Vegetation water demand prediction method and device based on multi-dimensional data, equipment and medium

Abstract

The invention discloses a vegetation water demand prediction method and device based on multi-dimensional data, equipment and a medium. The method comprises the following steps: constructing a time sequence data set based on all ecological environment indexes of target vegetation; inputting the time sequence data set into a water demand prediction model comprising a long short-term memory (LSTM) network layer and a prediction calculation layer; the LSTM network layer performs feature extraction processing on the time sequence data set to obtain an implicit state, and the implicit state can represent a nonlinear mapping relation which is extracted from the time sequence data set and has the strongest correlation with the water demand of the target vegetation; and the prediction calculation layer calculates the target water demand corresponding to the target vegetation through a nonlinear fusion algorithm and an implicit state. According to the scheme provided by the embodiment of the invention, for the target vegetation, the dynamic change of the water supply and demand of the target vegetation in various ecological system environments is comprehensively reflected based on the multi-dimensional ecological index data and the LSTM network, so that the prediction accuracy of the water demand is improved.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of vegetation water demand prediction technology, and in particular to a method, apparatus, equipment and medium for predicting vegetation water demand based on multidimensional data. Background Technology

[0002] In practical desertification control and ecological restoration, the lack of scientific and systematic methods for predicting vegetation water demand often relies on the experience and judgment of managers or single monitoring indicators, leading to low water use efficiency and difficulty in accurately determining irrigation timing and volume. This not only wastes water resources but also hinders healthy vegetation growth. Therefore, developing a technology that can accurately predict the water demand of desert vegetation is of great significance for improving water resource utilization efficiency and promoting desertification control.

[0003] Current water demand prediction methods mainly fall into two categories: one is index-based calculation methods using remote sensing and meteorological data. These methods have the advantage of acquiring data rapidly and over a wide area, but their limitation lies in relying on remote sensing image inversion results and lacking characterization of the physiological characteristics of vegetation and underground soil moisture conditions. The other category is drought assessment methods based on single environmental factors or traditional indices. While these methods have achieved some success in farmland scenarios, their data sources are often limited to a single dimension of meteorology or soil, making it impossible to dynamically predict the future water shortage risk of the vegetation. Especially in desert ecosystems, where vegetation types are complex and root distribution varies significantly, a single factor cannot comprehensively reflect the water demand characteristics of the vegetation. In other words, existing water demand prediction methods have poor accuracy and applicability, making it difficult to meet the needs of desertification control and ecological restoration. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and medium for predicting vegetation water demand based on multidimensional data, which can effectively improve the accuracy of predicting vegetation water demand under different ecological environments.

[0005] In a first aspect, embodiments of this application provide a method for predicting vegetation water demand based on multidimensional data, including:

[0006] Obtain ecological and environmental indicators corresponding to the target vegetation, wherein the ecological and environmental indicators include at least meteorological data, soil data, vegetation cover (VC), normalized difference vegetation index (NDVI), and desertification risk index (DRI) of the area where the target vegetation is located. All the aforementioned ecological and environmental indicators were time-series aligned and standardized to construct a time-series dataset; The time series dataset is input into a pre-trained water demand prediction model, wherein the water demand prediction model includes a long short-term memory (LSTM) network layer and a prediction computation layer. The LSTM network layer performs feature extraction processing on the time series dataset to obtain hidden states, wherein the hidden states can represent the nonlinear mapping relationship extracted from the time series dataset that is most strongly correlated with the water demand of the target vegetation. The prediction calculation layer calculates the target water demand corresponding to the target vegetation using a nonlinear fusion algorithm and the implicit state.

[0007] In some embodiments, the water demand prediction model is trained according to the following steps: Input the preset training set into the initial model and output the first water demand; The loss value is calculated based on the first water demand, the corresponding actual water demand, and the preset loss function; When the loss value is greater than or equal to a preset loss threshold, the Adam optimizer is used to adjust the model parameters of the initial model. The training set is then input into the model after the model parameters have been adjusted for retraining, and a new loss value is recalculated until the new loss value is less than the loss threshold. The model corresponding to the loss value less than the loss threshold is then determined as the water demand prediction model.

[0008] In some embodiments, the time-series dataset includes multiple feature sequences, and each feature sequence includes multiple dimensions of the ecological environment indicators acquired at the same time. A loss value is calculated based on the first water demand, the corresponding actual water demand, and a preset loss function, obtained according to the following formula: ; in, L The loss value, N The number of the feature sequences. For the first i The first water demand is output after the feature sequence is input into the initial model. for The corresponding actual water demand.

