农业大棚温室智能化自动控制

1、农业大棚温室智能化自动控制摘要：确定控制温室作物生长历来使用约束优化或应用人工智能技术，解决了轨迹的问题。已被用作经济利润的最优化研究的主要标准，以获得足够的作物生长的气候控制设定值。本文针对温室作物生长的问题，通过分层控制体系结构由 一个高层次的多目标优化方法，在解决这个问题的办法是找到白天和夜间温度参 考轨迹（气候相关的设定值）和电导率（fertirrigation相关设定值）。的目标是利润最大化，果实品质，水分利用效率，这些目前正在培育的国际规则。结果 说明，选择从那些获得工业的温室，在过去的八年中示出和描述。关键词：农业；分层系统；过程控制；优化方法；产量优化1介绍现代农业是时下在质量

2、和环境影响方面的规定，因此它是一个自动控制技术的应 用领域已经增加了很多，在过去的几年里，温室产生的空气系统的是一个复杂的 物理，化学和生物学过程，同时，使具有不同的响应时间和模式的环境因素，其 特征在于由许多相互作用，它必须加以控制，以以获得最佳效果的种植者。作物 生长过程是最重要的，主要受周围环境的气候变量（光合有效辐射-PAR温度, 湿度，二氧化碳浓度，里面的空气），水和化肥，灌溉，病虫害提供量，和文化 的劳动力，如修剪和农药的治疗等等。温室是适合作物生长，因为它构成了一个 封闭的环境中，可以控制气候和肥料灌溉变量。气候和肥料灌溉是两个独立的系 统，不同的控制问题和目标。根据经验，不同作

3、物品种的水和养分的要求是已知 的，在实际上，第一个自动化系统控制这些变量。另一方面，市场价格的波动和 环境的规则，以提高水的利用效率或其他方面加以考虑， 减少肥料残留在土壤中 的（如硝酸盐含量）。因此，优化生产过程，可概括为一个温室大气系统的问题， 达到以下目标：的最佳作物生长（一个更大的生产与质量更好） 联营公司的成本（主要是燃料，电力和化肥，减少） ，减少残留物（主要是杀虫剂和离子在土 壤中），和水的利用效率的提高。许多方法已被应用到这个问题，例如，处理的 温室气候管理中的最优控制字段。2 M0优化作物生产一个MOffi化问题可以定义为寻找决策变量的向量，它满足约束条件和优化的 目标函数一

4、个向量，其元素。特点是竞争的措施，表现或目标的问题被视为MO优化问题，其中n目标姬（p）在变量的向量PC P的同时最小化或最大化。问题往往没有最佳的解决方案，同时优化所有目标，但它有一组作为一个 Pareto最优集。其中一个折衷的解决方案可以选自已知的不理想的或不占主导 地位的替代解决方案设置一个决策过程。不同的标准，如物理产量，作物品质， 产品质量，生产过程中的时间不同，或生产成本和风险，可配制于温室作物管理。 这些标准往往会产生有争议的的气候和肥料灌溉要求，必须要解决的或明或暗地在所谓的战术层面上，种植者有几个相互冲突的目标做出决定。 该解决方案的这 个MOffi化过程，的pC P,是最佳

5、的日间和夜间的当前和未来的参考轨迹的温度， XTA导电性，XEC作物周期的其余部分。即，沿着优化的时间间隔内的空气温 度是一个向量，并沿着优化的时间间隔的电导率（E。是一个矢量。请注意，在 植物生长的PAR辐射（昼夜的条件）的影响下，进行光合作用过程。止匕外，温度 成为影响糖的生产速度通过光合作用，从而辐射和温度具有较高的辐射水平的方 式，对应于较高的温度达到平衡。所以，在昼夜条件下的温度维持在较高的水平 是必要的。在夜间条件下的植物都没有激活 （作物不生长），所以它不是必要的，以维持这样高的温度。出于这个原因，通常被认为是两个温度设定点：日间和夜 问。这是必要的，以反白显示，虽然在连续时间的

