针对农杆菌ATCC31749发酵法产凝胶多糖过程中产物质量浓度预测精度不高问题,提出一种基于模糊加权最小二乘支持向量机(least squares support vector machine,LSSVM)算法和机理模型相结合的混合建模新方法。首先通过添加模糊加权思想和混合核函数方法对LSSVM算法进行优化,并用优化后的LSSVM求解农杆菌ATCC31749发酵过程动力学模型,结合鸟群算法对动力学模型参数进行寻优;然后拟合出溶氧体积分数和各参数之间的关联函数模型,并代入到动力学模型,建立起以溶氧浓度作为关键控制变量的发酵动力学模型;最后,用鸟群算法对模型进行寻优,寻找使得发酵产物浓度最大的最优溶氧过程控制策略。实验仿真结果表明,混合模型的预测精度得到提高,产多糖期溶氧体积分数控制为52%时,产物质量浓度最大,为48.85 g/L。该研究所建立的农杆菌发酵过程混合模型及其溶氧优化结果,为发酵工业上进一步通过最佳溶氧控制策略来提高多糖产量提供了方向。
To solve the low prediction accuracy of curdlan concentration during Agrobacterium sp. fermentation, a new hybrid model based on fuzzy weighted LSSVM (least squares support vector machine) algorithm and a mechanism model was proposed. Firstly, the LSSVM algorithm was optimized by adding fuzzy weighted idea and using a mixed kernel function. The optimized LSSVM was used to solve the kinetic model of fermentation process of Agrobacterium sp. Bird swarm algorithm (BSA) was used to optimize the kinetic model parameters. The correlation functioned model between dissolved oxygen tension and various parameters was summarized and substituted into the kinetic model. A hybrid fermentation kinetic model that used dissolved oxygen tension as a key control variable was established. Finally, the BSA was used to find a controlling strategy for the optimal process of dissolving oxygen that maximized curdlan concentration. The results demonstrated that the prediction accuracy of hybrid model improved. When the dissolved oxygen concentration in the polysaccharide-producing period was 52%, the curdlan concentration reached the highest (48.85 g/L). The hybrid model of Agrobacterium fermentation process established in this study together with the results of optimized dissolved oxygen provided a direction for fermentation industries to further improve the yield of polysaccharides through optimal dissolved oxygen controlling strategy.
[1] ZHANG Hongtao,ZHU Li,LIU Dengfeng,et al. Model-based estimation of optimal dissolved oxygen profile in Agrobacterium sp.fed-batch fermentation for improvement of curdlan production under nitrogen-limited condition[J]. Biochemical Engineering Journal, 2015, 103: 12-21.
[2] LEE J H, LEE I Y, KIM M K, et al. Optimal pH control of batch processes for production of curdlan by Agrobacterium species[J]. Journal of Industrial Microbiology and Biotechnology, 1999, 23(2):143-148.
[3] SHIN H D, LIU LONG, KIM M K, et al. Metabolic engineering of Agrobacterium sp. ATCC31749 for curdlan production from cellobiose[J]. Journal of Industrial Microbiology & Biotechnology, 2016, 43(9):1 323-1 331.
[4] YU Xiaoqin, ZHANG Chao,YANG Liping, et al. CrdR function in a curdlan-producing Agrobacterium sp. ATCC31749 strain[J]. BMC Microbiology,2015, 15(1):25.
[5] WEST T P. Pyrimidine base supplementation effects curdlan production in Agrobacterium sp. ATCC 31749[J]. Journal of Basic Microbiology, 2010, 46(2):153-157.
[6] WEST T P, NEMMERS B. Curdlan production by Agrobacterium sp. ATCC 31749 on an ethanol fermentation coproduct[J]. Journal of Basic Microbiology, 2008, 48(1):65-68.
[7] 岳国君, 刘文信,刘劲松,等. Logistic模型模拟乙醇发酵产物动力学[J]. 农业工程学报, 2015, 31(5):280-286.
