A Fault Diagnosis Method based on TCN-LSTM-SE Neural Networks for Distributed PV Systems

Distributed photovoltaic power generation grid-connected systems add the variety and complexity of faults to the grid, so the research on fault diagnosis method with high diagnostic accuracy and practicality is of great significance for its safe and stable operation. The traditional modelling, control and detection methods are limited by the construction cost and susceptible to parameter fluctuations, which are difficult to adapt to future development needs. In this paper, we propose a fault diagnosis method based on the TCN-LSTM-SE model for distributed PV grid-connected systems and use it for three-phase unbalance and harmonic voltage prediction and fault diagnosis. Firstly, the SE attention mechanism is used to highlight fault features in the input data, then the TCN is chosen to perform feature extraction while maintaining the signal timing characteristics, the LSTM network finally memorizes the key features to complete the timing modeling. Simulation results show that the method has lower prediction error and higher diagnostic accuracy than mainstream network models, and the fault location and severity can be accurately known.