Energy communities (ECs) have emerged as key instruments for enabling local decarbonization and reducing energy costs through collective self-consumption and the integration of distributed energy resources (DERs). The optimal sizing of DERs such as photovoltaic (PV) panels and battery energy storage systems (BESS) is essential to maximize the economic benefits for ECs. However, when formulated as a mixed-integer linear programming (MILP) problem, the sizing of these DERs poses significant computational challenges, particularly as the number of EC members and the planning horizon grow.

Metaheuristic approaches offer a promising alternative by trading marginal solution quality for improved computational efficiency. Previous work has shown that Evolutionary Particle Swarm Optimization (EPSO) incorporates self-adaptive weights and evolutionary mechanisms such as mutation and selection, which can achieve solutions within a small percentage of the MILP optimum for EC sizing problems. However, the original EPSO implementation exhibited premature convergence and limited constraint handling, hindering its application to longer planning horizons and larger ECs.

This paper proposes an enhanced EPSO framework with several improvements designed to address these limitations. Dynamic scheduling of the inertia and cognitive/social parameters is introduced to progressively shift from exploration to exploitation. Adaptive penalty mechanisms adjust constraint violation weights based on population feasibility, while boundary control with bounce-back strategies ensures solution viability. Genetic algorithm operators, namely crossover and mutation, are integrated to maintain population diversity, and simulated annealing acceptance criteria help escape local optima. An elitist restart mechanism reinitializes stagnating particles near the best-known solution.

Furthermore, a hybrid MILP-EPSO methodology is proposed, where the MILP solver is first applied to a reduced time horizon to obtain a high-quality seed solution. This solution is then used as a warm-start for the EPSO population over the full planning horizon, combining the accuracy of deterministic optimization with the scalability of metaheuristic search. The approach is validated on a realistic case study comprising an EC with multiple consumers where PV systems and BESS are available for installation, using real energy consumption and production profiles. Numerical results demonstrate that the proposed hybrid approach achieves solutions with a cost gap below 1% relative to the MILP benchmark, while offering improved computational scalability for extended planning horizons.