An AI-driven grassland extreme climate simulation experimental device and ecological response prediction method

The grassland extreme climate simulation device, which integrates an experimental chamber, an undisturbed soil column planting unit, and an edge intelligent control unit, solves the problems of reproducing complex extreme climate events and monitoring ecological responses, and realizes real-time control and high-reliability prediction of grassland ecological responses.

CN123072269APending Publication Date: 2026-11-10NORTHWEST INST OF ECO ENVIRONMENT & RESOURCES CAS
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611166608.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-03
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing grassland extreme climate simulation devices are unable to reproduce complex extreme climate events, have incomplete monitoring of above-ground and underground responses, cannot participate in control in real time, and have low reliability of model extrapolation under small sample conditions.

Method used

An artificial intelligence-driven grassland extreme climate simulation experimental device integrates an experimental chamber, an undisturbed soil column planting unit, a climate execution unit, a multi-parameter three-dimensional monitoring unit, and an edge intelligent control unit. Through a physical-data dual-driven ecological response model and constrained predictive control, it achieves real-time monitoring and adaptive control of grassland ecological response.

Benefits of technology

It can reproduce complex extreme climate events such as heat waves superimposed on droughts, and rainstorms superimposed on strong winds, improving the synchronous monitoring accuracy of above-ground and underground responses and the predictive reliability of models under extreme conditions, reducing sample damage, and is suitable for long-term operation at field sites and laboratories.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123072269A_ABST
    Figure CN123072269A_ABST
Patent Text Reader

Abstract

This invention discloses an artificial intelligence-driven grassland extreme climate simulation experimental device and ecological response prediction method, belonging to the field of environmental simulation technology. It includes an experimental chamber, an undisturbed soil column planting unit, execution units for temperature, humidity, rainfall, wind speed, light intensity, and CO2 concentration, a multi-parameter three-dimensional monitoring unit, and an edge intelligent control unit. The edge intelligent control unit encapsulates the chamber environment, soil profile, canopy physiological data, and image data into a state vector with a quality mask, constructing a physical-data dual-driven ecological response model containing a carbon-water-energy conservation residual layer, data imputation branches, and a gated fusion layer. Based on the model prediction results, it continuously corrects the control quantities of each execution unit, achieving accurate reproduction of extreme climate scenarios such as combined heat waves, droughts, rainstorms, strong winds, cold and low light conditions, and increased CO2. This invention improves the repeatability of grassland extreme climate experiments, the integrity of response data, and the reliability of model extrapolation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of environmental simulation technology, and in particular to an artificial intelligence-driven grassland extreme climate simulation experimental device and ecological response prediction method. Background Technology

[0002] Global climate change has led to significant changes in the frequency, intensity, and duration of extreme climate events such as heat waves, extreme droughts, short-duration torrential rains, strong winds, and freeze-thaw anomalies. Grassland ecosystems, characterized by low canopies, concentrated root systems, rapid changes in soil moisture, and strong coupling of carbon and water processes, are highly sensitive to complex extreme climate events. To reveal the response mechanisms of grassland ecosystems to extreme climates, it is necessary to simultaneously reproduce the rapid transitions of climatic factors and the continuous responses of aboveground and belowground processes under controlled conditions.

[0003] Existing open-top air chambers, field heating greenhouses, and rain shelters can provide some degree of temperature increase or water control, but they are significantly affected by external background climate disturbances and cannot stably reproduce complex events such as heat waves superimposed with droughts, and heavy rains superimposed with strong winds, according to a predetermined script. Conventional artificial climate chambers are mainly designed for small potted plants or seed germination experiments and are usually unable to accommodate undisturbed soil columns of grassland containing complete root systems and soil profiles. They also lack simultaneous observation of near-surface wind fields, natural rainfall pulses, leachate, and root dynamics.

[0004] Existing ecological modeling methods mostly rely on offline modeling using remote sensing imagery or ground survey data. Purely physical process models have many parameters and are difficult to transfer to other regions; purely data-driven models are prone to overfitting under small sample extreme climate scenarios and may produce predictions that do not satisfy the conservation relationships of water, energy, or carbon flux. Model outputs usually lag behind the experimental control process and cannot participate in the real-time control of the experimental chamber actuators.