[8] 于晓萌,张京良,孙永超,等.基于发酵优化和动力学建立Levan果聚糖生产的数字化模型[J]. 食品与发酵工业, 2016,42(12):20-25.
[9] 徐莹, 何国庆. 弹性蛋白酶分批发酵动力学模型的建立[J]. 农业生物技术学报, 2007, 15(1): 138-141.
[10] GERNAEY K V, LANTZ A E, TUFVESSON P, et al. Application of mechanistic models to fermentation and biocatalysis for next-generation processes[J]. Trends in Biotechnology, 2010, 28(7):346-354.
[11] 贺晓冉,陈宸,金光石,等. 青霉素发酵过程的模型仿真与补料优化[J]. 化工学报, 2012, 63(9): 2 831-2 835.
[12] 李浩, 任嘉红,叶建仁,等. 吡咯伯克霍尔德氏菌JK-SH007分批发酵动力学模型的构建[J]. 南京林业大学学报(自然科学版), 2016, 40(2):173-178.
[13] 花亚梅,赵贤林,王效华,等. 基于改进BP神经网络的厌氧发酵产气量预测模型[J]. 环境工程学报, 2016, 10(10):5 951-5 956.
[14] 杨小梅, 刘文琦,杨俊. 基于分阶段的LSSVM发酵过程建模[J]. 化工学报, 2013, 64(9): 3 262-3 269.
[15] 朱湘临, 华天争. 基于混沌果蝇优化最小二乘支持向量机的秸秆发酵过程软测量建模[J]. 北京工业大学学报, 2016, 42(10):1 468-1 474.
[16] 沙如意, 楼坚,蔡成岗,等. 基于遗传算法优化BP神经网络的丙酮丁醇梭菌发酵预测模型研究[J]. 发酵科技通讯, 2016, 45(1): 23-26.
[17] 周梦舟, 吴珊,柳念,等. 基于神经网络对枯草芽孢杆菌富硒过程的建模研究[J]. 中国食品学报, 2016,(12):66-74.
[18] 公彦杰. 基于LS_SVM建立发酵过程动态模型的研究及软件实现[D]. 北京:北京工业大学, 2010.
[19] GHOVVATI M, KHAYATI G, ATTAR H, et al. Comparison across growth kinetic models of alkaline protease production in batch and fed-batch fermentation using hybrid genetic algorithm and particle swarm optimization[J]. Biotechnology & Biotechnological Equipment, 2015, 29(6):1 216-1 225.
[20] 董亚明. 融合机理模型与智能模型的复杂过程混合建模研究[D]. 上海:华东理工大学, 2015.
[21] 桑海峰. 基于机理知识与最小二乘支持向量机的诺西肽发酵过程混合建模方法[J]. 系统仿真学报, 2008, 20(2): 468-472.
[22] 黎兴宝. 基于机理知识与改进LS_SVM的谷氨酸发酵混合建模[D].无锡:江南大学, 2011.
[23] 徐飞.基于混合核函数的LSSVM发酵建模[D].大连:大连理工大学,2012.
[24] 刘春卫, 罗健旭. 基于混合核函数的PSO-SVM分类算法[J]. 华东理工大学学报(自然科学版), 2014, 40(1):96-101.
[25] 王立达.基于混合核函数的SVM及其应用研究[D].大连:大连海事大学,2016.
[26] 伍铁斌,刘云连,李新君,等.KPCA_模糊加权LSSVM预测方法及其应用[J].计算机测量与控制,2012,20(3):617-620.
[27] MENG Xianbing,GAO XZ,LU Lihua,et al. A new bio-inspired optimisation algorithm:Bird swarm algorithm[J]. Journal of Experimental & Theoretical Artificial Intelligence, 2015, 28(4): 673-687.
[28] 曾嶒,彭春华,王奎,等. 基于鸟群算法的微电网多目标运行优化[J]. 电力系统保护与控制, 2016, 44(13):117-122.
