A greenhouse environment control method based on adaptive control vector parameterization

Abstract

The application discloses a greenhouse environment control method based on adaptive control vector parameterization, comprising the following steps: constructing a nonlinear optimal control problem of a type of greenhouse environment dynamic constraint optimization based on constraint conditions in a greenhouse environment, and converting the problem into a nonlinear programming NLP problem of a type of limited dimension parameters; based on a control vector parameterization method, solving the NLP problem by fusing engineering constraint rules and a penalty function method aiming at different constraint conditions on fixed time grid nodes of a "coarse grid"; adaptively subdividing the fixed time nodes of the "coarse grid", and constructing new "refined" time grid nodes; based on the new "refined" time grid nodes, re-solving the NLP problem until a termination condition is met to stop the solving, so that the best solution result of the NLP problem is obtained. The application solves the problem of low calculation efficiency in the solving process of the control vector parameterization method for the dynamic optimal control problem in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of agricultural environment control technology, in particular to a greenhouse environment control method based on adaptive control vector parameterization. BACKGROUND

[0002] Greenhouse crop production is characterized by high input, high output, high efficiency and sustainable development, and has become one of the important symbols of modern agriculture, and advanced and applicable greenhouse environment control technology is the core element. Greenhouse environment control technology can achieve the purpose of high yield and high quality of greenhouse crops and reduce energy consumption, but its goal is not only to pursue crop yield and energy saving, but to maximize economic benefit. How to obtain the greenhouse environment control strategy of the maximum economic benefit goal is essentially a dynamic optimization control problem, which aims to meet the complex constraint conditions such as crop growth environment demand, environmental regulation facility operation limit, outdoor weather dynamic disturbance, and to obtain the optimal operation sequence of the regulation facility within a given production time, so as to maximize the economic benefit goal of greenhouse production.

[0003] The control vector parameterization method has the advantages of high solution accuracy and small scale of discretized nonlinear programming problem, and is one of the mainstream methods for solving optimal control problems. The existing control vector parameterization method only discretizes the control variable, and the time grid is uniformly fixed. In order to better approximate the optimal solution, it is necessary to densely divide the time grid, which will increase the variable dimension, resulting in low calculation efficiency, and even cause ill-conditioned problems, which cannot be solved. SUMMARY

[0004] The purpose of the present application is to provide a greenhouse environment control method based on adaptive control vector parameterization, which increases the effectiveness and feasibility of solving nonlinear programming problems, so as to obtain a control scheme that conforms to the actual control law of the greenhouse.

[0005] Technical scheme: In order to achieve the above purpose, the greenhouse environment control method based on adaptive control vector parameterization according to the present application comprises the following steps:

[0006] Step S1: based on the constraint conditions in the greenhouse environment, a class of nonlinear optimal control problems of greenhouse environment dynamic constraint optimization is constructed;

[0007] Step S2: based on the control vector parameterization method, a class of nonlinear programming NLP problems with finite dimension parameters is constructed, the control time domain [t0, t f ] is segmented first, forming a group of finite dimension time grid nodes, i.e. "coarse grid" fixed time nodes t k; under the "coarse grid" fixed time node, the control variable is discretized into a set of finite-dimensional input parameter vectors, thereby converting the nonlinear optimal control of a class of greenhouse environment dynamic constraint optimization in step S1 into a nonlinear programming NLP problem of a class of finite-dimensional parameters;

[0008] Step S3: based on the control vector parameterization method, the fixed time grid node t k The above, the engineering constraint rules and the penalty function method for different constraint conditions are fused, and the NLP problem in step S2 is solved;

[0009] Step S4: the slope analysis is carried out on the control variable trajectory curve generated in the solving process of step S3, so that the "coarse grid" fixed time node in step S2 is adaptively subdivided, and a new "refined" time grid node is constructed;

[0010] Step S5: returning to step S3, the NLP problem is solved again based on the new "refined" time grid node by the control vector parameterization method;

[0011] Step S6: steps S4-S5 are circularly executed until the algorithm termination condition requirement is met, so that the appropriate "refined" time grid node and the optimal solving result of the NLP problem in step S2 are obtained.

