Neural Network Approach for Availability Indicator Prediction

The principal aim of this research was to find out if artificial neuralnetworks could be employed to predict the availability factorfor water mains, distribution pipes and house connections.Modelling by means of artificial neural networks (ANNs) wascarried out using the Statistica 10.0 software package. Operatingdata from the years 1999–2005 were used to train the ANNswhile data from the next seven years of operation were usedto verify the model. The optimal model (characterized by thelowest mean-square error) contained 11 hidden neurons activatedby the exponential function. The linear function was usedto activate the 3 output neurons. 185 training epochs sufficed totrain the ANN, using the quasi-Newton method. The correlationbetween the availability indicator experimental values and themodelling results would remain high, amounting during modelverification to R2 = 0.740, R2 = 0.823, R2 = 0.992 for respectivelywater mains, distribution pipes and house connections. As theavailability indicator prediction example shows, the artificialneural networks are a promising tool enabling quick and easyanalysis of failure frequency. It is possible to train the ANN furtherand change the number of training epochs and hidden neuronsas well as the activation functions and training methods.