[0005] Furthermore, existing grassland climate simulation experiments mostly employ pre-set scripts and PID control, with the control target primarily being a single environmental factor such as temperature or humidity, without considering the grassland ecological response itself as a closed-loop constraint. When plant canopy temperature, soil water potential, or photosynthetic efficiency have entered dangerous ranges, the system often continues to operate according to the original script, which can easily cause irreversible damage to the samples and makes it difficult to obtain repeatable and interpretable threshold experimental data.

[0006] Therefore, there is an urgent need for a technical solution that combines controllable climate simulation devices, above-ground and underground three-dimensional monitoring, physical constraint artificial intelligence models, and constraint predictive control, so that the experimental chamber can not only passively execute climate scripts, but also adaptively correct control variables based on grassland conditions and model prediction results. Summary of the Invention

[0007] This invention provides an artificial intelligence-driven grassland extreme climate simulation experimental device and an ecological response prediction method, which addresses the problems of existing grassland extreme climate simulation devices that are unable to reproduce complex events, have incomplete monitoring of above-ground and below-ground responses, cannot participate in experimental control in real time with ecological models, and have low reliability of model extrapolation under small sample conditions.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an artificial intelligence-driven grassland extreme climate simulation experimental device and an ecological response prediction method, comprising: The experimental chamber comprises an undisturbed soil column planting unit, a climate control unit, a multi-parameter three-dimensional monitoring unit, and an edge intelligent control unit. The experimental chamber forms a sealed simulated space. The undisturbed soil column planting unit is used to accommodate grassland samples with native root systems and soil profiles. The climate control unit is used to create programmable temperature, humidity, rainfall, wind speed, light, and CO2 concentration conditions within the chamber. The multi-parameter three-dimensional monitoring unit is used to simultaneously collect environmental, soil profile, canopy physiological, and image data within the chamber.

[0009] The edge intelligent control unit includes a data synchronization module, a climate script generation module, a physics-data dual-driven ecological response model, a constraint predictive control module, a small-sample augmentation training module, a model compression and version latching module, and an anomaly-triggered reconstruction module. The data synchronization module performs unified timestamp registration, quality mask generation, and state vector encapsulation on sensor data. The physics-data dual-driven ecological response model uses the conservation relationships of water, energy, and carbon flux as a differentiable physical residual layer, and dynamically fuses it with the data residual branch through a gated fusion layer. The constraint predictive control module solves for the actuator control quantity of the next control cycle based on the model-predicted grassland response and uncertainty interval.

[0010] A method for simulating grassland extreme climates and predicting ecological responses using the aforementioned device includes: inserting undisturbed soil columns into planting units and installing sensors; running an adaptation program to form baseline samples; performing time synchronization, quality mask generation, and feature extraction on multi-source data; training a physics-data dual-driven ecological response model; generating single-factor or composite extreme climate scripts such as heat waves, droughts, rainstorms, strong winds, overcast and cold conditions, freeze-thaw cycles, or increased CO2; using a constraint prediction control module to close-loop regulate each execution module during script execution; and finally outputting ecological response prediction results, key driving factors, experimental reports, and model version records.

[0011] The beneficial effects of the technical solution provided by this invention include at least the following: The method of this invention integrates temperature, humidity, rainfall, wind speed, light, and CO2 concentration execution modules into the same experimental chamber, and arranges sudden, gradual, continuous, and recovery stages through a script generation module, which can reproduce complex extreme climate events such as heat waves superimposed on droughts and rainstorms superimposed on strong winds.

[0012] The method of this invention utilizes an undisturbed soil column planting trough, multi-depth soil sensors, root observation windows, leachate collection bottles, canopy sensors, and in-cabin environmental sensors to form a synchronous monitoring chain above and below ground, enabling environmental drives and ecological responses to have a unified time reference.

[0013] The method of this invention embeds the conservation relationships of water, energy and carbon flux into the model training and inference process, fits complex local responses through data-filled branches, and dynamically allocates the weights of physical branches and data branches according to the quality mask through a gated fusion layer, thereby improving the prediction reliability under conditions of sensor absence, small sample size and extreme scenarios.

