A method and system for adaptive regulation of mulberry growth under water level fluctuations in the drawdown zone
By collecting and simulating long-term environmental data of the drawdown zone, risk prediction and growth simulation models were established, and the optimal mulberry planting scheme was determined. This solved the problem of low survival rate and growth stability of mulberry trees in the drawdown zone due to the dual stresses of flooding and drought, and achieved efficient growth adaptation regulation of mulberry trees.
Patent Information
- Application Number
- CN202610274455.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-08
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-03-08
AI Technical Summary
When planting mulberry trees in the drawdown zone, existing technologies cannot effectively cope with the dual pressures of flooding and oxygen deficiency caused by rapid water level rise and soil drought caused by rapid water level drop, resulting in low survival rate and growth stability of mulberry trees.
By collecting short-term environmental data based on sensors, simulating long-term environmental data, establishing a risk prediction model, extracting environmental feature vectors, selecting experimental planting sites to plant different varieties of mulberry trees, collecting plant physiological data, establishing a growth simulator, outputting expected physiological indices, and determining the optimal planting plan in combination with preset planting requirements.
This study enabled a quantitative assessment of the future environmental stress risk at various geographical coordinates in the drawdown zone, extracted the tolerance of various mulberry varieties in real-world environments, determined the optimal planting site and planting variety, and solved the problem of low mulberry survival rate and growth stability caused by blind site selection and seed selection in existing technologies.
Smart Images

Figure CN122030169B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of ecological environment technology, specifically relating to a method and system for regulating the adaptive growth of mulberry trees under the fluctuation of water level in the drawdown zone. Background Technology
[0002] The drawdown zone refers to a special shoreline area formed by the periodic rise and fall of water bodies such as reservoirs and rivers. Vegetation restoration and ecological protection in this area play a crucial role in maintaining soil and water conservation, biodiversity, and shoreline stability. Due to the influence of multiple factors such as seasons, rainfall, and water management, the water level fluctuates frequently and significantly, making the plant growth environment extremely unstable and one of the most challenging areas in ecological restoration projects.
[0003] Currently, when planting mulberry trees in drawdown zones, the common practice is to select varieties in fixed locations or to use only water-tolerant or drought-tolerant varieties. These methods often focus only on the main stressors at a particular time, such as emphasizing flood tolerance while neglecting the root system's water requirements during the drying period, or conversely, prioritizing drought tolerance, making the plants highly susceptible to oxygen deficiency and death during flooding. As a result, regardless of the strategy chosen, the plants struggle to cope simultaneously with the dual pressures of flooding and oxygen deficiency caused by rapidly rising water levels and soil drought caused by rapidly falling water levels, resulting in consistently low survival rates and growth stability. Therefore, how to rationally arrange planting locations and variety combinations has become a key issue for the success of mulberry tree planting in drawdown zones. Summary of the Invention
[0004] To address the aforementioned problems, this invention provides a method and system for adaptive regulation of mulberry tree growth under fluctuating water levels in the drawdown zone, thereby resolving the issues described in the background section.
[0005] To achieve the aforementioned objectives, this invention proposes a method for adaptive regulation of mulberry tree growth under fluctuating water levels in the drawdown zone, comprising: Short-term environmental data is collected from preset geographic coordinates using sensors, and long-term environmental data is generated by simulating the short-term environmental data. Environmental feature vectors are extracted from long-term environmental data, a risk prediction model is established, and the environmental feature vectors are input into the risk prediction model to obtain the stress risk value of each geographic coordinate point. Select experimental planting sites at geographical coordinates, plant different varieties of mulberry trees at the experimental planting sites, and collect plant physiological data concurrent with short-term environmental data. Extract tolerance indicators for various mulberry varieties based on the plant physiological data. A growth simulator was established, and tolerance indicators were input into the growth simulator. The growth simulator performed differential simulation based on stress risk values and output the corresponding expected physiological indices. Based on the expected physiological indices and the preset planting requirements, the optimal planting plan is determined, including the best planting location and the best planting variety.
[0006] Furthermore, long-term environmental data is generated by simulating short-term environmental data, including: Historical meteorological data with a time span longer than short-term environmental data is acquired, and a gradient boosting model and a generative adversarial network are constructed. In the gradient boosting model, a loss function containing a prediction error term and a dynamic baseline regularization term is defined based on short-term environmental data and corresponding historical meteorological data. The dynamic baseline regularization term is used to penalize the high-frequency components of the predicted values of the gradient boosting model. The trained model finally outputs a natural fluctuation sequence, and a scheduling fluctuation sequence is generated based on short-term environmental data and the natural fluctuation sequence. The scheduling fluctuation sequence is input into the generative adversarial network for training. After training, the complete historical meteorological data is input into the gradient boosting model to generate a long-term natural fluctuation sequence. The generative adversarial network is then used to generate a long-term scheduling fluctuation sequence. The long-term natural fluctuation sequence and the long-term scheduling fluctuation sequence are used as long-term environmental data.
[0007] Furthermore, environmental feature vectors are extracted from long-term environmental data, including: Baseline rhythm features are extracted based on long-term natural fluctuation sequences. These features include dominant cycle features, trend features, and fluctuation amplitude features. Disturbance and impact features are extracted based on long-term scheduling fluctuation sequences. These features include impact frequency characteristics, average impact intensity, and impact variation characteristics. The natural abundant water sequence, the scheduled water increase sequence, the natural dry water sequence, and the scheduled water decrease sequence are extracted from the long-term natural fluctuation sequence and the long-term scheduled fluctuation sequence. The interaction coupling features include the mean of the product of the natural abundant water sequence and the scheduled water increase sequence, and the mean of the product of the natural dry water sequence and the scheduled water decrease sequence. The baseline rhythm features, disturbance impact features, and interactive coupling features are concatenated to obtain the environmental feature vector.
[0008] Furthermore, by retaining all values greater than zero in the long-term natural fluctuation sequence and setting the rest to zero, a natural abundant water sequence is obtained. By retaining all values less than zero in the long-term natural fluctuation sequence and setting the rest to zero, a natural low water sequence is obtained. By retaining all values greater than zero in the long-term scheduled fluctuation sequence and setting the rest to zero, a scheduled water increase sequence is obtained. By retaining all values less than zero in the long-term scheduled fluctuation sequence and setting the rest to zero, a scheduled water decrease sequence is obtained.
