Skip to Content
ProcessesProcesses
  • Article
  • Open Access

26 December 2025

19 Pages

Research on Bi-Level Optimal Scheduling Strategy for Agricultural Park Integrated Energy System Considering External Meteorological Environmental Uncertainty

,
,
,
,
and
1
Changchun Power Supply Company, State Grid Jilin Electric Power Co., Ltd., Changchun 130000, China
2
School of Electrical and Electronic Engineering, North China Electric Power University, Beijing 102206, China
*
Author to whom correspondence should be addressed.

Abstract

The Agricultural Park Integrated Energy System (APIES) is a key platform for integrating distributed renewable energy (DRE) with agricultural production. However, its economic operation and the stability of crop growth environments are severely challenged by bidirectional uncertainties from external meteorology. These include the inherent variability of wind-solar generation and critical agricultural loads, such as supplementary lighting and temperature control, a challenge that existing models with static environmental parameters fail to address. To solve this, a bi-level optimization scheduling model for APIES considering meteorological uncertainty is proposed. The upper layer minimizes operation costs by quantifying uncertainties via triangular fuzzy chance constraints, with core constraints on DRE output, energy storage charging-discharging, and load shifting, solved by YALMIP-Gurobi linear programming. The lower layer maximizes crop growth environment satisfaction using a dynamic weight adaptive mechanism and NSGA-II multi-objective algorithm. The two layers iterate alternately for coordination. Using a small agricultural park in Xinjiang, China, as a case study, the results indicate that the proposed two-layer optimal scheduling model reduces costs by 10.8% compared to the traditional single-layer optimization model, and improves environmental satisfaction by 4.3% compared to the fixed-weight two-layer optimization model.

1. Introduction

Modern agriculture is evolving toward high-yield, high-quality, and intensive production, with facility agriculture becoming a core carrier for improving crop productivity. However, the operation of such agricultural facilities relies heavily on energy-intensive loads—including supplementary lighting for photosynthesis, air-source heat pumps for temperature control, and irrigation pumps for humidity regulation—leading to a sharp increase in energy demand [1]. Meanwhile, the integration of distributed renewable energy (DRE) such as photovoltaic (PV) and wind turbine (WT) into agricultural parks has emerged as a critical path to address the “high carbon emission, low efficiency” issues of traditional agricultural energy use [2]. Against this backdrop, the Agricultural Park Integrated Energy System (APIES)—which synergizes DRE supply, agricultural load consumption, and energy storage regulation—has become a key solution for balancing agricultural production and low-carbon energy transition [3,4]. Nevertheless, APIES operation faces a pivotal technical bottleneck: external meteorological factors not only directly affect DRE output stability but also dynamically alter agricultural load demand. This bidirectional perturbation between “energy supply” and “agricultural load” driven by meteorology makes it extremely challenging to coordinate APIES’s operational economy and the stability of crop growth environments.
Against the dual backdrop of the global energy structure transitioning toward cleanliness and low carbon emissions and the accelerated advancement of agricultural modernization, the in-depth integration of agriculture and energy systems has become a critical pathway to achieving sustainable development. Traditional agricultural models face challenges such as low energy efficiency, high carbon emission pressure, and insufficient flexible regulation capabilities. However, the rise in distributed energy, intelligent control, and digital technologies has provided technical support for the innovation of agricultural energy systems [5]. As a new architecture integrating renewable energy, agricultural production loads, and energy storage equipment, the Agricultural Integrated Energy System (AIES) not only addresses the energy bottlenecks of traditional agriculture but also improves energy utilization efficiency through multi-energy complementarity and source-load interaction, offering crucial support for achieving carbon emission reduction goals [6].
Currently, the academic and industrial communities have conducted multi-dimensional research on agricultural energy systems. In terms of energy optimization [7] and configuration [8,9], researchers have established hierarchical multi-objective planning models that consider economic costs, environmental benefits, and energy utilization efficiency. By improving algorithms, they have realized the coordinated optimization of equipment configuration and operation strategies, providing configuration schemes for energy systems in different types of smart parks—such as industrial parks, agricultural parks, and ecological parks. In terms of load characteristic exploration and regulation, for typical agricultural loads such as greenhouse lighting, temperature-humidity control, and irrigation systems, scholars have developed refined demand models [10,11]. Through recursive least squares (RLS)-based parameter identification and flexibility assessment, they have proposed regulation strategies based on load aggregation, effectively enhancing the potential of agricultural loads to participate in power grid dispatching while ensuring crop growth demands [12].
In terms of technology integration and application, innovations span multiple domains: New technologies such as Light Detection and Ranging (LIDAR) have been used to reduce yaw errors in wind turbines, thereby improving the efficiency and reliability of wind power generation in agricultural parks [13]. autonomous agricultural robots (e.g., onion harvesters) have achieved energy-saving and safe operation through the coordination of path planning algorithms and physics-based battery management systems, addressing the energy supply challenges of small-scale fragmented farms [14]. blockchain technology has provided a trusted framework for peer-to-peer (P2P) trading of distributed energy, enabling residents in community microgrids to obtain monetary benefits through transactive energy markets [15]. and flexible electronic technologies have further expanded the application boundaries of agricultural energy systems in scenarios such as environmental monitoring and equipment condition monitoring [16].
Despite significant progress in existing research, several unresolved challenges remain: First, in APIES optimal scheduling research, the satisfaction of crop growth environments is often simplified as a static constraint, lacking dynamic coordination and optimization with energy scheduling costs [12]. Second, agricultural loads exhibit significant dependence on biological growth cycles and randomness, and existing models insufficiently characterize load flexibility across multiple scenarios, limiting the practicality of regulation strategies [17]. Third, the integration of new technologies with agricultural energy systems is still in the exploratory stage: for instance, while LIDAR can improve wind turbine performance, its long-term operational reliability and investment return (ROI) under agricultural park conditions require further verification. Similarly, the coordination mechanism between autonomous robots and existing agricultural energy infrastructure has not yet been standardized [18].
To address these key issues, this paper focuses on the optimized operation and technology integration of agricultural integrated energy systems. By integrating multi-source technologies and multi-objective optimization concepts, it aims to:
(1)
Construct a collaborative optimization framework for crop growth and energy scheduling suitable for agricultural greenhouses, promoting the intelligent transformation of facility agriculture.
(2)
Develop flexible regulation strategies for agricultural loads based on scenario awareness, improving the adaptability of load management to biological and environmental uncertainties.
(3)
To balance operational economy and crop growth environment satisfaction, a novel APIES economic-growth environment bi-level optimization model is proposed. This model enables the dual goals of ensuring economical operation and implementing precise regulation of crop growth environments.