[0014] The method of this invention does not only track environmental setpoints, but incorporates grassland ecological stress index, model uncertainty and actuator safety constraints into the constraint prediction control objective, so that the experiment can reduce irreversible damage to the sample while maintaining the authenticity of extreme climate, and improve the repeatability of control curves and response curves.

[0015] The method of this invention achieves real-time inference at the edge, continuous experimentation under network outage conditions, model snapshot rollback, and rapid adaptation in disaster or anomaly situations through small sample augmentation training, model compression, version latching, and anomaly-triggered reconstruction. It is suitable for long-term operation in field sites and laboratories. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the overall structure of the device according to an embodiment of the present invention.

[0018] Figure 2 This is a diagram of the data acquisition, model prediction, and closed-loop control architecture of an embodiment of the present invention.

[0019] Figure 3 This is a cross-sectional schematic diagram of the planting trough and sensor according to an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of the physical-data dual-driven ecological response model according to an embodiment of the present invention.

[0021] Figure 5 This is a flowchart illustrating the simulation experiment and model update process in an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of the composite extreme climate script and control target in an embodiment of the present invention.

[0023] In the diagram: 1. Experimental chamber; 2. Original soil column planting unit; 3. Climate execution unit; 5. Edge intelligent control unit; 6. Cloud and mobile terminal; 21. Grassland vegetation and original soil column; 22. Crushed stone filter layer; 23. Drainage pipe; 24. Leachate collection bottle; 26. Root observation window; 31. Light / heating unit; 32. Array of rain nozzles; 33. Variable speed fan; 34. Deflector plate; 41, 42, 43. Soil sensor; 44. Canopy sensor group; 45. Root observation window. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. Example

[0025] An artificial intelligence-driven grassland extreme climate simulation experimental device and ecological response prediction method.

[0026] like Figure 1 As shown, in one embodiment, the simulation experimental device includes an experimental chamber 1, an undisturbed soil column planting unit 2, a climate execution unit 3, a multi-parameter three-dimensional monitoring unit, and an edge intelligent control unit 5. The experimental chamber forms a sealed simulation space to accommodate grassland samples. The undisturbed soil column planting unit 2 is located within this sealed simulation space and is used to support undisturbed soil columns containing grassland vegetation, roots, and soil profiles. The experimental chamber 1 can be constructed using double-layer polycarbonate panels, fiberglass sandwich panels, transparent tempered glass, or other heat-insulating and transparent materials. The internal wall dimensions of the chamber can be set according to the experimental scale, for example, a length of 2.0 to 5.0 m, a width of 1.5 to 3.0 m, and a height of 2.0 to 3.0 m. A sealed door is provided on one side of the experimental chamber 1, an air vent with an electric valve is provided on the top or side wall, and a condensate drainage structure is provided on the bottom or side wall to discharge excess moisture after heavy rain or high humidity events.

[0027] The undisturbed soil column planting unit 2 includes one or more retractable planting troughs. The planting troughs are used to hold undisturbed soil columns 21 obtained from the target grassland, preserving native vegetation, litter layer, root system, and soil profile within the column. A gravel filter layer 22 and a drainage pipe 23 are installed at the bottom of the planting trough. The drainage pipe 23 is connected to a leachate collection bottle 24 for collecting leachate generated during heavy rain or snowmelt. Sensor insertion holes are provided at different depths on the sidewalls of the planting troughs, and a root observation window 26 is provided on at least one sidewall to allow observation of root growth and death processes via an endoscope, micro-root canal camera, or transparent window.

[0028] Climate execution unit 3 includes at least a temperature execution module, a humidity execution module, a rainfall execution module, a wind speed execution module, a light intensity execution module, and a CO2 concentration execution module. The temperature execution module may include a far-infrared radiation heating plate, an air-cooled or water-cooled evaporator, and a compressor-condenser unit for rapid heating, cooling, and freeze-thaw cycles. The humidity execution module may include an ultrasonic humidifier, a steam humidifier, a condenser dehumidifier, or a rotary dehumidifier. The rainfall execution module may include a water tank, a filter, a variable frequency water pump, a flow meter, a pressure sensor, a solenoid valve, and an array of rain nozzles, with each row of nozzles capable of independent opening and closing to simulate uniform rainfall, rainstorm pulses, or spatially heterogeneous rainfall. The wind speed execution module may include a variable speed axial flow fan, a honeycomb rectifier, and airflow guide plates to create a near-surface horizontal wind field. The light intensity execution module may include LED plant lights with blue, red, far-infrared, and white light channels to simulate sunrise and sunset, cloudy days with low light, or strong radiation scenarios. The CO2 concentration control module may include a CO2 gas source, a pressure reducing valve, a mass flow controller, and a porous gas distribution pipe.

