Water Flow Prediction in Perak River using Thin Plate Spline Basis Function Neural Network

Radial Basis Function Neural Network (ANN) technique has been found to be one of the most powerful tool use to predict the values of water discharge in Perak River. This technique has been proven to be the best alternatives to replace the previous forecasting technique such as Linear Regression Analysis and Flow Rating Curve which are less suitable to be applied to predict the non-linear stage and discharge data. The specific discharge data analysed from the developed Thin Plate Spline Basis function were important and crucial for the operational of river water management such as flood control system and construction of hydraulic structures, hence contribute towards the relevancy of this research paper. The data of the water level which were used as the input and discharge as the output were equally important for the training and testing purpose and those are taken for the three most recent years of 2011, 2012 and 2013. 780 data was used for the training whereas the remaining of 190 data was used for the testing purpose before run the analysis using the MATLAB software. At an optimal number of spread at 1.6607 and 30 hidden number the model architecture of using thin plate spline basis function showed a higher predictive performance than the normal Gaussian method at 0.986 for testing which is slightly lower than the training and Root Mean Square (RMS) of 2.310 which lower than the training due to the marginal difference in the minimum and maximum value of data. The comparison between the result obtained with the common kernel function used such as Gaussian shows that Thin Plate Spline Basis Function produce a more satisfactory result. Hence, the application of the thin plate spline basis function is recommended for the application in the other hydrology or non-hydrological field in future.