A novel model: Dynamic choice artificial neural network (DCANN) for an electricity price forecasting system

• The set of bad subsamples is defined.
• The IBSM is proposed to confirm the set of bad subsamples.
• A novel method called DCANN is able to select the proper input for each output.
• The Optimization Algorithm is utilized to update the original DCANN.
• The proposed model has a minimum MAPE in the simulations conducted in four seasons.

Big data mining, analysis, and forecasting always play a vital role in modern economic and industrial fields. Thus, how to select an optimization model to improve the forecasting accuracy of electricity price is not only an extremely challenging problem but also a concerned problem for different participants in an electricity market due to our society becoming heavily reliant on electricity. Many researchers developed hybrid models through the use of optimization methods, classical statistical models, artificial intelligence approaches and de-noising methods. However, few researchers aim to select reasonable samples and determine appropriate features when forecasting electricity price. Based on the Index of Bad Samples Matrix (IBSM), a novel method to dynamically confirm bad training samples, and the Optimization Algorithm (OA), DCANN and Updated DCANN are proposed in this paper for forecasting the day-ahead electricity price. This model is a hybrid system of supervised and unsupervised learning and creatively applies the idea of deleting bad samples and searching quality inputs to develop and learn, which is unlike BPANN, RBFN, SVM and LSSVM. Numerical results show that the proposed model is not only able to approximate the actual electricity price (normal or high volatility) but also an effective tool for h-step-ahead forecasting (h is less than 10) compared to benchmarks.

Figure optionsDownload as PowerPoint slide