Fault diagnosis and removal for hybrid power generation systems based on an ensemble deep learning diagnostic method with self-healing strategies

Abstract

An ensemble deep learning-based diagnostic method with self-healing strategies is proposed to realize the fault diagnosis and removal for a hybrid power generation system. The hybrid power generation system is composed of ammonia-based hydrogen sources (AHSs), proton exchange membrane fuel cells (PEMFCs), and ammonia-hydrogen fueled internal combustion engines (AHICEs). The deep learning diagnostic method with self-healing strategies is first employed to implement fault diagnosis under various operation loads, which combines convolutional neural networks (CNN) and bidirectional long short-term memory networks (BiLSTM). Based on the CNN-BiLSTM method, the average diagnostic accuracy reaches 98.62% at 100% load, and 99.8% at 20% load, 99.54% at 40% load, 99.32% at 50% load, 98.96% at 60% load, and 98.81% at 80% load, respectively. In comparison with other methods such as the support vector machine (SVM), back propagation neural network (BPNN), least squares support vector machine (LSSVM), CNN, gated recurrent unit (GRU), and LSTM method, the CNN-BiLSTM method achieves the highest accuracy of 99.16%. After the fault diagnosis is implemented, these faults are removed based on the self-healing strategy, where the removal rate exceeds over 99% for both single and multiple faults, demonstrating that the robustness of the self-healing strategy under various fault scenarios.