中原工学院学报

为提升织物染色配方的智能预测效果，采用反射光谱配色方法，以染色织物的分光反射率R(λ)作为输入数据，以染料配方为输出数据，分别构建了基于深度学习的卷积神经网络(CNN)和灰狼优化算法(GWO)优化的卷积神经网络(GWO-CNN)的智能配色模型。为全面评估模型性能，选取拟合度R²和实际配方与预测配方的平均绝对误差作为核心评价指标，模型训练与验证结果表明：GWO-CNN配色模型的拟合度R²为0.991,优于单一CNN配色模型的0.964;在对3种染料配方的预测中，GWO-CNN模型的预测结果与真实值之间的平均绝对误差分别为0.006、0.004和0.004,小于或等于CNN模型的0.007、0.005和0.004。由以上结果可以得出以下结论：采用GWO-CNN配色模型可有效提升织物染色配方的预测效果。

To improve the intelligent prediction of fabric dyeing formulas, this paper adopts the reflection spectrum color matching method, taking the spectral reflectance R(λ) of dyed fabrics as input data and the dye formulas as output data, and constructs intelligent color matching models based on the deep learning model Convolutional Neural Network(CNN) and the Convolutional Neural Network optimized by the Grey Wolf Optimizer(GWO-CNN) respectively. To comprehensively evaluate the models' performance, the coefficient of determination(R²) and the mean absolute error(MAE) between predicted and actual dye formulas were selected as key metrics. The training and validation results demonstrate that the GWO-CNN model achieved an R² value of 0.991, outperforming the standalone CNN model's 0.964. In predicting the concentrations of three dyes, the GWO-CNN model exhibited MAEs of 0.006, 0.004, and 0.004 respectively, all lower than the CNN model's 0.007, 0.005, and 0.004. The research results demonstrate that the application of the GWO-CNN color matching model can effectively enhance the prediction of fabric dyeing formulas.