[0009] In some embodiments, the target water demand includes multiple second water demands over consecutive time periods. After the prediction calculation layer calculates the target water demand corresponding to the target vegetation using a nonlinear fusion algorithm and the implicit state, the method further includes: Obtain the target water demand within a preset time period, and divide all the target water demand into low water demand samples, medium water demand samples, and high water demand samples; Select a number of preset candidate water demand thresholds and calculate the true positive rate and false positive rate of the samples corresponding to each candidate water demand threshold. Plot an ROC curve based on the true positive rate and false positive rate associated with all the candidate water demand thresholds. Calculate the Youden index corresponding to each of the candidate water demand thresholds, wherein the Youden index is the difference between the false positive rate and the true positive rate corresponding to the candidate water demand threshold; Among all the low water demand samples, the candidate water demand threshold corresponding to the Youden index with the largest value is determined as the first candidate threshold, wherein the first candidate threshold represents the lower limit of mild water shortage; Among all the high water demand samples, the candidate water demand threshold corresponding to the Youden index with the largest value is determined as the candidate second threshold, wherein the candidate second threshold represents the upper limit of high water scarcity; A target first threshold is selected from all the candidate first thresholds, and a target second threshold is selected from all the candidate second thresholds, wherein the sum of the Youden index corresponding to the target first threshold and the Youden index corresponding to the target second threshold is the largest among the sums of any candidate first threshold and any candidate second threshold, and the target second threshold is greater than the target first threshold; When the target water demand is greater than or equal to the target second threshold, calculate the first target irrigation water volume, and generate the first irrigation early warning information based on the target water demand and the corresponding first target irrigation water volume. When the target water demand is greater than or equal to the target first threshold and less than the target second threshold, the second target irrigation water volume is calculated, and the second irrigation early warning information is generated based on the target water demand and the corresponding second target irrigation water volume.

[0010] In some embodiments, after generating irrigation early warning information based on the target water demand and the corresponding target irrigation water volume, the method further includes: Periodically acquire ecological environment indicators for the target vegetation after irrigation with the target irrigation water volume, and calculate the actual water demand based on the ecological environment indicators; The difference between the actual water demand and the target water demand predicted before irrigation is defined as the feedback error. Based on the feedback error, the weight matrix of the prediction calculation layer in the water demand prediction model is adjusted to obtain a new water demand prediction model. The new water demand prediction model was retrained based on the training set.

[0011] In some embodiments, the LSTM network layer performs feature extraction processing on the time-series dataset to obtain the hidden states, which are then calculated according to the following formula: ; in, For the output parameters of the forget gate, The output parameters of the input gate, For the LSTM network layer in t The memory unit of a moment for t The memory representation corresponding to the time-series dataset at the specified time point. For the LSTM network layer in t The memory unit at time -1 for t The implicit state at time. for t The implicit state at time -1 The output parameters of the output gate. For the time series dataset, It is the Sigmoid activation function. It is the hyperbolic tangent function. , , and This is the weight matrix. , , and This is a bias term.

[0012] In some embodiments, the prediction calculation layer calculates the target water requirement corresponding to the target vegetation using a nonlinear fusion algorithm and the implicit state, and obtains the result according to the following formula: ; in, for t The target water demand is predicted at any given time. This is the weight matrix. This is a bias term.

[0013] Secondly, embodiments of this application provide a control device, including at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, which, when executed by the at least one control processor, enable the at least one control processor to perform the vegetation water demand prediction method based on multidimensional data as described in the first aspect.

[0014] Thirdly, embodiments of this application also provide an electronic device, including the control device of the second aspect.

[0015] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer-executable instructions for performing the vegetation water demand prediction method based on multidimensional data as described in the first aspect.

[0016] This application provides a method, apparatus, device, and medium for predicting vegetation water demand based on multidimensional data. The method includes: acquiring ecological environment indicators corresponding to a target vegetation, wherein the ecological environment indicators include at least meteorological data, soil data, vegetation cover (VC), normalized difference vegetation index (NDVI), and desertification risk index (DRI) of the area where the target vegetation is located; performing time-series alignment and standardization on all the ecological environment indicators to construct a time-series dataset; inputting the time-series dataset into a pre-trained water demand prediction model, wherein the water demand prediction model includes a long short-term memory (LSTM) network layer and a prediction computation layer; the LSTM network layer performing feature extraction processing on the time-series dataset to obtain a hidden state, wherein the hidden state can represent the nonlinear mapping relationship extracted from the time-series dataset that is most strongly correlated with the water demand of the target vegetation; and the prediction computation layer calculating the target water demand corresponding to the target vegetation through a nonlinear fusion algorithm and the hidden state. According to the solution provided in the embodiments of this application, for the target vegetation, based on multi-dimensional ecological indicator data and LSTM network, the dynamic changes of water supply and demand under various ecosystem environments are comprehensively reflected, so as to improve the prediction accuracy of water demand. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of a vegetation water demand prediction method based on multidimensional data provided in one embodiment of this application; Figure 2 This is a structural diagram of a control device provided in another embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] It is understandable that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, or the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0020] In practical desertification control and ecological restoration, the lack of scientific and systematic methods for predicting vegetation water demand often relies on the experience and judgment of managers or single monitoring indicators, leading to low water use efficiency and difficulty in accurately determining irrigation timing and volume. This not only wastes water resources but also hinders healthy vegetation growth. Therefore, developing a technology that can accurately predict the water demand of desert vegetation is of great significance for improving water resource utilization efficiency and promoting desertification control.

