Modeling and forecasting the electricity clearing price: A novel BELM based pattern classification framework and a comparative analytic study on multi-layer BELM and LSTM

With the deregulation of power market and the increasing penetration of renewable energy, the core role of demand side management (DSM) has become even more prominent. In this sense, there is an urgent need for all market participants to identify the pivotal aspects of electricity market price fluctuation effectively and anticipate its future trend. For certain applications such as DSM, considering the high volatility and nonlinear of real-time electricity price, we can approximate the interval prediction to achieve the multi-classification that relies on critical pattern recognition of entire category of the price sequence. Therefore, this paper presents a study on the utilization of a novel electricity price classification framework which consists of Bayesian extreme learning machine (BELM) model, minimum redundancy maximum relevance (MRMR) algorithm, and multivariate sequence segmentation (MSS). Considering many advantages of deep learning structure in capturing the hierarchical and sophisticated characteristics of multidimensional sequence, the multi-layer BELM (ML-BELM) model is also extended and utilized to the modeling. To demonstrate the potential of the pattern classification framework, the proposed approaches are evaluated using hourly clearing price cases from Canada Ontario and New York electricity market. In particular, we investigate the performance of different classifiers regarding the 3 multi-classification and higher dimensional classification modes with respect to various scenarios in terms of precision, recall, AUC (Area Under roc Curve) score, and F1-measure indicators. The findings suggest that the proposed pattern classification framework can obtain satisfactory forecasting results provided that suitable scheme is utilized to the pattern segmentation and feature ranking process.