Method for predicting and regulating soil salt content in stress-resistant cultivation of crops in saline-alkali soil

Abstract

The application relates to the technical field of saline-alkali soil improvement, and discloses a salt content dynamic prediction and regulation method for crop stress-resistant cultivation soil in saline-alkali soil, which predicts the soil salt content of the saline-alkali soil through a target soil salt content prediction model (including a ConvLSTM encoder, a physical constraint feature extraction layer and a Transformer encoder) to obtain a target soil conductivity prediction value and a target soil volume water content prediction value; then, the target soil conductivity prediction value, the target soil volume water content prediction value, a current soil volume water content and a preset salt content threshold are taken as inputs, a multi-objective optimization algorithm is used to solve a multi-objective optimization model which takes minimizing the soil conductivity exceeding value and the irrigation water consumption of the saline-alkali soil as a target and is constrained by the total daily irrigation amount and the dynamic soil volume water content, a target irrigation strategy is obtained, and the saline-alkali soil is irrigated; thus, a closed loop from prediction to regulation can be realized, and the saline-alkali soil management effect is improved.

Description

Technical Field

[0001] This invention relates to the field of saline-alkali land improvement technology, and relates to, but is not limited to, a method for predicting and regulating soil salinity dynamics in saline-alkali land for the cultivation of crops with stress resistance. Background Technology

[0002] In the cultivation of crops in saline-alkali land, different crops and different growth stages have different salt tolerance thresholds, which are also affected by factors such as soil texture, groundwater level, and meteorological conditions. Currently, sensors and communication equipment can be deployed in saline-alkali land to collect real-time data on soil moisture, salt content, and nutrients (fertility), which is then transmitted to a monitoring platform via wireless network. This allows for monitoring of the saline-alkali land and irrigation when the salt content exceeds a preset threshold, thus regulating soil salinity. However, the spatial generalization ability of the sensors in this approach is limited, leading to certain errors in the monitoring results and affecting management effectiveness. Summary of the Invention

[0003] In view of this, embodiments of the present invention provide a method for predicting and regulating soil salinity dynamics in the cultivation of crops in saline-alkali land, which can predict the soil electrical conductivity of saline-alkali land and determine irrigation strategies based on the prediction results, thereby improving management effectiveness.

[0004] The specific technical solutions of this invention are as follows: The first aspect of this application provides a method for predicting and regulating soil salinity dynamics in saline-alkali land crops for stress-resistant cultivation, including: Acquire monitoring data of saline-alkali land within a preset time period, and preprocess and time-series align the monitoring data to obtain time-series input data; the monitoring data includes at least soil electrical conductivity, soil volumetric water content, soil temperature, meteorological data, and irrigation event data; Based on time-series input data, the soil electrical conductivity at a predetermined soil depth within a predetermined time period is predicted using a target soil salinity prediction model, yielding predicted values ​​for both target soil electrical conductivity and target soil volumetric water content. The target soil salinity prediction model comprises a ConvLSTM encoder, a physical constraint feature extraction layer, and a Transformer encoder, arranged sequentially. The ConvLSTM encoder extracts spatiotemporal features; the physical constraint feature extraction layer incorporates prior laws of soil water and salt transport as soft constraints into the physical residual term during training; and the Transformer encoder performs long-term time-series dependency modeling and outputs predicted values. A multi-objective optimization model is constructed. The multi-objective optimization model aims to minimize the excessive soil electrical conductivity and irrigation water consumption in saline-alkali land, and is subject to the constraints of daily total irrigation volume and dynamic soil volumetric water content. The daily total irrigation volume constraint is that the irrigation water consumption in the future preset time period is less than or equal to the preset daily maximum irrigation volume. The dynamic soil volumetric water content constraint is that the soil volumetric water content in each time period in the future preset time period is within the safe constraint range. Using the predicted values ​​of target soil electrical conductivity, target soil volumetric water content, current soil volumetric water content, and preset salinity threshold as inputs, a multi-objective optimization model is solved through a multi-objective optimization algorithm to obtain the target irrigation strategy. Based on the target irrigation strategy, the saline-alkali land is irrigated. The target irrigation strategy is used to indicate the irrigation water volume for each irrigation period within a preset future time period.

[0005] In some embodiments, constructing a multi-objective optimization model includes: Using irrigation water consumption in each time period within a future preset time period as decision variables, a dual objective function is constructed with the goal of minimizing the excess value of soil electrical conductivity and irrigation water consumption in saline-alkali land. The prediction uncertainty of the target soil volumetric water content was determined by the target soil salinity prediction model, and a dynamic soil volumetric water content constraint was constructed based on the prediction uncertainty. Based on the constraints of dynamic soil volumetric water content, daily irrigation volume, and dual objective functions, a multi-objective optimization model is determined.

[0006] In some embodiments, the dual objective function is represented by the following formula: in, This represents the predicted soil electrical conductivity value for time period i. To preset the salt threshold, Let be the irrigation water consumption for the i-th time period.

