Hybridization of a Recurrent Neural Network by Quadratic Programming for Combinatory Optimization: Case of Electricity Supply in a University Campus

Joule loss is a major concern in the power grid because it can lead to significant energy waste. There are a number of methods for minimizing joule loss, but they are still incomplete because they don’t take into account the changing trend of electricity demand in buildings. In this paper, we propose a novel approach to minimizing joule loss using a hybridization of a recurrent neural network (RNN) and quadratic programming (QP). The RNN is used to learn the temporal dynamics of the electricity demand, while the QP is used to solve the combinatory optimization problem of minimizing joule loss. The recurrent neural network set up is able to predict the needs of the different campus premises in order to make the system flexible to the needs of consumers. It contains 5 layers: 3 of the Long short term memory type, 1 of the dropout type and 1 of the dense type for a total of 86,910 parameters to be determined during training. It was trained and validated using American hourly electricity power data normalized between 0 and 1. We model the power losses by Joule effect occurring during the transport from the production sources and to the consumers with quadratic programming. Once the optimization model was formulated based on electric laws, it was solved to determine the optimal current flows that should circulate in one part of the network. After simulation, the analysis of the results obtained shows that the proposed solution produces good results to the extent that by integrating variations in user needs, it makes it possible to determine the configuration which reduces hourly losses. To improve the accuracy of the proposed system one could use a data set containing a few more parameters affecting the consumption of the load such as meteorological datas.