Research Of Chemical Process Fault Diagnosis Methods Based On ESN

With the improvement of the automatic level of chemical production system, the scale and complexity of the chemical process are becoming bigger and higher. Once the fault happened, the production efficiency of chemical process would be reduced or the whole system would be forced to stop, even a loss of lives would be bore when the fault was serious. However, the technology of fault diagnosis can detect, locate and segregate the failure timely, which can reduce or avoid the loss effectively. So the research on fault diagnosis can contribute to increase the security and reliability in chemical process.Echo State Network(ESN) that has small amount of calculation and good ability of short-term memory is a new type of neural network. For the better fault detection and fault diagnosis results, some works have been done using the TE process as simulation platform to acquire data on the basis of ESN and One-Class Support Vector Machines(OCSVM). The main works that have been done in this paper as follows:It's difficult to detect the fault whose mean and/or variance have/has not obvious changes in chemical process. According to the characteristics of this type of fault, a fault feature extraction method, MCUSUM-PCA-CJESN, has been put forward in this paper. This method firstly acquires the historical cumulative information of the process data using MCUSUM to amplify the tiny changes of the process data, which can rich the information of the process data in time dimension. Then Principal Component Analysis(PCA) will be used to reduce the data dimension and computational work. At last, the fault feature will be extracted by using the Circular Jumping Echo State Network(CJESN) readout layer's ability that it can capture the difference of data characteristics in different time.Since the parameters selection has an effect on the fault detection rate and fault diagnosis rate in fault diagnosis, so Adaptive Fruit Fly Optimization Algorithm(AFOA) is proposed on the basis of Fruit Fly Optimization Algorithm(FOA), and is used to optimized parameters in fault diagnosis. AFOA can set the initial fruit fly swarm location by the reciprocal of the range(s) of the optimized variable(s), which overcomes the problem that it's difficult to determine the location.To achieve the purpose of fault detection and fault diagnosis in chemical process, MCUSUM-PCA-CJESN and One-Class Support Vector Machines(OCSVM) are combined to introduce the fault detection and diagnosis method based on CJESN-OCSVM in this paper. In order to improve the detection rate and recognition rate, the AFOA will be used to optimize the accumulated step in MCUSUM and the four key parameters. Finally, the fault detection and fault diagnosis method, CJESN-OCSVM, will be tested based on TE process to verify its feasibility and viability.