[0012] Advantages: the present application has the following advantages: 1, the present application carries out refined adjustment on the adaptive time grid, and inserts, deletes and reconstructs a new time grid through iteration of the previous coarse grid point, so that the problem of accurate positioning of important time nodes in the prior art is solved, and the method is relatively convenient, the node position is accurate, the control precision is high, and it is beneficial to practical application;

[0013] 2, in order to fuse the engineering constraint rules and the penalty function method for processing different constraint conditions, the adaptive time grid refinement method based on slope analysis is designed under the control vector parameterization method framework, the effectiveness and feasibility of solving the nonlinear programming problem are increased, the greenhouse environment dynamic constraint optimization problem can be better solved, the time node of the control variable responding to outdoor weather changes can be accurately positioned, the control precision is improved, and the control scheme conforming to the actual control law of the greenhouse is obtained. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 It is a method flowchart of the present application;

[0015] Figure 2 It is a control vector parameterization diagram under the control of the present application segment constant discretization;

[0016] Figure 3 It is a slope information analysis diagram of the present application. DETAILED DESCRIPTION

[0017] The technical solutions of the present application will be described in detail below in combination with embodiments and drawings.

[0018] As shown in the drawings, the greenhouse environment control method based on adaptive control vector parameterization comprises the following steps: Figure 1

[0019] Step S1: based on the constraint conditions in the greenhouse environment, a class of nonlinear optimal control problems of greenhouse environment dynamic constraint optimization is constructed;

[0020] Step S2: based on the control vector parameterization method, a class of nonlinear programming NLP problems of finite-dimensional parameters is constructed, and the control time domain [t0, t f ] is segmented to form a set of finite-dimensional time grid nodes, i.e. "coarse grid" fixed time nodes t k ; under the "coarse grid" fixed time nodes, the control variables are discretized into a set of finite-dimensional input parameter vectors, so as to convert the class of nonlinear optimal control of greenhouse environment dynamic constraint optimization in step S1 into a class of nonlinear programming NLP problems of finite-dimensional parameters;

[0021] Step S3: based on the control vector parameterization method, the NLP problem in step S2 is solved at the fixed time grid nodes t k ;

[0022] Step S4: the slope of the control variable trajectory curve generated in the solving process of step S3 is analyzed, so as to adaptively subdivide the "coarse grid" fixed time nodes in step S2 and construct new "refined" time grid nodes;

[0023] Step S5: return to step S3, and based on the new "refined" time grid nodes, the NLP problem is solved again by the control vector parameterization method;

[0024] Step S6: steps S4-S5 are executed in a loop until the termination condition requirement is met, so as to obtain appropriate "refined" time grid nodes and the best solution result of the NLP problem in step S2.

[0025] The specific process is as follows:

[0026] The step S1 of constructing a class of nonlinear optimal control problems of greenhouse environment dynamic constraint optimization based on the constraint conditions in the greenhouse environment comprises establishing a greenhouse environment mathematical model, defining variables, designing constraint conditions, and establishing a single-objective optimization function with the optimal economic benefit, wherein:

[0027] ​The variables are defined as follows: the greenhouse environmental factors are state variables x, including temperature, humidity and CO2 concentration; the heating operation, ventilation operation and CO2 supply are control variables u; and the outdoor environmental factors are external inputs d, including air temperature, humidity, CO2 concentration and solar radiation intensity.

[0028] The constraints include explicit constraints and implicit constraints, wherein the explicit constraints include the range constraints of the control variables u and the path constraints of the state variables x, which are described by equations or inequalities, and the implicit constraints refer to the artificial limit requirements for the control variables in the greenhouse engineering.

[0029] The single-objective optimization function is the economic profit, i.e. the crop economic income minus the total cost of energy consumption.

[0030] The mathematical model of the greenhouse environment is described as follows:

[0031]

[0032]

[0033]

[0034]

[0035] In the equations (1)-(4), t is the time, the photosynthetic rate of crop physiology is the transpiration rate respectively.