[0007] In some embodiments, a dynamic soil volumetric water content constraint is constructed based on prediction uncertainty, including: Based on the uncertainty of forecasting, the forecast confidence level for each future period is determined; The safety constraint range for soil volumetric water content is adjusted based on the prediction confidence level to obtain the adjusted safety constraint range. Based on the adjusted safety constraint range, a dynamic soil volumetric water content constraint is constructed.

[0008] In some embodiments, the prediction confidence level for each future time period is determined using the following formula: in, Let be the prediction confidence level for the i-th time period. The preset adjustment coefficient, Let represent the prediction uncertainty for the i-th time period.

[0009] In some embodiments, the safety constraint range for soil volumetric water content is adjusted based on the prediction confidence level to obtain the adjusted safety constraint range, including: When the prediction confidence level is less than or equal to a preset threshold, the safety constraint interval is narrowed using the following formula: in, This is the lower limit of the adjusted safety constraint range. Let be the prediction confidence level for the i-th time period. This is the lower limit within the safety constraint range. The preset shrinkage strength coefficient, This refers to the upper limit of the adjusted safety constraint range. This represents the upper limit of the safety constraint range.

[0010] In some embodiments, a target irrigation strategy is obtained by solving a preset multi-objective optimization model using a multi-objective optimization algorithm, including: The initialization involves a population of multiple individuals, where each individual represents a candidate irrigation strategy. Perform iterative optimization, call the target soil salinity prediction model, determine the predicted soil electrical conductivity and soil volumetric water content for each individual, and based on the predicted soil electrical conductivity, predicted soil volumetric water content, current soil volumetric water content and preset salinity threshold, determine the soil electrical conductivity exceedance value, irrigation water consumption and total constraint violation degree for each individual. Based on the soil conductivity exceeding the standard value, irrigation water consumption and total constraint violation degree corresponding to each individual, the constraint dominance rule is used to perform non-dominated ranking of multiple individuals and determine the crowding distance of individuals in each non-dominated layer. Based on the non-dominated ranking and crowding distance, selection, crossover and mutation operations are performed on multiple individuals to obtain the offspring population. The offspring population is merged with the current population, and the individuals included in the merged population are screened based on non-dominated ordering and crowding distance to obtain a new population. When the number of optimization iterations is greater than or equal to the preset number threshold, the final population is obtained, and the multiple individuals in the final population that are at the first non-dominated frontier and have a total constraint violation of zero are identified as the Pareto optimal irrigation strategy set. Based on the current growth stage of the crop, a target irrigation strategy is selected from the Pareto optimal irrigation strategy set.

[0011] In some embodiments, selecting a target irrigation strategy from a set of Pareto optimal irrigation strategies based on the current crop growth stage includes: If the current growth stage is the seedling stage, then select the irrigation strategy with the smallest soil electrical conductivity exceeding the standard value from the Pareto optimal irrigation strategy set as the target irrigation strategy. If the current growth stage is mid-growth, then select the irrigation strategy with the minimum weighted sum of soil electrical conductivity exceeding the standard value and irrigation water consumption from the Pareto optimal irrigation strategy set as the target irrigation strategy. If the current growth stage is the mature stage, then the irrigation strategy with the minimum irrigation water consumption is selected from the Pareto optimal irrigation strategy set as the target irrigation strategy.

[0012] In some embodiments, after irrigating saline-alkali land based on a target irrigation strategy, the method further includes: Soil data after irrigation were collected and used as post-intervention samples to be fed into the target soil salinity prediction model for incremental learning, so as to dynamically optimize the target soil salinity prediction model.

[0013] In some embodiments, soil data is used as post-intervention samples and input into a target soil salinity prediction model for incremental learning to dynamically optimize the target soil salinity prediction model, including: The post-intervention samples are stored in a pre-defined experience playback buffer. Training batches are obtained by sampling and mixing samples from the experience replay buffer and the original training set, and incremental training of the target soil salinity prediction model is performed based on the training batches. The proportion of the post-intervention samples in the training batches is greater than or equal to the preset proportion. When calculating the loss, the weight assigned to the post-intervention samples is greater than the weight assigned to the original training samples.

