DATA DRIVEN RULE-BASED PEAK SHAVING ALGORITHM FOR SCHEDULING REFRIGERATORS

The increasing use of thermostatically controlled loads (TCLs) like refrigerators poses a significant challenge to the grid due to their potential to increase peak demand. This study introduces a novel rule-based peak-shaving algorithm to effectively manage these loads. The algorithm operates in two modes: day-ahead and real-time. In the day-ahead mode, Long Short-Term Memory (LSTM) neural networks are utilized to forecast demand and generation. A Parameter tuned Grey Wolf Optimizer (GWOP) is proposed and employed to determine the optimal generation for the initial timestep of the scheduling period. The GWOP is tuned using a brute-force grid search method to optimize its parameters. In the real-time mode, the algorithm dynamically adjusts refrigerator operations based on real-time mismatch calculations between predicted demand and generation. Dynamic flexibility thresholds are employed to determine the optimal operation of refrigerators during peak and off-peak periods. This approach aims to minimize energy consumption while maintaining thermal comfort. The algorithm's performance was evaluated using real-world data from the Spanish Transmission Service Operators (TSO). The results demonstrate a significant reduction in peak demand and total energy consumption. The algorithm with dynamic flexibility achieved a substantial 18.89% reduction in peak demand and a notable 12.12% decrease in total energy consumption.