A PINN neural network-based evapotranspiration simulation method and system

Abstract

This invention proposes a PINN neural network-based evapotranspiration simulation method and system, belonging to the technical field of evapotranspiration calculation and simulation. The method includes: acquiring multi-source meteorological data and high-resolution remote sensing image data for the study area and constructing an input dataset; standardizing the input dataset to generate a remote sensing image raster and a meteorological data raster, respectively; inputting the raster into a trained PINN model, which uses multi-layer convolution to extract features from the remote sensing image raster and the meteorological data raster to generate a feature map representing the evapotranspiration process; embedding the surface energy balance equation into a loss function to implement physical constraints on the model; combining the extracted features with the physical constraint results to generate a spatiotemporal distribution map of evapotranspiration. The spatiotemporal distribution map is visualized using GIS technology to obtain a map of evapotranspiration intensity and a map of its changing trends for the study area. This method addresses the lack of physical consistency in traditional methods and enhances their universality.

Description

Technical Field

[0001] The present invention belongs to the technical field of evapotranspiration calculation simulation, and in particular relates to an evapotranspiration simulation method and system based on a PINN neural network. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Evapotranspiration (ET) is a crucial component of Earth's surface energy and water cycles, significantly impacting agricultural irrigation, ecological balance, and climate change. Traditional ET estimation methods rely primarily on empirical formulas (such as the FAO-56 Penman-Monteith equation) or physical models (such as the SEBAL surface energy balance model). These methods, which utilize meteorological observations, remote sensing imagery, and surface characteristic parameters, can effectively account for the physical mechanisms of ET. However, these methods are highly dependent on data quality and completeness; missing or inaccurate data can significantly reduce model accuracy. Furthermore, traditional models suffer from complex parameterization processes and low computational efficiency, making them inadequate for dynamic monitoring at the regional scale.

[0004] With the development of artificial intelligence and remote sensing technology, deep learning, particularly convolutional neural networks (CNNs), has been widely used in evapotranspiration simulation research. By extracting high-order features from input data, CNNs can capture complex spatiotemporal relationships from remote sensing imagery and meteorological data, significantly reducing reliance on domain expertise and improving computational efficiency. However, most deep learning models are data-driven and lack the constraints of physical mechanisms, which limits their generalizability in data-scarce scenarios. Furthermore, these models are often viewed as "black boxes," making it difficult to explain the relationship between their predictions and the physical process of evapotranspiration. This poses new challenges to the mechanistic study of evapotranspiration. Summary of the Invention

[0005] To overcome the shortcomings of the aforementioned existing technologies, the present invention provides a PINN neural network-based evapotranspiration simulation method and system. Using the SEBAL surface energy balance equation as a physical constraint, this method combines the nonlinear feature extraction capabilities of CNNs with the physical interpretability of physics-informed neural networks (PINNs) to achieve efficient and accurate evapotranspiration simulation. PINNs integrate physical constraints (such as energy conservation and water balance) into the loss function, achieving a fusion of data-driven and physical mechanisms.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] The first aspect of the present invention provides a PINN neural network-based evapotranspiration simulation method;

[0008] A PINN neural network-based evapotranspiration simulation method, comprising:

[0009] Obtain multi-source meteorological data and high-resolution remote sensing image data of the study area and construct input datasets;

[0010] Standardize the meteorological data and remote sensing image data in the input dataset to generate remote sensing image grids and meteorological data grids of the study area respectively;

[0011] The remote sensing image grid and meteorological data grid are input into the trained PINN model, and the dynamic changes of vegetation index are used to correct and refine the evapotranspiration simulation results, and the spatiotemporal distribution map of evapotranspiration in the study area is output;

[0012] The spatiotemporal distribution of evapotranspiration is visualized based on GIS technology to obtain the evapotranspiration intensity map and change trend map of the study area.

[0013] Among them, the PINN model uses multi-layer convolution to extract features from remote sensing image grids and meteorological data grids to generate feature maps representing the evapotranspiration process;

[0014] The surface energy balance equation is embedded into the loss function to implement physical constraints on the model;

[0015] The extracted features are combined with the physical constraint results to generate the spatiotemporal distribution map of evapotranspiration.

[0016] As a further technical solution, the multi-source meteorological data includes temperature, precipitation, relative humidity, wind speed and radiation data in the study area.

[0017] As a further technical solution, the process of standardizing the meteorological data and remote sensing image data in the input data set to generate remote sensing image rasters and meteorological data rasters of the study area includes: unifying the data formats of meteorological data and remote sensing image data, eliminating outliers, projecting, spatially resampling, time series interpolation and filling missing values; and performing atmospheric correction and radiation correction on the remote sensing image data; and finally generating remote sensing image rasters and meteorological data rasters of the study area that are consistent in time and space.