2. APIES Framework and Uncertainty Handling

2.1. Core Composition of APIES

The APIES centered on greenhouse production mainly includes small source-grid-load-storage structures such as distributed new energy, power routers (PR), agricultural flexible loads, and energy storage devices, as shown in the framework diagram in Figure 1. APIES consists of three core modules:
Figure 1. Framework diagram of APIES.
(1)
Energy Supply Side: Distributed photovoltaic (PV), wind turbine (WT), and electrical energy storage (EES).
(2)
Agricultural Load Side: Supplementary lighting lamps (for light environment regulation), irrigation pumps (for humidity regulation), ventilators (for carbon source supplementation), and air-source heat pumps (for thermal environment regulation).
(3)
Regulation Hub: Realizes multi-port energy management based on an electrical energy router to coordinate the interaction between energy supply and load.

2.2. Distributed Energy Resource Modeling

(1)
The power output model of a photovoltaic (PV) generation system can be expressed as a function of solar irradiance and ambient temperature:
P t P V = P S P V × I t I s t c × [ 1 − α × T t − T s t c ]
where P t PV is the output electrical power of the PV system at time interval t. P S P V is the rated output electrical power of the PV system under standard conditions, taken as 30 kW. I s t c is the solar irradiance under standard test conditions, taken as 1000 W/m2. I t is the solar irradiance at time interval t. Ambient temperature under standard conditions, taken as 25 °C. T t is the temperature at time t. α is the power temperature coefficient of the PV cell, taken as −0.004 °C−1.
(2)
The output power of a wind turbine (WT) is primarily influenced by factors such as wind speed, cut-in wind speed, rated wind speed, and cut-out wind speed. Its output model can be represented as:
P t W T = 0 0 ≤ v t ≤ v i n ∪ v t ≥ v o u t P S W T v R 3 − v i n 3 v t 3 − v i n 3 v i n ≤ v t ≤ v R P S W T v R ≤ v t ≤ v o u t
where P t W T is the output power of the wind turbine at time interval t. P S W T is the rated output power of the wind turbine under ideal conditions, taken as 50 kW. ν t is the wind speed at time t. v o u t , v R , v i n are the cut-out wind speed, rated wind speed, and cut-in wind speed, respectively, with values taken in this paper as 25 m/s, 14 m/s, and 3 m/s.
(3)
The energy storage unit in the park is primarily electrical energy storage. The operational constraints of the electrical energy storage (EES) include storage capacity constraints, charge/discharge power constraints, and charge/discharge state constraints. Furthermore, to ensure stable operation of the EES, the remaining energy at the end of one scheduling period must equal the initial energy. Additionally, considering the lifespan of the energy storage device, a constraint on the daily number of charge/discharge cycles is added.
E 1 s = E 24 s E _ s ⩽ E t s o c ⩽ E ¯ s P ¯ e ⩽ P t e ⩽ P ¯ e 0 ⩽ S t + R t ⩽ 1 P t E = P t − 1 E + P t cha η s cha − P t dis η s dis
where E _ s and E ¯ s are the lower and upper limits of the EES State of Charge (SOC), respectively. P ¯ e and P ¯ e are the lower and upper limits for EES charging power, respectively. S t and R t are binary variables indicating the charging and discharging states of the energy storage device in each time period (0 for inactive, 1 for charging or discharging). P t E is the remaining capacity of the EES at time interval t. P t cha , P t dis represent the charging and discharging power of the EES at time interval t. η s cha and η s dis represent the charging and discharging efficiencies of the EES.

2.3. Model of Energy-Consuming Equipment

To accurately match crop growth needs, quantitative models for four types of core loads are constructed:

2.3.1. Supplementary Lighting Load Model

The demand of crops for the light environment is also time-sensitive, and the length of light exposure can affect the normal growth of crops. Therefore, high pressure sodium lamps are used for daytime light compensation, and the conversion calculation formula between light intensity and power consumption is shown as follows:
P t 1 = − 48 10 4 ( I R ) 2 + 7 I R − 0.41
P t s l = A d t l i g h t P t 1
where Let P t 1 denote the power of supplementary lighting per square meter inside the greenhouse during time period t. I R represents the amount of artificial light, which is converted to the artificial photosynthetic light intensity I p s by the relation: I p s = 11.57 I R . Let P t s l be the total supplementary lighting power in the greenhouse during period t. A is the total area of the greenhouse. d t l i g h t is a binary variable that indicates whether the supplementary lighting is in operation at time t . and I i n denotes the total solar radiation intensity inside the greenhouse, with the detailed calculation procedure provided in reference [19].
D = L i n t e n s i t y T L i g h t 3600 / 10 6
where Let D denote the daily total photosynthetically active radiation in mol/(m2·d). L i n t e n s i t y represents the light intensity in μ m o l / m 2 · s , and T L i g h t is the duration of light exposure.
The amount of artificial light is given by the following equation:
I R = min ( L intensity , I t L S P ) − I i n
where I t L S P is the light saturation point.

