Purpose Anomaly detection is the process of identifying behaviors or events that do not meet expectations in data.High-power pulse energy systems are crucial technologies in numerous large-scale scientific devices.The...Purpose Anomaly detection is the process of identifying behaviors or events that do not meet expectations in data.High-power pulse energy systems are crucial technologies in numerous large-scale scientific devices.These systems utilize capacitors as energy storage units and can assess the condition of energy modules by analyzing discharge waveforms.Methods The range of values for different types of discharge waveforms varies significantly,and their distribution shows notable deviations.In this paper,we propose a long short-term memory-based encoder–decoder(LSTM-ED)framework for waveform anomaly detection,which effectively addresses the aforementioned issues.Results and conclusion Our method is divided into two stages,utilizing the mapping and recognition capabilities of the LSTM-ED model during the prediction stage.In the detection stage,we identify anomalies by setting different statistical thresholds for different circuits and types of discharge waveforms.Finally,a case study was conducted using real-time monitoring data from the energy module to validate the effectiveness of the proposed method.The results demonstrated that our approach can effectively identify anomalies across different types of discharge waveforms.展开更多
文摘Purpose Anomaly detection is the process of identifying behaviors or events that do not meet expectations in data.High-power pulse energy systems are crucial technologies in numerous large-scale scientific devices.These systems utilize capacitors as energy storage units and can assess the condition of energy modules by analyzing discharge waveforms.Methods The range of values for different types of discharge waveforms varies significantly,and their distribution shows notable deviations.In this paper,we propose a long short-term memory-based encoder–decoder(LSTM-ED)framework for waveform anomaly detection,which effectively addresses the aforementioned issues.Results and conclusion Our method is divided into two stages,utilizing the mapping and recognition capabilities of the LSTM-ED model during the prediction stage.In the detection stage,we identify anomalies by setting different statistical thresholds for different circuits and types of discharge waveforms.Finally,a case study was conducted using real-time monitoring data from the energy module to validate the effectiveness of the proposed method.The results demonstrated that our approach can effectively identify anomalies across different types of discharge waveforms.
