RAINFALL FORECASTING USING ARTIFICIAL NEURAL NETWORK

Abstract

Rainfall forecasting is the application of science and technology to predict the state of rainfall for a given location. This is done by collecting the quantitative data of the rainfall and using scientific understanding of the rainfall process to forecast the future conditions. In India, Rainfall forecasting is done by Indian Meteorological Department (IMD), New Delhi which provides the real-time monitoring and statistical analysis of district-wise daily rainfall. Several research works have been done using different methodologies of which the ANN technique is the fastest and provides reliable solutions. In this dissertation, ANN methodology is applied for forecasting of rainfall in Delhi region. Here, ANN methodology is used to forecast rainfall using various configuration of the models. This configuration depend on the various structural parameters, such as, number of hidden layers, number of neurons in each layer, activation functions and training backpropagation algorithms. These models are categorized according to the training algorithms, namely Levenberg-Marqurdt backpropagation algorithm (LM), Bayesian regularization backpropagation (BR) algorithm and Scaled Conjugate backpropagation (SC) algorithm. Seven models are there in each category. These models have been trained and tested. The results give two models with least value of performance parameter, ‘mse’, one from LM and BR each with three hidden layers with 10 number of neurons in each layer. Then, the forecast of the selected models have been checked for validation which give the satisfactory results for ANN based forecasting of rainfall in Delhi region.