6、过程优化，解次了在离散的时问间隔为一个优化地平线化，且（k）项（该层是可变的，代表剩余的时间段， 直到结束的农业季节）。因此，解向量，其中k是当前离散时间瞬间获得。需要注意的是，对于提出的优化问题，温室作物生产的模型是必需的，以估 计内的的气候行为和作物的生长，该算法通过不同的步骤，并涉及不同的功能目 标决策变量。温室内的微气候的动态行为是涉及能量转移（辐射和热）和质量平衡（水蒸汽通量和二氧化碳浓度）的物理过程的组合。另一方面，主要取决于作 物的生长和产量，在其他情况中，如灌溉和化肥，在温室内的温度，PAR鬲射，CO2?ft度。因此，无论是气候条件和作物生长的相互影响，其动态行为特征，可 以通

7、过不同的时间尺度。其中XCL = XCL （t）是一个n1的维向量的温室气候状态变量的（主要的内 部空气的温度和湿度，二氧化碳浓度，PAR辐射，土壤表面温度，盖温度，和植 物温度），XGR= XGR（叔）是作物生长状态变量（主要是数量的主茎上，叶面积 指数（LAI）或表面土壤面积的叶片，总干物质代表所有植物成分的根，茎节点 N2-维向量，叶，花和果实，不包括水，水果干物质生物量的水果，不包括水， 和成熟的果实干物质或成熟果实生物量的积累），U = U （t）是m维向量输入变 量（天然通风孔和加热系统，在这项工作中），D = D （t）是干扰（外界温度， 湿度，风速和风向，室外辐射，雨）邻维向量

8、，V = V （t）的一类q维向量，系统变量的（蒸腾，缩合，和其他进程有关），系统常数，C是r维向量，t是时间， XCL i和XGR在初始时刻ti , i是已知的状态整箱整箱（t）是一个非线性函 数的基础上的传质和传热的结余的fgr =的fgr （t）是一个非线性函数的基础上 的植物的基本的生理过程。地中海地区，已开发了线性和非线性模型的物理定律。 这些模型可以发现深 解释拉米雷斯阿里亚斯，罗德里格斯，Berenguel和费尔南德斯（水模），拉米雷斯-阿里亚斯等。（增长模型），罗德里格斯等人。（气候模式），罗德里格斯和Berenguel。这些模型过于复杂，这里详述，但主要的增长模型方程问题 的

9、目标和最终的MOffi化问题的解释在下面的章节将描述。这些方程将用来展示 如何在不同的目标（成本函数）表示为决策变量的函数的优化问题 （目前和未来 的温度和EC的设定值）。利润最大化利润的计算作为新鲜水果的销售收入，并关联到他们的生产成本之间的差异 VPR（t）是产量估计从市场的销售价格，XFFP （t）是获得作物生长模型的 VCO （T）的新鲜水果生产，所产生的费用由供热，电力，化肥，水， t是时间，ti 是作物周期的初始时间，th是最新的收获时间，同时选择由种植者。请注意，在实践中，有多个番茄作物收获在生长季节。出于这个原因，日式代表了最新的 收获时间。另一种方法是考虑在未来的收获时间(T

10、N),成本函数，并再次重新启 动优化过程，一旦前收割工作已经产生。这两种替代品的有效期为多收获。 收入 取决于番茄果实的价格(千克-1, ?公斤-1),收获日期，并在每表面单位鲜重 的产量(公斤米2)。价格政策需要市场模型或历史数据，这是一个非常困难的 预测问题。下面的小节描述如何新鲜水果生产，XFFP(T),以及工艺成本，压控振荡器(T),可以预计相关的决策变量。质量最大化利润最大化，虽然可以被理解为主要目标从种植者的角度来看， 这不能总是 被用来作为唯一的一个。种植者通常属于合作社或农业社会，有利于引入园艺产 品进入市场。这些协会修复的政策，优质的产品，根据不同的市场需求，因此， 种植者必

