Research On Equipment Anomaly Detection Method Based On Deep Learning

With the continuous development of information science and technology in the world today,China has entered the era of Industry 4.0.Today’s industrial fields have a large number of equipments operating together to produce products that meet people’s daily needs.In modern industry,process safety has become a key issue of increasing concern.With the widespread use of IoT technology in industrial production,a large number of sensor devices are deployed in industrial production equipment.These sensor devices will periodically collect the state values of the devices in operation according to established rules,and collect the device values into the data center,in.Since large-scale,high-speed sensing data streams hide important information that can reflect the operating conditions of the equipment,this provides an important data premise for modern stateful maintenance to identify potential failures of the equipment.However,due to the large-scale and complicated modern large-scale equipment,some disturbances during the operation of the equipment are not transmitted to the cooperative equipment,which makes the equipment relationship complex and variable,and it is difficult to use a unified model to indicate the running state of the equipment.In recent years,with the continuous development and advancement of deep learning technology,deep learning technology has become the mainstream solution for fault detection.This paper presents an online fault detection model based on long and short memory neural networks.Since the interaction between power generation devices is transitive,there is often a delay-related phenomenon between the sensor data.This phenomenon reduces the effect of feature extraction.Therefore,the curve alignment method is used to extract the sensor data,and then based on the Long-Short Time Memory Neural Network(LSTM)training fault detection model.However,in the actual industrial production environment,the equipment is affected by many factors(external environment,self-degradation,etc.),and the sensor data generated by it is also time-varying.For the change of the operating state of the power plant equipment,the model also needs time.The change of adapting to the state of the device finally realized the online detection of the device failure and the online update of the model by means of the sliding window technology.This paper is based on real power plant sensing data for experimental verification.Experimental results show the effectiveness of the proposed method.