A Novel Binary Classification Neural Network Optimized by the Mosquito Mating Swarm Optimization Algorithm for Predicting Microgrid Operational Modes

Integrating renewable energy sources presents technical challenges due to their variable nature, particularly in predicting and managing microgrid operational modes. Accurate identification of grid states—interconnected or islanded—is essential for maintaining stability and optimizing performance under fluctuating environmental conditions to meet energy demand. This work proposes a bio-inspired, optimized binary classification model based on Multi-Layer Perceptron Artificial Neural Networks (MLP-ANN), with the architecture and hyperparameters tuned using the novel Mosquito Mating Swarm Optimization (MMSO) algorithm, inspired by mosquito mating behavior and swarm dynamics. The model employs an MLP-ANN with a variable number of hidden layers and neurons per layer, configured to maximize classification accuracy by dynamically adjusting parameters, including the learning rate and regularization coefficients. Training utilizes k-fold cross-validation on experimental microgrid data. The MMSO approach is benchmarked against Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Grey Wolf Optimizer (GWO) to validate its effectiveness. Results show that the MMSO-optimized MLP-ANN achieved an 86.34% recall, 98.96% precision, and 92.29% accuracy, while minimizing the Mean Squared Error to 0.0206. The MMSO-optimized MLP-ANN model achieved competitive classification performance compared to the other algorithms evaluated; although no statistically significant differences in recall were observed among the optimizers (p = 0.22), the MMSO achieved the lowest MSE (0.0206). The MMSO was the only algorithm capable of discovering a four-layer architecture hidden within the same search space, evidencing superior exploration of deeper architectural regions of the solution space. These findings demonstrate the model’s capacity to predict microgrid operational modes under variable conditions, highlighting the potential of integrating bio-inspired algorithms with neural networks for energy management systems. This approach could enhance the efficiency and reliability of integrating renewable energy sources into dynamic energy systems.