Abstract:
This research objectives were to develop the tournament selection of genetic algorithm (GA)
for forecasting rainfall with artificial neural network (ANN) based on 3 principles; 1) normalized
geometric ranking (NGR), 2) roulette wheel selection (RWS) and 3) tournament selection (TS). The
research was divided into 2 parts. The first part was the development of forecasting model under the
simulation of 1,100 data with iteration 1,000 epochs, and the second part was the forecasting by
model of the previous one. For the latter part, the actual data were brought to divide into 2 sets;
namely training and testing data set at 77% and 23%, respectively. Then, the artificial neural network
model developed in the tournament selection and the artificial neural network model using the original
selection of Wang et al. (2017), in aspect of forecasting efficiency by mean absolute error (MAE),
mean absolute percentage error (MAPE), root mean square Error (RMSE), and coefficient of
determination (R)2. The input variables of artificial neural network were relative humidity, wind
speed, zonal wind, meridional wind, evaporation, minimum air temperature, maximum air
temperature and average temperature. The output variable was rainfall of Central Thailand. The total
data of 1,100 was seasonally analyzed. The results showed that the forecasting model developed by
the tournament selection of genetic algorithm was more effective than the model with original
selection of Wang et al. (2017) in every season.