Optimizing energy management in reconfigurable distribution networks: Integrating Hybrid Transformer-CNN with wind turbines and electric vehicles

A primary principle of modern distribution networks is to maintain the voltage profile within permissible limits while minimizing power losses. However, challenges arising from the uncertainty of electric vehicles (EVs) and wind power plants necessitate proper distribution network reconfiguration to maintain stability. Therefore, operators require fast and reliable tools for managing network configuration. In this research, a data-driven framework based on a Hybrid Transformer-CNN (HT-CNN) model is presented for network reconfiguration. In this framework, wind speed is first predicted using a TCN-BiGRU model. Then, the main HT-CNN model uses this prediction, along with load data and EV charging/discharging profiles, to simultaneously estimate bus voltages, switch statuses, and network losses. The proposed model, trained with wind, load, and EV charging data on the IEEE 33-bus system, reduced the loss error to 1.46 kW and the voltage error to 0.00417 per-unit, while predicting switch statuses with 94.96% accuracy. These errors are significantly lower than the standard 5% margin in network planning, confirming the model’s high reliability for identifying optimal configurations with minimal losses. In this study, EV charging and discharging data were simulated in MATLAB. This data, along with predicted wind data, was used to solve a linearized network reconfiguration problem in GAMS. Subsequently, the GAMS output was used to train and test the neural network model in Python. Finally, the accuracy and results of the methods were analyzed and validated using DIgSILENT software.