[0021] Current water demand prediction methods mainly fall into two categories: one is index-based calculation methods using remote sensing and meteorological data. These methods have the advantage of acquiring data rapidly and over a wide area, but their limitation lies in relying on remote sensing image inversion results and lacking characterization of the physiological characteristics of vegetation and underground soil moisture conditions. The other category is drought assessment methods based on single environmental factors or traditional indices. While these methods have achieved some success in farmland scenarios, their data sources are often limited to a single dimension of meteorology or soil, making it impossible to dynamically predict the future water shortage risk of the vegetation. Especially in desert ecosystems, where vegetation types are complex and root distribution varies significantly, a single factor cannot comprehensively reflect the water demand characteristics of the vegetation. In other words, existing water demand prediction methods have poor accuracy and applicability, making it difficult to meet the needs of desertification control and ecological restoration.

[0022] To address the aforementioned problems, this application provides a method, apparatus, device, and medium for predicting vegetation water demand based on multidimensional data. The method includes: acquiring ecological environment indicators corresponding to the target vegetation, wherein the ecological environment indicators include at least meteorological data, soil data, vegetation cover (VC), normalized difference vegetation index (NDVI), and desertification risk index (DRI) of the area where the target vegetation is located; performing time-series alignment and standardization on all the ecological environment indicators to construct a time-series dataset; inputting the time-series dataset into a pre-trained water demand prediction model, wherein the water demand prediction model includes a long short-term memory (LSTM) network layer and a prediction computation layer; the LSTM network layer performing feature extraction processing on the time-series dataset to obtain a hidden state, wherein the hidden state can represent the nonlinear mapping relationship extracted from the time-series dataset that has the strongest correlation with the water demand of the target vegetation; and the prediction computation layer calculating the target water demand corresponding to the target vegetation using a nonlinear fusion algorithm and the hidden state. According to the solution provided in the embodiments of this application, for the target vegetation, based on multi-dimensional ecological indicator data and LSTM network, the dynamic changes of water supply and demand under various ecosystem environments are comprehensively reflected, so as to improve the prediction accuracy of water demand.

[0023] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0024] refer to Figure 1 , Figure 1 This is a flowchart illustrating the steps of a vegetation water demand prediction method based on multidimensional data according to an embodiment of this application. This embodiment provides a vegetation water demand prediction method based on multidimensional data, which includes, but is not limited to, the following steps: Step S10: Obtain the ecological environment indicators corresponding to the target vegetation. The ecological environment indicators include at least the meteorological data, soil data, vegetation cover (VC), normalized difference vegetation index (NDVI), and desertification risk index (DRI) of the area where the target vegetation is located.

[0025] Specifically, the meteorological data in this embodiment includes temperature T (including daily average temperature and extreme values), precipitation P, wind speed U, and net solar radiation. Potential evaporation capacity of PET. PET is calculated using the following formula: ; in, The saturated vapor pressure slope (kPa / °C) Net solar radiation (MJ / / day), Soil heat flux (MJ / / day), The wet / dry constant (kPa / °C) Temperature (°C) 2-meter high wind speed , The vapor pressure difference is expressed in kPa.

[0026] In this embodiment, meteorological data is collected daily, and a 7-day sequence is recorded and stored in a MySQL table. For example, the sequence corresponding to the meteorological data is: [time, T, P, U, Rn, PET].

[0027] Specifically, the soil data in this embodiment includes soil moisture content (SMC), field capacity (FC), wilting water content (WP), available soil water capacity (AWC), soil texture, and permeability coefficient (k). Here, available soil water capacity (AWC) is the difference between field capacity (FC) and wilting water content (WP).

[0028] It is understood that in this embodiment, the soil data is collected daily, and the 7-day sequence is recorded and stored in a MySQL table. For example, the sequence corresponding to meteorological data is: [time, region ID, SMC, FC, WP, AWC, k, soil texture].