2.3.2. Irrigation Load Model

Affected by the growth characteristics of crops, the irrigation load is concentrated, and irrigation is concentrated in specific periods of a day. In addition, the water requirement of crops is a dynamically changing indicator, which is mainly determined by the type of crops, growth stage, and growing region. The crop irrigation demand in this paper refers to that in Reference [20].
E T c = K c ⋅ E T 0
K c = K c t a b + [ 0.04 ( U 2 − 2 ) − 0.04 ( r ( H min ) − 45 ) ] ( h 3 ) 0.3 ]
E T 0 = 0.408 Δ ( R n − G ) + γ 900 T + 273 U 2 ( e s − e a ) Δ + γ ( 1 + 0.34 U 2 )
where E T c denotes the crop water requirement. E T 0 denotes the reference evapotranspiration. K C denotes the crop coefficient. K c t a b denotes the crop coefficient under standard conditions at different growth stages. r ( H min ) denotes the average value of the daily minimum relative humidity in this growth stage. h denotes the height of the crop in this growth stage. R n denotes the net surface radiation, with the unit MJ/(m2·d). G denotes the soil heat flux, with the unit MJ/(m2·d). T denotes the daily average air temperature. e s denotes the saturation vapor pressure. e a denotes the actual vapor pressure. Δ denotes the slope of the saturation vapor pressure-air temperature relationship curve at T . γ denotes the psychrometric constant. U 2 denotes the wind speed at 2 m height.
The formula for calculating the power consumption required for irrigation is shown in Equation (11):
P t i r = d t w a t e r W ρ g Q H η w a t e r
where P t i r denotes the irrigation power consumption at time period t. d t w a t e r is the decision variable for irrigation. W is the irrigation weight, used to represent the crop water demand in different time periods. ρ is the density of water. g is the gravitational acceleration. H is the head. Q is the flow rate, and its calculation formula is Q = E T c A ρ t , m3/s. η w a t e r is the efficiency of the water pump. The start and stop of the irrigation system are determined by the binary variable d t w a t e r .
The calculation of irrigation weight is as shown in the formula:
W i = W b + ω t i ⋅ λ 1 + ω m i ⋅ λ 2 ∑ i | d t w a t e r = 1 ( W b + ω t i ⋅ λ 1 + ω m i ⋅ λ 2 )
where W b is the irrigation weight. ω t i and ω m i are the temperature factor and soil moisture factor of the i time period, respectively. λ 1 , λ 2 are the weight proportions of temperature and moisture.

2.3.3. Ventilation Load Model

Intermittent ventilation is currently the most commonly used ventilation control method for greenhouses. By setting the time-periodic ventilation frequency and duration, it can help improve the crop growth microenvironment, periodically supplement carbon dioxide to promote the accumulation of photosynthetic products, and alleviate the problem of carbon source deficiency in photosynthesis under enclosed environments. The calculation of ventilation power consumption is shown as follows:
P total = P fan ⋅ t on ⋅ N T ⋅ η
where P total is the average power consumption. P fan is the rated power of a single fan. t on is the duration of a single ventilation, in h. N is the ventilation frequency. T is the total time period duration, in h. η is the system comprehensive efficiency coefficient (including variable frequency loss, mechanical transmission efficiency, etc.).

2.3.4. Greenhouse Heat Load Model

Considering that solar radiation provides light intensity to crops while also making the greenhouse absorb part of the heat, and that atmospheric exchange in the greenhouse will carry away part of the heat, the heat load model of the greenhouse is as follows:
T t + 1 i n = T t i n + μ ⋅ Q t s p + Q t s l − Q t l o s s c a i r ρ a i r v g h
Q t l o s s = ∑ j = 1 n u j A j ( T t i n − T t o u t ) + ∑ i = 1 k u i A i ( T t i n − T t o u t ) + 0.5 k w i n d v g h N ( T t i n − T t o u t )
Q t s l = I t ⋅ A ⋅ τ
where T t i n is the temperature inside the greenhouse at time t. Q t s p is the heat supply. Q t s l is the radiant heat. μ is a flexible coefficient. In this paper, the temperature difference and the day-night temperature threshold are set, and it can be adaptively adjusted to prevent temperature from exceeding the limit. c a i r is the specific heat capacity of air. ρ a i r is the air density. v g h is the volume of the greenhouse. u j is the heat transfer coefficient of the j part of the envelope structure. A j is the surface area of the j part of the envelope structure. u i is the ground heat transfer coefficient of the i area. A i is the reference range area of the i greenhouse ground calculation point from the peripheral structure. T t o u t respectively represents the outdoor temperature of the greenhouse in time period t, in °C. k w i n d represents the wind factor. N represents the air change rate per hour. I t is the solar radiation intensity. A is the contact area. τ is the light transmittance of the contact material.

2.4. Uncertainty Quantification Method

The triangular fuzzy chance-constrained method is selected in this study due to its optimal alignment with the inherent fuzzy uncertainties and physiological-environmental constraint requirements within APIES, balancing theoretical rigor and model tractability [21]. Aiming at the fluctuation of wind-solar output and agricultural loads caused by meteorological environments (temperature, radiation, wind speed), triangular fuzzy chance constraints are adopted to describe uncertainty. Considering the general applicability of triangular fuzzy chance constraints in uncertainty analysis for wind and photovoltaic power output, this paper processes the uncertainties in wind-photovoltaic output and load by referencing the relevant calculations in the literature [22], as detailed below:
P ˜ t p v = k p 1 , 1 , k p 3 P t p v P ˜ t w t = k w 1 , 1 , k w 3 P t w t P ˜ t l o a d = k l 1 , 1 , k l 3 P t l o a d
where P ˜ t p v , P ˜ t w t and P ˜ t l o a d , and are fuzzy representations of photovoltaic output, wind output, and electrical load. k p 1 , k p 3 , k w 1 , k w 3 , k l 1 and k l 3 are proportional coefficients. P t p v , P t w t and P t l o a d are predicted output and load of photovoltaic and wind power, respectively, load.

