冷水机组故障特征优化选择

Feature-optimizing selection for chiller fault detection and diagnosis

从现场应用的角度,针对冷水机组典型故障,提出了一种特征选择(FS)的方法,选择少量获取成本低的特征表征故障,达到以最低成本的传感器投入获得最优的故障检测与诊断(FDD)性能,从而节省FDD成本。首先,在现场应用的约束下,对64个原始特征进行特征初选,选择出传感器成本低和对故障敏感程度高的16个特征;然后,基于互信息的FS模型对这16特征进行特征中选,确定故障指示特征的最佳个数;最后,基于灰色聚类分析的FS模型再对这16特征进行特征终选,确定具体的特征种类。使用ASHRAE RP-1043故障实验数据和基于支持向量机的FDD工具验证了提出FS方法的有效性。

From the perspective of field applications, a feature selection(FS) method was proposed for chiller fault detection and diagnosis(FDD). The purpose was to obtain an optimal diagnostic performance and to save initial FDD costs. The technological paths were as follows: first, under the constraint of field applications, the original 64 features were selected preliminarily, 16 features obtained by low-cost sensors were selected;second, these 16 features were selected medially based on mutual information, in order to determine the optimal number of features describing the typical chiller faults;last, the specific feature subsets were determined based on grey clustering analysis. The experimental data from the ASHRAE RP-1043 and the FDD tool based on support vector machine were used to evaluate the proposed FS method.