Innovative predictive energy management and sizing in hybrid microgrids: from research to industrial application

The energy sector is increasingly adopting renewable sources like solar and wind power. While these sources are necessary for sustainability, they introduce instability due to their dependence on weather conditions. To handle this variability, industry is adopting microgrids. local networks capable of generating, storing, and managing their own electricity. The primary challenge of this new approach lies in management: operators must efficiently decide when to store energy or generate backup power, together with multiple other technical and economical constraints. This thesis addresses these challenges by developing practical systems that bridge the gap between academic research and real-world industrial applications, aiming to create predictive tools that lower costs while ensuring reliability. This research began with a concrete industrial problem at a high-power EV charging station in San Sepolcro, Italy. Facing a weak grid connection and high demand, the site required a custom management system. This initial work produced a semi-predictive logic that successfully managed basic operation but proved inefficient during unexpected demand spikes because it could not anticipate future events. To improve this, the research developed a hybrid forecasting algorithm to estimate electrical load twenty-four hours in advance. By developing a hybrid combination of statistical models (SARIMAX) together with neural networks, the system achieved high accuracy, validated against highly fluctuating data from the island of Tilos. With reliable predictions available, the study designed two advanced control strategies. The first, a Predictive Rule-Based system, uses forecasts to calculate exactly how much energy the battery must reserve for upcoming peaks as well as well as scheduling the main power generation source. This system is robust enough for standard industrial controllers. The second strategy uses Mixed-Integer Linear Programming (MILP), a mathematical optimization method that calculates the absolute optimal schedule to minimize costs. To further support decarbonization, the research investigated hydrogen-fuelled internal combustion engines as clean backup generators. Experimental testing revealed that standard models (in industry and research) assuming constant efficiency caused significant errors in cost calculations. Consequently, a detailed model was created to accurately map efficiency changes based on engine speed and load. Finally, year-long simulations compared these strategies. The results showed that while the complex optimization system (MILP) yields the best financial savings, the simpler Predictive RuleBased system offers the stability and reliability required for immediate industrial use. These developed frameworks provide a practical toolkit for making independent, green energy systems viable for everyday application.