蓝藻水华预报模型及基于遗传算法的参数优化

蓝藻水华预报是应对水危机,保障水资源供给的一项重要工作。以太湖北部三湾(竺山湖、梅梁湾、贡湖)为研究对象,采用动态空间环境建模技术,构建了蓝藻水华预报模型,并通过实地观测建立了模拟的初始参数集。利用2008年04-09月太湖水环境、气象等实测数据,采用遗传算法优化叶绿素a浓度预报模型中敏感度较高的4个参数。研究结果表明,该模型在蓝藻水华空间分布的预报上达到了一定的精度;采用遗传算法能全面、高效地进行参数优化,降低了模拟结果的相对残差,提高了模型预报精度。     

基于环境流体动力学模型的浅水草藻型湖泊水质数值模拟

针对内蒙古乌梁素海面临的污染现状,基于三维环境流体动力学模型(EFDC)的计算模式,将其与CE-QUAL-ICM模型耦合,模拟了不同情景下的乌梁素海藻类、总氮、总磷、化学需氧量年际和季节变化规律,模型中不仅考虑了风速和蒸发对模型的影响,也加入了挺水植物密度、高度、直径等形态指标,以此反映水生植物存在的区域中植物吸收降解污染物质、风速对底部应力、流场变化等因素对模型模拟结果的影响。研究结果表明:考虑挺水植物分布的耦合模型能够很好的模拟藻类和污染物质在乌梁素海内的年际、季节变化过程,模拟结果更接近于实际,模拟值与实测值间的相对误差基本控制在30%以内,大部分相对误差已控制在20%以内;降低入湖污染物质负荷直接影响着湖区内水质浓度的变化。另外,对于浅水或挺水植物密集而无法进入取得信息的区域,利用所建耦合模型能够为这些区域湖泊规划、管理、修复提供依据。     

基于遗传算法的渔情预报HSI建模与智能优化

鱼类栖息地适宜性指数模型(HSI)基于鱼类分布与海洋环境之间存在的非线性关系而构建。然而,海洋环境因子之间存在着传统方法无法消除的相关性,导致获取的HSI参数较难准确表达环境因子与渔场之间的复杂关系。基于遗传算法(GA),自动消除海洋环境因子之间的相关性,构建了一种通用的鱼类HSI建模与智能优化框架(GeneHSI)。GeneHSI框架的核心是HSI建模空间向遗传算法空间的映射以及GA适应度函数的构建。该函数构建的思想是HSI预测的渔场概率与商业捕捞获取的渔场概率之间的累计误差值达到最小化。GeneHSI由待解问题构建、GA初始化和GA优化策略3部分组成。利用随机生成的标准化海洋环境数据与渔场概率数据,验证了GeneHSI模型框架的有效性。研究表明,GeneHSI能够有效优化HSI的建模并能自动获取HSI参数。不同限制条件下,遗传算法获取的HSI具有较大的差异,其中一般优化策略下获取的HSI参数最差;不等式、等式和上下界条件下,GeneHSI优化过程显著地更加合理,因此获取的HSI参数也更准确。此外,100、1000、5000和10000样本量下的优化建模表明,GeneHSI具有处理海量样本数据的能力。     

基于PCA和SOM模型的龙感湖水质时空动态研究

为评估湖泊渔业模式转型阶段水环境的时空动态,选择长江中下游典型湖泊龙感湖为研究地点,于2017—2018年对该湖的黄梅水域和宿松水域进行周年季度水质监测,通过主成分分析(PCA)和自组织特征映射人工神经网络(SOM)模型定量分析了水体理化参数的时空变化特征,采用综合营养状态指数法(TLI)对水体富营养化状况进行了评价。PCA分析结果表明,龙感湖宿松水域和黄梅水域的水质差异较小,季节动态明显。全湖氨氮夏季平均浓度高达0.64 mg/L;总氮夏季平均浓度为2.30 mg/L,冬季平均浓度为1.04 mg/L;叶绿素a夏季平均含量达95.28μg/L,秋季平均浓度为28.30μg/L; pH夏季最高,达9.27;总磷冬季最高,平均为0.22 mg/L; TLI指数表明龙感湖除秋季属于轻度富营养水体外,其他3个季节均属于中度富营养状态。SOM模型结果具有可视化强的优点,能够更清晰和直观地反映龙感湖水质的时空分布动态。围栏拆除和禁渔等管理措施有助于湖泊渔业环境修复和资源恢复,建议对渔业模式转型后的湖泊生态系统变化进行长期跟踪监测评估。     

