基于深度学习框架的供水管网压力异常实时诊断方法

为有效减少供水管网的漏失,诊断异常压力数据十分必要。在数据采集与监视控制(SCADA)系统中引入异常诊断模型,提出一种基于深度学习框架的异常诊断方法。该方法先利用卷积神经网络模型(CNN)对压力进行预测,再计算压力预测值与实际值的误差并进行离群点诊断,若异常值持续时间较长,则可能发生漏失。以D市供水管网模型为例,利用该模型对16个监测点压力数据进行诊断并与ARIMA诊断模型进行对比。结果表明,CNN模型能够准确地诊断供水管网的压力异常数据。

Real-time Diagnosis Way of Abnormal Pressure in Water Supply System Based on Deep Learning Framework

Abnormal pressure diagnosis is essential to reduce leakage in water distribution system. A abnormal diagnosis model based deep learning framework was presented to embed in supervisory control and data acquisition (SCADA) monitoring system. Pressure was firstly predicted by convolutional neural network (CNN) in this model. Secondly, outliers were diagnosed by calculating the absolute errors of real and predicted pressures values. If abnormal values keep appearing, leakage might be happened. A case study of city D was employed to verify this model. In this study, 16 monitors data were used to diagnose abnormal values. By contrast with ARIMA model, the results demonstrate that the CNN is more precise diagnosis abnormal pressure values of water supply system.