[0029] In this embodiment, the NDVI is obtained from MODIS images, and the NDVI is calculated according to the following formula: ; Where NIR stands for near-infrared band and RED stands for red band.

[0030] In this embodiment, the NDVI and VC are collected daily or every 3 days, and the corresponding sequence is [time, region ID, vegetation ID, name, NDVI, VC].

[0031] It is understood that, in this embodiment, the ecological environment indicators for the target vegetation may also include root depth Rd, leaf area index LAI, drought tolerance coefficient DTC, transpiration coefficient Kc, and historical irrigation records. These indicators correspond to parameters related to vegetation water demand in desertification ecosystems. This embodiment obtains different types of multi-dimensional ecological environment indicators based on different ecosystems, providing an effective data foundation for subsequent calculations of more targeted and accurate target water demand.

[0032] Specifically, the Desertification Risk Index (DRI) in this embodiment is calculated according to the following formula: ; in, and The weighting coefficient is represented by PET, which represents potential evapotranspiration and reflects maximum water evaporation. P represents precipitation (if P=0 on a given day, the 3-day average precipitation is used instead to avoid dividing by zero). For soil water shortage, Calculated using the following formula: FC stands for field water holding capacity, SMC for soil moisture content, and AWC for soil available water capacity.

[0033] It is understandable that the DRI calculated in this embodiment can comprehensively consider water shortage risk by integrating vegetation, meteorological, and soil data. After calculating the DRI, the DRI is combined with meteorological data, soil data, vegetation cover VC, and normalized difference vegetation index NDVI as inputs to the prediction model. This can effectively capture the impact of extreme environments such as desert sandstorms and hot and dry conditions on the water demand of the target vegetation, providing an effective data foundation for obtaining accurate target water demand in the future.

[0034] Step S20: Perform time-series alignment and standardization on all ecological and environmental indicators to construct a time-series dataset.

[0035] It is understood that after calculating the DRI, this embodiment integrates the DRI with meteorological data, soil data, VC, NDVI, etc. (including time series alignment and standardization processing) to form a time series unified dataset. This time series dataset is used as the input data for the water demand prediction model for dynamic prediction modeling of the water demand index WDI.

[0036] Specifically, the steps for constructing a time-series dataset include: aligning and standardizing data from different sources using time and spatial location as indices to ensure consistency between meteorological, soil, and vegetation data and DRI at the same time and in the same region; constructing a time-series dataset using time t as the sequence unit, for example, the expression for a time-series dataset is Xt=[Tt, Pt, PETt, Rnt, SMCt, FCt, WPt, AWCt, VCt, NDVIt, LAIt, DTCt, Kct, DRIt], where meteorological data (T, P, PET, Rn) reflect external climate conditions, soil data (SMC, FC, WP, AWC) characterize groundwater status, vegetation parameters (VC, NDVI, LAI, DTC, Kc) reflect vegetation physiological characteristics, and DRI serves as an environmental constraint factor to comprehensively characterize regional desertification risk.

[0037] In addition, the method provided in this embodiment also includes filling in missing data or observation gaps by using moving average or linear interpolation methods. All fused time series datasets are stored in a MySQL database in time series format, with fields including [time, region ID, vegetation ID, T, P, PET, Rn, SMC, FC, WP, AWC, VC, NDVI, LAI, DTC, Kc, DRI], providing high-quality input data support for subsequent LSTM dynamic prediction models.

[0038] Step S30: Input the time series dataset into the pre-trained water demand prediction model, wherein the water demand prediction model includes a long short-term memory (LSTM) network layer and a prediction calculation layer.

[0039] Additionally, it should be noted that the water demand prediction model in this embodiment also includes a data preprocessing layer. This layer is used to normalize and standardize the multidimensional input data (i.e., the time series dataset), unifying the dimensions of different physical quantities, eliminating data scale differences, and improving the stability of model training. The specific processing method is as follows: a piecewise linear normalization method is used to establish a transformation relationship based on different variable intervals, and corrections are made in conjunction with desert type and temperature conditions to obtain standardized input data. The calculation formula for the standardized input data obtained from the time series dataset is as follows: ; in, These are the original observations; These are the minimum and maximum values ​​of the variable, respectively. The values ​​are standardized and range from [0, 1]. The input to this layer is the original time-series dataset. The data includes meteorological data, soil data, vegetation data, and drought risk index (DRI), and the output is a standardized time-series dataset. This step achieves the fusion and scaling of multidimensional data, avoiding the impact of high-dimensional feature bias on model training.

