Optimal Operation of Microgrid Through Loss-Guided Neural Network-Based Approach

The uncertainty in microgrid systems introduces significant difficulties in their operation, such as increasing the operational costs when using conventional optimization algorithms. In this study, a Deep Neural Network (DNN)-based optimization model is proposed to minimize the operation costs of a microgrid system. The input layer of the DNN captures comprehensive system information, which is processed through hidden layers. The output layer provides optimal set-points for controllable distributed generation units, aiming to minimize operational costs while fulfilling operational constraints. A customized loss function has been developed to guide the DNN's training process without the need for any labels or datasets. Moreover, the uncertainties of demand and market price signals are considered during the training process to manage the system dynamics in real-time operations of the microgrid. Unlike the conventional backpropagation approach, this study employs Particle Swarm Optimization (PSO) to update the DNN's parameters, ensuring faster convergence and a more stable training process. Finally, the efficacy of the proposed method is verified under conditions both with and without system uncertainty.