[0009] Furthermore, a risk prediction model is established, including: A reference plant model representing the average physiological properties of mulberry trees was established. Virtual stress data of the reference plant model was generated by simulating short-term environmental data. The environmental stress index was calculated based on the virtual stress data. A training dataset was constructed based on the environmental feature vector and environmental stress index in the short-term environmental data. A dual-stream feature embedding layer is constructed, in which a rhythm embedding network and an impact embedding network are set. The baseline rhythm features in the training set are input into the rhythm embedding network, and the rhythm embedding network outputs the rhythm context vector. The perturbation impact features are input into the impact embedding network, and the impact embedding network outputs the impact basis vector. Construct a gating network, input the rhythm context vector into the gating network, and the gating network outputs a modulation gating vector. Perform element-wise multiplication between the modulation gating vector and the impulse basis vector to generate the modulated impulse vector. The rhythm context vector, the modulated impact vector, and the interactive coupling features are concatenated to form a fused feature vector. The fused feature vector is then input into the prediction network, which outputs a risk prediction value. The environmental feature vectors of long-term environmental data for all geographic coordinate points within the drawdown zone are input into the trained risk prediction model to obtain the stress risk value for each geographic coordinate point.
[0010] Furthermore, virtual stress data for a reference plant model is generated based on short-term environmental data simulations, including: Based on the water level, soil temperature and preset root depth in the short-term environmental data, the oxygen concentration in the root zone is calculated, and the effective water volume sequence in the root zone is calculated based on the water level, soil moisture, rainfall and evaporation. The oxygen concentration in the root zone is compared with the hypoxia stress threshold to generate the hypoxia stress index, and the effective water volume in the root zone is compared with the drought stress threshold to generate the drought stress index sequence. Traverse the effective water volume sequence and drought stress index sequence in the root region, identify all stress periods that exceed the duration threshold, and use the duration and cumulative stress value of all stress periods as virtual stress data.
[0011] Furthermore, tolerance indicators for various mulberry varieties were extracted based on plant physiological data, including: Stress analysis was conducted on various mulberry varieties to obtain corresponding actual stress data. Based on the actual stress data, the stress period of mulberry trees at the experimental planting site was determined. Physiological fluctuation data during the stress period was extracted from plant physiological data. The actual stress data and physiological fluctuation data were clustered, and the data belonging to the same cluster in the clustering results were analyzed to determine the corresponding tolerance index.
[0012] Furthermore, the growth simulator performs differentiated simulations based on stress risk values and outputs corresponding expected physiological indices, including: The stress risk value and tolerance index of mulberry trees at the experimental planting site were concatenated into a joint input feature. The expected physiological index was calculated based on the measured physiological data of mulberry trees at the experimental planting site. A training dataset was constructed based on the joint input feature and the expected physiological index. Construct a fully connected network, train the fully connected network using the training dataset, and obtain a growth simulator; The stress risk value of the geographical coordinate point is combined with the tolerance index of the mulberry variety to form a joint input feature. The joint input feature is then input into the growth simulator to output the corresponding expected physiological index.
[0013] Furthermore, based on the preset planting requirements, an optimal planting plan is determined, including the best planting location and the best planting variety, including: An optimization model is constructed and solved with the objective function of maximizing the sum of expected physiological indices of the entire planting area. The constraints of the optimization model include planting uniqueness constraint, regional coverage constraint, variety diversity constraint, minimum survival threshold constraint, and cost budget constraint.
[0014] This invention also provides a system for regulating the adaptive growth of mulberry trees under water level fluctuations in the drawdown zone. This system is used to implement the methods described above, and includes: The first acquisition module collects short-term environmental data from preset geographic coordinates using sensors, and then simulates and generates long-term environmental data based on the short-term environmental data. The first prediction module extracts environmental feature vectors from long-term environmental data, establishes a risk prediction model, and inputs the environmental feature vectors into the risk prediction model to obtain the stress risk value of each geographical coordinate point. The second data acquisition module selects experimental planting sites at geographical coordinates, plants different varieties of mulberry trees at the experimental planting sites, and collects plant physiological data concurrent with short-term environmental data. Based on the plant physiological data, it extracts tolerance indicators for various mulberry varieties. The second prediction module establishes a growth simulator, inputs tolerance indicators into the growth simulator, performs differentiated simulation based on stress risk values, and outputs the corresponding expected physiological index. The plan generation module determines the optimal planting plan, including the best planting location and the best planting variety, based on the expected physiological index and preset planting requirements.
[0015] The beneficial effects of this invention are as follows: This invention first collects short-term environmental data using sensors and simulates long-term environmental data. It then extracts environmental feature vectors and constructs and applies a risk prediction model, thereby quantitatively assessing the future environmental stress risk at various geographical coordinates in the drawdown zone. Next, by planting different varieties of mulberry trees at experimental planting sites and collecting their physiological data, it extracts indices of the tolerance of various mulberry varieties under real environmental stress. Subsequently, by establishing a growth simulator, it combines the tolerance indices of each variety with the stress risk values at each coordinate point for simulation, and optimizes the decision based on preset planting requirements. This determines the optimal planting plan, including the best planting location and the best planting variety. Notably, this invention solves the problem in existing technologies where the inability to accurately predict and quantify the dual stresses of flooding and oxygen deficiency and drought in the drawdown zone leads to blind site selection and planting, resulting in low mulberry survival rates and growth stability. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating the steps of a method for adaptively regulating mulberry tree growth under fluctuating water levels in a drawdown zone, as described in this invention. Figure 2 This is a schematic diagram of the risk prediction model of the present invention; Figure 3 This is a schematic diagram of the planting scheme generated by the present invention; Figure 4 This is a schematic diagram of the structure of a mulberry tree growth adaptation regulation system under water level fluctuations in the drawdown zone according to the present invention. Detailed Implementation
[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0018] It is understood that the terms "first," "second," etc., used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are used only to distinguish one element from another. For example, without departing from the scope of this application, a first script may be referred to as a second script, and similarly, a second script may be referred to as a first script.
[0019] like Figure 1 As shown, a method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone includes: S1: Collect short-term environmental data from preset geographic coordinates using sensors, and simulate and generate long-term environmental data based on the short-term environmental data.
[0020] Within the drawdown zone, representative geographical coordinate points with varying elevations and soil conditions are selected. At these points, multiple sensors capable of monitoring parameters such as water level, soil moisture, and temperature are deployed. Based on this, short-term environmental data is collected over a period of one or two years. Subsequently, long-term environmental data is generated through simulation based on the short-term data. This long-term environmental data is used to predict the future stress risks to the drawdown zone. The method for generating the long-term environmental data will be described in detail later.
[0021] S2: Extract environmental feature vectors from long-term environmental data, establish a risk prediction model, input the environmental feature vectors into the risk prediction model, and obtain the stress risk value of each geographical coordinate point.
[0022] After generating long-term environmental data, the next step is to assess the potential risks of these future environmental scenarios to mulberry tree growth. Stress risk is not a simple linear function of environmental indicators, but a complex result determined by the duration, frequency, and coupling effects of multiple environmental factors. Furthermore, long-term data contains numerous complex extreme event patterns not observed in short-term observations. Therefore, simply using formulas cannot accurately assess the risks of these new patterns. This invention aims to establish a risk prediction model to learn the mapping relationship between environmental fluctuation patterns and actual physiological stresses from real short-term observations, and then use this pattern to assess the risk values brought about by various complex future environmental scenarios.
