基于iForest-Adaboost的核电厂一回路故障诊断技术研究

传统的故障诊断方法如主元分析方法与BP神经网络等在复杂非线性系统中存在泛化能力较差、故障识别准确度较低的问题。而孤立森林(iForest)算法使用孤立树划分思想识别异常数据,可适用于非线性系统的状态监测;Adaboost算法是一种基于组合分类思想的提升算法,可通过多个弱分类器的叠加,使整体算法具有较好的泛化能力。因此采用i Forest算法与Adaboost算法建立iForest-Adaboost核电厂一回路故障诊断系统,使用GSE实时仿真平台与福清核电站一号机组仿真数据测试。测试结果表明,iForest算法相比于主元分析与QTA阈值法可以更快识别出系统异常,Adaboost算法相比于BP神经网络与支持向量机方法具有更高的故障识别准确率。

Research on Fault Diagnostic Technology of Primary Loop of Nuclear Power Plant Based on iForest-Adaboost

The traditional fault diagnostic methods such as principal component analysis and BP neural network are with poor generalization capability and low fault identification accuracy in complex nonlinear systems. The isolation forest (iForest) algorithm uses the idea of isolation tree partition to identify the abnormal data, which is applicable to the state monitoring of nonlinear systems. The Adaboost algorithm is a boosting algorithm based on the idea of combined classification, and the overall algorithm has better generalization capability through the superposition of multiple weak classifiers. Therefore, the isolation forest algorithm and Adaboost algorithm are used in this paper to establish the iForest-Adaboost primary loop fault diagnostic system for nuclear power plants, and the GSE real-time simulation platform and the simulation data in unit 1 of Fuqing Nuclear Power plant are used for the test. The test results show that the isolation forest algorithm can identify the system anomalies faster than the principal component analysis and QTA threshold algorithm. The Adaboost algorithm has a higher fault identification accuracy than BP neural network and support vector machine algorithm.