Adaptive Regularized ELM and Improved VMD method for Multi-step ahead Electricity Price Forecasting Grid

This paper proposes a hybrid machine learning algorithm for multi-step ahead electricity price forecasting problem. The non-stationary time series data like electricity price, needs robust learning model for prediction of future market price to effective operation of the market based power system. In this research, an adaptive regularized extreme learning machine (ARELM) is proposed with adaptive weight updation in the hidden layers based on both structural risk minimization and empirical risk minimization. The Ant Colony Optimization (ACO) algorithm is applied for optimizing the initial weights and thresholds between input layer and hidden layer of ARELM model. To enhance the overall prediction accuracy of ARELM, a new improved Variational Mode Decomposition (IVMD) is employed to decompose the pricing data into several intermediate frequency modes thereby eradicate stochastic components. Two real-time electricity price series of Australia and India are adopted for multi-step ahead prediction and the results are compared with other learning models available in the literature.