Prosumers Energy Consumption Forecasting: Leveraging LSTM and XGBOOST with Spatial and Weather Features

The energy transition towards green energy systems at the residential level leads to an unpredictable load on the power distribution grid. This is due to the simultaneous role of the endusers as electricity producers and consumers. Therefore, accurate and up-to-date energy consumption forecasting in Low Voltage (LV) grid has become critical for energy system operators. In this paper, we utilized Extreme Gradient Boosting (XGBOOST) and Long Short-Term Memory (LSTM) models to forecast residential energy consumption and quantify the impact of spatial and weather features on forecasting accuracy using anonymized real-time measurement data from LV substations. Our study demonstrates the feasibility of using anonymized data for forecasting and highlights the importance of weather and spatial features in improving forecasting accuracy through feature importance.