A study on using two-stage multiple support vector machines to forecast medium-term electricity market prices

Using a two-stage multiple Support Vector Machine (SVM) model as the foundation, this work proposes a unique forecasting approach focused on the medium-term electricity Market-Clearing Price (MCP) forecasting issue. The model aims to efficiently capture the complex nonlinear dynamic features of electricity market prices in order to provide accurate price forecasting information to electricity market participants and decision makers. Firstly, we introduce the core idea of the two-stage SVM, where the first stage SVM is used for the initial fitting of the data, while the second- stage SVM further fine-tunes the model to accommodate price fluctuations under different market conditions. Through this two-stage modelling approach, we are able to better address the non-linear and dynamic nature of electricity market prices. In this study, we pay special attention to two key SVM control parameters: the coefficients of the Gaussian radial basis kernel function (σ and the regularization constant (γ). The selection and optimization of these parameters is a key factor in the success of this model. We propose a method for parameter tuning using test data to ensure optimal performance of the SVM model. By optimizing the parameters, we can minimize the Mean square Absolute Error (MAE) of the SVM model and thus improve its accuracy in price forecasting. By applying it in medium-term electricity market price forecasting, we can better understand market trends and price fluctuations.