3. Bi-Level Optimal Operation Model of APIES

With the goal of “dual coordination of economy and growth environment satisfaction”, the model adopts an alternating iteration mechanism of “upper-layer cost optimization—lower-layer environment optimization” to achieve dynamic balance. The upper layer is formulated as a typical Mixed-Integer Linear Programming (MILP) model, for which the Gurobi solver can accurately and efficiently obtain the global optimal solution to minimize costs. The lower layer adopts the NSGA-II algorithm to effectively handle the complex, multi-objective, and nonlinear optimization problem of crop growth environment satisfaction. This dual-layer solution structure is established accordingly in this paper. Through the two-layer iterative framework proposed in this paper, combined with the NSGA-II algorithm and the Gurobi solver, the upper-layer objective can achieve the theoretical minimum within the optimal range of the lower layer, thereby obtaining the optimal solution for the two-layer optimization model (Figure 2).
Figure 2. Two-layer solution flow chart.

3.1. Upper-Level Optimization Model

3.1.1. Upper-Level Objective Function

The upper level aims to minimize the operation cost of the APIES constructed in this paper, and the objective function is:
min F = ∑ t = 1 24 ( C t g r i d + C t r . e + C t s o c + C t p u n i s h )
C t g r i d = P t b u y C b u y − P t s e l l C s e l l C t r . e = P t pv C p v + P t w t C w t C t s o c = ( P t c h a + P t d i s ) C s o c C t p u n i s h = ( P t p v . p r e − P t p v ) C 1 + ( P t w t . p r e − P t w t ) C 2  
where F is the total operation cost of the IES. C t g r i d is the transaction cost between the IES in the agricultural industrial park and the main grid at time t. C t r . e denotes the operation and maintenance (O&M) cost of renewable energy at time t. C t s o c represents the O&M cost of electrical energy storage at time t. C t p u n i s h is the penalty cost for wind curtailment and PV curtailment at time t. C b u y is the time-of-use (TOU) price. C s e l l is the electricity selling price. C p v and C w t are the O&M cost coefficients of PV and wind turbines, respectively. C s o c is the O&M coefficient of electrical energy storage. C 1 and C 2 are the penalty coefficients for PV curtailment and wind curtailment, respectively. P t p v is the PV power at time t. P t w t is the wind power at time t. P t b u y and P t s e l l are the power for purchasing and selling electricity at time t, respectively. P t c h a is the charging power at time t. P t d i s is the discharging power at time t. P t p v . p r e is the predicted PV output at time t. P t w t . p r e is the predicted wind power output at time t.

3.1.2. Upper-Level Constraints

(1)
Renewable Energy Output Constraints
In APIES, the distributed PV and wind turbines have the following output constraints:
0 ≤ P t r . e ≤ P t r . e . pre ≤ P max r . e P t r . e . cur = P t r . e . pre − P t r . e ≤ P max r . e . cur
where P t r . e denotes the renewable energy output at time t. P t r . e . pre is the predicted output of renewable energy at time t. P max r . e is the maximum output of renewable energy. P t r . e . cur represents the curtailment power of renewable energy at time t. P max r . e . cur is the maximum curtailment power.
(2)
Electrical Energy Storage (EES) Output Constraints
E 1 s = E 24 s E _ s ⩽ E t s o c ⩽ E ¯ s P ¯ e ⩽ P t e ⩽ P ¯ e 0 ⩽ S t + R t ⩽ 1 P t E = P t − 1 E + P t c h a η s c h a − P t d i s η s d i s
where E _ s and E ¯ s are the lower and upper limits of the EES state of charge (SOC), respectively. P ¯ e and P ¯ e are the lower and upper limits of the EES charging/discharging power, respectively. S t and R t are binary variables, indicating the charging/discharging states of the energy storage device at each time period (0 for non-operation, 1 for charging or discharging). P t E is the remaining capacity of EES at time t. P t c h a and P t d i s are the charging and discharging powers of EES at time t, respectively. η s c h a and η s d i s represent the charging and discharging efficiencies of EES.
(3)
Power Balance Constraint
P t r . e + P t b u y − P t s e l l − P t c h a r g e + P t d i s c h a r g e = P t l o a d
where P t l o a d is the electrical load power of the greenhouse at time t.
Considering the influence of external meteorological environment uncertainties on the output of distributed new energy and loads during agricultural production, the triangular fuzzy chance constraint is used to analyze the uncertainty. The system’s credibility fuzzy chance constraints are shown as follows, and the method in Reference [23] is adopted to convert the spinning reserve constraint under triangular fuzzy parameters into a clear equivalent class.
C r P ˜ t p v + P ˜ t w t + P t b u y − P t s e l l − P t c h a r g e + P t d i s c h a r g e ≥ P ˜ t l o a d ≥ α
C r P ˜ t p v + P ˜ t w t + P t b u y − P t s e l l − P t c h a r g e + P t d i s c h a r g e ≤ P ˜ t l o a d ≥ α
where C r ⋅ is the credibility expression. α is the credibility degree.
(4)
Load Shifting Constraint
For supplementary light loads, the part that exceeds the upper limit of the set value is defined as a shiftable load, and the constraint is as follows:
λ τ P t , min s l ⩽ P t t r a n , s l ⩽ λ τ P t , max s l
where P t , max s l and P t , min s l are the maximum and minimum values of the shiftable supplementary light load, respectively. P t t r a n , s l is the shifted supplementary light load power. The binary variable λ indicates the time-shifting state in a certain period, and λ τ = 1 means that time shifting occurs in period τ.
In addition, it is necessary to ensure that after time-shifting the supplementary light load, the light environment in each period meets the physiological state of the crop, that is, to ensure that the light intensity in each period is between the light compensation point and the light saturation point.

