区域级电力负荷数据特性研究与短期预测

Data Characteristics and Short-term Forecasting of Regional Power Load

为提高区域级负荷预测的精准度,通过对电力负荷数据特性的研究,建立了基于Attention机制的GRU(gate recurrent unit)与Catboost(gradient boosting and categorical features)混合预测模型。利用低层次信息(智能电表数据)建模去解决高层次(区域级负荷)问题,提高了模型的预测精准度。结合MAPE-RW(mean absolute percentage error-reciprocal weight)算法将模型融合并搜索最佳权重,构建了最佳混合预测模型。该思想框架在澳大利亚实际住宅智能电表数据上进行了测试,并使用不同的基准与方法对模型性能进行了比较。结果表明,所提出的针对区域级不同时段负荷的预测模型具有较强的兼容性,且预测精度最高达到95.56%。

A hybrid prediction model of GRU(Gate Recurrent Unit) model based on the Attention mechanism and the Catboost is proposed for improving the accuracy of regional load forecasting. Firstly, based on the full analysis of power load data, the corresponding GRU prediction model under the Attention mechanism and the Catboost prediction model are established. Secondly, the low-level information(the smart meter data) is modeled to solve the high-level(the regional level load) problem in order to improve the accuracy of the model. Finally, combined with the MAPE-RW(mean absolute percentage error recommended weight) algorithm, the above models are fused and the best weight is searched to build the best hybrid prediction model. This idea framework is tested on the smart meter data of the actual residential buildings in Australia, and the model performance is compared with different benchmarks and methods. The results show that the proposed load forecasting model has a strong compatibility with the regional level load forecasting model in different periods, and the forecasting accuracy reaches 95.56%.