Research And Application Of Anomaly Detection Method For Time Series Data Of Industrial Equipment

Equipment safety is directly related to the smooth operation of economic construction,the safety of people’s lives and property,and social stability.This thesis aims at the complex problems of quality and safety management caused by the large-scale,variety,large-quantity,and high-parameter conditions of industrial equipment.Analyze the monitoring data of industrial equipment,establish an intelligent detection model for industrial equipment abnormalities and health status,detect data abnormalities,and diagnose the health and abnormal probability of the equipment to ensure the stable and safe operation of industrial equipment.This thesis takes industrial equipment time series data as the research object.Aiming at the problems of traditional industrial equipment anomaly detection methods such as weak versatility,accuracy and low data utilization,a model of industrial equipment time series data prediction and anomaly detection including the prediction model EGA and the anomaly detection model PIS(the latter depends on the prediction results of the former)is proposed to improve The prediction accuracy and anomaly detection accuracy of industrial sensor data,and apply the model to the national key research and development projects participating in the research and development.The main contents of this thesis are as follows:(1)Generalize and summarize the algorithm models of time series data prediction and anomaly detection at home and abroad,and analyzing the problems of insufficient generalization ability and low accuracy of traditional prediction models,as well as incomplete anomaly classification and low accuracy of traditional anomaly detection models.Analyze the characteristics of time series data and anomaly classification methods,and propose a prediction-based time series data anomaly detection algorithm to improve the existing problems of the current model.(2)Propose an industrial equipment time series data prediction model EGA(EWT-GRU-Attention).Aiming at the problem of poor versatility in traditional time series prediction methods,the EWT empirical wavelet decomposition method is used to decompose the time series data.The attention mechanism optimizes the GRU model in order to improve the prediction accuracy of the recurrent neural network.By training and adjusting the parameters of the EGA prediction model in the real data set,comparative experiments verify the effectiveness and accuracy of the combined prediction model.(3)Aiming at the shortcomings of the existing methods,a PIS(Predict-Isolation Forest-SAX)detection model for time series data anomaly of industrial equipment is proposed,which divides time series data anomalies into single-point abnormal values,fluctuating abnormal points,and abnormal sequences.The residual sequence is obtained using the prediction results of the EGA prediction model and the real observations,and it were analyzed in combination with the isolated forest and the improved SAX character aggregation representation to detect different types of anomalies in the original time series sequence.Experiments on the PIS model on the real data set show that the model has a good anomaly detection effect.(4)According to the actual scenarios of the typical industrial equipment detection and monitoring cloud platform,apply the above-mentioned prediction-based time series data anomaly detection model to the research and development project of industrial equipment detection and monitoring cloud service platform.Using relevant software research and development technology,combined with software engineering practice methods,design and implement industrial equipment time series data prediction and anomaly detection subsystems.