[0014] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, monitoring data of saline-alkali land within a preset time period is first acquired, and the monitoring data is preprocessed and time-series aligned to obtain time-series input data. Then, based on the time-series input data, the soil electrical conductivity at a preset soil depth within a future preset time period is predicted using a target soil salinity prediction model to obtain the predicted values ​​of target soil electrical conductivity and target soil volumetric water content. The target soil salinity prediction model includes a ConvLSTM encoder, a physical constraint feature extraction layer, and a Transformer encoder arranged sequentially. The ConvLSTM encoder is used to extract spatiotemporal features. The physical constraint feature extraction layer is used to introduce the prior laws of soil water and salt transport as soft constraints into the physical residual term during the training phase. The Transformer encoder is used to perform long-term time-series dependency modeling and output the predicted values. Next, a multi-objective optimization model is constructed. This model aims to minimize the excessive soil electrical conductivity and irrigation water consumption in saline-alkali land, and is constrained by daily total irrigation volume and dynamic soil volumetric water content. The daily total irrigation volume constraint is that the irrigation water consumption within a preset future time period must be less than or equal to a preset maximum daily irrigation volume. The dynamic soil volumetric water content constraint is that the soil volumetric water content at each time period within the preset future time period must be within a safe constraint range. Finally, using the predicted target soil electrical conductivity, the predicted target soil volumetric water content, the current soil volumetric water content, and the preset salinity threshold as inputs, the multi-objective optimization model is solved using a multi-objective optimization algorithm to obtain the target irrigation strategy. Based on this strategy, the saline-alkali land is irrigated. The target irrigation strategy indicates the irrigation water consumption for each irrigation time period within the preset future time period. Thus, the integration of ConvLSTM encoder, Transformer encoder and physical constraint feature extraction layer can improve the accuracy of soil salinity prediction. Then, by combining multi-objective optimization algorithm, irrigation strategies that save water resources under the premise of the lowest soil salinity exceedance rate are selected, realizing closed-loop decision-making from prediction to regulation and improving the management effect of saline-alkali land. Attached Figure Description

[0015] 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 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, wherein: Figure 1 This is a flowchart illustrating the method for predicting and regulating soil salinity dynamics in saline-alkali land crops for stress-resistant cultivation, as provided in an embodiment of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. The following embodiments are used to illustrate the present invention, but are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0018] It should be noted that the terms "first, second, and third" used in the embodiments of the present invention are only used to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of the present invention described herein can be implemented in an order other than that illustrated or described herein.

[0019] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which these embodiments of the invention pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0020] Figure 1 This is a flowchart illustrating a method for dynamic prediction and regulation of soil salinity in saline-alkali land crops under stress resistance, provided in an embodiment of the present invention. The method can be executed via a control device, which includes at least one of a personal computer, laptop computer, smartphone, tablet computer, and portable wearable device; this embodiment does not limit the specific device used.

[0021] like Figure 1 As shown, the method for predicting and regulating soil salinity dynamics in saline-alkali land crops under stress resistance provided in this embodiment of the invention may include steps S101-S104.

[0022] S101. Obtain monitoring data of saline-alkali land within a preset time period, and preprocess and align the monitoring data with the time series to obtain time series input data.

[0023] For example, the control device can monitor the saline-alkali land by using multiple sensors deployed at monitoring points in the saline-alkali land and irrigation equipment, and then filter the monitoring data within a preset time period (such as the past 72 hours) from the data obtained from the monitoring to complete the acquisition of monitoring data.

[0024] In some embodiments, the monitoring data includes at least soil electrical conductivity, soil volumetric water content, soil temperature, meteorological data, and irrigation event data.

[0025] For example, the sensor may include at least one of a soil conductivity sensor, a soil volumetric water content sensor, a soil temperature sensor, an air temperature and humidity sensor, a wind speed sensor, a rainfall sensor, and a solar radiation sensor. In practical applications, the control device uses the soil conductivity sensor, soil volumetric water content sensor, soil temperature sensor, air temperature and humidity sensor, wind speed sensor, rainfall sensor, and solar radiation sensor to collect real-time data on soil conductivity, soil volumetric water content, soil temperature, and meteorological data (such as air temperature, air humidity, wind speed, rainfall, and solar radiation) at different soil depths (e.g., 0–10 cm, 10–20 cm, 20–40 cm) in saline-alkali land. The control device can also collect irrigation time data for saline-alkali land through irrigation equipment.

[0026] In some embodiments, the control device performs data cleaning, missing value interpolation, and timestamp alignment on the monitoring data to complete the preprocessing and time-series alignment of the monitoring data, thereby obtaining time-series input data.

[0027] For example, the control device can remove outliers from the monitoring data that exceed the physical range (such as soil moisture content greater than 100%), and fill in missing data using linear interpolation or the average of previous and subsequent time points to obtain preprocessed data. Then, the timestamps of the sensor data (such as soil electrical conductivity, soil volumetric water content, soil temperature, and meteorological data) and irrigation event data in the preprocessed data are aligned to the same time base, such as Coordinated Universal Time (UTC), and resampled and aggregated at preset time intervals (such as 1 hour) to generate initial time-series input data for different soil depths at each time point. The initial time-series input data includes sensor time-series data and irrigation event time-series data. Sensor time-series data at different soil depths for each time point include features such as soil electrical conductivity (EC), volumetric water content (VWC), soil temperature, air temperature, air humidity, wind speed, rainfall, and solar radiation. Each feature dimension is subjected to min-max normalization, mapping its numerical range to the [0, 1] interval to obtain normalized sensor time-series data. For irrigation event time-series data, it can be converted into a continuous time-series signal based on pulse coding and decaying convolution to obtain the processed irrigation signal. Finally, the normalized sensor time-series data and the processed irrigation signal are fused to obtain the time-series input data.

