ARTIFICIAL NEURAL NETWORKS TO MONITOR IRRIGATION SYSTEMS, SIMULATIONS AND APPLICATIONS

Abstract

Monitoring water management, especially in irrigation systems need to be efficient in order to understand their fitness in a particular time frame, looking at this by collecting field data could be time consuming and expensive. Therefore, modelling the whole system using factors effecting its inflow and outflow could be really useful. ANN can be used to monitor and predict irrigation systems by using primary and secondary data. Using ANN, it is possible to build nonlinear relationships between input and output parameters without relying on physical process. In this study we simulate the outflow pattern of “Yoda wewa” Sri Lanka considering six parameter model. Comprising inflow, rainfall, evaporation, temperature, relative humidity and wind speed. The model was obtained by performing the training with various transfer functions, learning rates and momentum coefficients. The results found to be capable of predicting water outflow pattern of “Yoda wewa” up to 66% of confidence. The success rate heavily depends on the parameters used and the density of the data in the study. Few parameters were not being able to be collected from the site itself. Hence it is necessary that a proper ground truth data to come in as inputs for the model. It is desirable as an extension of this work to include soil structure and underlying topography. The model could be tested for several Irrigational systems in order to make substantial recommendation of applying the method for different types of Irrigational System.