Abstract:
Fuzzy C means (FCM) clustering is a conventional clustering method. As an unsupervised learning method, FCM can make full use of historical data or real-time data, and detect and diagnose faults in process by establishing fuzzy similarity relation. However, in dealing with the industrial data, the clustering performance of FCM is lower due to strong noise and nolinear data. Sewage treatment process is a complex nonlinear industrial process and it is operated difficultly in long-term and stable operation. Therefore, it is quite necessary for monitoring sewage treatment process, detecting operational failures, and dealing with faults in time. This paper presents a fault detection method in sewage treatment process by combining desity weighted, kernel theory and possibility fuzzy C means (PFCM) clustering. In the proposed algorithm, the kernel function is utilized to map data into linear space for clustering analysis and eliminates the nonlinear influence. By introducing the concept of point density, the proposed algorithm iteration can be accelerated and the clustering accuracy is improved. Meanwhile, the possibility fuzzy C mean(PFCM) clustering is also maintained for the outlier robustness and the amount of calculation is reduced by introducing the sample variance. The simulation experiments are made via Benchmark Simulink Model-1(BSM1), in which there exist two types of fault:process fault and sensor fault and 13 effluent substances in water from the fifth biochemical reaction pool are used as raw data. The simulation runs for fourteen days and only the data in the last seven days are selected as experimental data. By using different categories of fault data clustering analysis and comparing with the fuzzy C means (FCM) clustering, the possibility fuzzy C mean (PFCM) clustering and the kernel possibility fuzzy C means (KPFCM) clustering algorithm, the proposed detection method can not only deal with the nonlinear problem, but also accelerate the convergence speed effectively.