Electrical Load Prediction Based on Integrated Learning GBDT Algorithm

The development of the national economy is closely linked to the production of electric energy, and excessive or insufficient production has a significant impact on the operational stability of the power system. Accurate load forecasting technology holds great practical significance in the context of an expanding power grid. Traditional load forecasting methods struggle to meet the demands of power grid regulation. To enhance the predictive performance of the conventional Gradient Boosting Decision Tree (GBDT) algorithm, this paper employs a regression tree model to forecast four types of short-term power loads: predictions using the CART algorithm, predictions based on a single regression model utilizing the GBDT algorithm, power load forecasts based on time series and the GBDT algorithm, and predictions that additionally consider the inclusion of an influencing factor (temperature). The results show that the 37 days average absolute error, absolute deviation mean error and quadratic deviation mean error are reduced by 2.1 %, 38.9 % and 62.33 %, respectively, compared with the single base learning model. The model has good prediction accuracy and stability.