基于用户群体划分的多步短期负荷预测方法

Multi-step Short-term Load Forecasting Method Based on User Group Division

新型电力系统中负荷的多样性和随机性特征日趋明显,同时单步负荷预测需要参考一阶滞后特征,无法提前预测多个时间步的观测值,因此需要提出适用于负荷新特征的多步负荷预测方法。为此,提出基于用户群体划分的多步短期负荷预测方法。首先,使用最大相关最小冗余准则从影响用户负荷的外界因素中选取典型特征,并依据每位用户负荷与典型特征的相关系数对用户群体进行划分;然后为每个用户群体构建全局注意力模型,并在模型解码端迭代输入预测时刻已知信息进行辅助预测;最后,将各模型的输出在对应时间步上进行求和,从而实现对整个用户群体负荷的多步准确预测。以公开数据集进行算例分析的结果表明,所提方法在预测准确性和稳定性方面有较大优势。

The diversity and stochastic characteristics of loads in new power systems are becoming increasingly obvious,meanwhile,it is required that single-step load forecasting should make reference to first-order lag characteristics,which cannot predict the observations of multiple time steps in advance.Therefore,it is necessary to propose a multi-step load forecasting method which is applicable to the new characteristics of loads.In this paper,a multi-step short-term load forecasting method based on user group division is proposed.First,typical characteristics are selected from the external factors affecting user load by using the maximum correlation minimum redundancy criterion,and user groups are divided based on the correlation coefficient between each user load and the typical characteristics.Then,a global attention model is constructed for each user group,and the information known at the moment of iteratively inputting on the decoding side of the model is complementarily forecast.Finally,the output of each model is summed to achieve an accurate multi-step forecast of the load of the entire user group.The results of the case study with a publicly available dataset show that the proposed method has greater advantages in forecasting accuracy and stability.