Study On Key Technologies Of Hydraulic Model Driven Online Leak Monitoring In Urban Water Distribution Networks

Leakages in urban water distribution networks have become one of the most ever-intensifying problems faced by water utilities both home and abroad.To alleviate the socioeconomic stresses caused by leakages,it is necessary to explore effective methods in detecting leakages in water distribution networks,which contribute to shortening average leakage duration and reducing total water losses.Due to the fact that the hydraulic model provides key clues for leakage detection and identification,this thesis proposes a hydraulic model driven online lea:kage monitoring method as a rapid coping strategy to occurred leakages.First,a bayesian inference method is adopted to calibrate the hydraulic model until it represents the real system adequately.Then,under the evalua-tion of the leakage diagnosability criterion,the locations of limited number of pressure monitoring sites are optimized.Finally,according to the leak-induced pressure residuals,a sparse representa-tion classifier is learned to identify leakage zones on the basis of the leak sensitivity matrix.The detailed research content and innovations are summarized as follows:(1)To improve the precision of the hydraulic model,a bayesian inference method is proposed to calibrate the hydraulic model.By employing a novel two-level Markov chain Monte Carlo par-ticle filter method,demand and roughness are jointly estimated.Specifically,an improved particle filter with ensemble Kalrman filter modification to proposal density is adopted to estimate nodal de-mands,while differential evolution adaptive metropolis(DREAM)algorithm is utilized to estimate Hazen-Williams coefficients,respectively.The method is very capable of automatically distin-guishing data with different degree of usefulness,and significantly overcoming the compensating errors resulting from single-parameter calibration.(2)To guarantee sufficient leakage diagnosability while satisfy budget constraint,an opti-mization strategy for pressure monitoring site locations is proposed.The average mutual coher-ence of normalized reduced sensitivity matrix is selected as a diagnosability criterion for evaluating whether the monitoring data contains enough information for identifying potential leakages.Under the evaluation of the diagnosability criterion,a binary version of artificial bee colony algorithm en-hanced by genetic operators,including crossover and swap,is introduced to optimize the pressure monitoring site locations.The method is very capable of seeking a quasi-optimal monitoring site locations scheme with a maximum degree of diagnosability.(3)To extract key features in the monitored pressure residuals that give clues to the zone of occurred leakages,an infonmation-oriented sparse representation classifier is adopted for hydrauli-cally identifying leakage zones.Based on a set of region-labeled normalized leakage signatures,a class discriminative dictionary is learnt using LC-KSVD,so that the among-class distances of'the derived sparse representation maximize and within-class distances minimize.Simultaneously,a multi-label classifier over the sparse representation is acquired to determine an exact zone that the occurred leakage belongs to.According to the monitored pressure residuals,the method isvery capable of identifying suspicious leakage zones in near real time,which is fundamental to the subsequent leakage pinpointing.(4)To verify the effectiveness and practicality of the proposed hydraulic model driven online leakage monitoring method,a highly controllable water distribution network experimental platform is built.On the experimental platform,the calibration of hydraulic model,the optimization for pressure monitoring site locations,as well as the identification of suspicious leakage zones have been implemented,laying a solid fundamental for its future applications on real water distribution networks.