11、须适应其生产这些政策的过程中，为了达到一些最低限度的质量水平。 食品质量拥抱感觉属性很容易察觉到人的感官和隐藏属性，如健康和营养。在水果和蔬菜的感官性能由糖类，有机酸，挥发性化合物的量，以及颜色，形状和纹 理。然而，糖和酸那些反映整体一个水果口味喜好。对于番茄作物，可溶性固形 物已涉及到糖和可滴定酸度主要有机酸， 因此它们可以作为果实品质的指标。 坚 定的水果是另一种重要的质量参数链中的种植者经销商消费者。然而，一些作品已经表明，园艺蔬菜，如西红柿或鲜花，感官质量的一些重要参数是在冲突与产 量。番茄果实可溶性固形物，滴定酸度，果实硬度和大小可以使用下面的线性方 法(Dorais 等人，2001

12、年XTA(t)和XEC(T)(决策变量)和母口门口丫附 和 面包车博格，1991) (15) Y (T) = A + B (X (T) -G (X (T) 其中Y(t)为变量的计算(可溶性固形物，滴定酸度，果实硬度，或大小)，X(t)是相关的决策变量(XEC VSSol(T) (T),腹侧被盖区(T), vfs的(t) 的和XTA(t)的VFF (t)的，在Y (t)的系数，是一个常数增量，b为增量在 Y (t)的系数，在X (t)的单位的增量，并 G (X (t)的代表在Y (T),其中 有一个增量的X (t)的阈值是一个分段函数。水利用效率的最大化这个目标优化问题明确纳入环境有关的目的。在

13、半干旱的气候，如地中海 的，水是非常稀缺和昂贵的资源，主要是在一些一年四季。有些作者认为，在这 样的地区，是由生产力可用的水和用水效率使用。 这样，适当管理的水是必需的。 与显式包含这一目标，种植者可以选择提供的期望的耗水量， 在生长周期从帕累 托前沿的解决方案。这一目标的尝试使用的水量足以作物生长发育的密切关系，所提供的营养液的浓度。在本文中，水分利用效率被认为是类似的生物量的效率 之间的关系定义为新鲜水果的物质生产与供给的水。多目标优化问题所有这些目标中的变量是空气温度，XTA和/或欧盟，XEC （XFFP（T）的FSF （T）,西南（T）,腹侧被盖区（T）, VSSol （T）的功能，V

14、FS （T）, VFF （T）, 以及衡量的干扰，如PAR鬲射或二氧化碳浓度。也就是说，目标函数可以表示为 对于i = 1,2,3 ,是沿着优化的时间间隔内的空气温度的向量是一个向量沿着优 化的时间间隔的EG 是一个向量测的扰动具有沿水平优化预测。MO优化问题的解决提供了欧共体内的空气温度控制地平线其余的日间和夜间的设定轨迹。包定的日间和夜间的设定点定义，稳定状态模型的温室气候和番茄作物，总结在 Eqs.Although几种技术已被评估为解决 MO/t化问题，在这种情况下，一个目标 实现算法已被用于（序贯二次规划 SQP。确定每个目标的重点，通过使用权重， 按顺序在每个迭代修改。的约束被定义为

15、从专家的知识获得的最大和最小的温度 和EC值表明“最佳”番茄的生长温度和通过分析局部数据从历史系列。由此产 生的约束条件改变整个每年的时间与过去的二十年收集的数据的基础上设计的 图案。3多级递阶控制结构动态参与温室生产过程中呈现出不同的时间尺度上，如上所述，即内部温室 气候，作物快速动力学（即蒸腾作用，光合作用和呼吸作用），和缓慢的的作物发育（即作物生长和果实的变化）。因此，多层分级控制架构已经提出并使用 （Rodriguez等人，2003年和罗德里格斯等人，2008）作物生长控制层考虑到长期目标（市场价格，收获日期和所需的质量）和长期预测的增长状 态，使用修改后的模型（拉米雷斯-阿里亚斯等人

16、，2004）进行优化计算的温室 内温度的设定值轨迹和欧盟一起考虑控制范围内（通常是 65天为一个淡旺季 -260决策变量-或120天为一个漫长的赛季-480决策变量）。灌溉模型也已开 发，控制和优化的目的。长期天气预测，这是逻辑上具有较高程度的不确定性的要素之一，是使用 一个软件工具，访问由西班牙国家气象局的天气预测， 未来八天向前，产生模式 在几个指标（清晰度，最大，平均和最低气温，太阳辐射），在本地搜索历史气 候序列数据库生成模式，更好地适合。以这种方式，以所选择的序列作为短期大 气预报，估计作物周期的其余部分被从该短序列和使用从历史数据库中的一个数 据窗口生成。通过滚动的方法，在第二层进

