基于多源异构数据融合的短期电力负荷预测

Short-Term Power Load Forecasting Based on Heterogeneous Data from Multiple Sources

为解决多源异构数据集成的难题,设计包含数据集成、数据空间、数据演化、数据输出四个组件的短期电力负荷预测数据空间框架。为了建立多源异构数据与短期电力负荷的复杂变化因果关系,提出一种基于Gabor-LSTM-XGboost两阶段预测模型。第一阶段,构建基于Gabor云图特征提取及编码模型,将非结构化云图进行幅值特征级增强并进行编码融合。第二阶段,建立LSTM-XGBoost预测模型,将综合气象数据、历史负荷数据、日类型数据进行拼接后形成的长序列作为预测模型的输入,引入XGBoost增加正则化,进一步提高模型的鲁棒性。算例结果表明,所提出的模型能够较好地挖掘云图气象数据,有效提升短期电力负荷预测精度,特别是突变天气下的负荷预测精度。

In order to solve the problem of integrating multi-source heterogeneous data,a short-term power load forecasting data space framework was designed,which includes four components:data integration,data space,data evolution and data output.In order to establish the causal relationship between heterogeneous multi-source data and complex changes of short-term power load,a two-stage forecasting model based on Gabor-LSTM-XGBoost was proposed.In the first stage,the Gabor cloud image feature extraction and coding model was constructed to enhance the amplitude feature level of the unstructured cloud image and conduct coding fusion.In the second stage,the LSTMXGBoost prediction model was established,and the long sequence formed after the integration of comprehensive meteorological data,historical load data and daily type data were used as the input of the forecasting model,and XGBoost was introduced to increase regularization and further improve the robustness of the model.The calculation results show that the proposed model can effectively mine cloud map meteorological data and improve the accuracy of short-term power load prediction,especially the accuracy of load prediction under sudden weather.