3.2. Lower-Level Optimization Model

3.2.1. Lower-Level Objective Function

The lower-level optimization objective is the satisfaction degree of crop environmental demands in the greenhouse, mainly including the growth environment demands for temperature, light, humidity, and ventilation. The objective function is as follows:
max F = W 1 ⋅ G + W 2 ⋅ D + W 3 ⋅ E + W 4 ⋅ W
where W i is the weight of each factor. G is the daily accumulated temperature of crops. D is the daily light integral (DLI) of crops. E is the crop irrigation amount. W is the crop ventilation demand.
The calculation of the weight W i is shown as follows:
W i S , R = α i S ⋅ β i R ∑ j = 1 4 α j S ⋅ β j R
α i S ∈ α i j 4 × 4
β i R ∈ β i j 4 × 4
where α i S is the basic weight of the crop growth stage. α i j 4 × 4 is a 4 × 4 array constructed based on the proportion of the above four environmental parameters (temperature, light, humidity, and ventilation) in different periods. β i R is the regional correction factor. β i j 4 × 4 is an array constructed by dividing crop growing regions into four major categories.
Meanwhile, to adapt to real-time environmental changes in practical applications, an adaptive coefficient γ i t can be introduced. Its value primarily depends on the crop’s growth environment and varies across different growth stages and energy consumption periods, as the requirements for growth conditions differ both between developmental phases and within distinct time intervals.
W i ′ S , R , t = W i S , R ⋅ γ i t ∑ j = 1 4 W j S , R ⋅ γ j t
where γ i t can adjust and correct the weight according to the real-time change in any growth environment parameter.

3.2.2. Lower Layer Constraints

(1)
Daily Accumulated Temperature Constraint
The crop’s demand for the thermal environment can be measured by the daily accumulated temperature. The daily accumulated temperature constraint in the crop environmental satisfaction degree is:
G min ≤ G ≤ G max
G = ( T o − T d ) 24 ∑ t = 1 24 R gd
R gd = ( T t gh − T d ) T o − T d , T d ≤ T t gh ≤ T o   0 ,   T t gh < T d , T u < T t gh ( T u − T t gh ) T u − T o , T o ≤ T t gh ≤ T u
where G is the safety range for daily accumulated temperature. G max and G min are the upper and lower limits of the safety range, set to 15.5 °C and 18.5 °C, respectively. R gd is the growth rate. T t gh is the greenhouse temperature at time t. T d , T o , and T u are the minimum, optimal, and maximum temperatures for crop growth, respectively.
(2)
Daily Total Photosynthetically Active Radiation Constraint
The daily total photosynthetically active radiation represents the crop’s demand for the light environment. Its constraint is as follows:
D min ≤ D ≤ D max
where D is the safety range for the daily total photosynthetically active radiation. D min and D max are the lower and upper bounds of D, set to 18 and 24 mol ( m 2 ⋅ d ) , respectively.
(3)
Irrigation Amount Constraint
Adequate water supply not only promotes crop photosynthesis and nutrient uptake but also facilitates root system development. The minimum irrigation threshold should be sufficient to meet crop water requirements and prevent physiological disorders caused by drought, while the maximum threshold should avoid hazards such as root hypoxia and disease proliferation due to excessive moisture. Therefore, the constraint for the irrigation amount is set as follows:
E min ≤ E ≤ E max
where E max and E min are the upper and lower bounds of the irrigation amount, set to 120% and 80% of the total irrigation requirement, respectively.
(4)
Ventilation Constraint
In agricultural greenhouses, ventilation is one of the key measures for regulating the internal climate and ensuring crop growth. Appropriate ventilation frequency can effectively improve the physiological environmental quality for crops. However, ventilation requirements differ between daytime and nighttime, aiming to ensure air circulation or prevent excessive heat loss. Thus, the constraints for ventilation frequency are set as follows:
W d min ≤ W d ≤ W d max W n min ≤ W n ≤ W n max
where W d and W n represent the ventilation frequency during daytime and nighttime, respectively. W d max and W d min are the upper and lower limits for daytime ventilation frequency. W n max and W n min are the upper and lower limits for nighttime ventilation frequency.

3.3. Solution of the Bi-Level Optimization Model

This paper employs a bi-level optimization model to address the energy management problem in greenhouses. The upper-level optimization focuses on energy cost minimization and is solved as a linear programming problem using the YALMIP toolbox with the Gurobi solver. The lower-level optimization deals with crop growth environmental satisfaction. Based on the NSGA-II multi-objective genetic algorithm, which is suitable for optimization problems involving multi-agent trade-offs [24,25,26], it maximizes environmental satisfaction under the cost budget constraint provided by the upper level, generating a Pareto optimal solution set.
In each iteration, the upper level passes the current electrical load and cost budget to the lower level, allowing a moderate cost to overrun to optimize environmental parameters. The lower level returns the environmental parameter combination with the highest satisfaction degree, which is used to update and correct the electrical load. The upper and lower levels optimize alternately until the changes in both levels between two consecutive iterations are less than 0.1% or the maximum number of iterations is reached.

