基于向量自回归模型的水质异常检测研究

Water quality abnormity inspection and detection via the vector auto-regressive model

水质异常检测对保障用水安全具有重大意义。为了准确有效地判断水质异常,提出基于向量自回归（VAR）模型的多参数融合水质异常检测算法。VAR模型是自回归（AR）模型的一种扩展。通过AR模型和VAR模型跟踪和预测水质背景数据,计算预测残差,与设定阈值比较判断水质是否异常。结果表明,与基于AR模型的水质异常检测算法相比,基于多参数融合的VAR模型在水质背景数据跟踪上具有更好的准确性,能够实现较高的异常检出率和较低的异常误报率。

The paper intends to propose a multi-parameter fusion algorithm to detect and deteminate the water quality abnormity pattern based on the vector auto-regressive（ VAR） model. The VAR model is by nature an extension of auto-regressive AR model usually adopted to track and predict the background data of the water quality. As a matter of fact,the model can be used for calculating and comparing the residual error with the set threshold to assess and judge if the water quality is abnormal. In order to test and verify the performance of the water quality abnormality detection algorithm,it is necessary to superimpose the inverted Ushaped abnormity on the basis of the actual monitoring data to simulate the abnormal variation of the water quality parameters.What is more,the normalized method can also be taken to make a dimensionless transformation of the water quality data with the results in which the mean value is 0 with the variance being 1. To achieve the purpose,it is unnecessary to stack the abnormal situation of the first 10 000 groups,rather,it is necessary to obtain the parameters of AR model and VAR model through necessary training. And the last 50 000 groups of data superposition abnormity should be used to contrast and compare the deviation between the predicted data and the actual data. Last of all,by setting the different thresholds,it wuld be possible to make out the corresponding relation between the false alarming rate and the detection rate under different conditions that have been obtained.What has been said above can help us to succeed in drawing the receiver operating characteristic（ ROC） curve with the false alarm rate as the horizontal axis and the detection rate as ordinate to evaluate the effect of the water quality abnormity detection. The experimental results show that VAR model based on the multi-parameter fusion has higher accuracy in tracking the water quality background data,and therefere,beneficial to achieve higher detection rate but lower false positive rate as compared with the water quality abnormity detection algorithm based on AR model.