考虑最小平均包络熵负荷分解的最优Bagging集成超短期多元负荷预测

Optimal Bagging Ensemble Ultra Short Term Multi-energy Load Forecasting Considering Least Average Envelope Entropy Load Decomposition

多元负荷预测技术是保证综合能源系统(integrated energy system,IES)供需平衡与稳定运行的关键基石。但具有强随机性与波动性的IES负荷加剧了超短期多元负荷准确预测的难度。为此,提出考虑最小平均包络熵负荷分解的最优Bagging集成超短期多元负荷预测方法。构建基于最小平均包络熵的变分模态分解参数优化模型,将IES多元负荷分解为本征模态分量集合;基于统一信息系数法筛选多元负荷预测的日历、气象与负荷强相关特征;结合负荷本征模态分量集合、日历规则、气象环境与负荷数据,构建Bagging集成超短期多元负荷预测模型,并建立基于平均绝对百分比误差与决定系数的集成策略优化模型,进而得到最优集成策略与最终预测结果。以美国亚利桑那州立大学坦佩校区IES为对象展开仿真验证,结果表明,所提方法的电、热、冷负荷预测平均绝对百分比误差分别为1.9486%、2.0585%、2.5331%,相比其他预测方法具有更高准确率。

Multi-energy load forecasting technology is the key cornerstone to ensure the supply and demand balance and stable operation of integrated energy system(IES).However,IES load with strong randomness and volatility aggravates the difficulty of accurate ultra short term multi-energy load forecast.Therefore,the optimal Bagging ensemble ultra short term multi-energy load forecasting method considering least average envelope entropy load decomposition is proposed.The parameters optimization model of variational mode decomposition based on least average envelope entropy is constructed,and the multi-energy load of IES is decomposed into the set of intrinsic mode functions;the strong correlation characteristic of calendar,weather and load of multi-energy load forecasting are filtered based on the uniform information coefficient method.Combined with the IMFs set of load,calendar rules,meteorological environment and load data,the Bagging ensemble ultra short term multi-energy load forecasting model is constructed,the ensemble strategy optimization model is constructed based on the mean absolute percentage error and R-square,and then the optimal ensemble strategy and final forecast results are also obtained.Simulation verification is carried out with IES of Arizona State University Tempe Campus as the object.The results show that the mean absolute percentage error of the proposed method in electric,heat and cooling load forecasting is 1.948 6%,2.058 5% and 2.5331%,respectively,which has higher accuracy than other forecast methods.