[0023] Environmental feature vectors are representative features in long-term environmental data that influence mulberry tree growth. Risk prediction models are machine learning models used to predict the stress risk value of each geographic coordinate point based on the environmental feature vectors in long-term environmental data. A higher stress risk value indicates a greater impact of that geographic coordinate point on mulberry tree growth.
[0024] S3: Select experimental planting sites at geographical coordinates, plant different varieties of mulberry trees at the experimental planting sites, and collect plant physiological data concurrent with short-term environmental data. Extract tolerance indicators for various mulberry varieties based on the plant physiological data.
[0025] Selected locations from geographical coordinates were chosen as experimental planting sites. Different varieties of mulberry trees were planted at these sites, and plant physiological data, including photosynthetic rate, chlorophyll content, and root activity, were continuously collected. Tolerance indices were quantitative values derived from the plant physiological data by analyzing its fluctuations under different stress events, representing the mulberry tree's resistance or resilience to specific stresses.
[0026] S4: Establish a growth simulator, input tolerance indicators into the growth simulator, the growth simulator performs differentiated simulation based on stress risk values, and outputs the corresponding expected physiological index.
[0027] The growth simulator is a predictor used to simulate the growth status of plants. Based on the stress risk value of geographical coordinates, it simulates the future growth of different varieties of mulberry trees and outputs the expected physiological index of the mulberry tree. The higher the expected physiological index, the better the mulberry tree can grow.
[0028] S5: Based on the expected physiological index and combined with the preset planting requirements, determine the optimal planting plan, including the best planting location and the best planting variety.
[0029] Traditional planting decisions often tend to concentrate the best varieties in areas with the lowest risk and best expected growth. However, this can lead to vegetation loss in higher-risk areas, failing to meet the requirements for overall coverage and ecological stability in drawdown zone ecological restoration. To address this issue, this embodiment constructs a mathematical optimization model for planting decisions. The core optimization objective of this model is to maximize the sum of the expected physiological indices of all mulberry trees in the entire drawdown zone. Constraints include ensuring coverage in certain risk areas to achieve soil and water conservation and avoiding large-scale planting of a single variety to improve ecosystem stability. By solving the optimization model, the optimal planting location and optimal variety are derived, thus generating the optimal planting plan.
[0030] In this embodiment, long-term environmental data is generated by simulating short-term environmental data, including: Historical meteorological data with a time span longer than short-term environmental data is acquired, and a gradient boosting model and a generative adversarial network are constructed. In the gradient boosting model, a loss function containing a prediction error term and a dynamic baseline regularization term is defined based on short-term environmental data and corresponding historical meteorological data. The dynamic baseline regularization term is used to penalize the high-frequency components of the predicted values of the gradient boosting model. The trained model finally outputs a natural fluctuation sequence, and a scheduling fluctuation sequence is generated based on short-term environmental data and the natural fluctuation sequence.
[0031] If data is collected directly from sensors over many years, the time cost is too high, making it difficult to meet the needs of short-term planning. While traditional interpolation or simple repetition methods can extend the data length, they cannot generate combinations of extreme events that have never occurred before, and they easily overlook the complex characteristics of the drawdown zone water level being driven by both natural and artificial regulation, resulting in insufficient authenticity of the extended data.
[0032] To address this issue, this embodiment decomposes environmental data into two parts: relatively smooth natural fluctuations driven by meteorological factors, and drastic fluctuations caused by reservoir scheduling. Subsequently, by learning and simulating the patterns of these two fluctuations separately, highly realistic long-term environmental data is generated. The specific steps are as follows.
[0033] The sensor collected short-term environmental data daily for two consecutive years. Simultaneously, it obtained daily rainfall, sunshine duration, and evaporation data for the past two years from the hydrological and meteorological department, serving as historical meteorological data. Based on this data, this embodiment constructed two collaborative deep learning models: a gradient boosting model and a generative adversarial network.
[0034] When training the gradient boosting model, historical meteorological data that corresponds perfectly with two years of short-term environmental data is used as input, with short-term environmental data serving as the training target. The gradient boosting model employs a special loss function, consisting of two weighted components: a prediction error term (e.g., mean squared error) and a dynamic baseline regularization term. The prediction error term encourages the model's predictions to approximate the actual short-term environmental data. The dynamic baseline regularization term penalizes the high-frequency components of the model's predictions, forcing the model to focus on learning relatively smooth low-frequency trends driven by meteorological factors in the data. In this way, the trained gradient boosting model essentially becomes a low-frequency trend extractor. Therefore, the model's predictions of short-term environmental data are defined as a natural fluctuation sequence. Correspondingly, the difference between the original short-term environmental data and the model's predicted natural fluctuation sequence is considered to be a high-frequency scheduling fluctuation sequence mainly caused by factors such as artificial reservoir scheduling.
[0035] The scheduling fluctuation sequence is input into the generative adversarial network for training. After training, the complete historical meteorological data is input into the gradient boosting model to generate a long-term natural fluctuation sequence. The generative adversarial network is then used to generate a long-term scheduling fluctuation sequence. The long-term natural fluctuation sequence and the long-term scheduling fluctuation sequence are used as long-term environmental data.
[0036] The scheduling fluctuation sequence is used as a real data sample and input into the generative adversarial network (GAN) for training. The GAN consists of a generator and a discriminator. The generator's task is to learn the statistical distribution and temporal pattern of the scheduling fluctuation sequence and generate fake scheduling fluctuation sequences, while the discriminator's task is to distinguish between real and fake scheduling fluctuation sequences. Through adversarial training, the GAN ultimately gains the ability to generate highly realistic scheduling fluctuation data.
[0037] After training both models, historical meteorological data is used to predict the historical meteorological data for the next two years. The prediction results are then input into a gradient boosting model, which outputs the corresponding natural fluctuation sequence. Specifically, to address the uncertainty of future climate, methods such as climate model downscaling or random weather generators are employed to generate multiple possible meteorological scenarios for the next two years. Subsequent steps, including long-term environmental data generation and risk assessment, will be performed separately for each scenario, thus obtaining a probability distribution of stress risk values rather than a single, deterministic value. In the final optimization decision, risk avoidance strategies can be adopted, such as using the 90th percentile of the risk value as the optimization input, or aiming to maximize the expected value of the expected physiological index. Additionally, a random noise vector is input into the generator of a generative adversarial network (GAN), which generates fluctuation values over a two-year time span, thereby obtaining a long-term scheduling fluctuation sequence. Specifically, the random noise vector is a numerical vector generated by the computer according to a preset probability distribution, such as a standard normal distribution or a uniform distribution. Finally, both the long-term natural fluctuation sequence and the long-term scheduling fluctuation sequence are used as long-term environmental data.
