基于Akima-LMD和GRNN的短期负荷预测

Short-term load forecasting based on Akima-LMD and GRNN

传统局域均值分解(LMD)对极值点采用了滑动平均值处理得到局域均值函数和局域包络函数,易造成分解的分量过平滑而影响精度。为了减小过平滑影响,采用Akima插值法代替滑动平均值法处理局域函数来改进LMD算法,针对电力系统负荷序列的非平稳性和非线性,利用改进LMD算法进行序列分解得到若干分量,再利用广义回归神经网络(GRNN)预测各个分量的趋势,叠加各分量趋势得到负荷序列总趋势。GRNN神经网络较传统神经网络训练速度快、精度高,能很好地预测非线性序列。算例分析表明,改进LMD结合GRNN的方法较经验模态分解(EMD)结合GRNN的方法在短期电力负荷预测中有更高的预测精度。

Conventional local mean decomposition（ LMD） algorithm employs the moving average method to obtain the local mean function and local envelope function for processing the extreme points of original sequence,which readily causes the component over-smoothness and effects the precision of decomposition. In order to reduce the influence of the over-smoothness,Akima interpolation method is used instead of sliding average method to improve LMD algorithm for dealing with the local function. For non-stationary and nonlinear power system load sequence,the sequence firstly decomposes several components by using the improved LMD algorithm. Then by means of generalized regression neural network（ GRNN） method,the trend of each component is forecasted. Finally,these predicted ones can be added to constitute the trend of the original sequence. Compared with the traditional neural network,the GRNN algorithm has the advantages of faster speed in training time and high precision,which can predict the nonlinear sequence well. The analysis of numerical example demonstrates that the improved LMD algorithm with the GRNN method enjoys higher accuracy than the empirical mode decomposition（ EMD） algorithm with GRNN method in the forecasting of the short-term load.