| The proper functioning of industrial equipment is the basis for safeguarding industrial processes.Diagnosing industrial equipment through effective anomaly detection technology can improve the detection efficiency of employees.More importantly,it can avoid safety accidents caused by accidental equipment damage.Because the industrial process is complex,involves a lot of equipment,and has a large number of sensors for real-time monitoring,it is very difficult to conduct comprehensive inspections only by workers according to empirical rules.Thanks to the continuous improvement of computer and abnormal detection technology,the traditional manual on-site supervision has gradually become a large-scale online operation and maintenance to monitor the industrial data generated by the sensor in real time.In order to improve the effect of equipment anomaly detection,this paper optimises the network model and automatic threshold selection for anomaly detection respectively.,and implements an anomaly detection system.The work of this paper is divided into four elements as follows:(1)By analyzing the characteristics of sensor data,an unsupervised anomaly detection model based on BiGRU-Attention network is constructed.First,the two directions of information are learned and complemented using the BiGRU network.Second,the equipment may cause fluctuations and noises in the operating data due to external or internal factors.By introducing attention mechanism,mining feature dimension information,and assigning more attention to important feature variables to highlight the impact of key features on prediction results.Two actual industrial monitoring data sets are selected,and the comparative experimental analysis verifies that the model is better than other comparative experimental models in anomaly detection.(2)An anomaly detection method based on PSO bigru attention model is proposed to optimize the bigru attention model and solve the problem of difficult model parameter adjustment.Particle swarm optimization algorithm is used to automatically optimize the number of neurons and other parameters of bigru attention model,so as to save manual parameter adjustment time and obtain more optimal network parameters as much as possible.(3)A method for automatic threshold selection for anomaly detection based on the LOF-POT model is proposed.By introducing the LOF algorithm and the POT model,combined with the proximity method and the statistical method,the abnormal score is comprehensively considered,and the two are combined as the final abnormal judgment result.Experiments show that the anomaly detection results of the automatic threshold method are better than the fixed threshold,and have good applicability.(4)Design and implement anomaly detection system for industrial equipment.Carry out demand analysis and system implementation for equipment anomaly detection,and the anomaly detection model in this paper is applied to the industrial equipment anomaly detection system.The system not only supports uploading and querying sensor data,but also supports users to create different anomaly detection requirements according to anomaly detection requirements.The network and abnormal threshold judgment method realizes functions such as management and analysis of industrial equipment sensor data.. |