[0040] Specifically, the training method for the water demand prediction model in this embodiment includes, but is not limited to, the following steps: Input the preset training set into the initial model and output the first water demand; The loss value is calculated based on the first water demand, the corresponding actual water demand, and the preset loss function; When the loss value is greater than or equal to the preset loss threshold, the Adam optimizer is used to adjust the model parameters of the initial model. The training set is then input into the model after the model parameters have been adjusted for retraining, and a new loss value is recalculated until the new loss value is less than the loss threshold. The model corresponding to the loss value that is less than the loss threshold is determined as the water demand prediction model.

[0041] Specifically, the loss value is calculated based on the first water demand, the corresponding actual water demand, and a preset loss function, using the following formula: ; in, L The loss value. N The number of feature sequences, For the first i The first water demand is output after the feature sequence is input into the initial model. for The corresponding actual water demand.

[0042] It is understood that in this embodiment, the training time step of the training model is set to 7 days, the hidden layer degree is 100, and a preset training set (i.e., a historical time series dataset, including...) is used. N The initial model is trained using a set of feature sequences, with the training labels being the measured or calculated historical water demand. Mean squared error is used as the loss function, and the Adam optimization algorithm is employed for parameter updates. During training, the model is considered to have reached convergence when the loss threshold is less than the target value or the number of training iterations exceeds a preset threshold, thus obtaining the trained water demand prediction model.

[0043] It should be noted that the historical actual water demand (WDI) is calculated using the following formula: ; Where, TAW = (FC - WP) × ×1000, where TAW is the available water volume in the root zone; =SMC× ×1000, For the current soil water storage, The WDI value represents the root zone depth. A WDI value closer to 1 indicates a more severe water shortage, while a WDI value closer to 0 indicates sufficient water.

[0044] In step S40, the LSTM network layer performs feature extraction processing on the time series dataset to obtain the hidden state. The hidden state can represent the nonlinear mapping relationship with the strongest correlation to the water demand of the target vegetation extracted from the time series dataset.

[0045] In step S50, the prediction calculation layer calculates the target water demand corresponding to the target vegetation using a nonlinear fusion algorithm and implicit states.

[0046] It should be noted that the LSTM network layer performs feature extraction on the time-series dataset to obtain the hidden states, which are then calculated using the following formula: ; in, For the output parameters of the forget gate, The output parameters of the input gate, For LSTM network layers in t The memory unit of a moment for t The memory representation corresponding to the time-series dataset. For LSTM network layers in t The memory unit at time -1 for t The implicit state of a moment. for t The hidden state at time -1 The output parameters of the output gate. It is a time series dataset. It is the Sigmoid activation function. It is the hyperbolic tangent function. , , and This is the weight matrix. , , and This is a bias term.

[0047] It is understood that the LSTM network layer in this embodiment includes memory units, input gates, forget gates, and output gates. The memory units... For storing integrated soil moisture accumulation and vegetation evapotranspiration responses, the forgetting gate This model is used to simulate the vegetation's ability to retain memories of past drought conditions. During prolonged droughts, the model maintains a high memory value. When rainfall or irrigation events occur, a forgetting gate automatically decays old information, determining how much historical information to discard. (Input gate...) Used to control the intensity of new meteorological and soil characteristics entering the system, and to update the "moisture reserve status"; output gate. To determine the generation intensity of the final water demand index, the Long Short-Term Memory Neural Network (LSTM) is used to predict the water demand of the target vegetation in a time series. This breaks through the limitations of static calculation in existing technologies, enabling the water demand prediction model to dynamically capture the temporal dependence of meteorological data, soil data and vegetation moisture changes, thus achieving accurate prediction of the target water demand.

[0048] It should be noted that, Figure 1 The target water demand in step S50 is calculated using the following formula: ; in, for t The target water demand is predicted in real time. This is the weight matrix. This is a bias term.

[0049] Understandable The value range of is [0, 1]. A higher numerical value indicates a greater degree of water deficit in the target vegetation. By comparing the prediction results with the theoretical water requirement of the vegetation, the model can dynamically adjust the prediction results, providing an effective data foundation for subsequent tiered predictions of water demand.

