Overhead conductor thermal rating using neural networks

This paper presents a neural network approach for predicting dynamic thermal rating of high voltage transmission lines. For the integration of intermittent renewable energy, developing a reliable and accurate measurement tool is important to maximize power line utilization. In this research, distributed Power Donuts have been utilized to collect transmission line thermal and other related information. Along with environmental data such as wind speed and weather ambient temperature, this information has been used for training and testing of a neural network predictor. Due to the inherent non-linearity properties, predicting conductor thermal behavior is extremely complex and challenging. This paper proposes a novel method using a Finite Impulse Response (FIR) and Back Propagation (BP) neural network to predict the conductor thermal behavior. A FIR neural network introduces a short-term memory model, which can mimic the correlation between previous relevant data to the conductor temperature in near future. The BP neural network provides a supervised learning method to train the collected data and performs accurate prediction. A simulation toolkit is developed and experiments are conducted on data collected from real environments. The predicted values for up to one hour have been compared with the IEEE738 standard and the collected data from the power donuts. The outcome indicates accurate prediction and provides an alternative to the existing transmission line thermal measurement methodology.