Stochastic and Artificial Neural Network Models for Reservoir Inflow Prediction
Kote, A. S
MetadataShow full item record
The performance ofartificial neural network (ANN) model was evaluated by applying it to the observed time series of Pawana reservoir in Upper Bhima River Basin, Maharashtra. The ANN model results were compared with conventional univariate auto regressive integrated moving average (ARIMA) models. Suitability of time lagged recurrent networks (TLRN) with time delay, gamma and laguarre memory structures was investigated for predicting seasonal (June to October) reservoir inflow with a monthly time step. The performance of back propagation through time (BPTT) algorithm trained networks for various inputs was compared with genetic algorithm (GA) trained networks. Due to large variation in the observed time series, transformation of the observed series to normal distribution was also tried and found that the network predicted better. The validation of the models was performed using comparison of the principal statistics, goodness-of-fit measures, time series and scatter plots. Encouraging results indicated that the logarithmic transformed, BPTT trained TLRN resulted in better and reliable forecasts of high and low inflows (extreme) compared to GA trained neural networks as well as ARIMA models.
- Article