4. Case Analysis

4.1. Basic Data

Given cucumber’s status as a primary cultivated fruit and vegetable in northern China and its high sensitivity to growth conditions, it is selected as the research subject in this study. To verify the validity of the bilayer model constructed in this paper, a cucumber greenhouse in a small agricultural park in Xinjiang, China, was selected, with the setting period of cucumbers as the time node. Referring to the physiological requirements of cucumbers to various growth environments at this growth stage, the validity of the method constructed in this paper for meeting the physiological requirements of crops and the economy of the model were analyzed and verified. With 24 h a day as one running cycle and a time step of 1 h, the analysis was conducted. The parameters of the equipment in the greenhouse are detailed in reference [27].

4.2. The Impact of Uncertainty on the Operation of APIES

Based on triangular fuzzy chance-constrained programming, this study analyzes the impact of different confidence levels on system economics as well as key influencing factors in crop growth environmental satisfaction—specifically, daily thermal integral and daily total photosynthetically active radiation.
Regarding the economic and low-carbon performance of the system, the operational costs and renewable energy utilization rates under different confidence levels are analyzed. As shown in Figure 3a, as the confidence level increases, the system operational costs exhibit an upward trend, while a high renewable energy utilization rate is maintained across all confidence levels. Considering that the selected cucumber growth period falls within the fruit-setting stage, the daily thermal integral and daily total photosynthetically active radiation are taken as examples to analyze the variation patterns of these two key growth environmental state variables under different confidence levels. Figure 3b reveals that the daily total photosynthetically active radiation gradually increases with rising confidence levels, whereas the daily thermal integral only begins to increase when the confidence level exceeds 0.75. Nevertheless, under all confidence levels, both key growth environmental variables remain within the specified threshold ranges.
Figure 3. The Impact of Different Confidence Levels on System Performance. (a) Chart of changes in operating costs and utilization of new energy; (b) Trend chart of daily accumulated temperature versus daily total photosynthetically active radiation.
It is also important to note that the increasing trends of daily thermal integral and daily total photosynthetically active radiation with higher confidence levels are among the primary reasons for the rise in system operational costs.

4.3. Comparison of Case Study Results

To verify the impact of different operational schemes on system economy and environmental satisfaction under external environment uncertainties, several scenarios with a confidence level of 0.9 are established for comparative analysis:
The optimization results for each scenario are presented in Table 1.
Table 1. Optimization Results for Each Scenario.
Scenario 1: Single-level optimal scheduling considering the optimal environment matching mechanism.
Scenario 2: Bi-level optimal scheduling with the growth environment satisfaction set to an average weight.
Scenario 3: Building upon Scenario 2, with adaptive environmental weights and time-shifting processing of the supplementary lighting load.
As observed in Figure 4 and Figure 5, the optimal scheduling under the environmental matching mechanism results in a significant total load. The thermal and ventilation loads are relatively evenly distributed over time. In contrast, the supplementary lighting load is primarily concentrated during daytime periods, specifically from 09:00–11:00 and 17:00–18:00, when the natural light intensity is insufficient to meet crop requirements. Consequently, supplementary lighting is necessary during these intervals, and its demand exhibits a clear negative correlation with solar irradiance. Similarly, the irrigation load, which is also concentrated in two distinct periods but with fairly uniform water volume, shows no significant variation across its operating intervals.
Figure 4. Optimization Results Chart for Scenario 1.
Figure 5. Load Distribution for Scenario 1.
A comparative analysis of Figure 4 and Figure 6 reveal that Scenario 2 achieves a notable reduction in total load demand compared to Scenario 1, with a decrease of 20%. Despite this reduction, the overall load profile shape remains highly similar. Furthermore, as evidenced by the data in Table 1, the bi-level optimization model reduces system operating cost and electricity purchase by 10.56% and 8.38%, respectively, compared to Scenario 1. It is noteworthy that the environmental satisfaction in Scenario 2 is 5.9% lower than that of the optimal environment matching mechanism. However, by ensuring that physiological demand thresholds for crops are met and selecting a Pareto-optimal solution from the economy-growth environment multi-objective framework, the proposed strategy achieves an optimal trade-off between system operating cost and crop growth requirements. This demonstrates the capability of the bi-level optimization model to realize the synergistic optimization of system economic operation and the agricultural growth environment.
Figure 6. Optimization Results Chart for Scenario 2.
Building upon Scenario 2, Scenario 3 incorporates dynamic time-shifting optimization for the supplementary lighting load. As illustrated in Figure 7, this strategy results in a more flattened load profile. By flexibly scheduling the lighting load, Scenario 3 effectively reduces the peak demand by 27.77% and 24.39% compared to Scenarios 1 and 2, respectively, thereby significantly mitigating peak-to-valley pressure. Furthermore, Scenario 3 employs a bi-level optimization framework that operates under the actual weights of various growth environmental factors, all of which are integrated into the satisfaction assessment. The upper-level optimization focuses on economic objectives, achieving reductions in overall energy costs and enhancing system low-carbon operation. Concurrently, the lower-level optimization refines the crop growth environment satisfaction, ultimately creating a more favorable growth micro-environment for the crops.
Figure 7. Optimization Results Chart for Scenario 3.
As can be concluded from Table 1, compared with the traditional single-objective scheduling model (Scenario 1), the proposed two-layer model—which optimizes both crop growth conditions and operating costs—reduces operating costs by 10.8%, while the environmental satisfaction index of crops only decreases by 1.6%. Therefore, it demonstrates superior performance in operational economy. Moreover, after introducing environmental factors, Compared to Scenario 2 the dynamic adaptation model improves the environmental satisfaction index of crops by 4.3%.