[0028] S102. Based on the time-series input data, the soil electrical conductivity at a preset soil depth within a preset future time period is predicted using the target soil salinity prediction model, thereby obtaining the predicted values ​​of the target soil electrical conductivity and the target soil volumetric water content.

[0029] Since soil water and salt transport in saline-alkali land is a complex process influenced by multiple physicochemical mechanisms such as water movement, solute diffusion, evaporation and transpiration, and ion exchange, to improve the prediction accuracy of the target soil salinity prediction model, the model includes a Convolutional Long Short-Term Memory (ConvLSTM) encoder, a Physically Informed Residual Block (PIRB) layer, and a Transformer encoder, arranged sequentially. The ConvLSTM encoder extracts spatiotemporal features; the PIRB layer incorporates prior laws of soil water and salt transport as soft constraints into the physical residual term during training; and the Transformer encoder performs long-term time-series dependency modeling and outputs predicted values.

[0030] In some embodiments, the target soil salinity prediction model is the trained initial soil salinity prediction model. During the training phase, the training sample set is input into the initial soil salinity prediction model. The ConvLSTM encoder in the initial soil salinity prediction model extracts local spatiotemporal features from the historical time-series data included in the training sample set. The physical constraint feature extraction layer decodes the implicit soil water and salt state based on the local spatiotemporal features extracted by the ConvLSTM encoder, and calculates the physical residual based on the prior laws of soil water and salt transport (such as the convection-diffusion equation). This physical residual is used as a soft constraint regularization term and constitutes the total loss function with the data loss between the predicted output and the true value of the initial soil salinity prediction model. The initial soil salinity prediction model is optimized through backpropagation. Afterward, the spatiotemporal features extracted by the ConvLSTM encoder are output to the Transformer encoder. The Transformer encoder performs long-term time-series dependency modeling on the spatiotemporal features and outputs the predicted soil electrical conductivity and soil volumetric water content. During the inference phase, the control device inputs time-series input data into the target soil salinity prediction model. Upon receiving the time-series input data, the model extracts spatiotemporal features from the data using a ConvLSTM encoder. Subsequently, the Transformer encoder models long-term temporal dependencies on these extracted features, capturing long-term relationships within the data. Finally, it outputs the predicted values ​​for target soil electrical conductivity and target soil volumetric water content at predetermined soil depths (e.g., 0–10 cm, 10–20 cm, 20–40 cm) within a future preset time period (e.g., the next 24 hours).

[0031] In some embodiments, the ConvLSTM encoder includes multiple stacked two-dimensional convolutional long short-term memory (ConvLSTM2D) units. For example, a ConvLSTM2D unit may have three layers. The ConvLSTM2D unit includes a gating mechanism based on convolution operations (including input gate, forget gate, and output gate) capable of extracting the spatiotemporal dependencies of input features. It should be noted that the structure of the ConvLSTM2D unit is prior art and will not be described in detail here.

[0032] The physical constraint feature extraction layer consists of a physical state decoder, a numerical differentiation module, a physical equation builder, and a source-sink term builder. The physical state decoder decodes the implicit soil electrical conductivity and soil volumetric water content from the spatiotemporal features extracted by the ConvLSTM encoder. The numerical differentiation module calculates the temporal and spatial derivatives of soil electrical conductivity and soil volumetric water content using differentiable difference operations (such as forward difference and central difference). The source-sink term builder processes irrigation event data, modeling it as exponentially decaying source terms. The physical equation builder calculates the physical residuals of salt transport based on the one-dimensional convection-diffusion equation, approximating water flux with the soil volumetric water content gradient.

[0033] The Transformer encoder consists of a multi-head self-attention mechanism layer, a feedforward neural network, layer normalization and residual connections, and a position encoding layer. The multi-head self-attention mechanism layer captures long-range dependencies within the sequence; the feedforward neural network layer performs a non-linear transformation on the output of the multi-head self-attention mechanism layer; the layer normalization and residual connection layer stabilizes the training process and accelerates convergence; and the position encoding layer injects positional information into each element of the sequence, providing temporal order information. It should be noted that the structure of the Transformer encoder is existing technology and will not be elaborated further here.

[0034] It is understood that the embodiments of this application, by constructing a hybrid model (i.e., a target soil salinity prediction model) that integrates a ConvLSTM encoder, a Transformer encoder, and a physical constraint feature extraction layer, can embed the physical prior knowledge of soil water and salt transport into the model training in the form of soft constraints. This makes the prediction results output by the target soil salinity prediction model (such as the predicted value of target soil electrical conductivity and the predicted value of target soil volumetric water content) conform to the basic laws of mass conservation and transport, thereby improving the prediction accuracy.

[0035] S103. Construct a multi-objective optimization model.

[0036] In some embodiments, the multi-objective optimization model aims to minimize the excess soil electrical conductivity and irrigation water consumption in saline-alkali land, and is constrained by the total daily irrigation volume and dynamic soil volumetric water content.

[0037] Among them, the daily total irrigation amount constraint is that the irrigation water consumption in the future preset period is less than or equal to the preset daily maximum irrigation amount; the dynamic soil volumetric water content constraint is that the soil volumetric water content in each period in the future preset period is within the safe constraint range.