[0029] like Figure 3 As shown, the multi-parameter three-dimensional monitoring unit deploys soil moisture-temperature-conductivity sensors, soil water potential sensors, and soil respiration acquisition interfaces at depths of 5cm, 20cm, and 50cm within the undisturbed soil column; an infrared canopy temperature sensor, a photosynthetically active radiation sensor, a chlorophyll fluorescence sensor, and an image acquisition device are deployed above the canopy; and an air temperature and humidity sensor, a CO2 concentration sensor, a wind speed sensor, and an atmospheric pressure sensor are deployed in the middle of the experimental chamber. Each sensor can be connected to the edge intelligent control unit 5 via RS485, CAN, Ethernet, analog input, or a wireless low-power network.

[0030] like Figure 2 As shown, the edge intelligent control unit 5 can be composed of a PLC, an industrial control computer, an embedded GPU module, or a single-board computer. The edge intelligent control unit 5 receives data from the multi-parameter three-dimensional monitoring unit, outputs control values ​​to the climate execution unit 3, and synchronizes model versions, experimental scripts, and report data with the cloud and mobile terminal 6. The data synchronization module in the edge intelligent control unit 5 adds a unified timestamp to each data channel, performs noise reduction, unit unification, outlier marking, and missing value marking, and generates a quality mask Qt. The quality mask Qt is used to indicate whether each data channel is available, drifting, delayed, or outside the physically reasonable range at a given time.

[0031] In one embodiment, the grassland state vector Xt formed by the data synchronization module includes an environmental sub-vector Et, a soil sub-vector St, a canopy sub-vector Ct, an image embedding sub-vector It, and a quality mask Qt. The environmental sub-vector Et includes temperature, relative humidity, CO2 concentration, wind speed, photosynthetically active radiation, rainfall intensity, and atmospheric pressure; the soil sub-vector St includes soil moisture content at different depths, soil temperature, soil electrical conductivity, soil water potential, and leachate volume; the canopy sub-vector Ct includes canopy temperature, leaf area index, photosynthetic efficiency, chlorophyll fluorescence parameters, and vegetation cover; and the image embedding sub-vector It is obtained from canopy images, root images, or near-field multispectral images by a lightweight convolutional network or image feature extractor.

[0032] like Figure 4 As shown, the physical-data dual-driven ecological response model includes an explicit ecological variable decoding layer, a physical residual layer, a data imputation branch, and a gated fusion layer. The explicit ecological variable decoding layer maps the grassland state vector Xt to interpretable variables such as leaf area index (LAI), soil moisture content (SMC), biomass, canopy temperature, photosynthetic efficiency, soil respiration flux, net ecosystem carbon exchange, and ecological stress index. The physical residual layer calculates residuals based on water balance, energy balance, and carbon flux balance. For example, the water balance residual can be composed of rainfall, drainage, evapotranspiration estimates, and soil moisture changes; the energy balance residual can be composed of net radiation, sensible heat flux, latent heat flux, and soil heat flux; and the carbon flux residual can be composed of total primary productivity, ecosystem respiration, and net carbon exchange.

[0033] The data imputation branch is used to learn nonlinear local responses that are difficult to describe by the physical residual layer, such as lodging caused by strong winds, canopy damage caused by rain splash erosion, short-term photosynthetic decline due to stomatal closure under heat waves, and structural differences between different grassland types. The data imputation branch can employ lightweight Transformer encoders, temporal convolutional networks, multilayer perceptrons, gated recurrent units, or combinations thereof. The gated fusion layer generates gating coefficients αt based on Xt and Qt, and fuses the physical branch output Yp and the data branch output Yd in the manner Yt = αtYp + (1-αt)Yd. When the quality mask indicates missing key sensors or high data noise, the gated fusion layer can increase the weight of the physical branch; when the data quality is high and local nonlinearities are significant, the gated fusion layer can increase the weight of the data imputation branch.

