Improving the performance of data-driven techniques through data pre-processing for modelling daily reservoir inflow

The effect of data pre-processing while developing artificial intelligence (AI) -based data-driven techniques, such as artificial neural networks (ANN), model trees (MT) and linear genetic programming (LGP), is studied for Pawana Reservoir in Maharashtra, India. The daily one-step-ahead inflow forecasts are compared with flows generated from a univariate autoregressive integrated moving average (ARIMA) model. For the full-year data series, a large error is found mainly due to the occurrence of zero values, since the reservoir is located in an intermittent river. Hence, all the techniques are evaluated using two data series: 18 years of daily full-year inflow data (from 1 January to 31 December); and 18 years of daily monsoon season inflow data (from 1 June to 31 October) to take into account the intermittent nature of the data. The relevant range of inputs for each category is selected based on autocorrelation and partial autocorrelation analyses of the inflow series. Conventional preprocessing methods, such as transformation and/or normalization of data, do not perform well because of the large variation in magnitudes, as well as the many zero values (65% of the full-year data series). Therefore, the input data are pre-processed into un-weighted moving average (MA) series of 3 days, 5 days and 7 days. The 3-day MA series performs better, maintaining the peak inflow pattern as in the actual data series, while the coarser-scale (5-day and 7-day) MA series reduce the peak inflow pattern, leading to more errors in peak inflow prediction. The results indicate that AI methods are powerful tools for modelling the daily flow time series with appropriate data preprocessing, in spite of the presence of many zero values. The time-lagged recurrent network (TLRN) ANN modelling technique applied in this study maps the inflow forecasting in a better way than the standard multilayer perceptron (MLP) neural networks, especially in the case of the seasonal data series. The MT technique performs equally well for low and medium inflows, but fails to predict the peak inflows. However, LGP outperforms the other AI models, and also the ARIMA model, for all inflow magnitudes. In the LGP model, the daily full-year data series with more zero inflow values performs better than the daily seasonal models.