非线性过程控制系统传感器故障检测新方法

针对电厂热力系统故障检测和定位准确性低的问题,提出了基于鲁棒输入训练网络(Robust Input-training Network, RITN)的传感器故障检测模型。采用带参数限制项的目标函数对网络进行训练,并在测试目标函数中引入影响因子,增加了模型训练精度,抑制了网络计算过程故障数据对正常值的影响,减小了残差污染,提高了模型准确性。以某300MW电厂热力系统20组测点为对象进行算例分析,通过反复的实验,结果表明,该模型能够更加准确的对非线性系统故障点进行检测和分离,并更加精确重构各变量真实值,验证了该模型用于非线性过程传感器故障检测的有效性和可靠性。

A New Sensor Fault Detection Method For Nonlinear Process Control System

For the problem of low accuracy of fault detection and location in the thermodynamic system of power plant, sensor fault detection model of nonlinear system based on robust input-training network is proposed. The objective function with parameters restriction term is used in the training process for avoiding the weights adjusting excessively and the influence factors are introduced into the objective function, which increases the accuracy of model training and the effects of network fault data for the normal are suppressed, induces the residual contaminations and increases the accuracy of sensor fault detection model. A case study with single-point fault and multi-point fault test is conducted to detect 20 points from the thermodynamic system in a 300MW unit, Through repeated experiments, the results show that, the model can detect and isolate the nonlinear system failure points and reconstruct the true values more accurately, The validity and reliability of the model for nonlinear process sensor fault detection is verified.