[0034] During model training, the loss function can include a prediction error term, a physical residual term, a gating smoothing term, a quality mask weighting term, and a control-related penalty term. The prediction error term is used to constrain the model output to be close to the observed values; the physical residual term is used to constrain the model to satisfy the conservation of water, energy, and carbon flux; the gating smoothing term is used to avoid drastic jumps in the gating coefficients between adjacent time steps without physical cause; the quality mask weighting term is used to reduce the influence of low-quality channels on the model parameters; and the control-related penalty term is used to reduce the prediction instability of the model near the actuator boundary.

[0035] To address the limited sample size in extreme climate events, the small-sample augmentation training module filters challenging examples from baseline, extreme event, and recovery period data. Challenging examples can be those with prediction errors exceeding dynamic thresholds, physical residuals higher than thresholds, rapidly changing ecological stress indices, or locally missing quality masks. The module first performs physical consistency screening on these challenging examples, then uses restricted Gaussian perturbation, time-series fragment reconstruction, scenario script perturbation, or conditional generation networks to generate augmented samples. The perturbation amplitude of the restricted Gaussian perturbation is limited by both the physical residual variance and sensor accuracy to avoid generating samples that do not conform to ecological principles.

[0036] The constrained predictive control module employs rolling time-domain control. Starting from the current time t, the controller, within a prediction window of 6 to 48 hours, combines the climate script objective, model-predicted ecological response, and actuator constraints to solve for the control variables for the next control cycle. The objective function may include environmental setpoint tracking error, grassland ecological stress index bias, model uncertainty penalty, actuator energy consumption, temperature slope constraint, rainfall pulse constraint, and wind speed safety constraint. The controller outputs control variables including heating power, cooling power, humidification capacity, dehumidification capacity, pump frequency, sprinkler head on / off combination, fan frequency, LED spectral channel intensity, and CO2 mass flow rate.

[0037] For example, in a heat wave superimposed with mild drought scenario, the climate script can be set to run for a 7-day baseline period, then on the 8th day at noon, raise the temperature from 25°C to 42°C at a predetermined slope and maintain it for 48 hours, while simultaneously lowering the relative humidity target to 25%RH, suspending rainfall, and keeping the light intensity constant, followed by a recovery period. The constraint prediction and control module continuously adjusts the warming slope during the warming phase based on canopy temperature, soil water potential, and model-predicted stress index. If the predicted ecological stress index exceeds the safety threshold, the warming time can be extended or the recovery period can be initiated earlier. If the measured temperature lags behind the target curve, the heating plate output can be increased and the ventilation rate reduced within power and safety constraints.

[0038] In the superimposed heavy rain and strong wind script, the climate script can be set to synchronize short-duration high-intensity rainfall with horizontal wind fields. The rainfall execution module controls the spatial distribution of rainfall according to the nozzle array, while the wind speed execution module forms the near-surface wind field according to the fan frequency and guide vane angle. The multi-parameter three-dimensional monitoring unit simultaneously records changes in soil moisture content, leachate volume, canopy lodging images, and soil respiration. The physics-data dual-driven model uses water balance residuals to constrain soil moisture changes and uses image embedding sub-vectors to supplement mechanical damage information.

[0039] The anomaly-triggered reconstruction module handles sensor anomalies, actuator mismatches, or disaster conditions. When any of the ecological stress index, actuator bias, sensor quality mask, or disaster probability exceeds a preset threshold, the anomaly-triggered reconstruction module freezes the current model and generates a snapshot model, recording the timestamp, model version, script number, and hash value. Within a preset time window after the anomaly is triggered, newly acquired data is appended as the anomaly-time feature vector. The model can then be fine-tuned only in the gating fusion layer or incomplete data branches, thus achieving rapid adaptation without disrupting the original model's physical structure. If the reconstructed model fails the physical consistency verification, the system rolls back to the snapshot model.

