Machine Learning–Based Demand Side Management for Industrial Load Scheduling with Solar Energy Integration

As the effects of volatile electricity prices, network constraints, and the growing use of renewable energy sources on industrial facilities become more significant, the traditional load scheduling approach that uses fixed operating plans is being challenged, and this has led to reconsideration of how load scheduling is managed in light of the convergence and interplay of these factors. Conventional methods fail to take into account hourly price fluctuations and the time-dependent availability of solar energy, leading to unnecessary operational costs and inefficient energy use in many instances. This work introduces a demand-side management framework that integrates machine learning with a practical load scheduling strategy for industrial applications. The study employs a Random Forest-based classification model to predict hourly operating decisions for seven industrial loads based on temporal variables, electricity tariffs, photovoltaic generation, and operational constraints. The model is trained on a synthetic but realistic dataset for a medium-scale industrial facility with an 80-kWp rooftop photovoltaic system. Based on these three objectives, a greedy multi-objective scheduling algorithm is then developed for minimizing peak-hour operation (to avoid this), maximizing use of available solar generation, while balancing shift operations as evenly as possible. The simulation results show that with the proposed approach, annual electricity costs can be reduced by 11.7%, the amount of time operating during peak hours could be cut down by 95.6%, and usage would increase from currently using about 42.3% to approximately 70%, while total operational hours or energy consumption remain constant. The findings indicate that effective industrial demand-side management can be attained by employing machine learning techniques that are straightforward and interpretable, in conjunction with domain-specific scheduling logic.