5. Conclusions

To address the intelligent energy management needs of Agricultural Park Integrated Energy Systems (APIES) with renewable energy under complex and variable external meteorological conditions, this paper proposes a collaborative optimization framework aimed at precise agricultural production regulation and enhanced system robustness. A bi-level operational model was constructed for the synergistic optimization of economy and growth environment in APIs. A case study of an agricultural park in Xinjiang, China, was conducted, yielding the following conclusions:
(1)
Robustness analysis based on triangular fuzzy chance constraints demonstrates that within the confidence interval of [0.5, 0.9], the primary physiological demands of crops—namely, daily accumulated temperature and daily total photosynthetically active radiation—are satisfied within their required thresholds. Furthermore, this approach enables a distributed renewable energy utilization rate exceeding 90%.
(2)
In terms of system economy and environmental satisfaction, the proposed optimal environment matching model with environmental factor weights reduces the operating costs and electricity purchase volume by 10.8% and 8.38%, respectively, compared to a traditional single-level scheduling model. Additionally, it improves environmental satisfaction by 4.3% compared to a bi-level model utilizing fixed average weights.
Aiming at the uncertainty of external meteorological conditions, this paper designs a two-layer optimization model integrating crop growth and energy scheduling. Its limitation lies in the relatively short scheduling period, which fails to fully reflect the sensitivity analysis of crops to environmental variations. Future research will focus on optimization studies across extended growth cycles and investigate performance under varying conditions using novel, high-efficiency multi-objective optimization algorithms. The schematic diagram illustrating the implementation of this method is shown in Appendix A.

Author Contributions

Conceptualization, Z.W. Investigation, Y.W. and L.X. Writing—original draft, H.S. Writing—review and editing, X.N. and H.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported in part by Science and Technology Project of State Grid Jilin Electric Power Company Limited “Unveiling the List and Appointing the Leader” (2023JBGS-01).

Data Availability Statement

The original contributions presented in the study are included in the article. further inquiries can be directed to the corresponding author.

Conflicts of Interest

Authors Zeyi Wang, Yao Wang, Hongyu Sun and Xueshan Ni were employed by the Changchun Power Supply Company, State Grid Jilin Electric Power Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Appendix A

Figure A1. Schematic diagram of the graphic and textual abstract.