17、行修改，降低不确定性的相关程度高。设定适应层在这一层中，被发送到下层为第二天的设定值被修改和更新，以避免不可行性问 题，并允许达到参考值。考虑在上层，短期内的天气预报（具有较低程度的不确 定性），当前状态的作物产生的轨迹，这些修改和短期种植者目标（考虑到他/她的技能和作物状态，这是必要的自由度，让种植者的分层控制系统进行交互）。 然后，该信息是用在上面描述的模型，以模拟的温室的行为，并评价，如果所提 供的设定点可以达到。在优化过程被重复修改（减少或增加设定值），根据仿真结果的约束。当设定点是可到达的，它们被发送到下层。气候控制和营养层从上层使用的温度和EC设定点，控制器计算的适当的控制信号，致

18、动器。 所开发的控制算法包括范围宽，从馈控制，自适应控制，预测控制，混合控制。 这显然是有限的引用列表和温度控制上的许多重要文件都没有提到，由于空间的限制。4。结论在这项工作中，一个 MOK化问题已经提出，温室作物生长管理测试，获得 三个目标：经济利益的最大化，果实品质，水分利用效率的折中解决方案。这个 优化方案已经集成到一个层次的控制架构，使日间和夜间的温度和EC通过整个作物周期（使用滚动战略）的设定值自动生成。结果表明短期和长期两个作物周 期的逻辑轨迹。在未来8年，提供实时的结果在工业温室进行建模，仿真，控制 和优化的温室作物生产工作总结研究。附录4 外文文献翻译一一原文Agricultu

19、ral greenhouses greenhouse intelligent automatic controlAbstract: The problem of determining the trajectories to control greenhouse crop growth has traditionally been solved by using constrained optimization or applying artificial intelligence techniques. The economic profit has been used as the mai

20、n criterion in most research on optimization to obtain adequate climatic control setpoints for the crop growth. This paper addresses the problem of greenhouse crop growth through a hierarchical control architecture governed by a high-level multiobjective optimization approach, where the solution to

21、this problem is to find reference trajectories for diurnal and nocturnal temperatures (climate-related setpoints) and electrical conductivity (fertirrigation-related setpoints). The objectives are to maximize profit, fruit quality, and water-use efficiency, these being currently fostered by internat

22、ional rules. Illustrative results selected from those obtained in an industrial greenhouse during the last eight years are shown and described.Keywords： Agriculture; Hierarchical systems; Process control; Optimization methods; Yield optimization1. IntroductionModern agriculture is nowadays subject t

23、o regulations in terms of quality and environmental impact and thus it is a field where the application of automatic control techniques has increased a lot during the last few years The greenhouse production agrosystem is a complex of physical, chemical and biological processes, taking place simulta

24、neously, reacting with different response times and patterns to environmental factors, and characterized by many interactions (Challa & van Straten, 1993), which must be controlled in order to obtain the best results for the grower. Crop growth is the most important process and is mainly influenced

25、by surrounding environmental climatic variables (Photosynthetically Active Radiation PAR, temperature, humidity, and CO2 concentration of the inside air), the amount of water and fertilizers supplied by irrigation, pests and diseases,and culture labors such as pruning and- pesticide treatments among

26、 others. A greenhouseis ideal for crop growing since it constitutes a closed environment in which climatic and Fertilizer irrigation variables can be controlled. Climate and Fertilizer irrigation are two independent systems with different control problems and objectives. Empirically, the water and n

27、utrient requirements of the different crop species are known and, in fact, the first automated systems were those that control these variables. On the other hand, the market price fluctuations and the environment rules to improve the water-use efficiency or reduce the fertilizer residues in the soil

28、 (such as the nitrate contents) are other aspects to be taken into account. Therefore, the optimal production process in a greenhouseagrosystem may be summarized as the problem to reaching the following objectives: an optimal crop growth (a bigger production with a better quality), reduction of the

