基于鲸鱼算法优化反向传播神经网络的中药安慰剂溶液颜色模拟处方预测  

Prediction of color simulation prescription for traditional Chinese medicine placebo solution based on whale algorithm-optimized back propagation neural network

作　　者：张三妹 林晓[1] 洪燕龙[1] 冯怡[1] 吴飞[1] ZHANG San-mei;LIN Xiao;HONG Yan-long;FENG Yi;WU Fei(Engineering Research Center of Modern Preparation Technology of Traditional Chinese Medicine,Ministry of Education,Innovative Research Institute of Traditional Chinese Medicine,Shanghai University of Traditional Chinese Medicine,Shanghai 201203,China)

机构地区：[1]上海中医药大学创新中药研究院,中药现代制剂技术教育部工程研究中心,上海201203

出　　处：《中国中药杂志》2024年第16期4437-4449,共13页China Journal of Chinese Materia Medica

基　　金：上海市卫健委科研基金项目(201940296)。

摘　　要：作为特定对象的模拟制剂,中药安慰剂研制过程中的颜色模拟既是重点又是难点。传统用大量试验进行处方筛选以及规律探索的方式,费时费力,因此颜色模拟处方的准确预测是中药安慰剂研制的方向。该文以图像法结合Matlab软件对安慰剂的颜色模拟处方进行高效精准预测。首先,针对中药安慰剂溶液,拍摄成像后通过Photoshop软件对图像的L*a*b*、RGB、HSV和CMYK 13个色度空间值进行提取,并对其进行相关性分析和归一化处理后构建13×9×3的反向传播(back propagation,BP)神经网络模型,随后利用鲸鱼算法(whale optimization algorithm,WOA)对初始权值和阈值进行优化,最后对优化后的WOA-BP神经网络进行3类代表性实例验证。通过训练预测结果可得,相比BP神经网络,WOA-BP神经网络能较好地对安慰剂色素配比进行预测,训练、验证、测试和全部的相关系数分别为0.95、0.87、0.95和0.95,接近1;各误差值均有所降低,最高可降低99.83%,3组实例验证的色差结果ΔE均小于3,进一步证实了WOA-BP神经网络的准确性与实用性。Traditional Chinese medicine(TCM)placebos are simulated preparations for specific objects and the color simulation in the development of TCM placebos is both crucial and challenging.Traditionally,the prescription screening and pattern exploration process involves extensive experimentation,which is both time-consuming and labor-intensive.Therefore,accurate prediction of color simulation prescriptions holds the key to the development of TCM placebos.In this study,we efficiently and precisely predict the color simulation prescriptions of placebos using an image-based approach combined with Matlab software.Firstly,images of TCM placebo solutions are captured,and 13 chromaticity space values such as the L∗a∗b∗,RGB,HSV,and CMYK values are extracted using Photoshop software.Correlation analysis and normalization are then performed on these extracted values to construct a 13×9×3 back propagation(BP)neural network model.Subsequently,the whale optimization algorithm(WOA)is employed to optimize the initial weights and thresholds of the BP neural network.Finally,the optimized WOA-BP neural network is validated using three representative instances.The training and prediction results indicate that,compared to the BP neural network,the WOA-BP neural network demonstrates superior performance in predicting the pigment ratios of placebos.The correlation coefficients for training,validation,testing,and the overall dataset are 0.95,0.87,0.95,and 0.95,respectively,approaching unity.Furthermore,all error values are reduced,with the maximum reduction reaching 99.83%.The color difference(ΔE)values for the three validation instances are all less than 3,further confirming the accuracy and practicality of the WOA-BP neural network approach.

关 键 词：中药安慰剂 颜色模拟 反向传播(BP)神经网络 色度空间值 鲸鱼算法 

分 类 号：R283.6[医药卫生—中药学]