基于纵横交叉支持向量机的制冷剂充注量故障诊断

Fault Diagnosis for Refrigerant Charge Based on Support Vector Machine with Crisscrossed Algorithm

针对离心式冷水机组的制冷剂充注量故障,本文采用了一种纵横交叉算法优化支持向量机进行故障诊断的方法,研究了一定范围内随机初始化训练支持向量机所需参数在连续迭代与更新方面的优化情况,分析了不同参数对于最后寻优结果的影响,使用Ashrae-1043-RP项目实验数据验证了该方法对于故障诊断准确率的提升。结果表明:相比划分网格寻优与遗传算法寻优,该方法有效提高了求解的精度,并避免了算法过早收敛于局部最优的情况;未优化时该种方法对于制冷剂充注过量或泄漏较为严重时,具有较高的区分度;参数优化后制冷剂故障诊断总体准确率由76.28%上升至97.52%,特别对于未优化前支持向量机不能有效区分的轻微故障情况有明显提升。

For refrigerant charge fault in centrifugal chiller,a fault detection and diagnosis(FDD)method based on crisscrossed support vector machine(SVM)has been established.The optimization situation of continuous iteration and update of the parameters required by the random initialization training support vector machine in a certain range is studied.The influence of different parameters on the final optimization result is analyzed.Finally,this paper uses the experimental data of Ashrae-1043-RP project to verify the improvement of the accuracy of the method for fault diagnosis.The results show that:compared with the optimization of the grid and the genetic algorithm,this method effectively improves the accuracy of the solution and avoids the premature convergence of the algorithm to the local optimum;this method is effective for refrigerant charging when it is not optimized.When the injection is excessive or the leakage is serious,it has a higher degree of discrimination;the overall accuracy of refrigerant fault diagnosis after parameter optimization increases from 76.28%to 97.52%,especially for minor faults that cannot be effectively distinguished by the support vector machine before optimization.