基于D-S证据理论的短期负荷预测模型融合

Short-term load forecast model fusion based on D-S evidence theory

在各种预测模型融合时确定各种模型的权重直接影响到预测精度。对3种不同的神经网络负荷预测模型分别建立了权重提取和权重融合的数学模型,并运用证据理论对3种预测模型的权重进行融合。通过对历史预测数据的分析,提取了证据理论的融合样本,并将信度函数的多重融合结果作为负荷预测模型权重,得到权重融合后待预测日的负荷预测结果。将权重融合模型的预测结果与单一模型的预测结果进行比较,结果表明权重融合后的模型具有更高的预测精度,提高了负荷预测的准确性。

The determination of weights in forecast model fusion affects greatly the forecast precision. The weight extraction model and weight fusion model are established respectively for three neural network load forecast models and fused by using evidence theory. The fusion samples of evidence theory are extracted based on the analysis of historical forecasts and the multi-fusion result of belief function is taken as the weight of load forecast model, by which the day load is forecasted. The load forecasted is compared with those by other independent models, which demonstrates that the model with weight fusion has higher forecast precision.