Modeling Of Air Temperature Using Hybrid Wavelet Transform - ANFIS - Support Vector Machine Computing Techniques

Abstract

The accurate modeling of average air temperature is a significant and much essential parameter in frame of reference for decision-making. Therefore, the characterization of such parameter is an important task. The information about the air temperature also helps in planning and management of water resources, irrigation, drought detection, tourism, health and issues of day to day life. In this study, a hybrid model consists of Wavelet - ANFIS has been developed for air temperature modeling. The results are compared with Wavelet - SVM, single ANFIS, and single SVM to confirm the superiority of the proposed model. To model average air temperature, ANFIS models were developed with different membership, namely generalized bell-shaped built-in membership function (GBELLMF), and Gaussian curve built-in membership function (GAUSSMF). Additionally, to check the result of modeling of average air temperature, SVM model was developed. To enhance the accuracy of modeling performance, single ANFIS and single SVM is integrated along with wavelet transformations were tested. Here wavelet transformation was used as pre-processing the data by capturing valuable information on various resolution levels. This study extends for seven stations in Karnataka state of India (Shimoga station, Raypura station, Linganmakki station, Honnali station, Hiriyur station, Bhadra station (B. R. Project) and Davanagere station) observed data of meteorological data like rainfall, wind speed, humidity and sunshine hour as input and as target average air temperatures are used for all the models. In the next phase, the influence of air pollutants along with the meteorological parameters has been investigated for average air temperature modeling for a specific Bhadra station in Karnataka state, India, which is near to industrial city. The obtained results were evaluated using Correlation Coefficient, Root Mean Square Error and Scatter Index. The performance of ANFIS, SVM, hybrid Wavelet - ANFIS and hybrid Wavelet - SVM is analyzed for modeling of average air temperature. Out of seven stations, station Linganamakki showed better performance with CC of 0.954, RMSE is 0.71and SI is 0.027 with hybrid Wavelet- ANFIS model (Gbell membership). Also for single Bhadra station, Hybrid Wavelet - ANFIS model with the parameter combinationiv (rainfall, wind speed, humidity, sunshine hour) for Db5 with level4 (2MF) and Gauss membership function is having the results of CC is 0.98, which is best in case of accuracy. The study reveals the higher accuracy of hybrid Wavelet - ANFIS in modeling air temperature for various meteorological and air pollutants input scenarios.