Fusion of CNN and Transformer for Line Loss Calculation and Sensitive Factor Analysis in Distribution Networks with Multi-Source Data

Calculating line loss in distribution networks is hard due to mixed data types and complex operating conditions. This study proposes a new method that combines convolutional neural networks (CNN) and Transformer models. The goal is to improve the accuracy of line loss prediction and support better management. The method is tested on a 10 kV distribution network in an industrial park. Key factors like load current, temperature, power factor, and transformer load rate are studied. A two-path model is designed. One path uses CNN to capture local features. The other uses Transformer to learn long-term dependencies. The two paths work in parallel and their results are combined. The model can effectively learn time and space patterns from multi-source data. Results show that it gives more accurate predictions.