基于BiLSTM-Attention-ERM层级修正模型的配电网母线净负荷预测OA

To address the issue of distributed photovoltaic output prediction deviation caused by the lack of measured meteorological data,which in turn affects the accuracy of bus net load forecasting,a meth-od that combines PV output conversion and hierarchical forecasting correction is proposed for distribution network bus net load forecasting.First,a reference power station with high correlation to the distributed PV output is selected.Weather types are determined by combining meteorological data with a clustering al-gorithm,and the optimal conversion coefficient is calculated based on historical output ratios.Next,a bidi-rectional long short-term memory(BiLSTM)-Attention model is constructed to directly forecast the PV output,bus load,and net load of the reference power station.The distributed PV output prediction result is then derived using conversion coefficients.Finally,a comprehensive correction model is built by integrating hierarchical prediction and empirical risk minimization(ERM).The direct prediction results are fed into this model to correct the net load forecast.Case studies using real data from southern China show that the proposed method significantly improves the accuracy of 10 kV bus net load forecasting in distribution net-works lacking precise meteorological data,compared to methods without conversion and correction.