ON DAM WATER SPILLAGE MODELLING USING 3-2-1 NEURAL NETWORK ARCHITECTURE

Abstract

Forecasting Dam Spillage is very important in reservoir operation and dam management. This hydro power plant is the main source of electrical energy and water, both for human consumption and for farm irrigation. When heavy rainfall occurs, heavy dam water spillage can have a great impact to the nearby communities below the dam. Monitoring the volume of dam water spillage is very crucial in the events of flood and disaster prevention programs for the safety of the communities and inhabitants especially of those living at the riverbanks. The objective of this paper is to present the use of 3-2-1 neural network architecture in forecasting the monthly dam water spillage. Three variables were used of which the average monthly rainfall and average monthly river inflow are considered the input variables and the average monthly water dam spillage is the output variable. One hundred twenty-one (121) months of data were gathered. Data from years 2013-2019 (84 months) were simulated that is 70% of the total number of data gathered and used as the training set, and the remaining 30% a total of 37 months of data records available in PAGASA and Pulangi IV HEPP were treated as the testing set that were used in determining the efficiency of the model. The Radial Basis Function Neural Network and the General Regression Neural Network were used for the calculation of the monthly dam water spillage forecast. The Root Mean Squared Error (RMSE) was used in determining the forecasting performance of the 3-2-1 architecture design. Results showed that forecasting the monthly dam water spillage volume using 3-2-1 neural network architecture model is possible and highly accurate. It also showed that the Radial basis neural network forecasts better when compared to using the General Regression Neural Network.

 

Keywords: Forecasting, Dam Spillage, Artificial Neural Network, Radial Basis Function Network, General Regression Neural Network