Prediction and demand control methodology for a distribution system using Artificial Neural Networks

The monitoring of the load assumes a fundamental role in contracting of locational economic amounts of the Transmission Network Use of System (TNUoS) tariff, in this way, it becomes attractive to use computational models of demand forecasting as tools, subsidizing more precise estimates of investments and enabling the attendance of the Reliability. In this context, the present work demonstrates a methodology to perform the reallocation of the TNUoS contracted in the transmission system, among downstream Distribution Substations (DSs), based on a mathematical model of short-term load forecasting, with a 24-hour horizon. Proposing a better daily control in the management of the TNUoS contracted together to the points of connection, idealizing the efficiency in the contracted demands and minimizations of penalties. The adopted model uses Artificial Neural Networks (ANNs) with Multilayer Perceptron architecture (MLP) and Gradient Descent Backpropagation learning algorithm. The research also allowed to perform a sensitivity analysis between the load demand variation and the temperature in the study horizon. The results obtained through the case study demonstrated that the proposed methodology, when necessary, provides conditions for the decision making regarding the reallocation of the load through medium voltage distribution network, between the points of connection with the Transmission System.