Information Fusion Model Based Improved Multi-Scale Convolutional Neural Network for Fault Diagnosis in EV V2G Charging Pile

With the rapid proliferation of electric vehicles, the demand for charging infrastructure is surging. Consequently, it becomes imperative to address the issue of maintenance for charging pile faults. To harness the multi-scale characteristic present in fault signals more effectively, this paper proposed an information fusion diagnosis model based on an improved multi-scale convolutional neural network for the automated diagnosis of switch tubes open-circuit in charging piles. The approach conducts feature extraction at different scales on the fusion information from the original fault signals through a parallel branching structure. Furthermore, a channel attention mechanism is incorporated in the final layer of the network branch. This mechanism gives precedence to different channel features, dynamically assigning weights to the features obtained from distinct filters. This enhances fault-related features, suppresses ineffectual ones, and subsequently elevates the model's performance. Ultimately, the different scale features acquired from each network branch are merged and subjected to classification and recognition using a Softmax classifier. Simulation results underscore the effectiveness of the proposed method in diagnosing switch tube open-circuit faults in the charging pile.