基于多信息融合的温室黄瓜肥水一体化灌溉系统研究

发布时间：2017-12-27 10:02

  本文关键词：基于多信息融合的温室黄瓜肥水一体化灌溉系统研究　出处：《南京农业大学》2016年博士论文　论文类型：学位论文

  更多相关文章： 多信息融合 温室黄瓜 灌溉施肥 机器视觉 营养液 图像分割 蒸腾 电导率

【摘要】：肥水一体化灌溉施肥是将灌溉和施肥有效结合的现代农业技术,具有可控性,可以精确控制肥水浓度、灌溉量和灌溉时间,不但提高肥水利用率,而且具有节水、节肥、增产及减少环境污染等优势,对我国设施农业发展具有重要的意义。现阶段我国肥水一体化灌溉技术还不够成熟,多数肥水一体化灌溉设备仍然根据经验采用定阶段、定时和定量的控制方式,没有将温室环境信息、作物生长信息和肥水信息融合到肥水灌溉控制系统中,导致肥水浓度和灌溉量无法精确变量控制。至今我国尚未研制出基于温室环境信息、作物生长信息和肥水信息的智能化肥水一体化灌溉系统。为了我国设施栽培实现智能化肥水一体化灌溉,本文以温室黄瓜为作业对象,研究基于多信息融合的温室黄瓜肥水一体化灌溉系统,探寻温室环境信息融合方法与黄瓜群体生长信息检测方法,设计多通道肥水一体化灌溉施肥机以及建立基于多信息融合的肥水电导率和灌溉量智能化控制模型。主要研究内容及结论如下:(1)综合考虑温室环境信息分布不均性问题,以ECO-WATCH数据采集器为核心,设计了温室环境信息检测系统。对多点同源环境信息采用自适应加权融合算法,将环境信息融合值作为肥水一体化灌溉系统的决策输入量,从而提高系统输入的可信度、容错性和可靠性。本研究为了验证自适应加权融合算法和平均融合算法的多传感器信息融合性能,选用3个评价参数分别为平均绝对误差、标准差和变异系数评价信息融合性能,结果表明自适应加权融合算法明显优于平均融合算法。(2)为了实现在线无损检测温室黄瓜群体生长参数,为肥水一体化灌溉系统提供具有代表性的黄瓜生长信息。采用机器视觉技术,捕获自然光环境下黄瓜群体冠层图像,通过超绿色-超红色分割算法、超绿色分割算法和归一化差异分割算法,分割黄瓜群体冠层区域图像,提取黄瓜群体冠层图像特征参数(冠层覆盖率、冠层幅长和冠层幅宽),并结合人工测量的黄瓜群体植株参数(茎秆高度、茎秆直径、叶面数量和坐果数量),构建黄瓜群体生长参数反演模型,实现在线无损检测温室黄瓜群体生长参数,并构建新的黄瓜群体验证图对反演模型性能进行验证。结果表明:在栽培方式为4行X4列、4行X3列和4行X2列时,反演模型反演值与测量值间线性相关达极显著水平,能够准确反演不同栽培方式的黄瓜群生长参数,其反演性能稳定。(3)为了多通道营养液与水源在线高效精准混合,实现肥水灌溉量与电导率精准控制,采用可编程控制器与HMI触控系统结合,设计了多通道肥水一体化灌溉施肥机。以温室黄瓜为作业对象,根据黄瓜营养液配方配制营养液,建立营养液稀释模型,为肥水混合控制提供依据。针对肥水混合时存在的长时滞性及均匀性问题,采用比例-模糊控制算法,实现肥水高效精准混合。结果表明:肥水混合控制算法稳定,变异系数小于2.5%,Ec值控制误差小于0.05 mS·cm-1;肥水灌溉均匀性指标表明:各通道Ec值和pH值的变异系数最大值分别为2.49 %和0.98 %,均匀度大于为98.16 %,整体肥水分布均匀系数大于97.98%,表明肥水灌溉系统的肥水混合均匀。(4)针对温室黄瓜肥水灌溉量在线智能控制问题,分别建立温室黄瓜蒸腾速率预测模型和椰糠日水分蒸发模型,并构建基于多信息融合的温室黄瓜灌溉量控制模型。本文提出一种基于小波变换和非线性自回归神经网络的黄瓜蒸腾速率时间序列非线性预测模型,采用温室环境参数与蒸腾速率的历史时间序列,建立各小波分解重构的低频和高频时间序列NARX子网络预测模型,其预测值合成可以准确预测蒸腾速率。结果表明:2层小波分解重构的时间序列的NARX子网络预测值合成值和未小波分解重构的原时间序列的NARX预测值与蒸腾速率测量值间相关性决定系数R2分别为0.974和0.856,平均绝对误差分别为4.42和10.09 g·h-1。WT-NARX预测性能优于相同网络结构的NAR和BP神经网络预测性能。Penman-Monteith方程模拟值与实测值间相关性显著,决定系数R2为0.900,标准误差SE为0.0083 g m-2·s-1,相对平均偏差RAD为36.42 %。根据日辐热积、日有效积温、日平均温度和日平均湿度,建立了椰糠日水分蒸发量回归方程,回归方程模拟值与测量值间相关性决定系数R2为0.956, SE为207.73 g·m--·d-1,RAD为8.15 %。基于多信息融合的温室黄瓜灌溉量控制模型,当时间尺度为15min和Id时,预测值与实测值间的RAD分别为11.6 %和3.41 %。试验结果表明:在试验期间(10天),温室黄瓜日耗水量实测值和模拟值间RAD为3.00 %。(5)针对温室黄瓜肥水电导率在线智能控制问题,综合时间信息、温室环境信息和肥水信息对黄瓜生长信息(茎秆高度、茎秆直径、叶面数量和坐果数量)的影响,分别构建以黄瓜生长阶段、辐射积、有效积温和肥水电导率累积值为影响因素的黄瓜长势指标,其应用于标准组长势评估时产生的偏离平均百分比分别为3.86%、6.15 %、14.19%和6.77%。根据各个影响因素的长势指标权重,采用加权融合方法,构建了三因素(生长阶段、累积辐热积和累积有效积温)和四因素(生长阶段、累积Ec值、累积辐热积和累积有效积温)的温室黄瓜长势多信息融合指标,其对标准组长势评估产生的偏离平均百分比分别为6.45 %和5.83 %。建立以多信息融合指标为输入量的电导率决策模型,实现肥水一体化灌溉施肥机电导率在线智能控制。
[Abstract]:The integration of water irrigation and fertilization is the modern agricultural technology effective combination of irrigation and fertilization, controllable, can accurately control the concentration of fertilizer, irrigation amount and time, not only to improve the utilization ratio of fertilizer, but also saving water and fertilizer, increase production and reduce environmental pollution and other advantages, has important significance for China's agricultural development. At the present stage of China's irrigation fertilizer integration technology is not mature enough, the majority of water irrigation equipment integration according to the experience of the stage still, timing and control method of quantitative, not greenhouse environment information, crop growth information and fertilizer information into fertilizer irrigation control system, resulting in the fertilizer concentration and amount of irrigation can not accurately control variables. Up