Mid-Long-Term Power Load Forecasting of Building Group Based on Modified NGO

The mid-long-term forecasting of load in existing building clusters has given relatively little consideration to the prediction of fixed power loads that do not actively participate in renewable energy consumption, which may lead to certain errors in the forecasting results of active renewable energy-consuming loads. Based on power supply dependency, this paper categorizes building electrical loads into fixed loads and those capable of actively consuming renewable energy. Following this categorization, a Modified Northern Goshawk Optimization algorithm (MNGO) is utilized to optimize the XGBoost model, ultimately establishing a mid-long-term load forecasting algorithm tailored for building groups. Initially, a Random Forest (RF) algorithm is deployed to filter the key feature factors influencing the accuracy of load forecasting. Secondly, the Northern Goshawk Optimization (NGO) algorithm is modified to optimize the XGBoost model for the electric load forecasting of building groups. A comparative analysis of the forecasting outcomes reveals that the XGBoost model, refined by the NGO algorithm, significantly diminishes the Mean Absolute Percentage Error (MAPE) and markedly escalates the coefficient of determination (R 2 ), thereby validating the efficacy of the proposed methodology. This approach not only furnishes data support for energy storage planning and ameliorates the capacity for new energy assimilation, but also ensures a stable power supply for buildings reliant on fixed electrical loads.