The increasing integration of distributed energy resources (DERs), such as rooftop photovoltaics (PV) and electric vehicle (EV) charging infrastructure, has introduced unprece-dented volatility into low-voltage (LV) distribution networks.As a result, obtaining precise short-term load forecasts (STLF) has grown increasingly vital yet difficult to achieve. Conventional forecasting approaches, ranging from statistical techniques to recurrent neural networks (RNNs), frequently fail to model the intricate nonlinear patterns and long-term temporal relationships characteristic of contemporary residential energy consumption.

This study proposes a novel deep learning framework based on a Transformer architecture—specifically utilizing the Patch-TST variant—to enhance predictive precision at the LV level. The proposed model incorporates historical load data alongside key exogenous variables, including meteorological conditions, calen-dar features, EV charging patterns, and solar power generation. To ensure a rigorous evaluation, the framework was tested across three operational scenarios: (1) standard residential demand,

(2) demand with EV penetration, and (3) a combined scenario featuring both EV and PV integration.

Experimental results, benchmarked against Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Ran-dom Forest models, demonstrate that the Patch-TST approach achieves the lowest Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The superior performance is attributed to the self-attention mechanism, which effectively models global temporal relationships and intricate feature interactions. Experimental results indicate that the proposed Transformer architecture offers a dependable solution for distribution network operators, facil-itating improved demand-side management, maintained voltage stability, and increased smart grid operational efficiency.

Index Terms—Low Voltage Load Forecasting, Transformer-Based Forecasting, Electric Vehicle Integration, Rooftop Solar Variability, Multivariate Time Series, Smart Grid.