Indian Journal of Science and Technology
Year: 2018, Volume: 11, Issue: 38, Pages: 1-9
Ernesto Aguilera1 , Ivan Amaya2* and Rodrigo Correa1
1 Escuela de Ingenierías Eléctrica, Electrónica y de Telecomunicaciones, Universidad Industrial de Santander, Colombia; [email protected], [email protected]
2 Escuela de Ingeniería y Ciencias, Tecnologico de Monterrey, México; [email protected]
*Author for correspondence
Escuela de Ingeniería y Ciencias, Tecnologico de Monterrey, México; [email protected]
Background/Objectives: In this work, we present an efficient design methodology that uses boosting algorithms to improve the accuracy of any given learning algorithm by combining the output of individual weak learners. Methods: First, a finite-difference time-domain model of a loaded rectangular wave guide yields the desired input-output response of a microwave heating system. Then, it is used to train neural networks used as weak learners in the boosting algorithm. Findings: The method is easy to implement and have a tendency not to over fit the training data. Data show that performance of the boosting algorithm increases with the number of neural networks. An example that uses 34 neural networks, with three hidden layers, fits 96 of 100 temperature profiles of the heating system with a previously defined root mean square error below 1°C. Applications: Two simple examples of inverse modelling problems of the heating system were solved efficiently using the output of the boosting algorithm.
Keywords: Boosting Algorithms, FDTD, Microwave Heating, Temperature Profile Estimation
Subscribe now for latest articles and news.