[0050] It should be noted that, in some embodiments, the target water demand includes multiple secondary water demands over consecutive time periods, during execution. Figure 1 Following step S50, the vegetation water demand prediction method based on multidimensional data provided in this application embodiment further includes, but is not limited to, the following steps: Step S61: Obtain the target water demand within a preset time period, and divide all target water demand into low water demand samples, medium water demand samples and high water demand samples. Step S62: Select multiple preset candidate water demand thresholds, and calculate the true positive rate and false positive rate of the samples corresponding to each candidate water demand threshold. Plot an ROC curve based on the true positive rate and false positive rate associated with all candidate water demand thresholds. Step S63: Calculate the Youden index corresponding to each candidate water demand threshold, where the Youden index is the difference between the false positive rate and the true positive rate corresponding to the candidate water demand threshold. Step S64: Among all the low water demand samples, the candidate water demand threshold corresponding to the Youden index with the largest value is determined as the first candidate threshold, where the first candidate threshold represents the lower limit of mild water shortage. Step S65: Among all the high water demand samples, the candidate water demand threshold corresponding to the Youden index with the largest value is determined as the candidate second threshold, where the candidate second threshold represents the upper limit of high water shortage. Step S66: Select a target first threshold from all the candidate first thresholds, and select a target second threshold from all the candidate second thresholds, wherein the sum of the Youden index corresponding to the target first threshold and the Youden index corresponding to the target second threshold is the largest among the sums of any candidate first threshold and any candidate second threshold, and the target second threshold is greater than the target first threshold; Step S67: When the target water demand is greater than or equal to the target second threshold, calculate the first target irrigation water volume, and generate the first irrigation early warning information based on the target water demand and the corresponding first target irrigation water volume; Step S68: When the target water demand is greater than or equal to the first target threshold and less than the second target threshold, calculate the second target irrigation water volume, and generate the second irrigation early warning information based on the target water demand and the corresponding second target irrigation water volume.

[0051] It is understood that the target water requirement calculated in this embodiment can provide effective data support for subsequently determining whether the target vegetation has irrigation needs, and calculating the specific irrigation amount when there are irrigation needs.

[0052] Specifically, in this embodiment, the determination of whether there is a graded interval for irrigation demand (i.e., the first target threshold and the second target threshold) is obtained according to the following method: taking the target water demand as the independent variable, selecting multiple preset candidate water demand thresholds from 0 to 1 in steps of 0.01, and calculating the true positive rate (TPR) (i.e., the number of correctly identified high water demand (or low water demand) samples) and the false positive rate (FPR) (i.e., the proportion of normal samples mistakenly identified as high water demand (or low water demand)) of the samples corresponding to each candidate water demand threshold, thereby obtaining a series of (FPR, TPR) points, based on the true positive rate and false positive rate associated with all candidate water demand thresholds. ROC curves are plotted for all (FPR, TPR) points, and the Youden index corresponding to each candidate water demand threshold is calculated. The Youden index is the difference between the false positive rate and the true positive rate corresponding to the candidate water demand threshold. The candidate water demand threshold corresponding to the largest Youden index value among all low water demand samples is determined as the first candidate threshold τ1, where the first candidate threshold τ1 represents the lower limit of mild water shortage. The candidate water demand threshold corresponding to the largest Youden index value among all high water demand samples is determined as the second candidate threshold τ2, where the second candidate threshold τ2 represents the upper limit of severe water shortage. A target first threshold is selected from all the candidate first thresholds τ1, and a target second threshold is selected from all the candidate second thresholds τ2. The sum of the Youden index corresponding to the target first threshold and the Youden index corresponding to the target second threshold (i.e., the joint Youden index) is the largest among the sums of any candidate first threshold and any candidate second threshold. The target second threshold is greater than the target first threshold. That is, the entire (τ1,τ2) grid is traversed to find the combination with the largest joint Youden index (i.e., the target first threshold and the target second threshold). The target first threshold and the target second threshold are used as the classification interval for determining whether there is an irrigation demand.

[0053] Specifically, the method for determining whether the target vegetation has an irrigation requirement in this embodiment includes: after determining the graded interval for judging whether there is an irrigation requirement, when the target water demand is detected to be greater than the target second threshold, it is determined that the target vegetation has an irrigation requirement and the degree of requirement is high water demand (immediate irrigation is required); when the target water demand is less than the target first threshold, it is determined that the target vegetation has no irrigation requirement; when the target water demand is greater than or equal to the target first threshold and less than the target second threshold, it is determined that the target vegetation has an irrigation requirement and the degree of requirement is medium water demand (moderate irrigation is sufficient).

[0054] Specifically, in this embodiment, all target water requirements are divided into low water requirement samples, medium water requirement samples, and high water requirement samples according to the following method: Low water requirement samples: vegetation grows well, with no obvious signs of water shortage, NDVI change rate ≥ 0, and vegetation grows healthily; Medium water requirement samples: vegetation is slightly short of water, growth is slightly affected, NDVI change rate is close to 0, or vegetation is slightly under water stress; High water requirement samples: vegetation is significantly short of water or growth is inhibited, NDVI change rate < 0, or vegetation is severely under water stress.