[0040] The model compression and version latch module is used to deploy trained models on edge devices. Model compression can employ INT8 quantization, channel pruning, or knowledge distillation. For sensitive parameters in physical residual layers and gated fusion layers, higher precision can be preserved or mixed precision can be used. The version latch module generates a version number, hash value, training sample time range, sensor configuration, script identifier, and validation results for each deployment, enabling experimental reports to trace the model's origin.

[0041] like Figure 5 As shown, the method for conducting experiments using the device of the present invention may include the following steps: S1, Select a target grassland plot, excavate an undisturbed soil column containing native vegetation and roots, place it in a planting trough, and install sensors. S2, Push the planting trough into the experimental chamber, run the adaptation period program to allow the vegetation to recover to a stable state, and collect baseline period data. S3, Perform time synchronization, denoising, missing value labeling, and state vector encapsulation on the baseline period data. S4, Train the physical-data dual-driven ecological response model and perform physical consistency verification. S5, Generate an extreme climate script according to the experimental objective. S6, Run the script and have the climate execution unit controlled in a closed loop by the constraint prediction control module. S7, Collect data during the extreme event period and the recovery period, and perform online calibration of the model. S8, Output an experimental report, key driving factors, response curves, and model version records.

[0042] like Figure 6As shown, the composite extreme climate script can include a baseline period, a rapid change period, an extreme duration period, and a recovery period. For different grassland types, the event intensity, duration, and recovery period length can be adjusted. For example, desert steppe can use a lower soil moisture threshold and a longer recovery period; meadow steppe can use a higher humidity target and a lower wind speed safety threshold. For alpine grasslands, diurnal freeze-thaw cycles and snow cover simulation modules can also be added. These parameter changes do not alter the core technical concept of embedding the ecological response model into climate closed-loop control.

[0043] This invention can also access historical data from satellites, drones, or field stations as prior data. Prior data is used to determine the target intensity and slope of change in climate scripts, and can also be used for model pre-training. Compared to multi-parameter monitoring data within the experimental chamber, remote sensing and drone data have a larger spatial scale. The edge intelligent control unit can encode them into image embedding sub-vectors or regional background vectors, and use quality masks to label their spatial resolution, timeliness, and registration error.

[0044] The key driving factors output by this invention can be obtained through counterfactual perturbation. Specifically, while keeping other inputs constant, the system performs small perturbations on variables such as temperature, humidity, soil water potential, light intensity, wind speed, or CO2 concentration, and compares the changes in the ecological stress index, photosynthetic efficiency, or soil respiration flux output by the model to obtain the Top-N key driving factors, threshold intervals, and confidence intervals. These results can be presented together with the control curves, response curves, and quality mask summaries in the experimental report.

[0045] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0046] The embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0047] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0048] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0049] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. An artificial intelligence-driven grassland extreme climate simulation experimental device, characterized in that, include: An experimental chamber, the interior of which forms a sealed simulated space for accommodating grassland samples; The undisturbed soil column planting unit is set up in the closed simulation space to support the undisturbed soil column containing grassland vegetation, root system and soil profile. The climate execution unit includes temperature execution module, humidity execution module, rainfall execution module, wind speed execution module, light intensity execution module, and CO2 concentration execution module; A multi-parameter three-dimensional monitoring unit is used to simultaneously collect cabin environmental data, soil profile data, canopy physiological data, and image observation data; The edge intelligent control unit is communicatively connected to both the climate execution unit and the multi-parameter three-dimensional monitoring unit.

2. The artificial intelligence-driven grassland extreme climate simulation experimental device as described in claim 1, characterized in that, in: The experimental chamber includes a transparent insulated enclosure structure, a sealed door, an air vent, and a condensate drainage structure; the original soil column planting unit includes a pull-out planting trough, a gravel filter layer at the bottom of the planting trough, a drainage pipe, a leachate collection bottle, sensor insertion holes at different depths on the side wall of the planting trough, and a root observation window.

