Development and optimization of deep convolutional neural network using Taguchi method for real-time electricity conservation system

Conservation of energy is imperative for the development of the country. The critical electrical energy situation is one of the immense challenges in the developing nations of the globe. Though there are many ways of preservations, one of the simplest and efficient ways is to turn off the electricity when it is not needed. Thus, in this study, we have developed an automatic smart system to enable the electricity turned on only when a human being is detected in Closed Circuit Television footage (CCTV). Human detection in the footage is done by classifying the frames with and without a human by Convolutional Neural Network (CNN) in an indoor environment. As the system is processing live CCTV footage, the reduction of training and overall processing time of the model is done by Extended Orthogonal array tuning method (E-OATM) which tunes the parameters and hyper-parameters of the model using Taguchi method. The constructed model is further pruned and the less important parameters are eliminated based on H-Rank algorithm to keep the model simple for fast processing. Finally, optimization of the network is done using Nadam technique, and the results (images) are fed into the Arduino UNO microcontroller. This controls the power supply that turns off/on the lights and fans only when human is unavailable/available respectively without the assist of a sensor. © 2022, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.