基于PCA-GA算法的贵州省水库标准化管理因子研究

Research on standardized management factors of reservoirs in Guizhou Province based on PCA-GA algorithm

摘要

依赖专家对标对表的创建工作中，存在专家在扣分区间内自由把控扣分程度的现象，针对所造成的因子重要性、所占总分比例、扣分区间一致，但其差异化不一致的问题，利用主成分分析法(PCA)以离散度作为表征指标，重构《贵州省水库标准化管理评价标准》(简写为《评价标准》),建立特征矩阵作为输入变量，利用GA遗传算法改进BP神经网络进行拟合验证。结果表明，《评价标准》因子离散程度越高，越能代表创建实际情况；GA算法通过建立解释目标与解释适应度核函数之间的映射关系，进一步提升了BP神经网络的泛化能力和精准度，改进后的GA-BP神经网络算法模型的拟合准确率达98.78%;重构后的《评价标准》拟合R²高达0.953,较重构前提升了0.107,有着更好的拟合精度。目前从因子离散度角度出发对水利工程达标创建标准解构、重组的研究相对较少，研究结论能够较好的辅助贵州省水库标准化管理创建、后续《评价标准》的修编等工作。

Abstract

In the creation of benchmark tables relying on experts, there exists freely controlling the degree of deduction within the deduction interval. In response to the problem of inconsistent differentiation caused by the importance of factors, the proportion of total scores, and the consistency of deduction intervals, principal component analysis(PCA) is used to reconstruct the “Evaluation Criteria” and establish a feature matrix as input variables. GA genetic algorithm is used to improve the BP neural network for fitting verification. The result indicate that the higher the degree of dispersion of the factors in the Evaluation Criteria, the more representative they are of the actual situation of creation; The GA algorithm further enhances the generalization ability and accuracy of the BP neural network by establishing a mapping relationship between the explanatory target and the explanatory fitness kernel function. The improved GA-BP neural network algorithm model has a fitting accuracy of 98.78%; The fitted R² of the “Evaluation Criteria” after reconstruction is as high as 0.953, which has improved by 0.107 compared to before reconstruction and has better fitting accuracy. At present, there is relatively little research on the deconstruction and reorganization of water conservancy engineering standardization from the perspective of factor dispersion. The research conclusion of this article can effectively assist in the creation of standardized management for reservoirs in Guizhou Province and the subsequent revision of the “Evaluation Criteria”.

关键词

水库标准化 / 主成分分析法 / 离散度 / 遗传算法 / 重构标准

Key words

reservoir standardization / principal component analysis / dispersion / genetic algorithm / refactoring standards

Author summay

倪 康(1993—), 男, 工程师, 硕士, 主要从事水利水电工程研究。E-mail: [e-mail]

基金资助

机理-数据协同驱动下降雨诱发堆积层滑坡变形预测预警关键技术研究(KT202401)