Grid Imbalance Prediction Using Particle Swarm Optimization and Neural Networks

Abstract

Fluctuations in the power demand amounts, supply problems, uncertainty in weather conditions are known to cause power deviations in the real-time power market. The imbalance costs are reflected in the consumer prices in the partly liberated markets of the developing countries. Thus, the accurate short-run forecast of the electricity market trends is beneficial for both the suppliers and the utility companies to constitute a balance between the physical energy supply and commercial revenue. When both day-ahead market and intra-day market exist to respond to the power demand, forecasting the imbalances lead both the suppliers and the regulators. This study aims to optimize the grid imbalance volume prediction by integrating the Particle Swarm Optimization (PSO) and Long Short-Term Memory Recurrent Neural Networks (LSTM). The model is applied for 1 h, 4-h, 8-h, 12-h and 24-h ahead. The Mean Absolute Percentage Error (MAPE) is also calculated. As a result, The MAPE levels are found to be 27.41 for 24 h, 25.66 for 12 h, 26.77 for 8 h, 25.39 for 4 h, 9.25 for 1 h. Although improvements are foreseen both in the model and data, achievements of this study would reduce the imbalance penalties for the power generators, whereas, the regulators will organize the outages with a precise approach. Hence, the economic benefits will affect the trading prices in the long term.