Analysis Of A Zero-inflated Pipe Failure Prediction Model And Development Of Information System For Water Distribution System

Water distribution system is one of the most important urban infrastructures, and regarded as the blood vessels of the city. With the development of the water distribution system, the frequency of the pipe failures increased and the damage aggravated. These call for actions. Generally, the rehabilitation and replacement of pipes are reactive measures that are taken after a failure. However,"Prevention is better than cure". So it is better to build up strategies to prevent the pipe failures before they happen. Based on the Structural Hazard Information System, this paper collected the basic information and historic records of the water distribution system, and provided a reliable methodology of the management, maintenance and renewal of the system by a the data mining-based approach. This paper lay a foundation to develop a system to manage the water distribution system, to sort the information and to analyse the water distribution system data.First the paper established the Structural Hazard Information System. Based on the GIS techniques and Google Map, this system is a platform which helps the water utilities and customers manage the water distribution system more efficiently. Features and techniques of the system were introduced. And by the system, the basic water distribution system data including maintenance records were collected.Then the paper defined the pipe failure and the rate of pipe failure. A zero-inflated model was developed by considering the static factors and dynamic factors to forecast the pipe failures. By assuming that the data of pipe failures satisfied a mixed distribution composed of the Non-homogeneous Poisson distribution and Logistic distribution, the zero-inflated Poisson model can not only consider the influence of all the factors but also eliminate the influence of too much zero pipe failures. It was proved that the zero-inflated Poisson model contributed to a better imitative effect when the data with too many zeros, especially for a larger sample data set. Based on the present value function, the optimal time to replacement of water mains were defined. Combined with the forecasted expected number of failures for pipe i in year t obtained from ZIP model, we scheduled the pipes for renewal to maximize economic utility.We use a data set obtained from the system to illustrate the performance of the ZIP model. The score test showed that the number of zero failures was too large for a Poisson distribution to fit the data well. ZIP model was trained on the failure data from year2006to2013and the parameters obtained from training were used to forecast failures for validation of the year2014. The result showed that pipe age, length, diameter and number of known precious failure had an significant influence on pipe failure. It also showed the ZIP model was rather successful in estimating the total number of failures per year, up to92.62%. Forecast result is then presented in GIS Map. Then a5years pipe replacement schedule was made. There are33pipes needed to be replaced which are also presented in GIS Map.It indicated that the methodology and system developed in this paper may help to develop a suite of tools, which provide the most efficient system of maintenance and replacement of water distribution system, with the aim to guarantee the security of water supply.