Neural Network - Based Modelling and Prediction of Reservoir Inflow Dynamics in Real Time

Abstract

A neural network modeling approach is used to construct a real time catchment flow prediction model for river basin. The present study aims towards introducing the use of ANN technique to model and predict the monthly inflow of Jayakwadi Reservoir. The data pertaining to the years 1927-1971 have been explored to develop the predictive model with Tanh Axon nonlinearity and Levenberg Marquardt as a training algorithm. The Jayakwadi reservoir, one of the largest irrigation projects in Maharashtra with catchment area of 21,750 km2, is selected as the study area. ANN model was developed based on the historical inflow data of the Jayakwadi Reservoir and successfully predicted inflow with less errors. In this study six forecasting model approach are used, which are multilayer Perceptron (MLP), feed forward (FF), modular neural network (MNN), Jordan/Elman network, time-lag recurrent (TLRNN), CANFIS network method for the natural inflow at Jayakwadi utilizing the inflow data of the monitoring stations. Inflow data collected over the last 45 years is used to develop the model and examine its performance. 70% of data is used to develop the model and rest of data is utilized to test the models. Except CANFIS method in all five methods i.e. MLP, GFFNN, MNN, J/ENN, TLRNN results shows that coefficient of correlations are above 90% and in CANFIS method coefficient of correlation is 89%. The result of this study showed that ANN technique is capable with small computational effort and high accuracy of predicting the monthly inflow.