[0036]

[0037]

[0038] The indoor air heat energy transmission Q caused by natural ventilation is vent,q [W m -2 ], the water vapor pressure energy transmission s -1 and the CO2 energy transmission are respectively:

[0039] Q vent,q = (c cap,q,v U v +c ai,ou )(X T -V T ) (7),

[0040]

[0041]

[0042] Heat energy input from solar radiation Q rad,q [W m -2 ] is:

[0043] Q rad,q = c rad,q V rad (10);

[0044] The mathematical model of the greenhouse environment represented by formula (1)-(4) is described as a state space equation with respect to x, u, d, t:

[0045]

[0046] In formula (11), is the n x dimensional state variable of the greenhouse environment system; is the n u dimensional control variable, representing the action of the greenhouse environment control facility; is the n d dimensional external climate input;

[0047] The state variable x, the control variable u, and the external input d are respectively:

[0048]

[0049] In formula (1)-(12), X d is the dry matter of the greenhouse crop, with a unit of kg·m -2 ; X T is the indoor air temperature, with a unit of ℃; X h is the absolute humidity of indoor air, with a unit of kg·m -3 ; X c is the indoor CO2 concentration, with a unit of kg·m -3 ; U q is the heating rate, with a unit of W·m -2 ; U c is the CO2 supply rate, with a unit of kg·m -2 ·s -1 ; U v is the specific ventilation rate, with a unit of m·s -1 ; V rad is the outdoor solar radiation intensity, with a unit of W·m -2 ; V T is the outdoor air temperature, with a unit of ℃; V c is the outdoor CO2 concentration, with a unit of kg·m -3 ; V h is the outdoor air humidity, with a unit of kg·m -3 .

[0050] The constraint conditions of the control variable u and the state variable x are designed, specifically: the control variable u constraint condition includes heating rate U q , specific ventilation rate U v , CO2 supply rate U c The value range of the three control variables is limited by the equipment operation capacity, U q , U v , U c The upper and lower boundary values of the three control variables are respectively:

[0051] 0≤U q ≤150, W·m -2 (13),

[0052] 0≤U v ≤7.5E-3, m·s -1 (14),

[0053] 0≤U c ≤1.2E-6, kg·m -2 ·s -1 (15);

[0054] According to the state space equation requirement, the control variable u boundary constraint condition is uniformly described as:

[0055]

[0056] In formula (16), u max , u min Characterize the upper and lower boundary values of the control variable u constraint range;

[0057] The constraint condition of the state variable x includes temperature X T , absolute humidity X h , CO2 concentration X c The value range of the three greenhouse environment factors is X T , X h , X c The upper and lower boundary values are respectively:

[0058] 5≤X T ≤40, ℃ (17),

[0059] 0≤X Rh ≤90, % (18),

[0060] 0≤X c ≤2000, ppm (19),

[0061] X Rh =X h / X h,sat (20);

[0062] X in formula (18) Rh is the relative humidity of indoor air, which is converted from the absolute humidity X h of air by formula (20) Rh X h,sat is the saturated water vapor pressure of indoor air;

[0063] According to the state space equation requirements, the state variable x boundary constraint conditions are uniformly described as:

[0064] x min ≤x≤x max (21);

[0065] In formula (21), x max , x min represent the upper and lower boundary values of the constraint range of the state variable x.

[0066] The greenhouse environment dynamic constraint optimization problem based on the single-objective optimization function J is described as:

[0067]

[0068] Φ(X d (t f ))=c pr1 +c pr2 X d (t f ) (23),

[0069] L(x,u,t)=c q ·U q +c co2 ·U c (24);

[0070] In formulas (22)-(24), t is the time, t0 is the starting time; t f is the ending time; X d (t f ) is the crop dry matter value at the ending time; x is the state variable, u is the control variable, and d is the external input;

[0071] In formula (23), the crop economic income function Φ(X d (t f )) is a fitting function of the crop market price c pr1 , c pr2 and the crop dry matter X d (t f ) at the ending time;

[0072] In formula (24), c q , c co2are the price coefficients of heating and CO2 supply, respectively. The energy cost function L(x, u, t) consists of the heating cost (c q ·U q ) and the CO2 supply cost (c co2 ·U c );

[0073] In equation (22), the crop economic income is subtracted from the total energy cost to obtain the crop economic profit value J.

