植物生长建模是一个涉及到植物学、数学、计算机科学、生态学等多学科知识的交叉研究领域，在农林学、计算机图形学、生态环境等学科中都具有广泛的应用前景，已经越来越受到国内外多学科领域专家的关注。其建模方法根据研究尺度（如细胞、器官、植株、群落等）、环境胁迫程度等分类标准的不同有不同的划分方式。一种快速区分模型差异的方法就是看所建模型是否利用领域知识。基于领域知识构建的模型称为知识驱动的模型（Knowledge-Driven Model, KDM），而完全基于数据（没有使用任何领域知识）构建的模型称为数据驱动的模型(Data-Driven Model, DDM)。

近年来，一些研究者在多个研究领域开展了融合两类建模方法的研究，但是在植物生长建模领域的相关研究却很少。对于植物这一复杂生命体的建模，要完全根据植物生长机理对其生长发育过程进行精确的建模是相当困难的。此外，多年来农学相关研究已获得了植物生长的大量知识，并已建立了单变量与植物产量之间的定量或定性关系（部分已知关系的知识驱动的模型）。因此，本篇论文的研究目的有两个：首先，考虑如何利用这些已有农学知识，而不是应用完全“黑箱”的数据驱动模型，探索如何有效融合这两类模型，提出新的植物生长建模方法；其次，基于提出的新的建模方法，开展温室番茄和密闭系统生菜作物生长的实例研究，并对该方法进行验证。本篇论文主要研究工作和贡献总结如下：

1. 提出了一种基于知识与数据共同驱动的植物生长建模方法。为了充分利用知识驱动模型和数据驱动模型各自的优点，提出了一种基于知识与数据共同驱动的植物生长建模方法（Knowledge-and-Data-Driven Plant Growth Modeling Approach，以下简称KDDM），探讨了“加和”与“复合”两种不同耦合方式的模型融合方法，并结合植物生长建模背景给出了两种耦合方式的物理解释意义，针对耦合模型参数学习问题，提出了一种两阶段的参数估计方法。

2. 开展了基于知识与数据共同驱动的植物生长建模方法在温室番茄作物生长的实例研究。本文提出了一种应用于温室作物生长建模的基于知识与数据共同驱动的植物生长模型。该模型以植物功能结构模型（GreenLab）作为知识驱动的子模型，径向基函数（Radial Basis Function，RBF）神经网络作为数据驱动的子模型，模型耦合算子采用加和与复合的耦合方式，并对两种耦合方式进行了对比研究。最后，用温室番茄作物的真实数据进行了验证和测试。其研究结果表明与传统的KDM和DDM相比，KDDM方法具有如下优点：（a）保留知识驱动子模型的植物生长机理及生物参数，模型具备可解释性；（b）仅应用总干重生长过程数据学习，实现植物生长过程预测，同时得到不同器官类型（叶、茎和果）的干重预测；（c）数据驱动子模型可有效补偿知识驱动子模型的不确定性和误差；（d）能够适应不同数据采集条件，具有“柔性”的优点。此外，番茄实例研究结果证实了KDDM方法继承了KDM和DDM各自的优点。

3. 开展了基于知识与数据共同驱动的植物生长建模方法在密闭系统生菜作物生长的实例研究。本文提出了一种应用于密闭系统作物生长建模的基于知识与数据共同驱动的植物生长模型。该模型以结合TomSim光合作用模型（计算生物量产生）与植物功能结构模型GreenLab（计算生物量分配）作为知识驱动的子模型以模拟植物生长过程，分段线性经验模型作为数据驱动的子模型以模拟受控生态生保密闭系统舱内乘员二氧化碳呼出速率和氧气消耗速率，通过构建系统内二氧化碳和氧气浓度变化的质量平衡模型将知识驱动子模型和数据驱动子模型以加和的耦合方式进行了有效的融合，并用密闭系统生菜作物的真实数据进行了验证和测试。其研究结果表明KDDM 方法不仅能够对密闭系统生菜作物生长以及二氧化碳和氧气浓度变化规律的模拟，还可以通过模型计算的方式回答种多大的植物面积以满足1人的呼吸需氧量。生菜实例研究结果进一步证实了KDDM方法的有效性和实用性。

Plant growth modeling is a multidisciplinary field, involving botany, mathematics and computer science, which has been widely used for many fields, such as agriculture, forestry, computer graphics, ecology, and has been receiving more and more attention in all related fields recently. Modeling approaches vary in a number of aspects (e.g., the scale of interest, the level of description, the integration of environmental stresses, etc.). For a fast examination of the approach differences, there are two basic modeling approaches with respect to the degree of domain knowledge included, namely, ``knowledge-driven" and ``data-driven" modeling. The knowledge-driven modeling (KDM) approach relies mainly on the given domain knowledge. In contrast, the data-driven modeling (DDM) approach is capable of formulating a model solely from the given data without using any domain knowledge.