[0055] It should be noted that, in this embodiment, the target irrigation water volume is calculated using the following formula: ; in, for, For crop coefficients, Based on the baseline evaporation rate, To predict precipitation, The irrigated area is defined as follows: In this embodiment, when the target water demand is determined to be within the mild water shortage range (i.e., the target water demand is greater than or equal to the first target threshold and less than the second target threshold), supplemental irrigation is achieved by adjusting the sprinkler irrigation duration or flow rate. When the target water demand is determined to be within the severe water shortage range (i.e., the target water demand is greater than or equal to the second target threshold), the system employs a deep irrigation strategy to rapidly restore root zone water content. Simultaneously, the system records the soil moisture recovery curve and vegetation response after each irrigation for model feedback correction.

[0056] In addition, in some embodiments, after generating the first irrigation early warning information or the second irrigation early warning information in steps S67 and S68, the vegetation water demand prediction method based on multidimensional data provided in this application embodiment also includes, but is not limited to, the following steps: Step S71: Periodically acquire ecological environment indicators for the target vegetation after irrigation with the target irrigation water volume, and calculate the actual water demand based on the ecological environment indicators. Step S72: The difference between the actual water demand and the target water demand predicted before irrigation is determined as the feedback error. Step S73: Adjust the weight matrix of the prediction calculation layer in the water demand prediction model based on the feedback error to obtain a new water demand prediction model. Step S74: Retrain a new water demand prediction model based on the training set.

[0057] It is understood that this embodiment periodically acquires ecological environment indicators for the target vegetation after irrigation based on the target irrigation volume, and calculates the actual water demand of the target vegetation after irrigation based on the ecological environment indicators. The difference between the actual water demand and the target water demand predicted before irrigation is determined as the feedback error. Based on the feedback error, the weight matrix of the prediction calculation layer in the water demand prediction model is adjusted to obtain a new water demand prediction model. That is to say, this embodiment uses the feedback error calculated after irrigation to correct the model, which enables the model to continuously learn and adjust during application, thereby making the target water demand output by the model more and more accurate and adaptable to actual environmental changes.

[0058] It should be noted that, in this embodiment, the weight matrix of the prediction calculation layer in the water demand prediction model adjusted based on feedback error is calculated according to the following formula: ; in, For learning rate, This is the adjusted weight matrix for the prediction calculation layer. This is the weight matrix of the prediction calculation layer before adjustment. This is the feedback error.

[0059] like Figure 2 As shown, Figure 2 This is a structural diagram of a control device provided in one embodiment of this application. The present invention also provides a control device 200, comprising: The processor 210 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 220 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 220 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 220 and called by the processor 210 to execute the vegetation water demand prediction method based on multidimensional data according to the embodiments of this application. Input / output interface 230 is used to implement information input and output; The communication interface 240 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 250 transmits information between various components of the device (e.g., processor 210, memory 220, input / output interface 230, and communication interface 240); The processor 210, memory 220, input / output interface 230 and communication interface 240 are connected to each other within the device via bus 250.

[0060] In addition, this application also provides an electronic device, including the control device 200 described in the above embodiments.

[0061] In addition, this application embodiment also provides a storage medium, which is a computer-readable storage medium, storing a computer program that, when executed by a processor, implements the above-described vegetation water demand prediction method based on multidimensional data.

[0062] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof. The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate, and may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0063] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0064] The above provides a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of the present invention.

Claims

1. A method for predicting evapotranspiration of vegetation based on multidimensional data, characterized by, The method comprises the following steps: obtaining an ecological environment index corresponding to target vegetation, wherein the ecological environment index at least includes meteorological data, soil data, vegetation coverage VC, normalized difference vegetation index NDVI and desertification risk index DRI of a region where the target vegetation is located; performing time sequence alignment and standardization processing on all the ecological environment indexes to construct a time sequence dataset; inputting the time sequence dataset into a pre-trained water demand prediction model, wherein the water demand prediction model comprises a long short-term memory LSTM network layer and a prediction calculation layer; the LSTM network layer performs feature extraction processing on the time sequence dataset to obtain a hidden state, wherein the hidden state can represent a non-linear mapping relationship most strongly related to the water demand of the target vegetation extracted from the time sequence dataset; the prediction calculation layer calculates the target water demand corresponding to the target vegetation through a non-linear fusion algorithm and the hidden state.

2. The method for predicting evapotranspiration of vegetation based on multi-dimensional data according to claim 1, characterized in that, The water demand prediction model is trained according to the following steps: inputting a preset training set into an initial model to output a first water demand; calculating a loss value based on the first water demand, a corresponding real water demand and a preset loss function; when the loss value is greater than or equal to a preset loss threshold, adjusting model parameters of the initial model using an Adam optimizer, retraining the model after adjusting the model parameters, and recalculating a new loss value until the new loss value is less than the loss threshold, and determining the model corresponding to the loss value less than the loss threshold as the water demand prediction model.