[0018] As a further technical solution, during the training process of the PINN model, a supervised learning method is used to use the observed evapotranspiration data as labels and compare them with the model prediction results;

[0019] Improve model performance using data augmentation, learning rate adjustment, and batch normalization;

[0020] The mean square error and coefficient of determination were used as evaluation indicators of model performance to quantify the accuracy of the simulation results.

[0021] As a further technical solution, the loss function is:

[0022] ;

[0023] in, Represents the physical constraint loss, which is used to constrain the physical consistency of evapotranspiration simulation; Represents the error of the observed data, which is used to minimize the difference between the model prediction and the true data.

[0024] As a further technical solution, the process of using the dynamic changes of the vegetation index to correct and refine the evapotranspiration simulation results and generate a spatiotemporal distribution map of evapotranspiration includes:

[0025] The evapotranspiration results were coupled with the dynamic vegetation index to analyze the impact of vegetation type and dynamic changes on evapotranspiration;

[0026] Through regional zoning simulation, a detailed estimation of evapotranspiration characteristics under different vegetation cover types can be achieved.

[0027] As a further technical solution, the process of visualizing the spatiotemporal distribution map of evapotranspiration based on GIS technology and obtaining the evapotranspiration intensity map and change trend map of the study area is as follows:

[0028] Based on the generated spatiotemporal distribution maps of evapotranspiration, the interannual and seasonal trends of evapotranspiration were analyzed, and the driving effects of meteorological conditions and vegetation changes on evapotranspiration were evaluated.

[0029] The results were visualized using GIS technology, including regional evapotranspiration intensity maps and trend maps. The formula for evapotranspiration intensity change is:

[0030] ;

[0031] Where, is the difference in evapotranspiration before and after the change; is the total area of ​​the study area; is the time span.

[0032] A second aspect of the present invention provides an evapotranspiration simulation system based on a PINN neural network.

[0033] An evapotranspiration simulation system based on a PINN neural network, comprising:

[0034] The input dataset acquisition module is configured to: acquire multi-source meteorological data and high-resolution remote sensing image data of the study area and construct the input dataset;

[0035] The data grid generation module is configured to: perform standardization processing on the meteorological data and remote sensing image data in the input data set to generate a remote sensing image grid and a meteorological data grid of the study area respectively;

[0036] The spatiotemporal distribution map acquisition module is configured to: input the remote sensing image grid and the meteorological data grid into the trained PINN model, use the dynamic changes of the vegetation index to correct and refine the evapotranspiration simulation results, and output the spatiotemporal distribution map of the evapotranspiration in the study area;

[0037] The visualization display module is configured to: visualize the spatiotemporal distribution map of evapotranspiration based on GIS technology, and obtain the evapotranspiration intensity map and change trend map of the study area.

[0038] Among them, the PINN model uses multi-layer convolution to extract features from remote sensing image grids and meteorological data grids to generate feature maps representing the evapotranspiration process;

[0039] The surface energy balance equation is embedded into the loss function to implement physical constraints on the model;

[0040] The extracted features are combined with the physical constraint results to generate the spatiotemporal distribution map of evapotranspiration.

[0041] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the evaporation simulation method based on a PINN neural network as described in the first aspect of the present invention.

[0042] A fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the evaporation simulation method based on a PINN neural network as described in the first aspect of the present invention are implemented.

[0043] One or more of the above technical solutions have the following beneficial effects:

[0044] (1) This paper combines a convolutional neural network (CNN) with a physical information embedding network (PINN), leveraging the nonlinear modeling capabilities of data-driven methods while embedding physical constraints, thereby significantly improving the accuracy and interpretability of evapotranspiration simulations. The CNN feature extraction module automatically captures multi-scale spatial features from meteorological data and remote sensing imagery, and the PINN introduces physical constraints such as surface energy balance, addressing the lack of physical consistency in traditional methods.

[0045] (2) Through automated data preprocessing, model training, and evapotranspiration estimation, an integrated evapotranspiration simulation process was achieved, significantly reducing reliance on expert knowledge and manual intervention and improving simulation efficiency. In addition, by combining regional zoning with dynamic vegetation indices (such as NDVI), the impact of different vegetation types and dynamic changes on evapotranspiration can be reflected, further enhancing the applicability of the model.

[0046] (3) Combining deep learning technology with GIS analysis can generate evapotranspiration maps with precise spatial distribution and clear physical meaning, supporting temporal and spatial trend analysis and visualization of results. This integrated processing not only improves the accuracy and credibility of prediction results, but also provides efficient scientific support for water resources management, ecological and environmental protection, and related fields.

[0047] Advantages of additional aspects of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0049] Figure 1 This is a flow chart of the method of the first embodiment.

[0050] Figure 2 This is a system structure diagram of the second embodiment. DETAILED DESCRIPTION

[0051] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0052] It should be noted that the terms used herein are for describing particular embodiments only and are not intended to limit the exemplary embodiments according to the present invention.