[0038] For example, the daily irrigation total constraint can be expressed by the following formula 1: (Formula 1) in, Let represent the irrigation water consumption in the i-th time period. This is the preset maximum daily irrigation volume.

[0039] The dynamic soil volumetric water content constraint can be expressed by the following formula 2: (Formula 2) in, This is the lower limit within the safety constraint range. Let be the soil volumetric water content in time period i. This represents the upper limit of the safety constraint range.

[0040] In some embodiments, constructing a multi-objective optimization model includes: using irrigation water consumption in each period within a future preset time period as decision variables, constructing a bi-objective function with the objectives of minimizing the excess value of soil electrical conductivity and irrigation water consumption in saline-alkali land; determining the prediction uncertainty of the target soil volumetric water content prediction value through a target soil salinity prediction model, and constructing a dynamic soil volumetric water content constraint based on the prediction uncertainty; and determining the multi-objective optimization model based on the dynamic soil volumetric water content constraint, the daily irrigation total constraint, and the bi-objective function.

[0041] In some application scenarios, the biobjective function is expressed as shown in Equation 3 below: (Formula 3) in, This represents the predicted soil electrical conductivity value for time period i. To preset the salt threshold, Let be the irrigation water consumption for the i-th time period.

[0042] In some embodiments, a dropout layer can be included in the feedforward neural network of the Transformer encoder, enabling the target soil salinity prediction model to determine the prediction uncertainty of the target soil volumetric water content prediction value.

[0043] For example, during inference, the target soil salinity prediction model can perform multiple (e.g., 100) independent forward propagations on the same time-series input data. Due to the randomness of the Dropout layer, slightly different predicted values ​​of target soil volumetric water content will be obtained each time forward propagation. Thus, the variance or standard deviation of the predicted values ​​of target soil volumetric water content obtained from multiple forward propagations can be used as the prediction uncertainty.

[0044] In some embodiments, constructing dynamic soil volumetric water content constraints based on prediction uncertainty includes: determining prediction confidence levels for each future time period based on prediction uncertainty; adjusting the safety constraint range of soil volumetric water content according to the prediction confidence levels to obtain the adjusted safety constraint range; and constructing dynamic soil volumetric water content constraints based on the adjusted safety constraint range.

[0045] In some application scenarios, the prediction confidence level for each future time period can be determined using the following formula 4: (Formula 4) in, Let be the prediction confidence level for the i-th time period. The preset adjustment coefficient, Let represent the prediction uncertainty for the i-th time period.

[0046] In some embodiments, the safety constraint range for soil volumetric water content is adjusted according to the prediction confidence level, and the adjusted safety constraint range includes shrinking the safety constraint range when the prediction confidence level is less than or equal to a preset threshold.

[0047] In some application scenarios, when the prediction confidence is less than or equal to a preset threshold, the safety constraint interval is narrowed using the following formula 5: (Formula 5) in, This is the lower limit of the adjusted safety constraint range. Let be the prediction confidence level for the i-th time period. This is the lower limit within the safety constraint range. The preset shrinkage strength coefficient, This refers to the upper limit of the adjusted safety constraint range. This represents the upper limit of the safety constraint range.

[0048] It should be noted that the shrinkage strength coefficient is a preset value, which can be set according to actual needs. For example, the shrinkage strength coefficient can be 0.5.

[0049] Understandably, dynamic soil volumetric moisture content constraints can automatically tighten the constraints to avoid over-irrigation or under-irrigation.

[0050] S104. Using the predicted value of the target soil electrical conductivity, the predicted value of the target soil volumetric water content, the current soil volumetric water content, and the preset salinity threshold as inputs, the multi-objective optimization model is solved by a multi-objective optimization algorithm to obtain the target irrigation strategy, and the saline-alkali land is irrigated based on the target irrigation strategy.

[0051] The target irrigation strategy is used to indicate the amount of irrigation water used in each irrigation period within a preset future time period.

[0052] For example, a multi-objective optimization algorithm can be the Non-dominated Sorting Genetic Algorithm II (NSGA-Ⅱ).

[0053] In some embodiments, solving a preset multi-objective optimization model using a multi-objective optimization algorithm to obtain the target irrigation strategy includes: initializing a population comprising multiple individuals; wherein each individual represents a candidate irrigation strategy; performing iterative optimization, calling a target soil salinity prediction model, determining the predicted soil electrical conductivity and predicted soil volumetric water content for each individual, and based on the predicted soil electrical conductivity, predicted soil volumetric water content, current soil volumetric water content, and a preset salinity threshold, determining the soil electrical conductivity exceedance value, irrigation water consumption, and total constraint violation degree for each individual; and based on the soil electrical conductivity exceedance value, irrigation water consumption, and total constraint violation degree for each individual... Multiple individuals are ranked using constraint dominance rules, and the crowding distance of individuals within each non-dominated layer is determined. Based on the non-dominated ranking and crowding distance, selection, crossover, and mutation operations are performed on multiple individuals to obtain a progeny population. The progeny population is merged with the current population, and the individuals in the merged population are screened based on the non-dominated ranking and crowding distance to obtain a new population. When the number of optimization iterations is greater than or equal to a preset threshold, the final population is obtained. The multiple individuals in the final population that are at the first non-dominated frontier and have a total constraint violation of zero are identified as the Pareto optimal irrigation strategy set. Based on the current crop growth stage, a target irrigation strategy is selected from the Pareto optimal irrigation strategy set.