3. The artificial intelligence-driven grassland extreme climate simulation experimental device as described in claim 1, characterized in that, in: The temperature control module includes a far-infrared radiation heating plate and a refrigeration evaporator; the humidity control module includes an ultrasonic humidifier and a condenser dehumidifier; the rainfall control module includes a water tank, a variable frequency water pump, a flow meter, a pressure sensor, and an array of rainfall nozzles; the wind speed control module includes a variable speed axial flow fan, a honeycomb rectifier, and an airflow guide plate; the illumination control module includes an adjustable spectrum LED plant light; and the CO2 concentration control module includes a CO2 gas source, a mass flow controller, and a porous gas distribution pipe.

4. The artificial intelligence-driven grassland extreme climate simulation experimental device as described in claim 1, characterized in that, in: The multi-parameter three-dimensional monitoring unit includes soil moisture-temperature-conductivity sensors, soil water potential sensors, and soil respiration acquisition interfaces buried at different depths in the undisturbed soil column; infrared canopy temperature sensors, photosynthetically active radiation sensors, chlorophyll fluorescence sensors, and image acquisition devices installed above the canopy; and air temperature and humidity sensors, CO2 concentration sensors, wind speed sensors, and atmospheric pressure sensors installed inside the chamber.

5. The artificial intelligence-driven grassland extreme climate simulation experimental device as described in claim 1, characterized in that, in: The edge intelligent control unit includes a data synchronization module, a climate script generation module, a physical-data dual-driven ecological response model, and a constraint prediction control module. The data synchronization module encapsulates the data collected by the multi-parameter three-dimensional monitoring unit into a grassland state vector with a quality mask. The physical-data dual-driven ecological response model predicts the grassland ecological response and uncertainty interval based on the grassland state vector. The constraint prediction control module generates the actuator control quantity for the next control cycle based on the grassland ecological response, uncertainty interval, climate script, and the operating constraints of each execution module, so as to perform closed-loop regulation of the temperature, humidity, rainfall, wind speed, light intensity, and CO2 concentration. The data synchronization module registers each sensor channel according to a unified timestamp and generates the quality mask based on sensor self-test results, channel drift, missing values, physical threshold violations, and sampling delay. The quality mask, together with the cabin environment vector, soil profile vector, canopy physiological vector, and image embedding vector, constitutes the grassland state vector. The physical-data dual-driven ecological response model includes an explicit ecological variable decoding layer, a physical residual layer, a data imputation branch, and a gated fusion layer. The explicit ecological variable decoding layer maps the grassland state vector to at least one of leaf area index, soil moisture content, biomass, canopy temperature, photosynthetic efficiency, soil respiration flux, and ecological stress index. The physical residual layer calculates physical residuals based on water balance, energy balance, and carbon flux balance. The data imputation branch fits local responses not explained by the physical residual layer. The gated fusion layer generates gating coefficients based on the grassland state vector and a quality mask, and fuses the outputs of the physical residual layer and the data imputation branch with these gating coefficients. The data imputation branch employs one or more of the following: a lightweight Transformer encoder, a temporal convolutional network, or a multilayer perceptron. During training, the prediction error, physical residual, gating smoothing term, and quality mask weighting term are used together as the loss function to reduce model extrapolation bias caused by sensor absence or insufficient extreme event samples.

6. The artificial intelligence-driven grassland extreme climate simulation experimental device as described in claim 5, characterized in that, in: The edge intelligent control unit also includes a small sample augmentation training module, which is used to extract difficult sample data from baseline and extreme climate period data, perform physical consistency screening on difficult sample data according to water, energy and carbon flux conservation constraints, and generate augmented sample data through restricted Gaussian perturbation, time-series fragment recombination or conditional generation network to update the physical-data dual-driven ecological response model. The constraint prediction control module adopts a rolling time-domain control method. In each control cycle, the prediction window is from 6 to 48 hours in the future. Based on the climate script tracking error, grassland ecological stress index, actuator power constraint, temperature slope constraint, rainfall pulse constraint and wind speed safety constraint, the objective function is constructed and the next control variables of temperature, humidity, rainfall, wind speed, light and CO2 concentration are obtained by solving the objective function. The climate script generation module is configured to generate composite extreme climate scripts for heat waves, extreme droughts, rainstorms, strong winds, overcast and cold weather, freeze-thaw cycles, CO2 increases, and combinations of at least two of the above events; each climate script includes at least the event start time, duration, target intensity, slope of change, recovery period length, and ecological response monitoring frequency.

