大气腐蚀数据降维最优维度研究

目的确定金属大气腐蚀数据降维的最优维度。方法分别采用PCA、MDS、Isomap和LLE四种方法对大气腐蚀数据进行降维处理,并采用集成学习算法建立预测模型。针对不同的降维方法和近邻点个数计算,使用MAPE(Mean Absolute Percentage Error,相对百分误差绝对值的平均值)对预测结果进行评价,将最佳预测速率所对应的维度作为最优维度。结果不同的降维方法和近邻参数作用下,最优维度从2维到7维不等。流形学习方法对大气腐蚀数据进行降维的MAPE均小于线性降维方法。结论适合每种降维方法的最优维度可能是不同的,通过对MAPE进行对比,得到大气腐蚀数据在各种降维方法的最优维度。

Optimal Dimension of Dimensionality Reduction of Atmospheric Corrosion Data

The work aims to determine the optimal dimension for the dimensionality reduction of metals'' atmospheric corrosion data. The four methods such as PCA, MDS, Isomap and LLE were used for the dimensionality reduction of atmospheric corrosion data, and an ensemble learning algorithm was used to establish the prediction model. For different dimensionality reduction methods and the calculation of the number of neighbors, the mean absolute percentage error (MAPE) was used to evaluate the prediction results, and the dimension corresponding to the best prediction rate was used as the optimal dimension. Under the action of different dimensionality reduction methods and the neighbor parameters, the optimal dimension ranged from 2 to 7 dimensions. Manifold learning method was used for the dimensionality reduction of atmospheric corrosion data, and the resulting MAPE was less than that of the linear dimensionality reduction method. The optimal dimension for each dimensionality reduction method may be different. Finally, the optimal dimension of the atmospheric corrosion data processed by the four dimensionality reduction methods is obtained through the comparison of the MAPE values.