Authors: Manjesh Kumar, Assistant Professor Jaya Shukla, Professor Rajnish Bhasker
Abstract: Short-term electric price prediction is important in deregulated power markets and operations as well as planning processes as it aids in the bidding process, risk management and demand response programs. The growing infiltration of renewable energy sources, as well as switching variability of the loads, and market uncertainties, has brought about high nonlinearity and volatility in the electricity price dynamics, which restrain the applicability of traditional forecasting techniques. In solving such challenges, this paper suggests a hybrid deep learning forecasting structure combined with efficient feature selection mechanism to predict short-term price of electricity. The advanced feature selection methods are used in the proposed approach to determine the most informative market, demand, and generation-related variables and to lower the dimensions, as well as to remove redundant information. A hybrid deep learning model, which is a combination of the positive attributes of sequential and nonlinear learning structures, is subsequently trained exploiting the chosen features to absorb intricate temporal variations and price surges. An evaluation of the model by real-world data of the electricity market and a comparison with the traditional statistical methods and individual machine learning are conducted. The simulation outcomes prove that the suggested hybrid structure is more accurate in predictions, more robust, and converges faster, which is indicated by the lower error indicators like MAE, RMSE, and MAPE. In addition, the feature selection step will increase the interpretability and the computational efficiency of models without affecting prediction accuracy. The results attest to the fact that the suggested approach is highly applicable when it comes to short-term electricity price prediction in highly volatile and renewable-based power markets.
