【作者】 赵煜；

【导师】 王俊；

【作者基本信息】 浙江大学 ， 农业机械化工程， 2013， 博士

【摘要】 近些年来,由于水产业的迅猛发展,寻求一种能够快速、准确的对养殖水进行定性和定量检测的方法具有重要的意义。电子舌作为一种能够模拟人类味觉对液体样品特性进行分析检测的新型仪器,其在食品、医药以及环境工程等方面对液体综合信息进行定性和定量检测的能力不断得到了证实。模式识别在电子舌系统中起着至关重要的作用,而人工神经网络作为模式识别的一种,在电子舌领域有着广泛的应用前景;目前在电子舌系统中应用最广的是BP神经网络。然而,电子舌系统在养殖水中的应用前景尚未得到证实；而对多种神经网络对电子舌信号的分析能力的研究也少有涉及。因此,本文使用法国Alpha.MOS公司的α-ASTREE电子舌结合六种人工神经网络模型：BP神经网络、径向基神经网络(RBFNN)、广义回归神经网络(GRNN)、粒子群优化BP神网络(PSOBP)、遗传算法优化BP神经网络(GABP)和模糊神经网络(TSFNN),以不同养殖条件(养殖密度、养殖温度、喂食条件、增氧条件和酸碱度条件)下的金鱼养殖水为研究对象,分别建立了养殖水水质定性评价体系和定量评价体系；揭示了电子舌传感器与不同养殖条件以及电子舌传感器与金鱼养殖水的四项主要理化指标的内在关系,并在此基础上对传感器阵列进行了优化；分别考察了原始传感器阵列和优化传感器阵列对不同金鱼养殖水的定性和定量区分能力;分别考察了基于原始传感器阵列和基于优化传感器阵列的电子舌系统结合六种神经网络模型对不同金鱼养殖水的定性和定量检测能力。主要结论如下：(1)电子舌响应与养殖条件及四个主要理化指标均有显著关系：并将方差分析中各因素的F值相对较大的几根传感器挑选出来可用于不同金鱼养殖水的定性检测和定量检测的优化传感器阵列(BA、BB、CA、HA和JB)。(2)分别基于原始传感器阵列的电子舌响应信号和基于优化传感器阵列对不同养殖条件金鱼养殖水的电子舌响应信号进行主成分分析的结果表明：该电子舌系统对同一养殖条件不同养殖水平的金鱼养殖水有着良好的区分能力,且传感器阵列的优化提高了其区分能力。(3)分别基于原始传感器阵列和优化传感器阵列使用主成分分析对同一理化指标不同含量区间的养殖水样品进行区分的结果表明：电子舌响应和各理化指标之间呈现复杂且非线性的关系；该电子舌系统对同一理化指标不同含量区间的养殖水样品有良好的区分能力,优化后的传感器阵列能够提高其区分效果。(4)根据所建立的养殖水水质定性预测评价体系对六种神经网络的评价结果表明：BP网络可以全面、快速且准确的对金鱼养殖水的不同养殖条件进行定性预测。(5)根据所建立的养殖水水质定量预测评价体系对六种神经网络的评价结果表明：GRNN网络和RBFNN网络最适宜用于快速准确的对金鱼养殖水的理化指标进行定量预测。更多还原

【Abstract】 Nowadays, with the rapid development of aquaculture, it is crucial to find a fast and accurate method to detect aquacultural water quality qualitatively and quantitatively. Electronic tongue is a novel instrument, which can mimic human taste to analyze the characterization of liquid samples. It has been established that the electronic tongue was capable of obtain the comprehensive information of liquid both qualitatively and quantitatively in many fields, such as food area, medical area as well as environmental engineering area and so on. Pattern recognition plays an important role in electronic tongue system. The BP neural network, as a kind of artificial neural network, has been used widely with electronic tongue for liquid analysis. However, there was no literature on the application of detection of aquacultural water quality with electronic tongue, as well as the capability of other artificial neural networks with electronic tongue for aquacultural water quality. In this work, a α-ASTREE electronic tongue by Alpha.MOS (France) was employed with six kinds of artificial neural network, which was BP neural network (BP), Radial basis function neural network(RBFNN), Generalized regression neural network(GRNN), Particle swarm optimization BP neural network(PSOBP), Genetic algorithm to optimize BP neural network(GABP) and Takagi-Sugeno Fuzzy neural network (TSFNN), to detect goldfish water from different cultivation conditions (cultivation densities, cultivation temperatures, food supply, aerating conditions and pH conditions) qualitatively and quantitatively. Both the qualitative and quantitative evaluation systems for aquacultural water quality were established. Both the internal relations between the sensors and different cultivation conditions and the internal relations between the sensors and four main chemical parameters were revealed, and the sensors array was optimized. Both the classified capability and prediction capabiltiy based on original sensors array and optimal sensors array for different goldfish water was evaluated qualitatively and quantitatively. The main conclusions are as follows.(1) Significant differences between samples from different cultivation conditions and all sensors were significant to four main chemical parameters (content of nitrate, ammonia and dissolve oxygen and pH value). And the optimal sensors array (BA、BB、CA、HA and JB) was obtained for both qualitative and quantitative detection of different goldfish water.(2) Samples from different levels of same cultivation condition could be classified by PCA based on response of original sensors array. And the classification result was improved when sensors array was optimized.(3) The relation between sensors response and chemical contents was complex and nonlinear. Samples with one of four chemical parameters belong to different value range from same cultiviation conditions could be classified by PCA based on response of original sensors array. And the classification result was improved when sensors array was optimized.(4) According to the developed qualitative evaluation system for aquacultural water, the developed BP network was most suitable for comprehensive and qualitative detection of cultivation conditions of goldfish water fast and accurately.(4) According to the developed quantitative evaluation system for aquacultural water, the developed GRNN and RBFNN networks were most suitable for quantitative detection of goldfish water fast and accurately.更多还原