A comparison of ANFIS and MLP models for the prediction of precipitable water vapor

This paper aimed to compare the adaptive neuro fuzzy inference system (ANFIS) with multi layer perceptron (MPL) of artificial neural network (ANN) structure in estimating the precipitable water vapor (PWV) value. The estimation is based on the surface meteorological data as input from the Malaysian environment and the results of these models were compared with PWV observed by GPS. Two kinds of training data sets were provided to develop these models based on the data gathered from UKMB and UMSK stations at one-minute resolution. To perform of these models, a correlation coefficient (r), root mean square (RMSE) and percent error (PE) were employed. Results showed that the correlation coefficient (r), RMSE and PE of ANFIS model for UKMB station were 0.999, 0.018, and 0.023 and 0.979, 0.019, 0.028 for UMSK station, respectively. For MLP model, the values are 0.975, 0.337 and 0.390 for UKMB station and 0.978, 0.305 and 0.443 for UMSK station. Based on the above results, both models showed strongest correlation. However, RMSE and PE for MLP model are higher ~5.5% and 5.59% compared with the ANFIS model. This indicated that ANFIS model has better performance and can be proposed as an alternative method in estimating the PWV value where the GPS data in a specific location is absent.