【作者】 李宏文；

【导师】 屠康；

【作者基本信息】 南京农业大学 ， 农产品加工与贮藏， 2007， 硕士

【摘要】 本文以红富士苹果为原料，测定苹果在不同处理过程中(38℃，96h；38℃，72h；20℃，1440h；0℃，720h)及贮藏期间的呼吸强度及颜色变化数据，分别建立苹果呼吸强度和颜色变化(果实底色a，H°)的BP神经网络模型，可以实现对苹果贮前热处理及贮藏过程中(0-38℃)呼吸强度和颜色变化的模拟和预测。同时，利用遗传算法的全局搜索能力，优化BP神经网络模型的输入层和隐层之间连接的权值和阈值矩阵、隐层和输出层连接的权值和阈值矩阵，从而近一步提高苹果呼吸强度与颜色预测的精度和缩短网络的建模时间，提高模型的实际应用价值。1、利用三层前向传递BP神经网络及红富士苹果在不同处理过程中及其贮藏期间呼吸强度的数据，通过多维输入数据(温度、时间)与呼吸强度的相关分析，建立红富士苹果呼吸强度拟合与预测的BP神经网络模型。经验证，该模型可以很好的模拟苹果贮藏前热处理及贮藏过程中呼吸强度的变化。最后，通过该模型预测非试验阶段的呼吸强度，并做试验与预测值相对比，对比发现此BP神经网络模型预测精度达到90.16％，可以实现对未知试验输入数据的预测。2、利用三层前向传递BP神经网络模型对红富士苹果在贮藏前热处理及其贮藏期间颜色变化(果实底色a，H°)数据，以温度和时间作为神经网络的输入，颜色变化(果实底色a，H°)作为输出，建立颜色变化的BP神经网络模型，从而实现对颜色变化的模拟和预测。经验证，该模型可以很好的模拟苹果不同贮前热处理过程及其贮藏期间整个过程的颜色变化的曲线。最后，通过该模型预测非试验阶段的颜色变化，并做试验与预测值相对比，对比发现BP神经网络模型对果皮底色a预测精度达到89.95％，对果皮底色色调角H°的预测精度达到93.49％。3、利用遗传算法的全局搜索能力，把BP神经网络程序作为遗传算法的适应度函数，建立遗传神经网络模型(GA-BP)，把需要优化的BP神经网络的权值和阈值矩阵最为遗传算法的种群，通过遗传算法的选择、交叉、变异等步骤的操作，优化出BP神经网络最优的连接权值和阈值矩阵，从而达到提高BP神经模型的预测精度与缩短模型建模的时间，以提高模型的工作效率和预测精度。经验证，呼吸强度模型的平均预测精度达到95.01％，对颜色变化模型果实底色a和色调角H°的预测精度分别达到93.3％和95.79％。更多还原

【Abstract】 In this paper, we used the material of Red Fuji (Malus domestica Borkh) apple andmeasured the data of respiration rate and color on the process of apple pre-storage heattreatment and storage period (38℃, 96h; 38℃,72h; 20℃,1440h;0℃,720h). The model ofpredicting respiration rate and color(background color a~*, H°) was determined respectivelyby using the BP neural network and can realize the simulation and prediction of therespiration rate and color on the process of apple pre-storage heat treatment and storage.Meanwhile, the application of genetic algorithms global search ability to optimization ofthe weight and threshold of matrix of BP neural network can improve the predictionprecision and decrease the time of constitution model to increase the model applicationvalue.1, A neural network topological structure was determined and an artificial neuralnetwork model for predicting respiration rate of Red Fuji apple was established by using BPneural network during the apple per-storage heat treatment and storage period and thecorrelation between multi-dimension input data (temperature, time) and respiration rate. Byverification, the model can fit the change of respiration rate on the process of applepre-storage heat treatment and storage period. At last, we use the model to predict thenon-experimental respiration rate and by contrast the prediction precision of the model is89.95% and conclude the model can be used to predict the respiration rate.2, A neural network topological structure was determined and an artificial neuralnetwork model for simulation and predicting color(a~*, H°) of Red Fuji apple was established by using BP neural network during the apple per-storage heat treatment andstorage period and the correlation between multi-dimension input data(temperature, time)and color(a~*, H°) as output data. By verification, the model can fit thechange of color on the process of apple pre-storage heat treatment and storage period. Atlast, we use the model to predict the non-experimental respiration rate and by contrast theprediction precision of the background color a~* model is 89.95% and the predictionprecision of the background color H°model is 93.49%.3, We use the genetic algorithm global searching ability and hold the program of BPneural network as the genetic algorithm fitness function and the mode of Genetic Algorithmand Back Propagation Neural Network (GA-BPNN) was determined. And we use theneeded optimization of the weight and threshold of matrix of BP neural network as thegenetic algorithm population and find the optimization of the weight and threshold ofmatrix of BP neural network by the process of selection, crossover, mutation to improve theBP neural network prediction precision and decrease the model established time. Byverification, the respiration rate model’s average prediction precision is about 95.01%and thebackground color a~* model’s average prediction precision is about 93.3% and thebackground color H°model’s average prediction precision is about 95.79%.更多还原