[0074] The control vector parameterization method based on step S2 constructs a nonlinear programming NLP problem with a finite-dimensional parameter, specifically:

[0075] As Figure 2 illustrated, the schematic diagram of the control vector parameterization method based on piecewise constant discretization of the present application. Based on the control vector parameterization method, the entire control time domain [t0, t f ] is divided into Nc subintervals [t k-1 , t k ], where k = 1, 2,..., Nc, and the time nodes t k satisfy:

[0076] 0 = t0≤t1≤···≤t Nc-1 ≤t Nc = t f (25);

[0077] After Nc is determined, t k is a fixed time switching point, denoted as "coarse grid" fixed time node t k , where k = 0, 1,..., Nc-1;

[0078] Assuming that the jth component of the n u dimensional control variable is u j (t), j = 1, 2,..., n u , according to the piecewise constant approximation method, u j (t) is expressed as:

[0079]

[0080] where, denotes the value of the control variable u j (t) in the kth subinterval [t k-1 , t k ], and σ j k is a basis function representing ;

[0081] According to the piecewise constant method, σ j kIt is in the k-th subinterval [t] k-1 , t k The control component u in equation (26) is a constant value that remains unchanged within the equation. j (t), consisting of a set of data distributed at a fixed time node t k The parameter vector σ on k = 0, 1, ..., Nc-1 j =[σ j 1 ,σ j 2 ,...,σ j k ,...,σ j Nc Approximate representation;

[0082] In a set of finite-dimensional time grid nodes t k For k = 0, 1, ..., Nc-1, the control variable u is discretized into a finite-dimensional set of input parameter vectors σ. j j = 1, 2, ..., n u ;

[0083] The control variable u contains three control quantities U. q U v U c ;remember They represent the control quantity U respectively q U v U c The parameter vector after piecewise constant parameterization is expressed as: σ q =[σ q 1 , σ q 2 , ..., σ q Nc ], σ v =[σ v 1 , σ v 2 , ..., σ v Nc ], σ c =[σ c 1 , σ c 2 , ..., σ c Nc ];

[0084] remember The input parameter vector set is piecewise constantized, i.e., σ = [σ q ,σ v ,σ c ], then in the time domain [t0, t fThe control variable is continuous The state variable x(t|σ) is determined by the vector group σ distributed in the discrete time node t k (k=1,2,…,Nc-1) on the time node t k |σ);

[0085] The greenhouse environment dynamic constraint optimization problem represented by formula (22) is approximately converted into the nonlinear programming NLP problem of optimizing the finite-dimensional parameter array σ, which is described as:

[0086]

[0087] In formula (27), x(t|σ) represents the state variable x value calculated by the parameter group σ at time t.

[0088] The NLP problem in step S1 is solved by fusing the engineering constraint rules and the penalty function method aiming at different constraint conditions, and specifically:

[0089] Step S301: define the penalty function l1 and the engineering constraint rule, wherein the penalty function l1 is expressed as:

[0090]

[0091] In formula (28), x is the state variable, u is the control variable, and d is the external input; G i is the i-th inequality constraint value of the state variable x; and ρ is a constant value selected;

[0092] The state variable x includes the temperature X T , the absolute humidity X h , and the CO2 concentration X c , and the penalty values at the time node t k are respectively denoted as P(X T (t k |σ))、P(X h (t k |σ))、P(X c (t k |σ)), and after adding the penalty function item, the energy consumption cost function L(x(t|σ),σ,t) in the NLP problem changes to:

[0093]

[0094] In formula (29), X T is the indoor air temperature, with the unit of ℃; X h is the indoor air absolute humidity, with the unit of kg·m -3 ; and X c is the indoor CO2 concentration, with the unit of kg·m -3Nc is the number of sub-regions in the whole control time domain; is the input parameter vector set after piecewise constantization; respectively represent the control variables U q , U c is the parameter vector after piecewise constant parameterization;

[0095] Define engineering constraint rules, which are implicit constraint conditions contained in the greenhouse environment system, expressed in the form of engineering constraint rules, including:

[0096] Constraint rule between ventilation and CO2 supply operation: stop ventilation when CO2 is supplied;

[0097] Constraint rule for night operation: no CO2 supply.