29、associate costs (mainly fuel, electricity, and fertilizers), reduction of residues (mainly pesticides and ions in soil), and the improvement of the water use efficiency. Many approaches have already been applied to this problem, for instance, dealing with the management of greenhouse climate in the

30、optimal control field, e.g.Challa and van 2. MO optimization in crop productionAn MO optimization problem can be defined as finding a vector of decision variables which satisfies constraints and optimizes a vector whose elements represent objective functions The problems characterized by competing m

31、easures of performance or objectives are considered as MO optimization problems, where n objectives Ji(p) in the vector of variables p C P are simultaneously minimized (or maximized)。The problem often has no optimal solution that simultaneously optimize all objectives, but it has a set of suboptimal

32、 or non-dominated alternative solutions known as a Pareto optimal set , where a compromise solution may be selected from that set by a decision process. Different criteria, such as physical yield, crop quality, product quality, timing of the production process, or production costs and risks, can be

33、formulated within greenhouse crop management. These criteria will often give rise to controversial climate and 月巴料灌溉 requirements, which have to be solved explicitly or implicitly at the so-called tactical level, where the grower has to make decisions about several conflicting objectives. The soluti

34、on of this MO optimization process, p C P, is the optimal diurnal and nocturnal present and future reference trajectories of temperature, Xta, and electrical conductivity, XEC, for the rest of the crop cycle. That is, where is a vector of the inside air temperature along the optimization intervals,

35、andis a vector of the electrical conductivity (EC) along the optimization intervals. Notice that the plants grow under the influence of the PAR radiation (diurnal conditions), performing the photosynthesis process. Furthermore, the temperature influences the speed of sugar production by photosynthes

36、is, and thus radiation and temperature have to be in balance in the way that a higher radiation level corresponds to a higher temperature. So, under diurnal conditions it is necessary to maintain the temperature at a high level. In nocturnal conditions, the plants are not active (the crop does not g

37、row), so it is not necessary to maintain such a high temperature. For this reason, two temperature setpoints are usually considered: diurnal and nocturnal . It is necessary to highlight that although the process optimization is presented in continuous time, it is solved in discrete time intervals fo

38、r an optimization horizon, Nf(k) (this horizon is variable and represents the remaining intervals until the end of the agricultural season). Thus, the solution vectors and are obtained as where k is the current discrete time instant.Notice that, for the proposed optimization problem, a greenhouse cr

39、op production model is required in order to estimate the inner climate behavior and the crop growth through the different steps of the algorithm and relate the different function objectives to the decision variables. The dynamic behavior of the microclimate inside the greenhouse is a combination of

40、physical processes involving energy transfer (radiation and heat) and mass balance (water vapor fluxes and CO2 concentration). On the other hand, the crop growth and yield mainly depend, among other conditions such as irrigation and fertilizers, on the inside temperature of the greenhouse, the PAR r

41、adiation, and the CO2 concentration. Thus, both climate conditions and crop growth influence each other and their dynamic behavior can be characterized by different time scales. Hence, the crop growth in response to the environment can be described by two dynamic models, represented by two systems o

42、f differential equations with a time scale associatedto their dynamics, which can be represented bywhere Xcl=Xcl(t) is an nl-dimensional vector of greenhouse climate state variables (mainly the inside air temperature and humidity, CO2 concentration, PAR radiation, soil surface temperature, cover tem

43、perature, and plant temperature), Xgr=Xgr(t) is an n2-dimensional vector of crop growth state variables (mainly number of nodes on the main stem, leaf area index (LAI) or surface of leaves by soil area, total dry matter which representsall the plant constituents -root, stem, leaves,flower and fruit

44、-excluding water, fruit dry matter being the biomass of the fruits excluding water, and mature fruit dry matter or mature fruit biomass accumulation),U=U(t) is an m-dimensional vector of input variables (natural vents and heating system in this work), D=D(t) is an o-dimensional vector of disturbance