[0054] For example, the control device can randomly generate a population including multiple individuals. Then, for each individual, a target soil salinity prediction model is invoked to determine the predicted soil electrical conductivity and soil volumetric water content for each individual. Next, based on the predicted soil electrical conductivity and a preset salinity threshold, the excess soil electrical conductivity for each individual is determined. Based on the predicted soil volumetric water content and the current soil volumetric water content, the irrigation water consumption for each individual is determined. Simultaneously, by calculating the degree of constraint violation for each individual, the total constraint violation degree for each population is determined.

[0055] After obtaining the soil conductivity exceedance value, irrigation water consumption, and total constraint violation degree for each individual, the control device can perform non-dominated ranking of multiple individuals based on the constraint dominance rule. For example, for any two individuals A and B, if the soil conductivity exceedance value of individual A is less than that of individual B, the irrigation water consumption of individual A is less than that of individual B, and the total constraint violation degree of individual A is less than that of individual B, then individual A is determined to be constrained and dominates individual B. In this way, multiple individuals can be divided into multiple non-dominated layers, where individuals in the first non-dominated layer are not constrained and dominated by any other individual. Simultaneously, for each non-dominated layer, the control device can calculate the distance between each individual in that layer and the soil conductivity exceedance value and irrigation water consumption, and sum these distances after normalization to obtain the crowding distance of each individual. The larger the crowding distance, the sparser the solutions around the individual, and the more likely it is to contain better solutions.

[0056] Finally, based on the non-dominated ranking and crowding distance, the control device can perform selection, crossover, and mutation operations on multiple individuals to obtain the offspring population. The selection operation prioritizes individuals at higher non-dominated levels and with larger crowding distances based on the non-dominated ranking and crowding distance; the crossover operation generates new individuals by exchanging parts of the genes of two individuals; and the mutation operation increases population diversity by randomly altering certain genes of individuals. After merging the offspring population with the current population, the control device can again screen the individuals in the merged population based on the non-dominated ranking and crowding distance, prioritizing the retention of individuals at higher non-dominated levels and with larger crowding distances, while removing some individuals at lower non-dominated levels or with smaller crowding distances to obtain a new population. When the number of optimization iterations is greater than or equal to a preset threshold (e.g., 200 iterations), the control device obtains the final population. At this point, the multiple individuals in the final population that are at the first non-dominated frontier and have a total constraint violation of zero can be identified as the Pareto optimal irrigation strategy set. The Pareto optimal irrigation strategy set includes irrigation strategies that can meet the constraints of daily total irrigation volume and dynamic soil volumetric water content, while minimizing the excess soil electrical conductivity and irrigation water consumption in saline-alkali land.

[0057] It is understood that the embodiments of this application are based on a target soil salinity prediction model to predict soil salinity, and combined with a multi-objective optimization algorithm to select an irrigation strategy that saves water resources under the premise of the lowest soil salinity exceedance rate. This can realize a closed-loop decision-making from prediction to regulation and improve the management effect of saline-alkali land.

[0058] In some embodiments, selecting a target irrigation strategy from the Pareto optimal irrigation strategy set according to the current crop growth stage includes: if the current growth stage is the seedling stage, selecting the irrigation strategy with the smallest soil electrical conductivity exceeding the standard value from the Pareto optimal irrigation strategy set as the target irrigation strategy; if the current growth stage is the mid-growth stage, selecting the irrigation strategy with the smallest weighted sum of soil electrical conductivity exceeding the standard value and irrigation water consumption from the Pareto optimal irrigation strategy set as the target irrigation strategy; if the current growth stage is the maturity stage, selecting the irrigation strategy with the smallest irrigation water consumption from the Pareto optimal irrigation strategy set as the target irrigation strategy.

[0059] For example, during the seedling stage, crops are highly sensitive to soil salinity. Therefore, choosing an irrigation strategy that minimizes the excess soil electrical conductivity helps ensure that crops can grow in a suitable soil environment during the early stages of growth. In the mid-growth stage, crops grow rapidly, increasing their demand for water and nutrients. At this time, choosing an irrigation strategy that minimizes the weighted sum of the excess soil electrical conductivity and irrigation water usage can ensure the water needed for crop growth while minimizing the adverse effects of soil salinity. During the maturity stage, crop growth gradually slows down, and the demand for water decreases relatively. At this time, choosing an irrigation strategy that minimizes irrigation water usage helps conserve water resources and improve irrigation efficiency.

