Study On Anomaly Operation Data Andcorrection Method Of Heating Pipe Network

In the context of the rapid development of the Internet of Things in the 21 st century,the intelligentization of the heating system has provided many conveniences for the operation and control of the pipe network.The supporting information collection and transmission system has achieved the era of centralized heating "big data".Data mining will open up new exploration paths for reducing the consumption of heating energy and ensuring the safety of the operation of the pipe network.However,the emergence of big data has made the data quality problem more and more prominent.The erroneous data contained in the heating data has caused a lot of interference to the in-depth mining analysis.Therefore,the introduction of data correction technology in the heating field is very necessary.This paper firstly determines the focus of data correction on the processing of abnormal data,and proposes a three-step data correction method for abnormality detection,abnormal classification and abnormal replacement.The data correction framework kernel is constructed by the algorithm of isolation forest algorithm and BP neural network..Under the current situation that the actual heating data sample is missing and the operation feedback is incomplete,the simulation data set of the theoretical model is used to improve the data foundation for the data correction research.Firstly,the hydraulic calculation model and MKP solution method of heating space pipe network are introduced.The abnormal operation conditions are constructed and simulated by the spatial calculation pipe network model,and the variation law of pipe flow and node pressure is analyzed,and the simulation data set of abnormal operation conditions is constructed with the parameter change rate.In addition,a sample network operation data sample set is constructed manually based on all the results,and the abnormal data records of different modes are randomly inserted therein,and random errors are also added,so that the data model of the constructed sample set is richer and meets the needs of problem research.Then the data correction method is verified on the artificially constructed heating pipe network data sample set.The first step is to establish an anomaly detection model based on the isolation forest algorithm,which shows good anomaly detection performance on the sample set.The detection accuracy rate reaches 93%,the ROC curve is close to the upper left corner and the AUC value is as high as 0.987.In addition,through parameter analysis,it can be seen that the smaller scale sampling has little effect on the detection performance of the model.The second step is to establish an anomaly classification model based on BP neural network.The simulation data set of abnormal operation conditions is used as training samples.The model distinguishes process anomalies and measurement anomalies by learning different data models.The classification accuracy of the tested models can reach 98.7%.Then,using the abnormal data output by the anomaly detection model as input,the result of verifying the generalization ability of the model also indicates the efficiency of classification.In the third step,the data record whose abnormal classification mode is determined as "measurement abnormality" is replaced by the conditional mean value in the time series,and the reasonable data replacement value is estimated according to the continuity of the running data of the pipe network in time.Finally,the data correction process on the artificial sample set is realized,and the effectiveness of the method is verified.Finally,based on the actual heating pipe network-Yantai Qingquan Industrial Heating System 1# first station centralized heating pipe network operation data for data correction work,the corrected flow pressure curve is obtained,which proves the feasibility of the data correction method in this paper..