[0038] In this embodiment, environmental feature vectors are extracted from long-term environmental data, including: Baseline rhythm features are extracted based on long-term natural fluctuation sequences. These features include dominant cycle features, trend features, and fluctuation amplitude features.
[0039] Time series analysis is commonly used to extract parameters such as annual average inundation duration, maximum consecutive drought days, and water level fluctuation rates from environmental data through statistical analysis. As previously discussed, environmental data is composed of a superposition of real natural fluctuations and synthetic scheduling fluctuations. Conventional methods analyze these natural and scheduling fluctuations together, resulting in extracted features that cannot distinguish the root cause of stress. For example, an extremely high water level peak could be caused by continuous heavy rainfall or by emergency discharge from an upstream reservoir. The mechanisms and durations of these two factors' impact on plant physiology may be completely different, but conventional methods may generate the same feature values, obscuring the nature of the risk and reducing the accuracy of subsequent risk prediction models. Therefore, this embodiment proposes the following technical solution.
[0040] For each geographic coordinate point, a Fourier transform is applied to identify the period corresponding to the maximum peak at non-zero frequencies in the power spectral density function; this period is then used as the dominant periodic feature. Next, to assess the trend strength of the sequence, the Hearst exponent of the long-term natural fluctuation sequence is calculated, and its value is used as the trend feature. Finally, the long-term natural fluctuation sequence is divided by year, the standard deviation of the sequence within each year is calculated, and the average of the standard deviations for all years is taken as the fluctuation amplitude feature. The dominant periodic feature, trend feature, and fluctuation amplitude feature are combined to form the baseline rhythm feature.
[0041] Disturbance and impact features are extracted based on long-term scheduling fluctuation sequences. These features include impact frequency characteristics, average impact intensity, and impact variation characteristics.
[0042] For each geographic coordinate point, a long-term scheduling fluctuation sequence is used to extract its disturbance impact characteristics using event detection and statistical methods. Positive and negative impact thresholds are set, and the long-term scheduling fluctuation sequence is traversed to identify all fluctuation events with amplitudes exceeding the thresholds. The total number of events is then counted as the total impact frequency feature.
[0043] The positive and negative impact thresholds can be determined by combining statistical methods with expert experience. For example, a summary statistical analysis can be performed on the long-term scheduling fluctuation sequences of all geographic coordinate points in the entire study area to calculate the overall distribution characteristics of the dataset. For instance, the 90th or 95th percentile of all positive fluctuations can be set as the positive impact threshold, and the 10th or 5th percentile of all negative fluctuations can be set as the negative impact threshold.
[0044] The mean amplitude and standard deviation of the amplitude of all identified fluctuation events are calculated and used as the average impact intensity characteristic and the impact intensity variability characteristic, respectively. The total impact frequency characteristic, the average impact intensity characteristic, and the impact intensity variability characteristic are combined to form the disturbance impact characteristic.
[0045] The natural abundant water sequence, the scheduled water increase sequence, the natural dry water sequence, and the scheduled water decrease sequence are extracted from the long-term natural fluctuation sequence and the long-term scheduled fluctuation sequence. The interactive coupling features include the mean of the product of the natural abundant water sequence and the scheduled water increase sequence as the feature of the abundant water enhancement effect, and the mean of the product of the natural dry water sequence and the scheduled water decrease sequence as the feature of the dry water aggravation effect.
[0046] The baseline rhythm features, disturbance impact features, and interactive coupling features are concatenated to obtain the environmental feature vector.
[0047] In this embodiment, all values greater than zero in the long-term natural fluctuation sequence are retained, and the rest are set to zero to obtain the natural abundant water sequence. All values less than zero in the long-term natural fluctuation sequence are retained, and the rest are set to zero to obtain the natural dry water sequence. All values greater than zero in the long-term scheduled fluctuation sequence are retained, and the rest are set to zero to obtain the scheduled water increase sequence. All values less than zero in the long-term scheduled fluctuation sequence are retained, and the rest are set to zero to obtain the scheduled water decrease sequence.
[0048] In particular, to ensure the consistency of the dimensions of the feature data, the features extracted above need to be normalized.
[0049] In this embodiment, a risk prediction model is established, including: A reference plant model representing the average physiological properties of mulberry trees is established. Virtual stress data of the reference plant model is generated based on short-term environmental data. The environmental stress index is calculated based on the virtual stress data. A training dataset is constructed based on the environmental feature vector and environmental stress index in the short-term environmental data.
[0050] First, a standardized reference plant model is defined. This model does not correspond to any specific mulberry variety but represents the average physiological characteristics of the mulberry species. Specifically, the average root depth can be obtained by statistically analyzing the average soil depth containing the majority of root biomass through literature review and field profile excavation. Through pot flooding control experiments, the dissolved oxygen concentration at which key physiological indicators such as photosynthetic rate significantly decreases is determined to establish the critical oxygen concentration threshold for rhizosphere hypoxia stress. Through pot water-controlled drought experiments, the soil volumetric water content corresponding to the inflection point of a sharp decline in photosynthetic rate is determined to establish the critical available water content threshold for rhizosphere drought stress.
[0051] Subsequently, based on short-term environmental data collected from various geographic coordinate points and combined with the aforementioned reference plant model, the stress events that would occur if the reference plant model were planted at the geographic coordinate points were identified, thereby obtaining virtual stress data. This data includes the type, duration, and cumulative stress intensity of each stress event. The method for obtaining virtual stress data will be introduced later.
[0052] Next, the stress event data are nonlinearly aggregated and comprehensively evaluated to generate a stress risk value that comprehensively reflects the short-term environmental stress level. The stress risk value is not derived through simple linear weighting, but rather aims to simulate the complex cumulative effects of plant physiological damage. For example, the final impact of high-intensity, short-term shock stress differs from that of low-intensity, long-term chronic stress, and there are nonlinear additive or amplification effects. Using this method, a quantitative, highly generalized stress risk value is generated for each experimental plot, serving as the target label for subsequent risk prediction model training.
[0053] Specifically, when calculating the environmental stress index, this invention preferably employs a nonlinear damage accumulation model. For example, the stress risk value can be modeled as the sum of the damage contributions of all stress events, while the damage contribution of a single event... Its strength The product of the power function of the duration T, i.e. .in, This represents the virtual damage contribution value caused by the i-th independent stress event. This is the basic damage coefficient, used for scaling to ensure the final calculated index is within a reasonable range for model processing. Its value can be set using domain expert knowledge or after preliminary analysis of sample data. This represents the average stress intensity of the i-th stress event. This value is calculated from the average daily hypoxia stress index or drought stress index during the duration of the event, and then normalized. Represents the duration of the i-th coercive event, in days. It is the intensity response index, a constant greater than 1. The setting is intended to non-linearly amplify the effects of high-intensity stress events. It is a time-cumulative exponent, a constant greater than or equal to 1, used to describe the cumulative effect of damage over time. When This indicates that prolonged and continuous stress will have an accelerating effect on deterioration; for example, the root system's ability to recover will decrease exponentially after prolonged flooding.