to now, our country has not developed an intelligent fertilizer and water integrated irrigation system based on the information of greenhouse environment, crop growth and fertilizer. In order to realize the intelligent cultivation in China based on the integration of irrigation water fertilizer, greenhouse cucumber as the operation object, study the information fusion system based on the integration of water and Fertilizer on greenhouse cucumber irrigation, explore the greenhouse environment information fusion detection method and growth of cucumber population, design of multi channel integration and the establishment of fertilizer irrigation fertilizer fertilizer and irrigation quantity intelligent conductivity based on multi information fusion control model. The main research contents and conclusions are as follows: (1) considering the uneven distribution of greenhouse environment information, and taking the ECO-WATCH data collector as the core, the greenhouse environment information detection system is designed. The adaptive weighted fusion algorithm is applied to the multi point homologous environment information, and the environmental information fusion value is used as the decision input of the fertilizer and water integration irrigation system, so as to improve the credibility, fault tolerance and reliability of the system input. In order to study the multi-sensor information fusion algorithm and validation adaptive weighted average fusion algorithm of the fusion performance, using 3 evaluation parameters respectively, the average absolute error, standard deviation and variation coefficient to evaluate the information fusion performance, results show that the fusion algorithm is better than adaptive weighted average fusion algorithm. (2) in order to realize nondestructive testing of cucumber population growth parameters on line, it provides representative cucumber growth information for the integrated irrigation system of fertilizer and water. By using the machine vision technology, capture natural light environment of cucumber canopy image, through the super green - red, green super segmentation segmentation algorithm and normalized difference segmentation algorithm, the segmentation of cucumber canopy area image, extraction of cucumber canopy image features (canopy coverage, canopy width and canopy width, length) and parameters the cucumber group manual measuring (the number of stem height, stem diameter, leaf number and fruit), construct the cucumber population growth parameter inversion model, on-line nondestructive detection of greenhouse cucumber population growth parameters, and build on the performance of the inversion model to verify the New Cucumber population test. The results showed that when the 4 row X4 row, 4 row X3 column and 4 row X2 column were cultivated, the linear correlation between the inversion value and the measured value was very significant, which could accurately retrieve cucumber growth parameters of different cultivation methods, and the inversion performance was stable. (3) in order to achieve precise and precise control of irrigation quantity and conductivity for multi-channel nutrient solution and water source, a multi-channel fertilizer and water integrated fertilizer applicator is designed by combining programmable logic controller with HMI touch system. The