基于深度自编码器的核动力历史异常数据检测技术研究

Research on Historical Anomaly Data Detection Technology for Nuclear Power Plant Based on Deep Auto-Encoder

为解决核动力历史异常数据检测中存在的新的异常识别困难问题,基于重构误差的思路,提出基于深度自编码器的历史数据异常检测模型。该模型以某稳态运行工况下正常历史数据为学习对象,通过最小化正常数据重构误差目标训练模型,根据待测数据重构误差大小判断其是否异常。研究结果表明,深度自编码器对正常数据重构能力较好,对异常数据重构能力不足。因此,通过比较重构误差大小,深度自编码器能够有效检测出核动力历史异常数据,其性能优于一类支持向量机,可以为核动力装置状态评估提供相关依据。

In order to solve the problem of new anomaly identification difficulties in the detection of nuclear power historical anomaly data,according to the idea of reconstruction error,an anomaly detection model based on deep auto-encoder is proposed.The model takes the normal historical data under steady-state operating condition as the learning object,trains itself by minimizing the reconstruction error of the normal data,and judges whether the unknown data is abnormal according to the size of the reconstruction error.The research results show that the deep autoencoder has better ability to reconstruct normal data but insufficient ability to reconstruct abnormal data.Thus,by comparing the reconstruction error size,the deep autoencoder can effectively detect the historical abnormal data of nuclear power plant,and its performance is better than that of one-class support vector machine,which can provide relevant basis for the state evaluation of nuclear power plants.