[0054] This embodiment does not use conventional machine learning models, such as gradient boosting trees or standard multilayer perceptrons, when constructing the predictive model. This is because conventional machine learning models assume that all input features are independent or equivalent. However, in the scenario of fluctuating water levels in the drawdown zone, there is an asymmetric response relationship between the slowly changing background environment driven by natural rainfall and seasonal variations and the drastic anthropogenic disturbance caused by reservoir scheduling. Specifically, the destructive force of the disturbance impact is not constant but is strongly modulated by the baseline rhythm state at the time. For example, the ecological impact of flood discharge during the natural dry season on mulberry root systems is far greater than that of the same impact during the natural wet season. Conventional models cannot structurally reflect this asymmetric modulation relationship. They may learn from large amounts of data that the risk is higher when a certain rhythm and a certain impact occur simultaneously, but this learning is indirect and fuzzy, and cannot accurately model the amplification or suppression effect of rhythm on impact risk. Therefore, this embodiment proposes the following method to construct the predictive model.
[0055] A dual-stream feature embedding layer is constructed, in which a rhythm embedding network and an impact embedding network are set. The baseline rhythm features in the training set are input into the rhythm embedding network, and the rhythm embedding network outputs the rhythm context vector. The perturbation impact features are input into the impact embedding network, and the impact embedding network outputs the impact basis vector.
[0056] like Figure 2 As shown, the rhythm embedding network receives a baseline rhythm feature vector composed of dominant periodic features, trend features, and fluctuation amplitude features. Through a multilayer perceptron structure containing multiple fully connected layers and nonlinear activation functions, it maps the baseline rhythm feature vector into a high-dimensional rhythm context vector. The rhythm context vector is a deep abstract representation of the environmental background state. The shock embedding network receives a perturbation shock feature vector composed of shock frequency features, average shock intensity, and shock variation features. Through another multilayer perceptron with a similar structure, it maps the perturbation shock feature vector into a high-dimensional shock basis vector. This shock basis vector represents the original potential threat brought about by human scheduling.
[0057] It should be noted that the specific number of layers and the number of neurons in each layer of the multilayer perceptron can be determined using conventional experimental and validation methods familiar to those skilled in the art to achieve optimal performance. The rhythmic context vector and impact basis vector are not calculated using simple preset formulas, but rather learned autonomously by the model during training. At the start of model training, the parameters of these two embedding networks are randomly initialized. During training, the entire risk prediction model continuously adjusts all internal parameters based on the error between the predicted results and the actual stress risk values using a backpropagation algorithm. The goal of this process is to drive the rhythmic embedding network and the impact embedding network to learn how to perform the most effective feature transformations, thereby generating the rhythmic context vector and impact basis vector most useful for the final risk prediction task.
[0058] A gating network is constructed. The rhythm context vector is input into the gating network, and the gating network outputs a modulation gating vector. The modulation gating vector is then multiplied element-wise with the impulse basis vector to generate the modulated impulse vector.
[0059] A gated network is a fully connected layer with a sigmoid activation function. The gated network receives the rhythm context vector generated in the previous step as input. After computation, the gated network outputs a modulation gate vector, which has the same dimension as the impact basis vector, and due to the sigmoid function, each element has a value between 0 and 1. The modulation gate vector is then multiplied element-wise with the impact basis vector.
[0060] If an element in the modulation gating vector is close to 1, the original threat value at the corresponding position in the impact basis vector will be almost completely preserved. This simulates a situation where, under sensitive environmental conditions (such as the dry season), the potential threat of an impact is completely converted into an actual risk. If an element in the modulation gating vector is close to 0, the original threat value at the corresponding position in the impact basis vector will be almost attenuated to zero. This simulates a situation where, under insensitive environmental conditions (such as the wet season), the same impact threat will hardly pose an actual risk. Values between 0 and 1 represent different degrees of risk attenuation or amplification.
[0061] The rhythmic context vector, the post-modulation impact vector, and the interactive coupling features are concatenated to form a fused feature vector. This fused feature vector is then input into the prediction network, which outputs a risk prediction value.
[0062] The final fusion and prediction layer is constructed. The rhythmic context vector, the modulated impact vector, and the original interaction coupling feature vector (representing strong prior knowledge) are concatenated along the feature dimension to form the final fused feature vector. This fused feature vector encompasses the environmental background risk, the background-corrected impact risk, and the explicit interaction effect between the two. This fused feature vector is input into the prediction network, which consists of several fully connected layers and outputs a stress risk value between 0 and 1 through regression calculation. The training objective of the model is to minimize the mean squared error between the predicted stress risk value and the true stress risk value in the training dataset.
[0063] The environmental feature vectors generated from long-term environmental data for all geographic coordinate points within the drawdown zone are input into the trained risk prediction model to obtain the stress risk value for each geographic coordinate point.
[0064] After the model training is completed, the environmental feature vectors of the long-term environmental data generated for all geographic coordinate points within the drawdown zone are batch-input into the trained feature modulation interaction network. The feature modulation interaction network calculates the environmental feature vector for each geographic coordinate point and outputs a risk prediction value, which reflects the magnitude of the inherent threat posed by the environment at that geographic coordinate point to the growth of mulberry trees.
[0065] Virtual stress data for a reference plant model were generated based on short-term environmental data simulations, including: Based on the water level, soil temperature and preset root depth in the short-term environmental data, the oxygen concentration in the root zone is calculated, and the effective water volume sequence in the root zone is calculated based on the water level, soil moisture, rainfall and evaporation.
[0066] Since this is virtual stress data used to generate a reference plant model, the functional relationship between respiration oxygen consumption rate and temperature can be experimentally calibrated. Using daily collected environmental data such as water level, soil temperature, and preset soil porosity as inputs, and based on Fick's diffusion law, plant root depth is used as the variable to calculate oxygen diffusion distance. This simulates the rate of oxygen diffusion from the water-soil interface to the root zone in saturated soil. Specifically, a one-dimensional diffusion model based on Fick's second law is used, with the following formula: C represents oxygen concentration, t represents time, z represents soil depth, D represents the effective diffusion coefficient of oxygen in water (calculated using the Milington-Quick model, which is well-known in the art), and R(T) represents the root respiration oxygen consumption rate, the relationship between which and soil temperature T is represented by the Q10 model. Describe it, in which, At reference temperature The standard root respiration oxygen consumption rate is as follows. This is a temperature coefficient, representing the factor by which the respiration rate changes for every 10°C increase in temperature. The model boundary condition is that the oxygen concentration at the soil surface is equal to the dissolved oxygen concentration in the water.
