人工神经网络优化油莎豆油亚临界萃取工艺  

Optimization of subcritical butane extraction for tiger nut oil based on artificial neural network coupled with PSO

作　　者：邓淑君 郝琴 万楚筠[1,3] 郭婷婷 魏春磊[1,3] 郑明明 DENG Shu-jun;HAO Qin;WAN Chu-yun;GUO Ting-ting;WEI Chun-lei;ZHENG Ming-ming(Oil Crops Research Institute of the Chinese Academy of Agricultural Sciences,Wuhan 430062,China;Wuhan Polytechnic University,Wuhan 430023,China;Oil Crops and Lipids Process Technology National&Local Joint Engineering Laboratory,Wuhan 430062,China)

机构地区：[1]中国农业科学院油料作物研究所,湖北武汉430062 [2]武汉轻工大学,湖北武汉430023 [3]油料油脂加工技术国家地方联合工程实验室,湖北武汉430062

出　　处：《中国油料作物学报》2024年第5期1178-1186,共9页Chinese Journal of Oil Crop Sciences

基　　金：中国农业科学院科技创新工程(CAAS-ASTIP-2021-OCRI);支持企业技术创新发展项目(2021BLB151)。

摘　　要：为优化亚临界丁烷萃取脱皮油莎豆油工艺,采用单因素试验确定因素水平,中心复合表面设计(CCF)安排寻优试验,在此基础上分别构建了响应面(RSM)和反向传播人工神经网络(BP-ANN)模型,运用粒子群算法(PSO)对BP-ANN模型进行优化,并对RSM和PSO-BP-ANN模型的寻优结果进行了比较。结果表明,RSM模型优化的萃取条件为:料液比(脱皮油莎豆∶丁烷)1∶10.36 g/mL、萃取时间45 min、萃取温度30℃、坯料厚度0.5 mm;PSOBP-ANN模型优化的萃取条件为:料液比1∶10.67 g/mL、萃取时间40.10 min、萃取温度34℃、轧坯厚度0.5 mm。在最佳条件下,RSM模型预测提取率为91.63%,验证值为94.27%,相对误差2.56%;PSO-BP-ANN模型预测值为95.58%,验证值为95.14%,相对误差0.46%。采用人工神经网络耦合粒子群算法(PSO-BP-ANN)优化油莎豆油亚临界萃取工艺,具有提取率高、相对误差小等优势。本研究可为亚临界萃取技术在油莎豆油高效制取中应用提供参考。In order to optimize the subcritical butane extraction process for dehulled tiger nut oil,single factor experiment was taken to determine the levels of the factor,central composite face-centered design(CCF)was used to optimize the subcritical extraction conditions,based on which response surface methodology(RSM)and back propagation artificial neural network(BP-ANN)models were constructed,respectively.The BP-ANN was optimized by particle swarm optimization(PSO),and the optimization results of the RSM model and PSO-BP-ANN model were compared.The optimal extraction conditions optimized by RSM and PSO-BP-ANN models were as follows:solid-liquid ratio(dehulled tiger nut:butane)was 1:10.36 g/mL,incubation time for 45 min,extraction temperature was 30℃,the rolling thickness was 0.5 mm;the solid-liquid ratio was 1:10.67 g/mL,the extraction time was 40.10 min,the extraction temperature was 34℃,and the thickness of the rolled billet was 0.5 mm.Under the optimal con-ditions,the predicted extraction rate of the RSM model was 91.63%,the experimental result was 94.27%,and the relative error was 2.56%.The prediction value of the PSO-BP-ANN model was 95.58%,the validation value was 95.14%,and the relative error was 0.46%.The artificial neural network coupled particle swam optimization(PSO-BP-ANN)was used to optimize the subcritical extraction process of tiger nut oil,which had advantages of high ex-traction rate and small error.This study can provide a reference for the application of subcritical extraction technolo-gy in the efficient production of tiger nut oil.

关 键 词：反向传播人工神经网络 粒子群优化算法 亚临界丁烷萃取 脱皮油莎豆 工艺优化 

分 类 号：TS224.4[轻工技术与工程—粮食、油脂及植物蛋白工程]