3. The method for predicting evapotranspiration of vegetation based on multi-dimensional data according to claim 2, characterized in that, The time sequence dataset comprises a plurality of feature sequences, and one feature sequence comprises a plurality of dimensions of the ecological environment index obtained at the same time. The loss value is calculated based on the first water demand, the corresponding real water demand and the preset loss function, and is obtained according to the following formula: ； wherein, L is the loss value, N is the number of feature sequences, is the first water demand output by the initial model after input of the i th feature sequence, is the corresponding real water demand.

4. The method for predicting evapotranspiration of vegetation based on multi-dimensional data according to claim 1, wherein, The target water demand comprises a plurality of second water demands of consecutive time periods. After the prediction calculation layer calculates the target water demand corresponding to the target vegetation through the non-linear fusion algorithm and the hidden state, the method further comprises: obtaining the target water demand in a preset time period, and dividing all the target water demands into low water demand samples, medium water demand samples and high water demand samples; selecting a plurality of candidate water demand thresholds, calculating the true positive rate and the false positive rate of the samples corresponding to each candidate water demand threshold, and drawing an ROC curve based on the true positive rate and the false positive rate associated with all the candidate water demand thresholds; calculating the Youden index corresponding to each candidate water demand threshold, wherein the Youden index is the difference between the false positive rate and the true positive rate corresponding to the candidate water demand threshold; determining the candidate first threshold corresponding to the Youden index with the largest value in all the low water demand samples as the candidate first threshold, wherein the candidate first threshold represents the lower limit of light water shortage; determining the candidate second threshold corresponding to the Youden index with the largest value in all the high water demand samples as the candidate second threshold, wherein the candidate second threshold represents the upper limit of high water shortage. select a target first threshold value from all the candidate first threshold values and a target second threshold value from all the candidate second threshold values, wherein a sum of a Youden index corresponding to the target first threshold value and a Youden index corresponding to the target second threshold value is maximum in a sum of any one of the candidate first threshold values and any one of the candidate second threshold values, and the target second threshold value is greater than the target first threshold value; when the target water demand is greater than or equal to the target second threshold value, calculate a first target irrigation water amount, and generate first irrigation warning information based on the target water demand and the corresponding first target irrigation water amount; when the target water demand is greater than or equal to the target first threshold value and less than the target second threshold value, calculate a second target irrigation water amount, and generate second irrigation warning information based on the target water demand and the corresponding second target irrigation water amount.

5. The method for predicting evapotranspiration of vegetation based on multi-dimensional data according to claim 4, characterized in that, After generating the irrigation warning information based on the target water demand and the corresponding target irrigation water amount, the method further comprises: periodically obtaining an ecological environment index of the target vegetation after irrigation based on the target irrigation water amount, and calculating an actual water demand based on the ecological environment index; determining a feedback error as a difference between the actual water demand and the target water demand predicted before irrigation; adjusting a weight matrix of a prediction calculation layer in the water demand prediction model based on the feedback error to obtain a new water demand prediction model; retraining the new water demand prediction model based on a training set.

6. The method for predicting evapotranspiration of vegetation based on multi-dimensional data according to claim 1, wherein, The LSTM network layer performs feature extraction processing on the time series data set to obtain a hidden state, which is calculated according to the following formula: ； in, For the output parameters of the forget gate, The output parameters of the input gate, For the LSTM network layer in t The memory unit of a moment for t The memory representation corresponding to the time-series dataset at the specified time point. For the LSTM network layer in t The memory unit at time -1 for t The implicit state at time. for t The implicit state at time -1 The output parameters of the output gate, For the time series dataset, It is the Sigmoid activation function. It is the hyperbolic tangent function. , , and This is the weight matrix. , , and This is a bias term.

7. The method for predicting evapotranspiration of vegetation based on multi-dimensional data according to claim 1, wherein, The prediction calculation layer calculates the target water demand of the target vegetation by a nonlinear fusion algorithm and the hidden state, which is calculated according to the following formula: ； wherein, is t the target water demand predicted at the moment, is a weight matrix, is a bias term.

8. A control device characterized by comprising: The control device comprises at least one control processor and a memory connected in communication with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to perform the vegetation water demand prediction method based on multi-dimensional data according to any one of claims 1 to 7.

9. An electronic device, comprising: The control device of claim 8.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions for causing a computer to execute the vegetation water demand prediction method based on multi-dimensional data according to any one of claims 1 to 7.

Images