Optimizing green port operations with reinforcement learning and model predictive control

While playing a crucial role in global trade, seaports are polluting hubs. To mitigate greenhouse gas emissions from ships at berth, the present study incorporates a combination of solutions to meet ships’ demand, including shore-side electricity (Cold ironing), multi-energy storage, renewable generation, and artificial intelligence techniques for forecasting and optimizing renewable energy usage. Key contributions include the sizing of a port based on real ship traffic data and area from the ports of Genova and Trapani, the comparison of reinforcement learning and model predictive control for effective energy management, and the development of artificial intelligence-driven forecasting models for building demand, and photovoltaic and wind generation. Long Short-Term Memory and Time Delay Neural Networks are used to predict renewable generation and demand, while sensitivity analyses demonstrate the flexibility and efficiency of the Soft Actor Critic algorithm in handling forecast uncertainties. The hyperparameters of the artificial intelligence models are optimized on the data of the port of Genova, and tested on both the ports. Results highlight the strength of the developed control algorithms to achieve diesel usage reduction at berth up to 100%. Numerical findings prove the model predictive control ability to extract knowledge from input data and demonstrate the advantages of reinforcement learning in optimizing energy management under limited and uncertain information. The model predictive control and the Soft Actor Critic algorithms were able to achieve 95.9% and 94.4%, respectively, of the maximum achievable cash flow in the port of Genova, and 91.3% and 89.3%, respectively, in the port of Trapani. • Comparison of RL and MPC for managing smart port operations integrating RES. • Evaluation of the benefits of cold ironing for reducing greenhouse gas emissions. • Development of AI forecasting algorithms for RES and building loads. • Integrating AI predictions improves management and the economic outcome. • RL performs better in managing energy and handling uncertainties in energy forecasts.