[0098] Step S302: Based on the control vector parameterization algorithm, fuse the engineering constraint rules and the penalty function, and solve the NLP problem in step S2;

[0099] The control vector parameterization algorithm is realized by calling the fmincon toolbox of MATLAB, and the basic format of the fmincon command is as follows:

[0100] [x, fval] = fmincon (@objfun, u0, [], [], LB, UB, [], optNLP);

[0101] Where x and fval are the optimal solution of the control variable and its corresponding fitness value; objfun is the objective value calculation function; optNLP is the configuration parameter of the fmincon command; u0 is the initial value of the control variable; LB and UB are the upper and lower boundary values of the control variable, respectively;

[0102] The penalty function and engineering constraint rules in step S301 are realized by programming objfun;

[0103] Solve the NLP problem in step S2, and the specific implementation process is as follows:

[0104] Set the optimization time domain [t0, t f ], the number of segments Nc, the initial value u0 of the control variable and its boundary value LB, UB, and configure the fmincon command parameter optNLP;

[0105] Write the program of the objective value calculation function objfun:

[0106] Read the initial value u0 of the control variable to generate the parameter vector set σ; read the known outdoor climate data, i.e. the external input d, and configure it at the time node t k , k = 0, 1, …, Nc;

[0107] Based on the greenhouse environment mathematical model formula (1)-(4), using four / five order variable step Runge-Kutta method, sequentially solve the state variable value x on each time node t k (k=1,2,…,Nc);

[0108] Based on formula (22), (23), (29), respectively, calculate the crop economic income Φ, the energy consumption cost L containing the penalty term and the economic profit value J, carry out the reciprocal transformation to the economic profit value J, obtain the return value of the function objfun, that is, the fitness value fval, fval=1 / J;

[0109] Call the fmincon command to realize the solution of the NLP problem, obtain the optimal solution array [x,fval] of x, that is, the control variable corresponding to the fixed time node t k (k=0,1,…,Nc-1), denoted as the control variable trajectory, and fval is the optimal fitness value;

[0110] Substitute the optimal solution x into the target value calculation function objfun to calculate the optimal solution of the state variable value x and the economic profit value J.

[0111] Step S4 described that the control variable trajectory generated in step S3 is subjected to slope analysis, and a "refined" time grid is designed, specifically:

[0112] As Figure 3 shown is a slope information analysis schematic diagram of the present application, for the piecewise constant control variable trajectory curve, the left slope k of the time node t the right slope

[0113]

[0114]

[0115] When , delete the time node t k , and two time subintervals [t k-1 , t k ] and [t k , t k+1 ] are merged into [t k-1 , t k+1 ], and the control parameter on the new time subinterval is the average value of the control parameters of the two time subintervals, that is, (σ j k +σ j k+1 ) / 2;

[0116] When When, for the time subinterval [t] k , t k+1 The process is further subdivided, and time nodes are inserted using a bisection method, inserting a time node at the midpoint.

[0117] By selecting appropriate α and β values, slope analysis of the trajectory curve of the control variable can be performed, and the elimination and insertion of time grid nodes can be adaptively carried out to obtain a new "refined" time grid on the original "coarse network" time nodes.

[0118] Based on the new "refined" time grid nodes, the NLP problem is solved again using the control vector parameterization method. Steps S3-S5 are executed iteratively until the algorithm termination conditions are met, in order to obtain a suitable "refined" time grid node and the optimal solution to the NLP problem in step S2. The termination conditions include meeting the upper limit of the number of iterations, the minimum time interval, and the error tolerance of the target value.

[0119] This invention provides a fine-grained adaptive adjustment of the time grid and reconstructs a new time grid by iteratively inserting and deleting points from the previous coarse grid. This solves the problem of inaccurate positioning of important time nodes in existing technologies. The method is convenient to implement, provides precise node positioning, high control accuracy, and is beneficial for practical applications. Therefore, considering the complex constraints of the greenhouse environment, and integrating engineering constraint rules and penalty function methods for handling different constraints, an adaptive control vector parameterization method is designed to enhance the effectiveness and feasibility of the algorithm, which has significant practical implications. This invention can better solve the dynamic constraint optimization problem of the greenhouse environment, accurately locate the time nodes when the control variables respond to changes in outdoor weather, and improve control accuracy, thereby obtaining a control scheme that conforms to the actual control laws of greenhouses.

