Fault prediction of hybrid scraper based on optimized LSSVM-HMM

The hybrid power scraper has a harsh working environment, complex electrical system, strong coupling of fault causes, many types of faults, and mostly nonlinear data.Aiming at the problem that the traditional single method is difficult to accurately predict the fault of electric system of the scraper, a fault prediction method combining a least square support vector machine (LSSVM) and a hidden Markov model (HMM) is proposed and improved.Initially, the state data of the scraper at historical time is trained by LSSVM, the state data at current time is input into the trained LSSVM to predict the state data at future time, and then HMM models under different fault states are trained by historical data.Subsequently, the current state data and the state data predicted by LSSVM are imported into the trained HMM model to predict the state and its change trend of the scraper in the future.In view of the traditional empirical method to train LSSVM parameters and Baum-Welch method to select the HMM parameters, which are easy to fall into the shortcomings of the local optimal solution and the slow convergence speed, it is proposed to use Artificial Fish Swarm Algorithm (AFSA) in the LSSVM medium and HMM parameters to improve.It improves the parameter estimation performance of LSSVM and HMM, and obtains the optimal penalty parameter and radical basis function required by LSSVM.The data in the whole process is collected from a 14-ton hybrid scraper working in the mine site.The research results show that compared with the collected real state data, the state data of the scraper predicted by LSSVM has smaller error and higher coincidence.The accuracy of fault diagnosis of the optimized LSSVM-HMM method is 91.1%.This method can accurately predict the fault of the electric system of the hybrid electric scraper and its state changing trends.