Data Rate Prediction for Broadband Powerline Communication Based on Machine Learning

The increasing integration of Distributed Energy Resources into medium and low voltage grids requires a Smart Grid enhancement. To enable last mile communication in low voltage grids, Broadband Powerline Communication is one of the preferred technological options. Although it provides many operational advantages, the planning of Powerline Communication networks is still challenging. Existing planning tools so far rely on rough assumptions in critical planning steps, e.g. the calculation of data rates, so that methods for accurate data rate prediction in planning procedures are still to be developed. We therefore now present a laboratory setup to investigate the relations of data rates and Signal-to-Interference-and-Noise Ratio of Broadband Powerline technologies that are based on Orthogonal Frequency-Division Multiplexing. For the purpose of investigating the frequency-selective sensitivity of the data rate on fluctuating channel conditions and unknown bit loading schemes, we generate and measure data rates and signal spectrum simultaneously and subsequently train Machine Learning models to learn the relations. Applying five different techniques considered suitable for regression tasks, Gradient-Boosted Trees were identified as the most accurate method. As a result, our models showed sufficient accuracy to enable Powerline Communication planning based on frequency-selective channel estimations.