Multi-sensor Fault Detection for Natural Gas Combined Cycle Power Plants Based on Multiple Robust Input Training Neural Network Models

Abstract

In order to enhance the accuracy and reliability of multiple sensor measurements in natural gas combined cycle (NGCC) power plants, a multi-sensor fault detection method based on multi-model robust input training neural network is proposed in this paper, in which multiple robust input training neural network (RITNN) models are built and prioritized for the purpose of sensor faults reconstruction and monitoring. The relationship between the models are set in terms of serial or parallel connections. The influence of numerous failure data with significant errors can be effectively inhibited by virtue of reliable sensor data calculated from cooperative multi-model, such that the accuracy and reliability of fault detection is greatly improved. In addition, the process for sensor fault detection is presented to establish a complete fault dectection system. The proposed method was evaluated in a 200 MW NGCC power plant, where the multi-sensor fault detection was conducted and the results were compared with those from single RITNN model and single input training neural network (ITNN) model detection.The proposed method demonstrated higher accuracy in multiple sensor failure cases than the single RITNN model or the single ITNN model.

Keywords

• multiple models
• ritnn
• fault detection
• ngcc