Claims

1. A method of greenhouse environment control based on adaptive control vector parameterization, characterized in that, The method comprises the following steps: Step S1: based on the constraint conditions in the greenhouse environment, a nonlinear optimal control problem of greenhouse environment dynamic constraint optimization is constructed; It includes establishing a mathematical model of the greenhouse environment, defining variables, designing constraint conditions, and establishing a single-objective optimization function for optimal economic benefit, wherein: Define variables: the greenhouse environment factors are state variables x, including temperature, humidity and CO2 concentration; heating operation, ventilation operation and CO2 supply are control variables u; outdoor environment factors are external input d, including air temperature, humidity, CO2 concentration and solar radiation intensity; The constraints include explicit constraints and implicit constraints, wherein the explicit constraints include control variable u value range constraints and state variable path constraints described by equations or inequalities, and the implicit constraints refer to artificial limit requirements for performing actions on control variables in the greenhouse engineering; Single-objective optimization function: the economic profit, i.e. the economic income of crops minus the total cost of energy consumption, is taken as the single-objective optimization function J; The mathematical model of the greenhouse environment is described as: （1）， （2）， （3）， （4）； In formulas (1) - (4), t is time, photosynthetic rate of crop physiological action φ phot,c [kg m -2 s -1 ]、 transpiration rate φ transp,h [kg m -2 s -1 ] respectively: （5）， （6）； Natural ventilation induced indoor air heat energy transport Q vent,q [W m -2 ] water vapor pressure energy transport φ vent,h [kg m -2 s -1 ] and CO2 mass energy transport φ vent,c [kg m -2 s -1 ] respectively: （7）， （8）， （9）； Heat energy input Q from solar radiation rad,q [W m -2 ] is: （10）； The mathematical model of the greenhouse environment represented by formula (1)-(4) is described as a state space equation with respect to x, u, d and t: （11）； In formula (11), is n x dimensional state variable of the greenhouse environment system; is n u dimensional control variable, representing the action of the greenhouse environment regulating facility; is n d dimensional external climate input quantity; The state variable x, the control variable u and the external input d are respectively: ， ， （12）； In formulae (1) to (12), X d is dry matter of greenhouse crop, unit: kg·m -2 ; X T is indoor air temperature, unit: ℃; X h is absolute humidity of indoor air, unit: kg·m -3 ; X c is CO2 concentration of indoor air, unit: kg·m -3 ; U q is heating rate, unit: W·m -2 ; U c is CO2 supply rate, unit: kg·m -2 ·s -1 ; U v is specific ventilation rate, unit: m·s -1 ; V rad is outdoor solar radiation intensity, unit: W·m -2 ; V T is outdoor air temperature, unit: ℃; V c is CO2 concentration of outdoor air, unit: kg·m -3 ; V h is outdoor air humidity, unit: kg·m -3 ; Step S2: Based on the control vector parameterization method, a class of finite-dimensional parameter nonlinear programming NLP problem is constructed, and the control time domain [t0, t f ] is segmented to form a set of finite-dimensional time grid nodes, i.e. "coarse grid" fixed time nodes t k ; under the "coarse grid" fixed time nodes, the control variable is discretized into a set of finite-dimensional input parameter vectors, so that the nonlinear optimal control of a class of greenhouse environment dynamic constraint optimization in step S1 is converted into a class of finite-dimensional parameter nonlinear programming NLP problem; Step S3: Based on the control vector parameterization method, the NLP problem in step S2 is solved at the fixed time grid nodes t k The engineering constraint rules and the penalty function method for different constraint conditions are fused, and the NLP problem in step S2 is solved. Step S4: slope analysis is performed on the control variable trajectory curve generated in the solving process of step S3, so that the "coarse grid" fixed time nodes in step S2 are adaptively subdivided to construct new "refined" time grid nodes; Step S5: return to step S3, and based on the new "refined" time grid nodes, the control vector parameterization method is used to solve the NLP problem again; Step S6: steps S4-S5 are executed in a loop until the algorithm termination condition requirement is met, so as to obtain appropriate "refined" time grid nodes and the best solving result of the NLP problem in step S2.

2. Greenhouse environment control method based on adaptive control vector parameterization according to claim 1, characterized in that, The constraint conditions of the control variable u and the state variable x are designed, specifically: the control variable u constraint condition includes heating rate U q , specific ventilation rate U v , CO2 supply rate U c The value range of the three control variables is limited by the equipment operation capacity, U q , U v , U c The upper and lower boundary values of the three control variables are taken as: （13）， （14）， （15）； According to the state space equation requirement, the control variable u boundary constraint condition is uniformly described as: （16）； In formula (16), u max , u min upper and lower boundary values characterizing the range of constraints of the control variable u; State variable x constraint conditions include temperature X T , absolute humidity X h , CO2 concentration X c The value range of three greenhouse environmental factors, X T , X h , X c The corresponding upper and lower boundary values are respectively （17）， （18）， （19）， （20）； X in formula (18) Rh is the relative humidity of the indoor air, which is converted from the absolute humidity X h of the air and the relative humidity X Rh of the air by formula (20) X h,sat is the saturated water vapor pressure of the indoor air; According to the state space equation requirements, the state variables The boundary constraint condition is uniformly described as: （21）； In formula (21), x max , x min characteristic state variable upper and lower boundary values of the constraint range.