References

  1. Zhang, H.; Luo, Y.; Fang, S.; Niu, T.; Yang, J.; Liao, R. Coordinated Energy-Carbon Management for Agricultural Integrated Energy System. IEEE Trans. Ind. Appl. 2025, 61, 4603–4612. [Google Scholar] [CrossRef] [Scilit]
  2. Liu, C.; Wang, C.; Yao, W.; Liu, C. Distributed Low-Carbon Economic Dispatching for Multiple Park-level Integrated Energy Systems Based on Improved Shapley Value. IEEE Trans. Ind. Appl. 2024, 60, 8088–8102. [Google Scholar] [CrossRef] [Scilit]
  3. Yu, S.; Zhou, S.; Chen, N. Multi-objective optimization of capacity and technology selection for provincial energy storage in China: The effects of peak-shifting and valley-filling. Appl. Energy 2024, 355, 122289. [Google Scholar] [CrossRef] [Scilit]
  4. Wang, K.; Yang, J.; Zhang, C.; Wen, F.; Lu, G. Joint energy-frequency regulation electricity market design for the transition towards a renewable-rich power system. Int. J. Electr. Power Energy Syst. 2024, 155, 109504. [Google Scholar] [CrossRef] [Scilit]
  5. Zhao, Z.; Guo, J.; Luo, X.; Lai, C.S.; Yang, P.; Lai, L.L.; Li, P.; Guerrero, J.M.; Shahidehpour, M. Distributed robust model predictive control-based energy management strategy for islanded multi-microgrids considering uncertainty. IEEE Trans. Smart Grid 2022, 13, 2107–2120. [Google Scholar] [CrossRef] [Scilit]
  6. Bakhshi, R.; Sandborn, P.A. A Return on Investment Model for the Implementation of New Technologies on Wind Turbines. IEEE Trans. Sustain. Energy 2018, 9, 284–292. [Google Scholar] [CrossRef]
  7. Chen, Z.; Li, Z.; Lin, D.; Xie, C.; Wang, Z. Multi-Time-Scale Optimal Scheduling of Integrated Energy System with Electric-Thermal-Hydrogen Hybrid Energy Storage under Wind and Solar Uncertainties. J. Mod. Power Syst. Clean Energy 2024, 13, 904–914. [Google Scholar] [CrossRef] [Scilit]
  8. Tang, Y.; Xun, Q.; Zheng, Z.; Min, F.; Deng, C.; Xie, J.; Yang, H. An Optimization Framework for Component Sizing and Energy Management in Electric-Hydrogen Hybrid Energy Storage Systems. IEEE Trans. Sustain. Energy 2025, 16, 2182–2196. [Google Scholar] [CrossRef] [Scilit]
  9. Sha, Z.; Cui, C.; Wang, Z.; Yu, F.; Yin, S.; Yang, Z.; Cao, C.; Huang, X.; Liu, Z. An Optimal Capacity Configuration Method for a Renewable Energy Integration-Transmission System Considering Economics and Reliability. Symmetry 2025, 17, 1880. [Google Scholar] [CrossRef] [Scilit]
  10. Ramírez-Carmona, M.; Rendón-Castrillón, L.; Ocampo-López, C.; García-Ríos, M.; Lamilla-Mendoza, X.; Piedrahíta-Pérez, S.; Rodríguez-Estrada, J.; Samaan-Salazar, V.; Urrea-López, S.; Valencia-Yepes, D.; et al. Anhydrous Ethanol Production from Discarded Fruits Using Fermentation and a Green Dehydration System. Processes 2025, 13, 1854. [Google Scholar] [CrossRef] [Scilit]
  11. Isemin, R.; Tabet, F.; Shevchenko, A.; Mikhalev, A.; Kuzmin, S.; Midekssa, M.J.; Milovanov, O.; Klimov, D.; Milovanov, K.; Faleeva, Y.; et al. Assessment of the Influence of Temperature and Exothermic Effects During Torrefaction on the Properties of Agricultural Waste. Processes 2025, 13, 3294. [Google Scholar] [CrossRef] [Scilit]
  12. Dong, Y.; Zhang, H.; Wang, C.; Zhou, X. A novel hybrid model based on Bernstein polynomial with mixture of Gaussians for wind power forecasting. Appl. Energy 2021, 286, 116545. [Google Scholar] [CrossRef] [Scilit]
  13. Xia, S.W.; Cai, L.Y.; Tong, M.Z.; Wu, T.; Li, P.; Gao, X. Regulation Flexibility Assessment and Optimal Aggregation Strategy of Greenhouse Loads in Modern Agricultural Parks. Prot. Control Mod. Power Syst. 2024, 9, 98–111. [Google Scholar] [CrossRef] [Scilit]
  14. Vyas, S.D.; Subash, V.K.; Sai Venkatesh, M.M.; Sinha, A.; Guha, A.; Dey, S. Energy- and Safety-Aware Operation of Battery-Powered Autonomous Robots in Agriculture. IEEE Trans. AgriFood Electron. 2024, 2, 51–58. [Google Scholar] [CrossRef] [Scilit]
  15. Bokkisam, H.R.; Singh, S.; Acharya, R.M.; Selvan, M.P. Blockchain-based Peer-to-Peer Transactive Energy System for Community Microgrid with Demand Response Management. CSEE J. Power Energy Syst. 2022, 8, 198–210. [Google Scholar] [CrossRef] [Scilit]
  16. Misra, V.; Bozkurt, A.; Calhoun, B.; Jackson, T.; Jur, J.S.; Lach, J.; Lee, B.; Muth, J.; Oralkan, Ö.; Öztürk, M.; et al. Flexible Technologies for Self-Powered Wearable Health and Environmental Sensing. Proc. IEEE 2015, 103, 665–681. [Google Scholar] [CrossRef] [Scilit]
  17. Yue, M.; Lambert, H.; Pahon, E.; Roche, R.; Jemei, S.; Hissel, D. Hydrogen energy systems: A critical review of technologies, applications, trends and challenges. Renew. Sustain. Energy Rev. 2021, 146, 111180. [Google Scholar] [CrossRef] [Scilit]
  18. Bødal, E.F.; Holm, S.E.; Subramanian, A.; Durakovic, G.; Pinel, D.; Hellemo, L.; Ortiz, M.M.; Knudsen, B.R.; Straus, J. Hydrogen for harvesting the potential of offshore wind: A North Sea case study. Appl. Energy 2024, 357, 122484. [Google Scholar] [CrossRef] [Scilit]
  19. Bernier, F.; Cao, J.; Cordy, M.; Ghamizi, S. PowerGraph-LLM: Novel Power Grid Graph Embedding and Optimization With Large Language Models. IEEE Trans. Power Syst. 2025, 40, 5483–5486. [Google Scholar] [CrossRef] [Scilit]
  20. Bin Poyen, F.; Hazra, S.; Sengupta, N.; Banerjee, S. Poyen’s Fuzzy Logic Controlled Automatic Irrigation (FCAI): Precision Irrigation Scheduling Scheme. IEEE Trans. Instrum. Meas. 2022, 72, 9501109. [Google Scholar] [CrossRef] [Scilit]
  21. Yue, Q.; Guo, P. Managing agricultural water-energy-food-environment nexus considering water footprint and carbon footprint under uncertainty. Agric. Water Manag. 2021, 252, 106899. [Google Scholar] [CrossRef] [Scilit]
  22. Kokila, A.; Deepa, G. Improved fuzzy multi-objective transportation problem with Triangular fuzzy numbers. Heliyon 2024, 10, e32895. [Google Scholar] [CrossRef] [Scilit]
  23. Blaauwbroek, N.; Nguyen, P.H.; Konsman, M.J.; Shi, H.; Kamphuis, R.I.G.; Kling, W.L. Decentralized Resource Allocation and Load Scheduling for Multicommodity Smart Energy Systems. IEEE Trans. Sustain. Energy 2015, 6, 1506–1514. [Google Scholar] [CrossRef] [Scilit]
  24. Gayibov, A.; Gasimov, V. Interactive Evolutionary NSGA-II with Machine Learning for Efficient Hyperparameter Optimization of Convolutional Neural Networks. In Proceedings of the 2025 6th International Conference on Problems of Cybernetics and Informatics (PCI), Baku, Azerbaijan, 26–28 August 2025; pp. 1–5. [Google Scholar] [CrossRef] [Scilit]
  25. Li, X.; Dong, H.; Li, H.; Li, M.; Sun, Z. Optimization Control of Front-End Speed Regulation (FESR) Wind Turbine Based on Improved NSGA-II. IEEE Access 2019, 7, 45583–45593. [Google Scholar] [CrossRef] [Scilit]
  26. Liu, R.; Yang, T.; Sun, G.; Lin, S.; Li, F.; Wang, X. Multi-Objective Optimal Scheduling of Community Integrated Energy System Considering Comprehensive Customer Dissatisfaction Model. IEEE Trans. Power Syst. 2023, 38, 4328–4340. [Google Scholar] [CrossRef] [Scilit]
  27. Liu, M.; Zhao, L.; Chen, J.; Wang, J.; Tian, M.; Li, Q. Research on Energy Management Strategies for Intelligent Agricultural Integrated Energy Systems Considering Multi-Stage Growth Requirements of Crops. Power Syst. Technol. (In Chinese) [CrossRef]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.