In recent years, to take advantage of both the KDM and DDM approaches, studies on integrating these two types of modeling approaches have been conducted. Investigations on the successful application of this integrated approach in different research fields (e.g., chemical engineering, machine learning, etc.) have been reported. However, there are a few studies on the integration approach in plant growth modeling and applications. Although plants, as a typical bio-system, are highly complex and dynamic systems, and their growth and development mechanisms are difficult to understand and model completely and mathematically, a great deal of empirical knowledge concerning plant growth in agricultural and ecological sciences are discovered, and some known relationships or even mechanistic models (e.g., Process-Based Models, Functional-Structure Plant Models) have been developed. Therefore, in this thesis, utilization of these knowledge (including mechanistic models) and fusion between two types of approaches in plant growth modeling are our specific concern for developing new plant growth modeling approach. Besides, a tomato crop grown in greenhouse and a lettuce crop grown in a closed system are taken as case studies on plant growth modeling. The main research content and contributions of this thesis include as follows:

1. A knowledge-and-data-driven plant growth modeling approach. To take advantage of both the KDM and DDM approaches, we proposed a knowledge-and-data-driven plant growth modeling approach (abbreviated as KDDM) that consists of a ``knowledge-driven (KD)" submodel and a ``data-driven (DD)" submodel. The two types of submodels were integrated using a two-way coupling connection. Two cases of superposition and composition coupling operators in the KDDM approach and their physical explanations concerning plant growth were discussed. Besides, a two-stage parameter estimation method was proposed to address parametric identification of KDDM, which caused by the coupling operation between the two submodels.

2. A knowledge-and-data-driven modeling approach for simulating plant growth: A case study on tomato growth in greenhouse. We proposed a knowledge-and-data-driven plant growth modeling approach and applied it to tomato crop. In this work, a functional-structural plant growth model (GreenLab) was adopted as the KD submodel and a radial basis function network (RBFN) as the DD submodel. A tomato crop grown in greenhouse was taken as a case study on plant growth modeling. Two versions of the KDDM based on the superposition and composition coupling operators were developed. Finally, tomato growth data sets from twelve greenhouse experiments over five years were used to calibrate and test the models. In comparison with the existing KDM and DDM, the proposed KDDM approach presented several benefits: (a) The KDDM approach is able to preserve physically interpretable parameters and has explanatory power in predicting plant growth. (b) The proposed KDDM approach exhibits high accuracy in predicting the dry weights of leaves (including petioles), stems and fruits even when observational data on the organs are unavailable; this characteristic can greatly improve data collection efficiency (only requiring measures of the total dry weight). (c) The DD submodel in the KDDM approach can effectively compensate for the unknown part and error of the KD submodel due to uncertainty during plant growth. (d) This approach offers a high degree of modeling flexibility that it cannot only maximally utilize domain knowledge and ecological data to improve model performance, but it also effectively addresses situations associated with adding and/or missing variables or data. Besides, the case study confirms that the KDDM approach inherits advantages from both the KDM and DDM approaches.

3. A knowledge-and-data-driven modeling approach for simulating plant growth: A case study on lettuce growth in a closed system. We proposed a knowledge-and-data-driven plant growth modeling approach and applied it to lettuce crop. In this work, GreenLab, including a mechanistic process-based model (TomSim) for biomass production and its partitioning was adopted as the KD submodel and a piecewise linear empirical model (PLEM) for carbon dioxide production (CO2) and oxygen (O2) consumption by crew in the crew cabin of controlled ecological life support system (CELSS) integration test platform (CITP) as the DD submodel, the two submodels were integrated using a superposition coupling operators through the mass-balance model with metabolic stoichiometries, which was derived for CO2 and O2 concentration in CITP. A lettuce crop grown in a closed system was taken as a case study on plant growth modeling. Finally, real data from a two-30-day CELSS integrated test with lettuce were used to calibrate and test the model. The experimental results demonstrate that the proposed KDDM approach can not only provide strong predictions for dry weights of different types of organs as well as CO2/O2 concentration, but also answer how much planting area could meet the oxygen demand of one person based on the KDDM approach. Furthermore, the case study confirms the validity and usefulness of the KDDM approach.