3. The greenhouse environment control method based on adaptive control vector parameterization of claim 1, wherein, The greenhouse environment dynamic constraint optimization problem based on the single-objective optimization function J is described as: (22)， (23)， (24)； In formulas (22) to (24), t is the time, t0 is the initial time; t f is the final time; X d (t f ) is the crop dry matter value at the final time; x is the state variable, u is the control variable, and d is the external input. In formula (23), the crop economic income function Φ(X d (t f )) is a fitting function of the crop market price c pr1 , c pr2 , and the crop dry matter X d (t f ) at the termination time point; In formula (24), c q , c co2 are price coefficients of heating and CO2 supply, respectively, and the energy cost function L(x, u, t) is composed of heating cost (c q · U q ) and CO2 supply cost (c co2 · U c ). In formula (22), the economic income of crops minus the total energy consumption cost is the economic profit value J of crops.

4. The greenhouse environment control method based on adaptive control vector parameterization of claim 1, wherein, The control vector parameterization method in step S2 constructs a nonlinear programming NLP problem with a limited number of parameters, which is specifically: Based on the control vector parameterization method, the entire control time domain [t0, t f ] is divided into Nc subintervals [t k-1 , t k ], where k = 1, 2, …, Nc, and the time nodes t k satisfy: （25）； After Nc is determined, t k is a fixed time switching point, denoted as "coarse grid" fixed time node t k Here k = 0, 1, …, Nc-1. Assume n u The control variable The jth component of u j (t), j = 1, 2, …, n u According to piecewise constant approximation, u j (t) is expressed as: （26）， wherein represents a control quantity u j (t) a value within the kth sub-interval [t k-1 , t k ], σ j k are basis functions representing ​ According to the piecewise constant method, σ j k It is in the k-th subinterval [t] k-1 , t k The control component u in equation (26) is a constant value that remains unchanged within the equation. j (t), consisting of a set of data distributed at a fixed time node t k The parameter vector σ on k = 0, 1, ..., Nc-1 j =[σ j 1 ,σ j 2 ,...,σ j k ,...,σ j Nc Approximate representation; On a finite-dimensional time grid of nodes t k , k = 0, 1,..., Nc-1, the control variable u is discretized into a finite-dimensional input parameter vector set σ j , j = 1, 2,..., n u ; Control variable u, including three control variables U q , U v , U c ; record , , , respectively, control variable U q , U v , U c After the segmented constant parameterization, the parameter vector is expressed as: σ q =[σ q 1 , σ q 2 ,..., σ q Nc ], σ v =[σ v 1 , σ v 2 ,..., σ v Nc ], σ c =[σ c 1 , σ c 2 ,..., σ c Nc ]; Record The input parameter vector set after piecewise constantization, that is, σ = [σ q ,σ v ,σ c ], the continuous control variable u(t k ) in the time domain [t0, t f ] is determined by the vector set σ distributed on the discrete time nodes t k (k = 1, 2, …, Nc-1), and is expressed as u(t k |σ). The greenhouse environment dynamic constraint optimization problem represented by formula (22) is approximately converted into a nonlinear programming NLP problem for optimizing a limited number of parameter arrays σ, which is described as: （27）； In formula (27), x(t|σ) represents the state variable x value calculated by the parameter group σ at time t.

