Random fully connected layered 1D CNN for solving the Z-bus loss allocation problem

Abstract

Power loss allocation methods should be efficient enough to meet the needs of the customers on the bus and effectively calculate the losses from generators and consumers. In order to perform these tasks, a highly robust model is essential to distinguish between the effects of multi-consumers. This study presents a novel convolutional neural network (CNN) architecture that is highly effective for z-bus loss allocation. The proposed CNN architecture that uses the Z-bus matrix as input is 1D. Unlike traditional 1D CNN architectures in the literature, the fully connected layer (FCL) of the proposed method is randomized. Unlike Traditional FCL layers, randomized FCL's input weights and biases are not needed to be tuned. This makes the proposed 'Randomized Fully Connected Layered 1D CNN' architecture relatively fast and straightforward. Proposed Randomized Fully Connected Layered 1D CNN is trained in an end-to-end manner with a regression task for robust loss allocation. The performance of it is higher than other state-of-the-art methods. In addition to the fact that the proposed method's regression performance is very promising, the classifier performance is quite satisfactory thanks to the changes to be made in its output.