A multipurpose dam serves multiple modalities like agriculture, hydropower, industry, daily usage. Generally dam water level and inflow are changing throughout the year. So, multipurpose dams require effective water management strategies in place for efficient utilization of water. Discrepancy in water management may lead to significant socio-economic losses and may have effect on agriculture patterns in surrounding areas. Inflow is one of the dynamic driving factors in water management. So accurate inflow forecasting is necessary for effective water management. For inflow forecasting various methods are used by researchers. Among them Auto Regressive Integrated Moving Average (ARIMA) and Artificial Neural Network (ANN) techniques are most popular. Both of these techniques have shown significant contribution in various domains in regards to forecasting. But they have a common drawback in handling non-stationary inflow patterns. To address this drawback, in this work neural Reservoir Computing technique is used. In this work, Context reverberation network, also known as reservoir computing approach, is applied for inflow forecasting. It comprises of a dynamic neural reservoir. As the nature of a neural reservoir is dynamic, it can easily model complex nonstationary patterns along with stationary ones. Proposed model is applied on daily inflow data of Srisailam Dam which is a multipurpose dam. Here ARIMA and ANN models are compared with Reservoir Computing model. On various evaluation parameters Reservoir computing is proved better than ARIMA and ANN.
Evaluation of the BMA probabilistic inflow forecasts using TIGGE numeric precipitation predictions based on artificial neural network