基于GWR模型的东江水质空间分异与水生态功能分区验证

流域水生态功能分区研究是我国正在开展的一项重要工作,如何验证分区结果的合理性,是当前亟待解决的问题。采用地理加权回归(GWR)模型评估流域特征对东江水质的影响,验证水质及流域影响空间差异是否与一二级水生态功能分区结果吻合,并对比了GWR模型与普通最小二乘(OLS)模型性能,讨论了GWR在分区验证方面的应用价值及不足。结果显示:1)水质指标以及GWR模型局部解释率(Local R~2)均在一二级水生态功能分区间存在显著差异;2)相比OLS模型,GWR模型校正R~2更高,残差空间自相关指数Moran′s I更低。研究表明东江水生态功能分区结果能合理反映水陆耦合关系,有效解释水质空间差异。此外建议选择总氮(TN)而非溶解氧(DO)和总磷(TP)作为分区验证指标。GWR模型在分区结果验证中具有广泛应用前景。降低数据空间自相关影响及改善距离测度方法是未来GWR模型研究的难点问题。     

基于东北温带落叶阔叶林通量数据的BEPS 模型参数优化

控制其他参数为经验常数,利用迭代方法对主要光合作用参数最大羧化速率(V_{c max}) 及最大电子传递速率(J_max)进行不同数值组合,将得到的多组模拟结果的逐日总初级生产力(GPP)分别与东北帽儿山落叶阔叶林的通量观测数据进行比较,实现对小时步长BEPSHourly模型V_{c max}和J_max的参数优化.结果表明: 对于东北温带落叶阔叶林,当V_{c max}为41.1 μmol·m^-2·s^-1、J_max 为82.8 μmol·m^-2·s^-1时,模拟的2011年逐日GPP与观测数据比较的均方根误差(RMSE)最小,为1.10 g C·m^-2·d^-1,R²最高,为0.95.经过光合作用参数V_{c max}和J_max优化后,BEPSHourly模型能更好地模拟GPP的季节变化.     

基于FLUXNET观测数据与VPM模型的森林生态系统光合作用关键参数优化及验证

生态系统总初级生产力(GPP)是全球生态系统碳循环研究的重要组成部分.植被最大光能利用率(ε_max)是陆地生态系统GPP模拟的关键参数.本文基于植被光合模型(VPM)和全球通量网(FLUXNET)40个站点(179条站点年数据)的涡度相关通量观测数据,采用单因素轮换法对VPM模型进行参数敏感性分析,并利用交叉验证法对全球森林生态系统的光合作用关键参数进行优化和验证.结果表明:森林生态系统GPP模型受ε_max、光合最高温度(T_max)以及光合最适温度(T_opt)的影响最大;优化后的ε_max在不同植被类型之间存在明显差异,介于0.05～0.08 μmol CO₂·μmol^-1 PAR,常绿阔叶林>常绿针叶林>混交林>落叶阔叶林;优化后的森林生态系统T_max为38～48 ℃,T_opt为18～22 ℃;利用分植被类型优化后的模型关键参数,VPM模型可较好地模拟全球主要森林生态系统GPP的季节和年际变化.     