7. The artificial intelligence-driven grassland extreme climate simulation experimental device as described in claim 5, characterized in that, The edge intelligent control unit also includes a model compression and version latching module, which is used to perform INT8 quantization, channel pruning or knowledge distillation on the trained physical-data dual-driven ecological response model, and to write the version number, hash value, training data range and control script identifier to the compressed model. The edge intelligent control unit also includes an anomaly triggering reconstruction module. When any one of the ecological stress index, actuator deviation, sensor quality mask, or disaster probability exceeds a preset threshold, the anomaly triggering reconstruction module freezes the current model and generates a snapshot model. It appends the newly collected data within a preset time window after the anomaly trigger as a disaster-time or anomaly-time feature vector and only fine-tunes the gating fusion layer or data completion branch to generate a reconstruction model.

8. A method for simulating grassland extreme climate and predicting ecological response using the simulation experimental apparatus described in any one of claims 1-7, characterized in that, Includes the following steps: S1, insert the original soil column of the grassland into the original soil column planting unit, and install a multi-parameter three-dimensional monitoring unit; S2, run the adaptation period program to collect baseline data on the cabin environment, soil profile, canopy physiology and image observations; S3 performs time synchronization, quality mask generation, and feature extraction on the collected data to form a grassland state vector; S4, a physical-data dual-driven ecological response model is trained based on grassland state vectors; S5 generates an extreme climate script based on the target experimental objective; S6. During the extreme climate script operation, the grassland ecological response is predicted using the physical-data dual-drive ecological response model, and the climate execution unit is adjusted in a closed loop through the constraint prediction control module. S7 writes back the data collected during extreme weather events and the recovery period to the sample library and outputs ecological response prediction results, key driving factors, and experimental reports.

9. The method for simulating grassland extreme climate and predicting ecological response using the simulation experimental apparatus as described in claim 8, characterized in that, in: In step S3, the grassland state vector includes an environment sub-vector, a soil sub-vector, a canopy sub-vector, an image embedding sub-vector, and a quality mask sub-vector; the environment sub-vector includes at least temperature, relative humidity, CO2 concentration, wind speed, photosynthetically active radiation, and rainfall intensity; the soil sub-vector includes at least soil moisture content at different depths, soil temperature, soil electrical conductivity, soil water potential, and leachate volume; the canopy sub-vector includes at least canopy temperature, leaf area index, photosynthetic efficiency, or chlorophyll fluorescence parameters. In step S4, training the physical-data dual-driven ecological response model includes: using baseline period samples as the support set; using samples whose prediction errors exceed the dynamic threshold during extreme climate periods as hard sample samples; performing physical consistency screening on the support set and hard sample samples; training the physical residual layer, data imputation branch, and gated fusion layer based on the screened samples; and verifying whether the model output meets the ecological specific index threshold through independent recovery period samples. In step S5, generating an extreme climate script includes: determining the target intensity from historical meteorological records, target grassland ecological thresholds, and user-defined event levels; determining the change slope based on actuator performance; and pre-simulating the grassland stress trajectory based on the physical-data dual-driven ecological response model, automatically adjusting the duration, recovery period, or event combination sequence when the pre-simulation result exceeds the safety threshold. In step S6, if the deviation between the measured in-cabin climate factor and the target climate script exceeds the first threshold, the control quantity of the corresponding execution module is corrected first; if the predicted ecological stress index exceeds the second threshold, the event change slope is reduced or the recovery period is extended; if the mass mask shows that the key sensor has failed, the backup sensor is called or the model estimate is used for temporary closed-loop control. In step S7, the key driving factors are obtained by counterfactual perturbation of the grassland state vector and output in the form of variable name, threshold interval, response direction and confidence interval; the experimental report includes extreme climate script, actuator control curve, environmental measured curve, ecological response curve, model version number and quality mask summary.

10. An electronic device, characterized in that, It includes a processor, a memory, a data acquisition interface, and an actuator control interface. The memory stores a computer program. When the processor executes the computer program, it implements the method described in any one of claims 1 to 9. It receives data from the multi-parameter three-dimensional monitoring unit through the data acquisition interface and outputs control quantities to the climate execution unit through the actuator control interface.