5. The greenhouse environment control method based on adaptive control vector parameterization of claim 1, wherein, The NLP problem in step S1 is solved by fusing the engineering constraint rules and the penalty function method for different constraint conditions, which is specifically: Step S301: define the penalty function l1 and the engineering constraint rule, wherein the expression of the penalty function l1 is: (28)； In Equation (28), x is a state variable, u is a control variable, and d is an external input; G i is the ith inequality constraint value for the state variable x; is a constant value selected. State variables including temperature X T , absolute humidity X h , CO2 concentration X c , the penalty values at the time node t k are respectively denoted as P(X T (t k |σ))、P(X h (t k |σ))、P(X c (t k |σ), after adding the penalty function term, the energy consumption cost function L(x(t|σ),σ,t) in the NLP problem changes to: (29)； In formula (29), X T is indoor air temperature, unit ℃; X h is indoor air absolute humidity, unit kg·m -3 ; X c is indoor CO2 concentration, unit kg·m -3 ; Nc is the number of sub-regions divided in the whole control time domain; is the input parameter vector group after piecewise constantization; , , respectively represent the control variables U q , U c after piecewise constant parameterization; c q , c co2 are the price coefficients of heating and CO2 supply, respectively; Step S302: based on the control vector parameterization algorithm, the engineering constraint rule and the penalty function are fused to solve the NLP problem in step S2; The control vector parameterization algorithm is realized by calling the fmincon toolbox of MATLAB, and the basic format of the fmincon command is: [x,fval]=fmincon(@objfun, u0, [], [], LB, UB, [], optNLP); Where x, fval are the optimal solution of the control variable and its corresponding fitness value; objfun is the objective value calculation function; optNLP is the configuration parameter of the fmincon command; u0 is the initial value of the control variable; LB, UB are the upper and lower boundary values of the control variable, respectively; The penalty function and engineering constraint rules in step S301 are realized by programming of objfun; The NLP problem in step S2 is solved, and the specific implementation process is as follows: Setting the optimization time domain [t0, t f ], the number of segments Nc, the initial value u0 and its boundary value LB, UB of the control variable, configuring the fmincon command parameter optNLP; The program of the objective value calculation function objfun is written: Reading the initial value of the control variable u0, generating the parameter vector set σ; reading the known outdoor climate data, i.e. the external input d, which is arranged at the time node t by interpolation method k Nc = 1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36, Based on the mathematical model of greenhouse environment (1) ~ (4), using four / five order variable step Runge-Kutta method, the state variable value x on each time node t k (k = 1,2,…, Nc) is sequentially solved. Based on equations (22), (23), and (29), the economic income of crops Φ, the energy consumption cost L with the penalty term, and the economic profit value J are calculated, respectively. The economic profit value J is transformed by taking the reciprocal, and the return value of the function objfun is obtained, which is the fitness value fval. fval = 1 / J; The optimal solution array of the NLP problem is obtained by calling the fmincon command, and the array [x, fval] is obtained, where x is the control variable corresponding to the fixed time node t k (k =0,1,…, Nc-1) is obtained, denoted as the control variable trajectory, and fval is the optimal fitness value. The optimal solution x is substituted into the objective value calculation function objfun to calculate the optimal solution of the state variable value x and the economic profit value J.

6. Greenhouse environment control method based on adaptive control vector parameterization according to claim 5, characterized in that, The definition of engineering constraint rules in step S301 is as follows: the implicit constraint conditions contained in the greenhouse environment system are expressed in the form of engineering constraint rules, including: The constraint rules between ventilation and CO2 supply operation: stop ventilation when CO2 is supplied; The constraint rules of night operation: no CO2 supply.

7. The adaptive control vector parameterization based greenhouse environment control method according to claim 1, characterized in that, The slope analysis of the control variable trajectory generated in step S3 is performed in step S4 to design the "refined” time grid, which is as follows: For piecewise constant control variable trajectory curves, define the left slope k of the time node t , the right slope of the time node t (30)， (31)； When the time node t k is deleted, the two time sub-intervals [t k-1 , t k ] and [t k , t k+1 ] are merged into [t k-1 , t k+1 ], and the control parameter on the new time sub-interval takes the average of the control parameters of the two time sub-intervals, i.e., (σ j k +σ j k+1 ) / 2; When time sub-interval [t k , t k+1 ] is subdivided, a time node is inserted, and a time node at a midpoint position is inserted using a bisection method. Selecting appropriate α, β values, the slope analysis of the control variable trajectory curve can be performed, and the elimination and insertion of time grid nodes are performed adaptively to obtain a new "refined” time grid on the original "coarse network” time nodes.

8. The adaptive control vector parameterization based greenhouse environment control method according to claim 1, characterized in that, The termination conditions according to step S6 include meeting the upper limit of the number of iterations, the minimum time interval, and the target value error tolerance.

Images