%0 Journal Article %T 基于二型模糊神经网络的NOx排放预测
NOx Emission Prediction Based on Type-2 Fuzzy Neural Networks %A 王懋譞 %A 李振国 %A 吴撼明 %A 邵元凯 %A 孟令伟 %A 刘强 %A 张旺 %J Operations Research and Fuzziology %P 427-439 %@ 2163-1530 %D 2024 %I Hans Publishing %R 10.12677/orf.2024.144412 %X 为提高柴油机NOx排放预测精度及减少试验成本，本文提出了二型模糊神经网络(Type-2 Fuzzy Neural Networks, T2FNNs)的柴油机NOx排放预测方法。首先利用随机森林算法进行了输入变量的重要性分析并约简优化了输入变量，然后对本文提出T2FNNs预测模型的参数分别采用梯度下降法和递推最小二乘进行训练寻优，基于寻优后的最优预测模型进行了训练和验证，并取得了较好的预测效果。在训练集和验证集上NOx的回归决定系数(R2)分别为0.985和0.962。最后对T2FNNs预测模型进行了泛化性测试，并与其他神经网络预测模型进行了对比。结果表明在测试集上其R2值为0.961，相较于训练集和验证集仅分别下降了2.4%和0.1%，说明T2FNNs预测模型具有极强的泛化能力。在与其他预测模型的对比中，T2FNNs预测模型也表现出了更好的性能特性。
To improve the accuracy of diesel engine NOx emission prediction and reduce the cost of testing, a Type-2 Fuzzy Neural Networks (T2FNNs) method for predicting diesel engine emission characteristics was proposed. Firstly, the importance analysis of input variables was conducted using the random forest algorithm, and the input variables were reduced and optimized. Then, the parameters of the T2FNNs prediction model proposed in this paper were trained and optimized using gradient descent method and recursive least squares. Based on the optimized optimal prediction model, training and validation were conducted, and good prediction results were achieved. The regression determination coefficients (R2) of NOx on the training and validation sets are 0.985 and 0.962, respectively. Finally, a generalization test was conducted on the T2FNNs prediction model, and it was compared with other neural network prediction models. The results showed that the R2 value on the test set was 0.961, which decreased by only 2.4% and 0.1% compared to the training and validation sets, respectively, indicating that the T2FNNs prediction model has strong generalization ability. In comparison with other prediction models, the T2FNNs prediction model also showed better performance characteristics. %K 柴油机， %K 排放预测， %K 二型模糊神经网络， %K 参数训练
Diesel Engine %K Emission Prediction %K Type-2 Fuzzy Neural Networks %K Parameter Training %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=94286