greenhouse cucumber was used as the operating object, the nutrient solution was prepared according to the formula of cucumber nutrient solution, and the dilution model of the nutrient solution was established to provide the basis for the control of the mixed fertilizer and water. In order to solve the problem of long time delay and uniformity in the mixing of water and water, the proportion - fuzzy control algorithm is used to achieve high efficiency and precision mixing of water and water. The results show that the mixed fertilizer control algorithm is stable, the coefficient of variation is less than 2.5%, the Ec value control error is less than 0.05 mS cm-1; show that fertilizer irrigation uniformity index: each channel coefficient variation of Ec value and pH value of the maximum values were 2.49% and 0.98%, uniformity is more than 98.16%, the overall water distribution uniformity coefficient greater than 97.98% indicates the fertilizer mixed fertilizer irrigation system. (4) in view of online intelligent control of greenhouse cucumber water and fertilizer irrigation, greenhouse cucumber transpiration rate prediction model and coconut bran daily water evaporation model were established respectively, and a cucumber irrigation control model based on multi information fusion was built. This paper proposes a prediction model of autoregressive neural network based on wavelet transform and nonlinear time series nonlinear transpiration, the greenhouse environment parameters and transpiration rate of the historical time series, the establishment of the wavelet decomposition and reconstruction of low frequency and high frequency time series NARX network prediction model, the prediction of synthesis can accurately predict transpiration rate. The results show that the predictive value of synthetic value and wavelet decomposition and reconstruction of the original time series prediction values of NARX and transpiration rate measurement correlation coefficient of determination R2 were 0.974 and 0.856 NARX sub network time series 2 level wavelet decomposition and reconstruction, the average absolute errors are 4.42 and 10.09 G - h-1. The predictive performance of WT-NARX is better than that of NAR and BP neural networks with the same network structure. The correlation between the simulated value and the measured value of Penman-Monteith equation is significant. The coefficient R2 is 0.900, the standard error SE is 0.0083 g m-2 s-1, and the relative average deviation RAD is 36.42%. According to daily heat accumulation, daily effective accumulated temperature, daily average temperature and daily average humidity, the regression equation of daily water evaporation of coconut chaff was established. The correlation coefficient between regression equation and measured value was R2, which was 0.956, SE was 207.73 G. M-- D-1 and RAD was 8.15%. In the greenhouse cucumber irrigation control model based on multi information fusion, when the time scale was 15min and Id, the RAD between the predicted and the measured values were 11.6% and 3.41% respectively. The test results showed that during the test period (10 days), the RAD between the measured daily water consumption of greenhouse cucumber and the simulated value was 3%. (5
【学位授予单位】：南京农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S642.2;S626
，

本文编号：1341239