme learning machine (ELM) initialization is proposed to build classification prediction models for different daily patterns, which effectively solves the problem of slow or even non-convergence of the traditional WNN. Case studies based on PJM market data show that the proposed approach outperforms other approaches, especially when the volatility of electricity prices is high.