基于蚁群遗传算法的DNA序列比对方法

蚁群遗传算法是在蚁群算法的基础上用遗传算法对其参数进行优化而产生的一种改进算法。把蚁群遗传算法应用于DNA序列比对上,结果表明这种新的序列比对算法是非常有效的。     

用遗传算法优化流加培养的底物流加轨迹

遗传算法（Genetic Algorithm,GA)j是把生物进化论和遗传学原理应用于工程优化而创造出来的新的优化算法,在复杂问题的优化方面显示出了优良性能。近年来GA开始应用于发酵工程领域,本文介绍了应用GA优化流加培养流加轨迹的原理和方法。     

Parameter optimization and sensitivity analysis for large kinetic models using a real-coded genetic algorithm

Dynamic modeling is a powerful tool for predicting changes in metabolic regulation. However, a large number of input parameters, including kinetic constants and initial metabolite concentrations, are required to construct a kinetic model. Therefore, it is important not only to optimize the kinetic parameters, but also to investigate the effects of their perturbations on the overall system. We investigated the efficiency of the use of a real-coded genetic algorithm (RCGA) for parameter optimization and sensitivity analysis in the case of a large kinetic model involving glycolysis and the pentose phosphate pathway in Escherichia coli K-12. Sensitivity analysis of the kinetic model using an RCGA demonstrated that the input parameter values had different effects on model outputs. The results showed highly influential parameters in the model and their allowable ranges for maintaining metabolite-level stability. Furthermore, it was revealed that changes in these influential parameters may complement one another. This study presents an efficient approach based on the use of an RCGA for optimizing and analyzing parameters in large kinetic models.     

Optimization of a feed medium for fed-batch culture of insect cells using a genetic algorithm

Insect cells have been cultured for over 30 years, but their application is still hampered by low cell densities in batch fermentations and expensive culture media. With respect to the culture method, the fed-batch culture mode is often found to give the best yields. However, optimization of the feed composition is usually a laborious task. In this report, the successful use of genetic algorithms (GAs) to optimize the growth of insect cells is described. A feed was developed from 11 different medium components, each used at a wide range of concentrations. The feed was optimized within four sets of 20 experiments. The optimized feed was tested in bioreactors and the addition scheme was further improved. The viable-cell density of HzAm1 (Helicoverpa zea) insect cells improved 550% to 19.5 x 10(6) cells/mL compared to a control fermentation in an optimized commercial medium. No accumulation of waste products was found, and none of the amino acids was depleted. Glucose was depleted, which suggests that even further improvement is possible. We show that GAs are a successful method to optimize a complex fermentation in a relatively short time frame and without the need of detailed information concerning the cellular physiology or metabolism.     

Estimation of parameters of a biochemically based model of photosynthesis using a genetic algorithm

Photosynthesis response to carbon dioxide concentration can provide data on a number of important parameters related to leaf physiology. The genetic algorithm (GA), which is a robust stochastic evolutionary computational algorithm inspired by both natural selection and natural genetics, is proposed to simultaneously estimate the parameters [including maximum carboxylation rate allowed by ribulose 1·5-bisphosphate carboxylase/oxygenase (Rubisco) carboxylation rate ( V _cmax), potential light-saturated electron transport rate ( J _max), triose-phosphate utilization (TPU), leaf dark respiration in the light ( R _d) and mesophyll conductance ( g _m)] of the photosynthesis models presented by Farquhar, von Caemmerer and Berry, and Ethier and Livingston. The results show that by properly constraining the parameter bounds the GA-based estimate methods can effectively and efficiently obtain globally (or, at least near globally) optimal solutions, which are as good as or better than those obtained by non-linear curve fitting methods used in previous studies. More complicated problems such as taking the g _m variation response to CO₂ into account can be easily formulated and solved by using GA. The influence of the crossover probability ( P _c), mutation probability ( P _m), population size and generation on the performance of GA was also investigated.     

