Bus- and Topology-Agnostic Load Forecasting-Based Fault Prediction Framework for Low Voltage Grids

Aging electrical infrastructure and rapid electrification demands large scale renewals and reinforcements of the power grid. Smart asset management and predictive maintenance are tools that may help the grid operators safeguard the power system by intelligently maintaining the grid and postponing the investment needs by allowing the system to be operated closer to its operational limits safely. In this paper, we propose a fault and critical event prediction framework for power grids, which leverages smart meter data to provide actionable insights into the future operation of both low and medium voltage assets. The framework's foundation is based on load forecasting for small low voltage consumer clusters, of typically 2-4 residential households or small industries. Even though this is an especially challenging task, our results demonstrate the efficacy of the framework's fault prediction capabilities when load forecasts are used for subsequent power flow calculations. The forecasting model is designed to be bus-semi-agnostic and topology-agnostic which not only enhances the computational efficiency of the framework but also provides essential information that may be used for short-term what-if analysis and power flow optimization. The proposed framework has been implemented in a real distribution system where it has been compared with, and outperforms, other methods. The results demonstrate that the proposed framework can predict overloads and under-voltages several hours ahead.