[0060] Understandably, selecting targeted irrigation strategies based on crop growth stages can make rational use of water resources and improve the stress resistance of crops in saline-alkali land.

[0061] To enable the target soil salinity prediction model to dynamically adapt to changes in the soil environment and the impact of control measures, in some embodiments, after irrigating saline-alkali land based on the target irrigation strategy, the method for dynamic prediction and control of soil salinity in saline-alkali land crop stress-resistant cultivation provided in this embodiment of the invention further includes: collecting the implemented soil data after irrigation, and inputting the implemented soil data as a post-intervention sample into the target soil salinity prediction model for incremental learning, so as to dynamically optimize the target soil salinity prediction model.

[0062] For example, soil data after irrigation can be collected by multiple sensors deployed at monitoring points in saline-alkali land and by irrigation equipment.

[0063] In some embodiments, inputting the implemented soil data as post-intervention samples into the target soil salinity prediction model for incremental learning to dynamically optimize the target soil salinity prediction model includes: storing the post-intervention samples in a preset experience replay buffer; sampling and mixing the experience replay buffer and the original training set to obtain a training batch, and incrementally training the target soil salinity prediction model based on the training batch; wherein the proportion of post-intervention samples in the training batch is greater than or equal to a preset proportion; and the weight assigned to the post-intervention samples is greater than the weight assigned to the original training samples when calculating the loss.

[0064] For example, the experience replay buffer is a first-in, first-out queue with a fixed capacity (e.g., N=1000 samples) used to store post-intervention samples to balance the distribution of new and old data. When incrementally learning the target soil salinity prediction model, M post-intervention samples can be randomly sampled from the experience replay buffer, and N historical samples can be randomly sampled from the original training set. The post-intervention samples and historical samples are mixed to form a training batch, and the target soil salinity prediction model is incrementally trained based on this training batch. To ensure that the target soil salinity prediction model can fully learn the soil change characteristics after irrigation, the proportion of post-intervention samples in the training batch is greater than or equal to a preset proportion (e.g., 50%). Simultaneously, to make the target soil salinity prediction model pay more attention to post-intervention samples when updating parameters, thereby accelerating the adaptation process of the target soil salinity prediction model to dynamic changes in the soil environment, a larger weight is assigned to the post-intervention samples when calculating the loss. For example, the weight of the post-intervention samples is 0.6, and the weight of the historical samples is 0.4.

[0065] Understandably, through incremental learning mechanisms, the target soil salinity prediction model can continuously optimize its prediction performance, more accurately reflect the real-time state of soil salinity, improve the reliability of subsequent irrigation strategy formulation, and thus improve the management effect of saline-alkali land.

[0066] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of the invention, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the invention. The sequence numbers of the above-described embodiments of the invention are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0067] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0068] In the several embodiments provided by this invention, it should be understood that the disclosed methods can be implemented in other ways. The methods disclosed in the several method embodiments provided by this invention can be arbitrarily combined without conflict to obtain new method embodiments. The features disclosed in the several method embodiments provided by this invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0069] The above description is merely an embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting and regulating soil salinity dynamics in saline-alkali land crops for stress-resistant cultivation, characterized in that, include: Acquire monitoring data of saline-alkali land within a preset time period, and preprocess and time-series align the monitoring data to obtain time-series input data; the monitoring data includes at least soil electrical conductivity, soil volumetric water content, soil temperature, meteorological data, and irrigation event data; Based on time-series input data, the soil electrical conductivity at a predetermined soil depth within a predetermined time period is predicted using a target soil salinity prediction model, yielding predicted values ​​for both target soil electrical conductivity and target soil volumetric water content. The target soil salinity prediction model comprises a ConvLSTM encoder, a physical constraint feature extraction layer, and a Transformer encoder, arranged sequentially. The ConvLSTM encoder extracts spatiotemporal features; the physical constraint feature extraction layer incorporates prior laws of soil water and salt transport as soft constraints into the physical residual term during training; and the Transformer encoder performs long-term time-series dependency modeling and outputs predicted values. A multi-objective optimization model is constructed. The multi-objective optimization model aims to minimize the excessive soil electrical conductivity and irrigation water consumption in saline-alkali land, and is subject to the constraints of daily total irrigation volume and dynamic soil volumetric water content. The daily total irrigation volume constraint is that the irrigation water consumption in the future preset time period is less than or equal to the preset daily maximum irrigation volume. The dynamic soil volumetric water content constraint is that the soil volumetric water content in each time period in the future preset time period is within the safe constraint range. Using the predicted values ​​of target soil electrical conductivity, target soil volumetric water content, current soil volumetric water content, and preset salinity threshold as inputs, a multi-objective optimization model is solved through a multi-objective optimization algorithm to obtain the target irrigation strategy. Based on the target irrigation strategy, the saline-alkali land is irrigated. The target irrigation strategy is used to indicate the irrigation water volume for each irrigation period within a preset future time period.

