基于气象累积和ACA-GRNN的短期电力负荷预测

Short-term electric load forecasting based on accumulated weather and ACA-GRNN

针对气象条件具有累积效应以及不同气象条件对负荷影响的程度不同的特点,采用一加权的几何距离公式来选取神经网络的训练样本,不仅加快了神经网络的训练速度,而且加强了神经网络的逼近能力。同传统的神经网络相比,广义回归神经网络的训练过程实际上是不断地调整平滑参数σ的过程,因此,σ的不同取值对网络的输出具有重要的影响。在优化广义回归神经网络的平滑参数σ时,采用基于蚁群种群的新型优化算法——蚁群算法来优化,在很大程度上减少了人为选择参数的主观影响。最后通过实例验证了该模型的有效性。

According to the features of accumulated effects of weather condition and different effects for load with different weather condition, a geometric distance formula with different weights is adopted in this paper, which can speed the training and strengthen the approximation of neural network. The training process of general regression neural network （GRNN） is in fact to adjust the smooth parameter σ compared with the traditional neural network, so that different values of σ have major impact on outputs of network. In the course of optimizing smooth parameter σ of GRNN, a new type optimization algorithm-Ant Colony Algorithm （ACA） based on population of ant colony is employed, which can reduce the subjective effects to a large extent in the course of selecting the parameters. Finally, an example is given to show the effectiveness of the model.