A Method for Evaluating the Accuracy of Neural Network Estimation Using Attractor Behavior
Previous neural network systems for temperature estimation are based on the enormous number of nodes and connections, and they provide accurate predictions. However, they are too large to implement into cheaper IoT systems. In this paper, we propose a new combined method of a small neural network with temperature attractor behavior to solve this challenge. Our small neural network mainly calculates stable temperature predictions, and the attractor behavior detects irregular temperature. According to the results of our yearly experiments in five major cities in Japan, we found that our method can predict accurate temperatures.
Authors: Huidong Tang, Yuichiro Mori, Masahiko Toyonaga
Published in: World Congress on Sustainable Technologies (WCST-2020)
- Date of Conference: 8-10 December 2020
- DOI: 10.20533/WCST.2020.0001
- ISBN: 978-1-913572-25-9
- Conference Location: London, UK