Abstract:The heavy overload of distribution transformers is one of the main causes of transformer failures and damage. Therefore, accurately predicting the operating conditions of distribution transformers is crucial for the safe and reliable operation of the power system. Due to the influence of various complex variables, the distribution load in distribution substations is often difficult to reliably model and estimate, resulting in certain uncertainties in the final prediction results. Traditional point predictions aim to predict a single optimal value, which cannot fully quantify the uncertainty of the predictions.This paper integrates multidimensional features and historical operating data of transformers, and models the load conditions in distribution substations using quantile regression methods. By combining conditional quantiles with general linear or nonlinear models, a probabilistic prediction model is constructed. Quantile regression can model the entire conditional distribution and provide more information compared to standard regression methods. Traditional point predictions, due to their inability to estimate the uncertainty of the prediction results, carry a certain risk for decision-making in business departments. In contrast, probabilistic predictions can not only provide the future optimal prediction point, similar to point predictions, but also provide information about the distribution of future predicted values. Probabilistic predictions, in the form of probability density, prediction intervals, or quantiles, can better estimate the uncertainty of the distribution load.Therefore, probabilistic predictions play a crucial role in accurately assessing the uncertainty of distribution load, ensuring the safety and reliable operation of the power system, and making informed decisions.