45、s (outside temperature and humidity, wind speed and direction, outside radiation, and rain), V=V(t) is a q-dimensional vector of system variables (related to transpiration, condensation, and other processes), C is an r-dimensional vector of system constants, t is the time, Xcl,i and Xgr,i are the kn

46、own states at the initial time ti, fcl=fcl(t) is a nonlinear function based on mass and heat transfer balances, and fgr=fgr(t) is a non-linear function based on the basic physiological processes of the plants.For the Mediterranean area, the authors have developed linear and nonlinear models using ph

47、ysical laws. These models are too complex to be detailed here, but the main growth model equations will be described in the following sections where the problem objectives and the final MO optimization problem are explained. These equations will be used to show how the different objectives (cost fun

48、ctions) are expressed as functions of the decision variables of the optimization problem (present and future temperature and EC setpoints).Maximization of profitsProfits are calculated as the difference between the income from the selling of the fresh fruits and the costs associated to their product

49、ionwhere Vpr(t) is the selling price of the production (estimated from the market), XFFP(t) is the fresh fruit production obtained from the crop growth model Vcos(t) are the costs incurred by heating, electricity, fertilizers, and water , t is the time, ti is the initial time of crop cycle, and th i

50、s the latest harvesting time, both selected by the grower. Notice that in practice, the tomato crop has multiple harvest during the growing season. For that reason, th represents the latest harvesting time in Eq.An alternative is to consider the next harvesting time (tn) in the cost function and res

51、tarting the optimization process again once the previous harvest has been produced. Both alternatives are valid for multiple harvest. The income depends on the price of tomato fruits ($kg-1,?kg-1), the harvesting dates, and on the yield in fresh weight per surface unit (kg m-2). The price policy req

52、uires market models or historical data, this being a very difficult prediction problem. The following subsections describe how the fresh fruit production, XFFP(t), and the process costs, Vcos(t), can be estimated and related with the decision variables,.Maximization of qualityAlthough maximizing the

53、 profits can be understood as the main objective from the growers point of view, this cannot always be used as the only one. The growers usually belong to cooperatives or agrarian societies that facilitate the introduction of the horticultural products into the market. These associations fix the pol

54、icies on quality products based on the different market requirements, and thus the growers must adapt their production process to those policies in order to reach some minimum quality levels. Food quality embraces both sensory attributes that are readily perceived by the human senses and hidden attr

55、ibutes such as healthiness and nutrition (Shewfelt, 1999). In fruits and vegetables, the sensory properties are determined by the amount of sugars, organic acids, and volatile compounds, as well as color, shape, and texture. However, sugars and acids are those reflecting overall taste preferences fo

56、r a fruit. For a tomato crop, soluble solids have been related to sugars ( Li et al., 2001 and Sonneveld and van der Burg, 1991) and titratable acidity to main organic acids ( Auerswald et al., 1999 and Sonneveld and van der Burg, 1991); thus they can be used as indicators of fruit quality. Firmness

57、 of the fruit is another important quality parameter in the chain grower-dealer-consumer. Nevertheless, some works have shown that in horticultural vegetables, such as tomato or flowers, some important parameters of sensory quality are in conflict with yield ( Dorais et al., 2001, Li et al., 2001 an

58、d Sonneveld and van der Burg, 1991). Hence, the fruit quality can be expressed as (14)where VSSol(t) is the soluble solids concentration in the fruit, Vta(t) is the titratable acidity in fruits, Vff(t) is the fruit firmness, Vfs(t) is fruit size, and wssol, wta, wff, and wfs are weighting parameters

59、.In tomato fruits, soluble solids, titratable acidity, fruit firmness and size may be related to Xta(t) and XEC(t) (decision variables) using the following linear approach (Dorais et al., 2001 and Sonneveld and van der Burg, 1991)(15)Y(t尸a+b(X(t)-g(X(t)where Y(t) is the variable to be calculated (so

60、luble solids, titratable acidity, fruit firmness, or size), X(t) is the related decision variable (XEC(t) for VSSol(t), Vta(t), Vfs(t); and Xta(t) for Vff(t), a is a constant increment coefficient in Y(t), b is the increment coefficient in Y(t) per unit of increment in X(t), and g(X(t) is a piecewis