Wide-Deep Neural Networks with Randomized Sparse Backpropagation Simplify Demand-Side Load Event Detection
Manasa Adusumilli, G. P. Ramesh, Vijayrani Katkam, Bhargavi Bhargavi, Piyush Kumar Pareek
Abstract
Identifying demand-side load events is a critical step in lowering energy use in a number of situations. Using wide-deep neural networks and randomised sparse backpropagation algorithms, this study presents a unique way to identify demand-side load events. To detect subtleties in the input data, we employ randomised sparse backpropagation and both shallow and deep network topologies. We also use the ReLU and sigmoid activation functions to improve the performance of our model. We use the Adam optimizer to train the proposed model, using a learning rate of 0.001 per batch and a batch size of 64. The results were compared to those of other popular technologies, such as support vector machines. According to the findings of our studies, the technique we utilised is more precise and efficient than any other option we investigated. The proposed approach combines wide-deep neural networks with randomised sparse backpropagation to simplify demand-side load event identification.