基于自组织映射−前馈神经网络和先知混合模型的短期负荷预测

Short-Term Load Forecasting Based on SOM-BP and Prophet Hybrid Model

  • 摘要: 为提高电力系统短期负荷预测精度,充分挖掘历史数据中的多维度信息,更好地克服历史数据缺失带来的不利影响,提出一种基于自组织映射-前馈神经网络和先知混合模型的短期负荷预测方法。首先通过SOM神经网络对历史非功率数据聚类计算得到相似日集合,而后采用相似日数据对BP神经网络进行训练得到单点负荷值预测结果。其次,重点考虑历史数据的周期性和时序变化趋势,基于Prophet时序模型对历史负荷数据进行周期非线性拟合。通过历史数据拟合误差反馈,调整优化模型的关键超参数,最后基于误差倒数法组合得到短期负荷预测结果。以某地区电力负荷数据作为算例验证,结果表明所提的改进预测模型预测精度更高,且在克服历史数据缺失和拟合非工作日负荷曲线等方面具有优势。

     

    Abstract: To improve the accuracy of short-term load forecasting in power system, fully exploit the multi-dimensional information in the historical data to better overcome the adverse effects caused by the lack of the historical data, a short-term load forecasting method based on the self- organizing maps and back propagation (abbr. SOM-BP) and Prophet hybrid model was proposed. Firstly, the similar day set was obtained by clustering the historical non-power data through SOM neural network, and then the BP neural network was trained with the similar day data to obtain the single point load value prediction results. Secondly, focusing on the periodicity and temporal trends of historical data, the Prophet temporal model was used to perform periodic nonlinear fitting on historical load data. Through historical data fitting error feedback, the key hyperparameters of the optimization model was adjust, and finally the short-term load forecasting results based on the combination of error reciprocal method were obtained. Finally, Taking the power load data of a certain region as an example for verification, the results show that the proposed improved prediction model has higher prediction accuracy and advantages in overcoming historical data deficiencies and fitting non- working day load curves.

     

/

返回文章
返回