[0067] The root respiration oxygen consumption rate is calculated based on the preset functional relationship between temperature and the root system. The daily oxygen concentration in the root zone is then the difference between the diffusion supply rate and the respiration oxygen consumption rate. Based on the soil moisture balance equation, using daily environmental data such as water level elevation, soil moisture, rainfall, and evaporation as inputs, the effective water content of the root zone at each geographic coordinate point is calculated and output daily.
[0068] Specifically, the water balance equation dW / dt=P+IER is used, where W is the soil moisture content in the root zone, P is the rainfall, I is the irrigation or lateral recharge (its value can be estimated using Darcy's law based on the hydraulic gradient between the root zone soil and the external water body), E is the evapotranspiration (which can be calculated using the Penman-Montes formula), and R is the deep percolation. The calculated moisture content is compared with field capacity and wilting coefficient to obtain the effective water volume.
[0069] The oxygen concentration in the root zone is compared with the hypoxia stress threshold to generate the hypoxia stress index, and the effective water volume in the root zone is compared with the drought stress threshold to generate the drought stress index sequence.
[0070] Hypoxia and drought stress thresholds are established as critical points for determining whether a plant enters a stress state. These thresholds are determined based on plant physiological studies or through controlled pot experiments. For example, the root zone oxygen concentration at which a significant decrease in the photosynthetic rate of a mulberry tree begins to occur is defined as the hypoxia stress threshold. The daily root zone oxygen concentration is compared to the hypoxia stress threshold. When the actual concentration is lower than the hypoxia stress threshold, the difference between the threshold and the actual concentration is defined as the daily hypoxia stress index; otherwise, the index is zero. Similarly, the daily effective water content in the root zone is compared to the drought stress threshold. When the actual water content is lower than the drought stress threshold, the difference between the threshold and the actual water content is defined as the daily drought stress index; otherwise, the index is zero.
[0071] Traverse the effective water volume sequence and drought stress index sequence in the root region, identify all stress periods that exceed the duration threshold, and use the duration and cumulative stress value of all stress periods as virtual stress data.
[0072] Set a time threshold, such as 3 days, and iterate through the daily hypoxia stress index time series and daily drought stress index time series for each geographic coordinate point to identify periods where the hypoxia stress index and / or drought stress index are continuously greater than zero and the duration exceeds the duration threshold. For each identified stress period, obtain the number of days of stress and the cumulative stress value composed of the sum of the daily values of the same type of stress index within the period.
[0073] In this embodiment, tolerance indicators for various mulberry varieties are extracted based on plant physiological data, including: Stress analysis was conducted on various mulberry varieties to obtain corresponding actual stress data. Based on the actual stress data, the stress period of mulberry trees at the experimental planting site was determined. Physiological fluctuation data during the stress period was extracted from plant physiological data. The actual stress data and physiological fluctuation data were clustered, and the data belonging to the same cluster in the clustering results were analyzed to determine the corresponding tolerance index.
[0074] The method for obtaining actual stress data is basically the same as the method for obtaining virtual stress data of the reference plant model. The difference is that the root length of the reference plant model is fixed, while the root length here needs to be determined according to the mulberry variety. In addition, the functional relationship between respiration oxygen consumption rate and temperature also needs to be recalibrated according to the mulberry variety.
[0075] Based on the above analysis, the stress period encountered by mulberry trees at each experimental planting site can be determined. For each determined stress period, the preceding non-stress cycle (e.g., 7 days) is defined as the baseline period. From the plant physiological data of the same period, the average photosynthetic rate, average transpiration rate, and average chlorophyll content of mulberry trees at each experimental planting site during the baseline period are extracted as physiological baseline values. Similarly, the average photosynthetic rate, average transpiration rate, and average chlorophyll content of mulberry trees during the stress period are extracted as physiological values during the stress period. The rate of change of the physiological values during the stress period relative to the physiological baseline values is calculated, and the rate of change of all physiological indicators of mulberry trees is combined into physiological fluctuation data.
[0076] For each mulberry variety, a cluster analysis dataset was constructed. Each data point in the dataset consists of two parts: the first part is the characteristics of a single stress event, including stress type, stress duration in days, and cumulative stress value; the second part is the physiological fluctuation data corresponding to the stress event. The K-means clustering algorithm was used to cluster the dataset, grouping data points with similar stress characteristics and physiological fluctuation responses into the same cluster.
[0077] Subsequently, statistical analysis was performed on the data within the same cluster to extract the rate of change of various physiological indicators of mulberry trees. It should be noted that tolerance indicators include multiple values, such as the rate of change in photosynthesis and the rate of change in root vitality. For example, statistical analysis could determine how much the average rate of change in photosynthesis of mulberry variety A would change after being subjected to a certain type of stress. The method used for statistical analysis of the data is existing technology and will not be described further here.
[0078] Using the above technical solution, stress events are automatically classified into different modes. Then, for each mulberry variety and each stress mode (i.e. each cluster), the average relative change rate of physiological indicators caused by all stress events under that mode is statistically analyzed. The vector formed by these average change rates is defined as the tolerance index of that variety under that stress mode.
[0079] In this embodiment, the growth simulator performs differentiated simulations based on stress risk values and outputs corresponding expected physiological indices, including: The stress risk value and tolerance index of mulberry trees at the experimental planting site were concatenated into a joint input feature. The expected physiological index was calculated based on the measured physiological data of mulberry trees at the experimental planting site. A training dataset was constructed based on the joint input feature and the expected physiological index.
[0080] First, a training dataset for the growth simulator is constructed based on stress risk values calculated from short-term environmental data and simultaneously calculated tolerance indices. For example, for mulberry tree samples planted at experimental planting sites, the stress risk values at their geographical coordinates and the aforementioned calculated tolerance indices are obtained. The stress risk values and tolerance indices are then concatenated to form joint input features. Simultaneously, at the end of the observation period, physiological indicators such as biomass, leaf area index, and root-to-shoot ratio of the mulberry tree samples are obtained through field measurements. Expected physiological indices are calculated based on preset weights. These weights can be defined according to the final planting goal; for example, increasing the weight of leaf area index for leaf production and increasing the weight of biomass and root-to-shoot ratio for ecological protection. Alternatively, expert decision-making methods such as the analytic hierarchy process (AHP) can be used to determine the weights, quantifying the importance of each physiological indicator. Construct a fully connected network, train the fully connected network using the training dataset, and obtain a growth simulator.
[0081] The growth simulator consists of one input layer, three hidden layers, and one output layer. The number of neurons in the input layer is the same as the dimension of the joint input features, and the output layer contains a single neuron. The growth simulator uses mean squared error as the loss function and continuously adjusts the network weights through backpropagation and gradient descent optimizers.
[0082] The stress risk value of the geographical coordinate point is combined with the tolerance index of the mulberry variety to form a joint input feature. The joint input feature is then input into the growth simulator to output the corresponding expected physiological index.
