基于机器学习代理的手动与自动开关在配电网中的优化配置

Fault management involves actions to restore interrupted customers as quickly as possible after a fault occurrence, which is facilitated by optimal switch placement. However, the switch optimization problem requires significant computational effort because of the nonlinearity and the large search space. Hence, the problem may be interactable for large-scale power distribution systems. This paper proposes a machine learning-based proxy to determine the optimal number and location of switches in real-world power distribution systems, including manual and automatic switches. The objective function comprises equipment costs and reliability indices, including the system average interruption frequency index (SAIFI), the system average interruption duration index (SAIDI), and the energy not supplied (ENS) index. The proposed model is a stacked ensemble model, in which convolutional neural network (CNN)-based models serve as base learners to capture complex and spatial information from their input data. The input data for the base learners are optimally selected by explainable artificial intelligence (XAI) tools from a large feature set of candidate installation points. The metamodel combines the base learners' predictions to determine the optimal points for installing switches and is a fully connected artificial neural network. This paper uses the Bayesian optimization algorithm to optimally construct the stacked ensemble model. We validate the proposed model alongside state-of-the-art learning-based and mathematics-based models using a real power distribution system. The simulation results demonstrate that the proposed model outperforms the state-of-the-art models in terms of computational complexity and optimality.