Research On Water Use Anomaly Monitoring Based On Isolated Forest Algorithm

With the wide deployment of wireless Internet of things monitoring equipment,the data obtained by the monitoring platform has increased rapidly,reaching an unprecedented scale.The automatic abnormal detection of data has become an important direction of the development of monitoring system.The monitoring ability of water resources has been paid attention by all walks of life.For the massive water data produced in daily life,it is an important means to realize intelligent monitoring of water resources by automating the abnormal data through machine learning.According to the characteristics of continuity and randomness of water use data,and the requirements of real-time and high accuracy of abnormal monitoring,this paper studies the technology of water use anomaly monitoring,and the main work includes:(1)The principle of monitoring water use anomaly by the isolated forest algorithm is studied,and the shortcomings of Iforest algorithm in training and detection stage are analyzed.(2)An improved W-iForest algorithm is proposed.The improved W-iForest algorithm is mainly aimed at the training of iTree(isolated binary tree)and the calculation of abnormal score in traditional iForest algorithm.Traditional iTree ignores the influence of data features on the degree of data abnormality in training.W-iForest algorithm gives different weights to each feature of data during W-iTree training,and selects features according to weight,and then divides data sets to complete W-iTree training.Compared with the traditional iForest algorithm,the improved w-iForest algorithm considers the influence of feature weight on path length when calculating outliers.The larger the weight,the smaller the length of the path represented,the more accurate the calculated results.In this paper,we use W-iForest algorithm to test four sets of data sets respectively.By comparing with traditional iForest algorithm,the improved algorithm accuracy and AUC value have been improved,and the missed rate and false alarm rate are reduced.The effectiveness and stability of the algorithm are verified.The real-time performance of the algorithm is tested.The difference of milliseconds level can be achieved under the data set in this paper often identified.(3)An automatic meter reading system is realized to monitor water use abnormity intelligently.The system needs analysis and solutions are given,and the overall architecture of the system is designed.The modules of terminal data collection,real-time monitoring and warning,unified management of water meter,terminal equipment installation and operation management and system management are realized.