[0083] Finally, for the geographic coordinates and mulberry variety to be evaluated, the stress risk value of the geographic coordinates and the tolerance index of the mulberry variety are first obtained. The stress risk value and tolerance index are concatenated into a joint input feature, which is then input into the growth simulator. The growth simulator outputs the expected physiological index.
[0084] The aforementioned technical solution does not directly utilize collected physiological data from various mulberry varieties to train an end-to-end model—that is, the risk prediction model directly outputs which geographical coordinates are suitable for planting which variety. Instead, a phased approach is adopted. This is because an end-to-end model tightly couples environmental assessment and variety selection—two different levels of decision-making—leading to poor model generalization and insufficient system flexibility. For example, when introducing a new mulberry variety, a large amount of data needs to be collected again and the entire complex model retrained. In contrast, the phased approach first uses the risk prediction model to perform a general environmental assessment, generating stress risk values. Subsequently, the growth simulator uses these environmental assessment results, combined with the tolerance characteristics of different varieties, to conduct differentiated simulations. This decoupled design makes the risk prediction model a reusable basic environmental assessment tool, while the growth simulator can flexibly incorporate the physiological parameters of any new variety for rapid assessment, greatly improving the scalability and practicality of the entire system.
[0085] In this embodiment, the optimal planting plan, including the best planting location and the best planting variety, is determined based on preset planting requirements, including: An optimization model is constructed and solved with the objective function of maximizing the sum of expected physiological indices of the entire planting area. The constraints of the optimization model include planting uniqueness constraint, regional coverage constraint, variety diversity constraint, minimum survival threshold constraint, and cost budget constraint.
[0086] Specifically, a binary decision variable matrix is constructed, where the decision variables... To represent a planting decision, if the decision variables are... If the value of is 1, it means that variety j is planted at geographical coordinate point i. If the decision variable A value of 0 indicates that variety j is not planted at geographic coordinate point i. A comprehensive benefit function is then constructed, the goal of which is to maximize the sum of the expected physiological indices of all selected planting decisions. The growth simulator has output the expected physiological indices for each combination of geographic coordinate point i and each variety j. Therefore, the comprehensive benefit function can be expressed as: .
[0087] The constraints include: each geographic coordinate point can plant at most one variety of mulberry tree, expressed as follows: for any geographic coordinate point i, Regional coverage constraints are implemented to address the requirement of uniform planting or coverage in specific areas as outlined in the pre-defined planting requirements. The planting area is divided into sub-regions such as safe zones and risk zones based on a planting risk map, and a minimum coverage threshold is set for the sub-regions where coverage needs to be guaranteed. , expressed as ,in This represents the total number of geographic coordinates within subregion R. The coverage constraint forces the subregion to allocate a sufficient number of plantings. There are also variety diversity constraints, setting upper and lower limits on the planting quantity of each mulberry variety to avoid large-scale planting of a single variety. Finally, there is a minimum survival threshold constraint, setting a minimum survival index threshold, which only applies when the expected physiological index is higher than a certain threshold. At the minimum survival index threshold, the corresponding The cost is set to 1; cost budget constraint, setting a unit planting cost for each variety j. and set a total budget , expressed as .
[0088] Then, the defined comprehensive benefit function and all mathematical constraints are integrated to form an integer linear programming model. A solver is used to solve the integer linear programming model to find the decision variable matrix that maximizes the comprehensive benefit function while satisfying all mathematical constraints, thereby obtaining the optimal planting scheme. The final planting scheme is shown below. Figure 3 As shown.
[0089] like Figure 4 As shown, this embodiment also provides a system for regulating the adaptive growth of mulberry trees under water level fluctuations in the drawdown zone. This system is used to implement the method described above, and includes: The first acquisition module collects short-term environmental data from preset geographical coordinates using sensors, and then simulates and generates long-term environmental data based on the short-term environmental data.
[0090] The first prediction module extracts environmental feature vectors from long-term environmental data, establishes a risk prediction model, and inputs the environmental feature vectors into the risk prediction model to obtain the stress risk value of each geographical coordinate point.
[0091] The second data acquisition module selects experimental planting sites at geographical coordinates, plants different varieties of mulberry trees at these sites, and collects plant physiological data concurrent with short-term environmental data. Based on this plant physiological data, tolerance indicators for various mulberry varieties are extracted.
[0092] The second prediction module establishes a growth simulator. The tolerance index is input into the growth simulator, which performs differentiated simulation based on the stress risk value and outputs the corresponding expected physiological index.
[0093] The plan generation module determines the optimal planting plan, including the best planting location and the best planting variety, based on the expected physiological index and preset planting requirements.
[0094] In the specific implementation of this invention, various thresholds, weighting coefficients, model hyperparameters, and empirical constants are involved, such as, but not limited to, hypoxia stress threshold, drought stress threshold, duration threshold, weighting coefficients (e.g., 0.4 and 0.6) in the stress risk value calculation formula, and minimum survival threshold constraints in the optimization model. Their specific values are not the sole limitation of this invention. Those skilled in the art should understand that the optimal values of these parameters can be adjusted and optimized based on specific application scenarios, geographical characteristics, mulberry variety characteristics, and expected planting goals, using conventional experimental methods, literature references, expert experience, or cross-validation techniques. For example, the critical point of physiological stress can be calibrated by setting up a series of gradient experiments, or the weights can be determined by statistical analysis of historical data. Therefore, reasonable selection or adjustment of these parameters without departing from the core technical concept of this invention should fall within the protection scope of this invention.
[0095] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0096] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.
[0097] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone, characterized in that, include: Short-term environmental data is collected from preset geographic coordinates using sensors, and long-term environmental data is generated by simulating the short-term environmental data. Environmental feature vectors are extracted from long-term environmental data, a risk prediction model is established, and the environmental feature vectors are input into the risk prediction model to obtain the stress risk value of each geographic coordinate point. Select experimental planting sites at geographical coordinates, plant different varieties of mulberry trees at the experimental planting sites, and collect plant physiological data concurrent with short-term environmental data. Extract tolerance indicators for various mulberry varieties based on the plant physiological data. A growth simulator was established, and tolerance indicators were input into the growth simulator. The growth simulator performed differential simulation based on stress risk values and output the corresponding expected physiological indices. Based on the expected physiological indices and the preset planting requirements, the optimal planting plan is determined, including the best planting location and the best planting variety.
2. The method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone according to claim 1, characterized in that, Long-term environmental data is generated by simulating short-term environmental data, including: Historical meteorological data with a time span longer than short-term environmental data is acquired, and a gradient boosting model and a generative adversarial network are constructed. In the gradient boosting model, a loss function containing a prediction error term and a dynamic baseline regularization term is defined based on short-term environmental data and corresponding historical meteorological data. The dynamic baseline regularization term is used to penalize the high-frequency components of the predicted values of the gradient boosting model. The trained model finally outputs a natural fluctuation sequence, and a scheduling fluctuation sequence is generated based on short-term environmental data and the natural fluctuation sequence. The scheduling fluctuation sequence is input into the generative adversarial network for training. After training, the complete historical meteorological data is input into the gradient boosting model to generate a long-term natural fluctuation sequence. The generative adversarial network is then used to generate a long-term scheduling fluctuation sequence. The long-term natural fluctuation sequence and the long-term scheduling fluctuation sequence are used as long-term environmental data.
