Fault Prediction Algorithm Research Based On Particle Filter

Fault prediction has practical value in timely maintenance of equipment, guaranteeing the reliability and safety of equipment Modern system equipment structures are complex, and are mostly non-linear systems influenced by the on-site noise. In order to ensure non-linear systems’safety and reliability, it has great significance to study the fault prediction of nonlinear systems. Particle filter algorithm is an efficient algorithm to solve the problem of nonlinear non-Gaussian system failure prediction, which an implementation method is based on Bayesian filter. The core idea of this paper is:use the particle filter algorithm to estimate the system states of the current moment after a period time, namely the predictive value sequence. Then use the time series similarity measure method to match the predictive value sequence and real observable value sequence of system. Using their similarities as the basis of fault prediction to set up the fault prediction model to monitor the system operation and to predict the potential fault or the fault will occur.The main contributions of this paper include:(1) In order to improve the accuracy of fault prediction, this paper puts forward an improved cosine similarity measure formula and the fault prediction method based on the similarity. From the analysis of the new angle of the similarities between the time-series data of normal work equipment and abnormal data caused by potential faults, it predicts the potential faults. Compared to the traditional cosine similarity measure formula, the improved cosine similarity measure formula in this paper can integrated measure their similarities from the direction and length these two angles for two isometric sequences, thus it makes up the disadvantage of cosine similarity formula that measures the similarities of two sequences only from direction. Compared to the traditional fault prediction method, the experimental results verify the feasibility of this method, and the method in this paper can predict the system fault more timely and accurately.(2) In order to consider the existing unequal length situation of the predictive value sequence and the observable value sequence, this paper introduces the Dynamic Time Warping (DTW) technology for different lengths of time series similarity measure, and proposes the particle-filter fault prediction algorithm based on DTW matching. Every once in a while, it can obtain the predictive value sequence by using the particle filter algorithm, and then predict the potential fault by comparing the similarities between the Dynamic Time Warping (DTW) algorithm and the true value sequence(two sequences lengths can be different). The experimental results show that this method can deal with the unequal length situation of the predictive value sequence and the observable value sequence, and thus improve the applicability of the fault prediction method based on particle filter.(3) Design a real-time online analysis and fault prediction software system. This system has equipment fault monitoring, fault early warning and other functions. The user can use this system to real-time monitor the faults. When the system discovers the equipment data occurring exceptions, the system gives a warning forecast and alarms, thus predict the faults in advance.