Skip to Main Content
A sensor fault detection method of nonlinear system based on robust input-training network was proposed. The objective function with parameters restriction term was used in the training process for avoiding the weights adjusting excessively and meanwhile the influence factors were introduced into the objective function in the testing process for the purpose of inhibiting the influence of failure data in the network calculation, which avoided the residual contaminations and increased the accuracy of sensor fault detection and data reconstruction. The fault detection process was presented and the effectiveness analysis proved the feasibility of the model in dealing with nonlinear problems. A case study with single-point fault and multi-point fault test was conducted to detect 20 points from the thermodynamic system in a 300MW unit. The simulation results of different methods showed that the RITN model in this paper can detect fault points more accurately and reconstruct the true values, improving the anti-interference ability and verifying the accuracy and reliability of the model.
Date of Conference: 24-26 June 2011