Fault Prediction Research Based On The Improved Extreme Learning Machine For Complex Industrial Process

With the rapid development of modern industrial production technology, the production requirements will become more and more stringent. Large scale, complexity, intelligent and digital are the characteristics of the modern production system. Because of the devices are interconnected with each other, and have high coupling degree, which makes the fault features of the complex system become uncertain and nonlinear. In order to ensure the safety of the complex industrial system, fault prediction become a necessary measure in recent years. Fault prediction is a more advanced security technology than fault diagnosis, which has gradually become a research focus recently. Therefore, we are aiming at the complex system characteristic of data processing in this paper, The mainly research about feature extraction, neural network are fault prediction are as follows:(1) According to the characteristics of time series data, this paper focuses on the changes of actual production process variables in the system. Differential Vector Kernel Principal Component Analysis (DV-KPCA) is proposed for feature extraction. Due to the complex relationship of the system, the kernel principal component analysis (KPCA) method are used to solve the problem of nonlinear mapping. The differential vector plus KPCA is proposed to reduce the dimension of process data and enlarge the feature difference. Through the differential operation between the input and the common vectors, eliminating redundant information and feature extraction.(2) In view of the improvement of the dynamic network, this paper improves the structure of the neural network based on the ELM algorithm, called DR-ELM. In DR-ELM, the self-feedback layer is constructed in the hidden layer to memorize the output of the hidden layer. According to the trend characters of stored historical information, the weights of the feedback layer is dynamically updated. By increasing the dynamic performance of ELM, improve the accuracy of the model prediction is improved and mapping the performance of the dynamic characteristics.(3) In order to cope with the training data it is arrived chunk-by chunk in industrial process, the online sequential extreme learning machine(OS-ELM) is used to construct as fault prediction model. A self-feedback unit and error feedback unit from the output of hidden layer and output layer is added to improve the accuracy of fault prediction. Finally, in order to verify the validity of the proposed method, the experiment of fault prediction and identification in TE process is carried out. The experimental results show that the method can achieve a higher accuracy of fault prediction.The research results show that the feature extraction method based on KPCA for time series data can effectively improve the classifier effect. The feedback layer in the neural network can compute the trend change information, enhance the success rate of fault prediction accuracy, which provide a new ideas for the fault prediction.