基于融合特征值提取的配电网工程数据分析算法

Analysis algorithm of distribution network engineering data based on fusion eigenvalue extraction

针对现有算法对配电网工程数据预测精度较低的问题,文中对数据特征融合与多元数据分析模型进行了研究。为了消除原始配电网数据仓库中的噪声并降低冗余度,引入了一种主成分分析算法(PCA)。该方法将原本相关的特征重新组合为互不相关的特征,以缩短后续数据分析的时间开销,从而避免了维数灾难。在进行造价预测时,利用特征融合后的实际工程数据对回声状态网络(ESN)加以训练,且在网络中加入了一种储备池结构来替代传统神经网络中的神经元,进而提升了网络的收敛速度。仿真结果表明,所提模型可以在降低数据维度的同时,全面地表征配电网工程的特性。且该算法对配网工程造价数据的预测精度较现有主流算法也有显著提升,与BP神经网络相比,其NMSE和MAPE分别提升了19.60%与2.76%。

Aiming at the problem of low prediction accuracy of existing algorithms for distribution network engineering data,data feature fusion and multivariate data analysis models are studied in this paper.In order to eliminate the noise in the original distribution network data warehouse and reduce the data redundancy,a Principal Component Analysis algorithm(PCA)is introduced.This method combines the original related features and fuses them into unrelated features,which shortens the time cost of subsequent data analysis and avoids the dimension disaster.When making cost prediction,the Echo State Network(ESN)is trained by using the actual engineering data after feature fusion.This network introduces a reserve pool structure to replace the neurons in the traditional neural network,thus improving the convergence speed of the network.The simulation results show that the model established in this paper can reduce the data dimension and fully reflect the characteristics of the acquisition and distribution network project.At the same time,the prediction accuracy of the algorithm for the distribution network project cost data has been significantly improved compared with the existing mainstream algorithms.Taking BP neural network as an example,the NMSE and MAPE of the algorithm have been improved by 19.60%and 2.76%respectively.