Protein structure prediction in a 210-type lattice model: parameter optimization in the genetic algorithm using orthogonal array

We have applied the orthogonal array method to optimize the parameters in the genetic algorithm of the protein folding problem. Our study employed a 210-type lattice model to describe proteins, where the orientation of a residue relative to its neighboring residue is described by two angles. The statistical analysis and graphic representation show that the two angles characterize protein conformations effectively. Our energy function includes a repulsive energy, an energy for the secondary structure preference, and a pairwise contact potential. We used orthogonal array to optimize the parameters of the population, mating factor, mutation factor, and selection factor in the genetic algorithm. By designing an orthogonal set of trials with representative combinations of these parameters, we efficiently determined the optimal set of parameters through a hierarchical search. The optimal parameters were obtained from the protein crambin and applied to the structure prediction of cytochrome B562. The results indicate that the genetic algorithm with the optimal parameters reduces the computing time to reach a converged energy compared to nonoptimal parameters. It also has less chance to be trapped in a local energy minimum, and predicts a protein structure which is closer to the experimental one. Our method may also be applicable to many other optimization problems in computational biology.     

基于遗传算法的土地利用优化：NSGA-Ⅱ和NSGA-Ⅲ的对比研究

土地利用优化通常要兼顾不同群体的多种要求,理论上是复杂的超多目标(4个及以上)优化问题。但实际操作中却往往被简化为多目标(2—3个)优化问题,通过一种流行的多目标优化算法第Ⅱ代非支配排序遗传算法(NSGA-Ⅱ)求解。究其原因是对超多目标优化算法认知的缺失和与多目标优化算法理论对比的匮乏。对NSGA系列中应用最广泛的多目标优化算法NSGA-Ⅱ和最新提出、面向超多目标优化的算法NSGA-Ⅲ进行探究,从理论和实验两方面对Ⅲ和Ⅱ进行对比,从而探究二者进行土地利用优化时的优劣。在理论上,对比两种算法原理的异同。在实验中,分别设计多目标(3个目标)和超多目标(13个目标)土地利用优化问题,利用两种算法进行求解。对实验结果采用四层架构、六大指标进行全面评价,以对比两种算法的可用性。理论对比发现,两个算法只有种群多样性保护的方法不同,其中NSGA-Ⅲ是基于与固定的参考点的距离,而NSGA-Ⅱ则是基于相邻解间的距离。通过实验对比发现,NSGA-Ⅲ在超多目标优化时运算速度快,且产生的最优方案实用价值更高,NSGA-Ⅱ在算法的有效性方面更有优势。     

Optimization of a fermentation medium using neural networks and genetic algorithms

Artificial neural networks and genetic algorithms are used to model and optimize a fermentation medium for the production of the enzyme hydantoinase by Agrobacterium radiobacter. Experimental data reported in the literature were used to build two neural network models. The concentrations of four medium components served as inputs to the neural network models, and hydantoinase or cell concentration served as a single output of each model. Genetic algorithms were used to optimize the input space of the neural network models to find the optimum settings for maximum enzyme and cell production. Using this procedure, two artificial intelligence techniques have been effectively integrated to create a powerful tool for process modeling and optimization.     

基于遗传算法特征选择的HBV再激活分类预测模型

探讨原发性肝癌患者精确放疗后乙型肝炎病毒(hepatitis b virus,HBV)再激活的危险特征和分类预测模型。提出基于遗传算法的特征选择方法,从原发性肝癌数据的初始特征集中选择HBV再激活的最优特征子集。建立贝叶斯和支持向量机的HBV再激活分类预测模型,并预测最优特征子集和初始特征集的分类性能。实验结果表明,基于遗传算法的特征选择提高了HBV再激活分类性能,最优特征子集的分类性能明显优于初始特征子集的分类性能。影响HBV再激活的最优特征子集包括:HBV DNA水平,肿瘤分期TNM,Child-Pugh,外放边界和全肝最大剂量。贝叶斯的分类准确性最高可达82.89%,支持向量机的分类准确性最高可达83.34%。