A developed short-term electricity price and load forecasting method based on data processing, support vector machine, and virus colony search

Abstract

Surely, electricity market participants need an accurate estimate for the price and load signals to properly manage their programs to increase their profits. Regarding the two-way interaction between suppliers and consumers, they are able to manage their required profile. Therefore, the individual price or load forecasting is not acceptable and cannot capture their related pattern. Therefore, this paper presents a new forecasting method for price and load signals in simultaneous configuration. This method consists of three main parts: the first part proposes two data preprocessing methods based on wavelet transform to remove noisy term and mutual information to select best features with maximum relevancy and minimum redundancy. The second part suggests a learning engine based on a least square support vector machine (LSSVM) with self-adaptive kernel functions and generalized autoregressive conditional heteroscedasticity (GARCH) time series to extract the linear and nonlinear patterns. Finally, the last part employs a new modified virus colony search algorithm (VCS) to efficiently set the LSSVM control parameters to use its all capacity. Simulations are carried out for three well-known Australia’s, New York’s, and Ontario’s electricity markets. The obtained results show the acceptable results.