Abstract
The amount of water lost in various leakage events in water networks in the UK not only generates economic loses for the water management companies, but more importantly, threatens supplies to households and businesses in times of severe drought. To address this problem, a novel forecasting approach is presented in this thesis that aims at predicting the leaks so that appropriate resources can be planned to handle them.A modified Fuzzy Evolving Takagi-Sugeno (Mod eTS) algorithm has been developed and applied to the leakage forecasting problem. The algorithm recursively clusters the data samples as soon as they become available and builds the fuzzy structure based on the generated clusters. Each new data sample is compared to already gathered data by calculating its potential. Based on that, the new data sample can modify the structure of existing clusters or initiate a new one. The modification may not only shift the cluster centre, but can also change the area of influence of the cluster, by adjusting its radius. The clusters are used to generate fuzzy If-Then rules through Takagi-Sugeno inference. The modified version of Recursive Least Squares algorithm is used to estimate parameters of the resulting linear equations, by taking into account the firing strength of the Fuzzy If-Then rules. This way the system is allowed to evolve, by constantly learning and adapting to changes in the data.
The algorithm has been applied to two sets of data: the leakage data from 8 regions of operation of one of the biggest water suppliers in the UK and to artificially generated time-series data using the Mackey-Glass process. The novel approach is evaluated and compared with a number of computational intelligence and widely accepted, statistical methods and consistently demonstrated highest accuracy for the leakage data sets (MASE of 1.263 as compared to 2.319 on average for other methods).It also preformed well on the Mackey-Glass time-series when both the accuracy and the complexity of the model were considered. Its performance demonstrated the potential to be further developed and applied in the industrial setting, not only in the water industry, but also in other areas.
Date of Award | 2016 |
---|---|
Original language | English |
Awarding Institution |
|
Supervisor | Dobrila Petrovic (Supervisor) & John Boylan (Supervisor) |