Application of a neural network approach for localization of problem areas of a centralized water supply system

In the presented work, the method of localization of vulnerable sections of the centralized water supply network using a neural network approach is considered. This method will allow you to identify areas where there is significant wear on pipes, leading to emergencies and water leaks. The paper discusses existing methods for assessing water quality using artificial intelligence. The adaptive neuro-fuzzy network ANFIS was used in the work. The input data are electrical conductivity, ppm, hardness, iron and flow rate in pipes of various diameters. The output value is the pipe wear value. The ANFIS network model alternately applied two types of training: back propagation and a hybrid method, aggregating backpropagation and least squares. The Fuzzy Logic Toolbox alternately applied membership functions: piecewise linear, Gaussian distribution, sigmoid curve, quadratic and cubic curves. A comparative analysis of the errors of different types of membership functions and learning algorithms was carried out. The coefficient of determination for the simulated and initial data is determined. The research was carried out on a section of the centralized water supply network, one of the districts of the city of Murom, Vladimir region, Russia.