3. The method for regulating the adaptive growth of mulberry trees under water level fluctuations in the drawdown zone according to claim 2, characterized in that, Extracting environmental feature vectors from long-term environmental data, including: Baseline rhythm features are extracted based on long-term natural fluctuation sequences. These features include dominant cycle features, trend features, and fluctuation amplitude features. Disturbance and impact features are extracted based on long-term scheduling fluctuation sequences. These features include impact frequency characteristics, average impact intensity, and impact variation characteristics. The natural abundant water sequence, the scheduled water increase sequence, the natural dry water sequence, and the scheduled water decrease sequence are extracted from the long-term natural fluctuation sequence and the long-term scheduled fluctuation sequence. The interaction coupling features include the mean of the product of the natural abundant water sequence and the scheduled water increase sequence, and the mean of the product of the natural dry water sequence and the scheduled water decrease sequence. The baseline rhythm features, disturbance impact features, and interactive coupling features are concatenated to obtain the environmental feature vector.
4. The method for regulating the adaptive growth of mulberry trees under water level fluctuations in the drawdown zone according to claim 3, characterized in that, To obtain a natural abundant water sequence, retain all values greater than zero in the long-term natural fluctuation sequence and set the rest to zero. To obtain a natural low water sequence, retain all values less than zero in the long-term natural fluctuation sequence and set the rest to zero. To obtain a scheduled water increase sequence, retain all values greater than zero in the long-term scheduled fluctuation sequence and set the rest to zero. To obtain a scheduled water decrease sequence, retain all values less than zero in the long-term scheduled fluctuation sequence and set the rest to zero.
5. The method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone according to claim 3, characterized in that, Establish a risk prediction model, including: A reference plant model representing the average physiological properties of mulberry trees was established. Virtual stress data of the reference plant model was generated by simulating short-term environmental data. The environmental stress index was calculated based on the virtual stress data. A training dataset was constructed based on the environmental feature vector and environmental stress index in the short-term environmental data. A dual-stream feature embedding layer is constructed, in which a rhythm embedding network and an impact embedding network are set. The baseline rhythm features in the training set are input into the rhythm embedding network, and the rhythm embedding network outputs the rhythm context vector. The perturbation impact features are input into the impact embedding network, and the impact embedding network outputs the impact basis vector. Construct a gating network, input the rhythm context vector into the gating network, and the gating network outputs a modulation gating vector. Perform element-wise multiplication between the modulation gating vector and the impulse basis vector to generate the modulated impulse vector. The rhythm context vector, the modulated impact vector, and the interactive coupling features are concatenated to form a fused feature vector. The fused feature vector is then input into the prediction network, which outputs a risk prediction value. The environmental feature vectors of long-term environmental data for all geographic coordinate points within the drawdown zone are input into the trained risk prediction model to obtain the stress risk value for each geographic coordinate point.
6. The method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone according to claim 5, characterized in that, Virtual stress data for a reference plant model were generated based on short-term environmental data simulations, including: Based on the water level, soil temperature and preset root depth in the short-term environmental data, the oxygen concentration in the root zone is calculated, and the effective water volume sequence in the root zone is calculated based on the water level, soil moisture, rainfall and evaporation. The oxygen concentration in the root zone is compared with the hypoxia stress threshold to generate the hypoxia stress index, and the effective water volume in the root zone is compared with the drought stress threshold to generate the drought stress index sequence. Traverse the effective water volume sequence and drought stress index sequence in the root region, identify all stress periods that exceed the duration threshold, and use the duration and cumulative stress value of all stress periods as virtual stress data.
7. The method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone according to claim 5, characterized in that, Tolerance indicators for various mulberry varieties were extracted based on plant physiological data, including: Stress analysis was conducted on various mulberry varieties to obtain corresponding actual stress data. Based on the actual stress data, the stress period of mulberry trees at the experimental planting site was determined. Physiological fluctuation data during the stress period was extracted from plant physiological data. The actual stress data and physiological fluctuation data were clustered, and the data belonging to the same cluster in the clustering results were analyzed to determine the corresponding tolerance index.
8. The method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone according to claim 7, characterized in that, The growth simulator performs differentiated simulations based on stress risk values and outputs corresponding expected physiological indices, including: The stress risk value and tolerance index of mulberry trees at the experimental planting site were concatenated into a joint input feature. The expected physiological index was calculated based on the measured physiological data of mulberry trees at the experimental planting site. A training dataset was constructed based on the joint input feature and the expected physiological index. Construct a fully connected network, train the fully connected network using the training dataset, and obtain a growth simulator; The stress risk value of the geographical coordinate point is combined with the tolerance index of the mulberry variety to form a joint input feature. The joint input feature is then input into the growth simulator to output the corresponding expected physiological index.
9. The method for adaptive regulation of mulberry tree growth under water level fluctuations in the drawdown zone according to claim 1, characterized in that, Determine the optimal planting plan based on preset planting requirements, including the best planting location and the best planting variety, including: An optimization model is constructed and solved with the objective function of maximizing the sum of expected physiological indices of the entire planting area. The constraints of the optimization model include planting uniqueness constraint, regional coverage constraint, variety diversity constraint, minimum survival threshold constraint, and cost budget constraint.
10. A system for regulating the adaptive growth of mulberry trees under water level fluctuations in the drawdown zone, used to implement the method for regulating the adaptive growth of mulberry trees under water level fluctuations in the drawdown zone as described in any one of claims 1-9, characterized in that, The system includes: The first acquisition module collects short-term environmental data from preset geographic coordinates using sensors, and then simulates and generates long-term environmental data based on the short-term environmental data. The first prediction module extracts environmental feature vectors from long-term environmental data, establishes a risk prediction model, and inputs the environmental feature vectors into the risk prediction model to obtain the stress risk value of each geographical coordinate point. The second data acquisition module selects experimental planting sites at geographical coordinates, plants different varieties of mulberry trees at the experimental planting sites, and collects plant physiological data concurrent with short-term environmental data. Based on the plant physiological data, it extracts tolerance indicators for various mulberry varieties. The second prediction module establishes a growth simulator, inputs tolerance indicators into the growth simulator, performs differentiated simulation based on stress risk values, and outputs the corresponding expected physiological index. The plan generation module determines the optimal planting plan, including the best planting location and the best planting variety, based on the expected physiological index and preset planting requirements.
Citation Information
Patent Citations
Method for ecologically remediating water-level-fluctuating zone through water logging resistant mulberries
CN108643125A
Hydro-fluctuation belt ecological restoration system and method
CN120465412A
