基于B-P人工神经网络的环境测点的优选

为了对有多项污染物的环境测点进行优选,提出了单项污染物对环境作用的“相对贡献率"和多项污染物对环境综合作用的“作用和贡献率"的新概念及其计算公式。将B-P神经网络原理与逐步聚类分析思想相结合,用于环境测点聚类优选。该方法用于成都市12个环境测点的优选结果符合客观实际。此外,它还具有简便实用、客观性好的特点。      

Consideration of fuzziness: Is it necessary in modelling fish habitat preference of Japanese medaka (Oryzias latipes)?

The present study aims to clarify the necessity and effectiveness of considering fuzziness in modelling fish habitat preference, and the advantages which would be achieved by considering it. For this purpose, genetic algorithm (GA) optimized habitat preference models under three different levels of fuzzification were compared with regard to prediction ability of the habitat use of Japanese medaka (Oryzias latipes) dwelling in agricultural canals in Japan. Field surveys were conducted in agricultural canals in Japan to establish a relationship between fish habitat preference and physical environments of water depth, current velocity, lateral cover ratio and percent vegetation coverage. The habitat preference models employed for testing the fuzzy-based approach were category model, fuzzy habitat preference model, and fuzzy habitat preference model with fuzzy inputs. All the models were developed at 50 different initial conditions. The effectiveness of the fuzzification in fish habitat modelling was assessed by comparing mean square error and standard deviation of the models, and fluctuation in habitat preference curves evaluated by each model. As a result, the effect of fuzzification appeared as smoother curves and was found to reduce fluctuation in habitat preference curves in proportion to the level of fuzzification. The smooth curves would be appropriate for expressing uncertainty in habitat preference of the fish, by which fuzzy habitat preference model with fuzzy input achieve the best prediction ability among the models. In conclusion, the present study revealed that there are two advantages of fuzzification: reducing fluctuations in habitat preference evaluation and improving prediction ability of the model. Therefore, the consideration of fuzziness would be appropriate for representing fish habitat preference under natural conditions.     

Dynamic reserve design with the union-find algorithm

When reserve networks are established over time, there is a risk that sites will be developed in areas planned for future reservation, reducing the effectiveness of reserves. We developed a dynamic reserve design model that maximizes the expected number of species conserved, taking account of the risk of future habitat loss and fragmentation. The model makes use of the union-find algorithm, which is an efficient method for maintaining a list of connected regions in a graph as nodes and edges are inserted. A simple extension of the algorithm allows us to efficiently determine, for each species, when a sequence of site selections results in a reserve in which the species can persist. The extension also allows us to determine when a sequence of deforestation events results in the species becoming non-viable. The dynamic reserve design model is much more effective than commonly used heuristics, particularly when multiple connected sites are required for species persistence. The model also is able to solve much larger problems with greater effectiveness than the only previous dynamic reserve design model that considered site connectivity relationships. The union-find algorithm has much scope for addressing ecological management problems in which dynamic connectivity needs to be considered.     

醛类化合物分子结构与对鼠急性毒性关系

通过检索美国国立职业与卫生研究所化学物质毒性效应登录 ( RTECS) 1998年版光盘系统 ,搜集了 88种醛类化合物对大鼠急性毒性口服 LD₅₀ 数据 ,并利用遗传神经网络建立了醛类化合物分子结构与对大鼠急性毒性关系的 BP神经网络模式 ,模式的交互检验相关系数达 0.83～ 0.87,具有较强的预报能力 .遗传神经网络是由遗传算法和神经网络耦合而成 ,文章详细讨论了遗传神经网络的构造以及利用遗传神经网络建立 QSAR模型的方法 .     

基于Kriging模型的载人航天某装置结构优化设计研究

对于处在振动环境中的载人航天某装置复杂结构,设计过程中存在最小化质量、同时最大化一阶整体振动频率的多目标优化问题,可以利用多目标遗传算法(NSGA-II)求解得到Pareto最优解集,但利用NSGA-II进行求解计算量大,对于复杂结构更甚。为了节省计算时间,提出基于Kriging代理模型的复杂结构多目标优化设计方法,首先通过拉丁超立方采样获得样本点,然后通过样本点创建Kriging模型,最后利用多目标遗传算法求解代理模型得到Pareto最优解集,方便设计师从中选择最优解,优化结果表明本文所述方法合理有效。     

电除尘器比电阻软件测算研究与应用

为了进一步提高静电除尘器的收尘效率、降低能耗,对粉尘比电阻这一重要影响因素进行了研究。依据粉尘比电阻值定义,建立了收尘极板上粉尘层压降模型,提出了利用压降模型和伏安拟合特性曲线数学表征式来在线测算真实工况下粉尘层比电阻的实现方法。以480C型阳极板、RS芒刺阴极线极配形式进行了模拟验证实验,结果表明测算所得的比电阻数值与实际值相符。最后对现有电除尘控制系统中基于比电阻值的反馈控制方法做了简要介绍。     

区域性应急物资储备级库间的协同选址模型研究

鉴于我国应急物资的多部门分散管理与分散储备的特点,研究了区域性应急物资储备体系构建中的应急物资储备级库协同选址问题。为了实现应急物资在不同级库间的协同储备管理与分散优化调拨,建立了应急物资储备级库选址的协同优化模型。用启发式算法通过MATLAB编程来实现,并用算例进行了验证。     

氯代脂肪烃构效关系的遗传神经网络模式

通过检索美国国立职业与卫生研究所化学物质毒性效应登录（RTECS）1998年版光盘系统,搜集了65种氯代脂肪烃化合物对大鼠急性毒性口服LD₅₀数据,并利用遗传神经网络建立了氯代脂肪烃分子结构与对大鼠急性毒性关系的BP神经网络模式对预报集8种氯代脂肪烃测试结果表明,所建立的网络具有较好的实用性遗传神经网络是由遗传算法和神经网络耦合而成讨论了遗传神经网络的构造以及怎样利用遗传神经网络建立构效关系（QSAR）模式