基于ANFIS改进的大气腐蚀环境缺失数据填补方法

目的针对大气腐蚀中重要环境数据缺失的复杂问题,提出一种相关因素(Relevance Factors)和自适应神经模糊推理系统(Adaptive Neuro-Fuzzy Inference System)结合的方法(RF-ANFIS)对缺失数据进行填补。方法首先采用相关因素方法计算缺失数据和多项环境因素间的相关程度,筛选出相关系数较大的因子,然后应用ANFIS构建缺失数据与所选环境因子的关系模型。最后以二氧化硫数据为具体对象,采用北京2015年的气象数据对所建立的模型进行检验。结果经过改进的RF-ANFIS模型在最优情况下样本均方误差为0.696,在14个测试样本中有13个相对误差在20%以内,针对有限样本的数据分析中更为适用。结论该方法有效提高了大气腐蚀环境数据缺失的填补精度,对在数据缺失情况下预测大气腐蚀速率具有重要意义。

A Imputation Method Based on Improved ANFIS for Environmental Data

Objective To propose a new method (RF-ANFIS) based on relevance factors and Adaptive Neuro-Fuzzy Infe-rence System to impute missing important environmental data on atmospheric corrosion.Methods The relevance degree be-tween missing data and a number of environmental factors was calculated through relevance factors. Factors of high relevance degree were selected; then a relationship model between missing data and environmental factors was built through ANFIS. Fi-nally, SO2 data was taken as the specified object to test the model according to atmospheric data of Beijing in 2015.Results The error of mean square of samples in the improved RF-ANFIS model was 0.696 in the best case. The relative error of 13/14 test samples was within 20%. It was applicable to data analysis of limited samples.Conclusion The new method effectively im-proves the accuracy of imputing environmental data in atmospheric corrosion. It is vital to predict atmospheric corrosion rate with missing data.