Research On Soft Fault Detection And Diagnosis Of Temperature Sensor In An Environmental Control System At A Subway Station

The environmental control system of subway stations aims to maintain a good subway operating environment and the waiting and working environment inside the station,but its energy consumption is high and there are a large number of sensors involved in its regulation.Once the sensor fails,the environmental control system will be difficult to efficiently ensure that the environment in subway tunnel and stations meets the requirements,and may lead to increased energy consumption.Therefore,it is necessary to strengthen the research on the detection and diagnosis methods of sensor faults in the environmental control system of subway stations.Due to the difficulty in obtaining monitoring data for the operation of the subway station environmental control system,this article uses TRNSYS software to build a simulation model of the subway station environmental control system based on relevant national standards and literature materials,and verifies the reliability of its simulation results.Furthermore,through simulation,the operation data of the subway environmental control system under normal and fault states of the sensors are obtained,A study was conducted on the detection and diagnosis of single and concurrent faults in the temperature sensors(fresh air,return air,and supply air temperature sensors)of the air handling unit in the subway environmental control system,involving types of faults such as deviation,drift,and decreased accuracy.Usually,a classification machine learning algorithm called Support Vector Data Description(SVDD)is applied for fault detection,but SVDD can only detect sensor faults that exist at the current moment,but cannot locate the faulty sensor.Therefore,based on SVDD,this article combines a mixed weighting strategy of distance and density with the idea of contribution graph,and proposes a support vector data description based on mixed weights(MW-SVDD),which improves the accuracy of fault detection models and fault detection accuracy,reduces the false alarm rate(FAR)of the model,and realizes the localization of fault sensors.Under single sensor failure,the average detection accuracy of the MW-SVDD model for various types of faults in fresh air,return air,and supply air sensors reached 86.3%,95.6%,and 93.8%,respectively,and all of them were able to locate the faulty sensor;In addition,the detection accuracy of the MW-SVDD model in the case of concurrent faults of various sensors is higher than that in the case of single sensor fault.However,for the case of dual sensor concurrent faults with significant differences in severity,it can only locate the sensor location with the highest degree of fault.In the aspect of fault diagnosis,after the MW-SVDD model is used to complete the detection and location of the faulty sensor,this paper uses the regression prediction ability of the short-term memory neural network(LSTM)to establish LSTM prediction models for different types of sensor faults,and proposes corresponding expert rules for sensor fault diagnosis according to the mathematical model of sensor soft faults.Compare the residual curve between the predicted and actual values of the LSTM prediction model with the expert rules for sensor fault diagnosis to determine the fault type of the target sensor.Both the LSTM prediction model and sensor fault diagnosis expert rules can identify the fault type of the target sensor,whether in the case of a single sensor fault or a concurrent sensor fault.Meanwhile,the LSTM prediction model provides the possibility for fault-tolerant operation of the subway station environmental control system.