2. The method according to claim 1, characterized in that, Construct a multi-objective optimization model, including: Using irrigation water consumption in each time period within a future preset time period as decision variables, a dual objective function is constructed with the goal of minimizing the excess value of soil electrical conductivity and irrigation water consumption in saline-alkali land. The prediction uncertainty of the target soil volumetric water content was determined by the target soil salinity prediction model, and a dynamic soil volumetric water content constraint was constructed based on the prediction uncertainty. Based on the constraints of dynamic soil volumetric water content, daily irrigation volume, and dual objective functions, a multi-objective optimization model is determined.

3. The method according to claim 2, characterized in that, The dual objective function is expressed by the following formula: in, This represents the predicted soil electrical conductivity value for time period i. To preset the salt threshold, Let represent the irrigation water consumption during the i-th time period.

4. The method according to claim 2, characterized in that, Dynamic soil volumetric water content constraints are constructed based on prediction uncertainties, including: Based on the uncertainty of forecasting, the forecast confidence level for each future period is determined; The safety constraint range for soil volumetric water content is adjusted based on the prediction confidence level to obtain the adjusted safety constraint range. Based on the adjusted safety constraint range, a dynamic soil volumetric water content constraint is constructed.

5. The method according to claim 4, characterized in that, The prediction confidence level for each future time period is determined using the following formula: in, Let be the prediction confidence level for the i-th time period. The preset adjustment coefficient, Let represent the prediction uncertainty for the i-th time period.

6. The method according to claim 4, characterized in that, The safety constraint range for soil volumetric water content is adjusted based on the prediction confidence level to obtain the adjusted safety constraint range, which includes: When the prediction confidence level is less than or equal to a preset threshold, the safety constraint interval is narrowed using the following formula: in, This is the lower limit of the adjusted safety constraint range. Let be the prediction confidence level for the i-th time period. This is the lower limit within the safety constraint interval. The preset shrinkage strength coefficient, This refers to the upper limit of the adjusted safety constraint range. This represents the upper limit of the safety constraint range.

7. The method according to claim 1, characterized in that, The target irrigation strategy is obtained by solving the pre-defined multi-objective optimization model using a multi-objective optimization algorithm, including: The initialization involves a population of multiple individuals, where each individual represents a candidate irrigation strategy. Perform iterative optimization, call the target soil salinity prediction model, determine the predicted soil electrical conductivity and soil volumetric water content for each individual, and based on the predicted soil electrical conductivity, predicted soil volumetric water content, current soil volumetric water content and preset salinity threshold, determine the soil electrical conductivity exceedance value, irrigation water consumption and total constraint violation degree for each individual. Based on the soil conductivity exceeding the standard value, irrigation water consumption and total constraint violation degree corresponding to each individual, the constraint dominance rule is used to perform non-dominated ranking of multiple individuals and determine the crowding distance of individuals in each non-dominated layer. Based on the non-dominated ranking and crowding distance, selection, crossover and mutation operations are performed on multiple individuals to obtain the offspring population. The offspring population is merged with the current population, and the individuals included in the merged population are screened based on non-dominated ordering and crowding distance to obtain a new population. When the number of optimization iterations is greater than or equal to the preset number threshold, the final population is obtained, and the multiple individuals in the final population that are at the first non-dominated frontier and have a total constraint violation of zero are identified as the Pareto optimal irrigation strategy set. Based on the current growth stage of the crop, a target irrigation strategy is selected from the Pareto optimal irrigation strategy set.

8. The method according to claim 1, characterized in that, Based on the current growth stage of the crop, a target irrigation strategy is selected from the Pareto optimal irrigation strategy set, including: If the current growth stage is the seedling stage, then select the irrigation strategy with the smallest soil electrical conductivity exceeding the standard value from the Pareto optimal irrigation strategy set as the target irrigation strategy. If the current growth stage is mid-growth, then select the irrigation strategy with the minimum weighted sum of soil electrical conductivity exceeding the standard value and irrigation water consumption from the Pareto optimal irrigation strategy set as the target irrigation strategy. If the current growth stage is the mature stage, then the irrigation strategy with the minimum irrigation water consumption is selected from the Pareto optimal irrigation strategy set as the target irrigation strategy.

9. The method according to claim 1, characterized in that, Based on the target irrigation strategy, after irrigating saline-alkali land, the method further includes: Soil data after irrigation were collected and used as post-intervention samples to be fed into the target soil salinity prediction model for incremental learning, so as to dynamically optimize the target soil salinity prediction model.

10. The method according to claim 9, characterized in that, Soil data from the intervention was used as post-intervention samples and fed into the target soil salinity prediction model for incremental learning to dynamically optimize the model. This included: The post-intervention samples are stored in a pre-defined experience playback buffer. Training batches are obtained by sampling and mixing samples from the experience replay buffer and the original training set, and incremental training of the target soil salinity prediction model is performed based on the training batches. The proportion of the post-intervention samples in the training batches is greater than or equal to the preset proportion. When calculating the loss, the weight assigned to the post-intervention samples is greater than the weight assigned to the original training samples.

Images