Stochastic designing of a hybrid system with hydrogen energy management based-fuel cell using machine learning and improved arithmetic optimization algorithm for building electrification

This paper presents a new stochastic design structure for a hybrid energy system that includes photovoltaic (PV), wind turbine (WT), and fuel cell (FC) with hydrogen storage (PV/WT/FC) to meet a building's electrical demand, addressing the challenges posed by uncertainties in renewable power and building load. To overcome these challenges, a two-point estimate method (2m + 1 PEM) is used to model the uncertainties. The 2m + 1 PEM is considered over other methods, such as Monte Carlo simulations, due to fewer representative data points, making it more computationally efficient. The Improved Arithmetic Optimization Algorithm (IAOA), which incorporates Rosenbrock's direct rotational method, is employed to minimize the system's net present cost (SNPC) while maintaining the high reliability as probability of demand not-supplied (PDNS). The IAOA is selected due to its enhanced ability to avoid premature convergence of traditional AOA. Additionally, machine learning is applied through a multilayer perceptron artificial neural network (MPANN) to forecast meteorological data such as irradiance, wind speed, and temperature, and also building demand, providing accurate inputs for the system design. The MPANN is applied because of its superior ability to capture non-linear relationships and forecast time-series data more accurately compared to simpler models. The simulation results show that the hybrid PV/WT/FC combination, offers the lowest SNPC and the highest reliability, compared to other system configurations. The deterministic results indicated a 23.07 % decrease in reliability and a 17.86 % increase in SNPC for a maximum PDNS of 5 % when using forecasted data compared to the real data-based design. Additionally, the stochastic results indicated a 5.54 % increase in SNPC and a 12.50 % reduction in reliability due to the incorporation of system uncertainties. The IAOA's effectiveness is demonstrated through comparison with well-known algorithms, showing better convergence speed and solution quality. This research provides an efficient solution for designing hybrid energy systems, and addressing uncertainties.