[0053] In the absence of conflict, the embodiments of the present invention and the features thereof may be combined with each other.

[0054] This paper proposes a PINN neural network-based evapotranspiration simulation method and system. This method uses a neural network to model the surface energy balance equation and, combined with actual observation data, accurately simulates the evapotranspiration process. This improves simulation accuracy and computational efficiency, effectively addressing the complex parameterization process and low computational efficiency of traditional models, which make them difficult to meet the needs of dynamic monitoring at the regional scale. The following describes the present invention in detail with reference to specific examples.

[0055] Example 1

[0056] This embodiment discloses a PINN neural network-based evapotranspiration simulation method;

[0057] like Figure 1 As shown, a PINN neural network-based evapotranspiration simulation method includes:

[0058] Step S1, obtaining multi-source meteorological data and high-resolution remote sensing image data of the study area and constructing an input dataset;

[0059] Daily meteorological data for the study area, including temperature, precipitation, relative humidity, wind speed, and radiation, are collected. High-resolution remote sensing imagery (such as NDVI and LAI) is acquired to dynamically monitor changes in vegetation cover. A long-term, multi-scale input dataset is constructed using meteorological observation stations, meteorological reanalysis data, and remote sensing imagery to ensure that the spatial and temporal resolutions meet the requirements for evapotranspiration simulation. Furthermore, auxiliary geographic data (such as land use type maps and soil property data) are obtained to provide environmental variables for regional evapotranspiration simulation.

[0060] Step S2, standardizing the meteorological data and remote sensing image data in the input data set to generate a remote sensing image grid and a meteorological data grid of the study area respectively;

[0061] Standardization of meteorological and remote sensing image data involves standardizing data formats, removing outliers, performing projection transformations, spatial resampling, time series interpolation, and filling in missing values. To ensure data consistency and accuracy, atmospheric and radiation corrections are performed on remote sensing image data, and a spatially and temporally consistent remote sensing image grid and meteorological data grid are generated for the study area. Obtaining these grids facilitates efficient subsequent spatial analysis and modeling. Furthermore, the spatial resolution of the remote sensing image grid and meteorological data grid is standardized to 500 meters, and the temporal resolution is adjusted to daily. Data outliers are removed, and interpolation is performed to fill in missing values, further ensuring data accuracy.

[0062] Step S3: input the remote sensing image grid and the meteorological data grid into the trained PINN model, use the dynamic changes of the vegetation index to correct and refine the evapotranspiration simulation results, and output the spatiotemporal distribution map of evapotranspiration in the study area;

[0063] During PINN model training, 1 / 10 of the image pairs in the target area image library were randomly selected as samples (if there were more than 1000 pairs, 1000 pairs were selected). Evapotranspiration observations obtained using the eddy covariance method were used as labels to generate training data. The data was randomly divided into a training set and a validation set with a ratio of 9:1.

[0064] The key variables of surface energy balance are extracted and embedded as physical constraints; the key variables of surface energy balance include net radiation flux (Rn), sensible heat flux (H), latent heat flux (LE) and soil heat flux (G).

[0065] Establish energy conservation constraints:

[0066] ;

[0067] Physical constraints serve as auxiliary training objectives to enhance the physical consistency of model predictions.

[0068] The trained PINN model then extracts features from the input data through multi-layer convolution, generating a feature map representing the evapotranspiration process. During this feature extraction process, a two-branch network with shared weights is used to extract deep features from the dual-temporal input data (meteorological and remote sensing).

[0069] In step S31, a dual-branch network with shared weights is used to extract deep features from meteorological data and remote sensing imagery data, respectively. The shared weights indicate that the convolution kernel parameters of the two paths in the dual-branch network are identical, indicating that the two paths have the same feature extraction capabilities and are used to capture the similarities between the two types of input data.

[0070] In the process of extracting bi-temporal input data, branch 1 is used to process meteorological data (such as temperature, wind speed, and other rasterized inputs), and branch 2 is used to process remote sensing image data (such as NDVI and LAI).

[0071] The input of each branch i=1,2 can be represented as a 3D tensor:

[0072] ;

[0073] in and is the spatial dimension, is the number of channels.

[0074] During feature extraction, for the input tensor , the convolution layer applies K convolution kernels (weight sharing), each convolution kernel size is k×k×C, to generate the output feature map :

[0075] ;

[0076] in For the l+1 The output feature map of the layer; is the weight of the j-th convolution kernel; is the bias term; is the activation function.

[0077] A nonlinear activation function is applied after convolution to introduce nonlinear expression capabilities:

[0078] ;

[0079] Where, is a nonlinear activation function, is the maximum value function.

[0080] Subsequently, batch normalization is performed on the output feature map to stabilize the training. The formula is as follows:

[0081] ;

[0082] in are the mean and variance of the current batch; is the normalized result, F is the feature (or activation value) of the current input, and is the output obtained by the convolution operation; is a very small constant used to prevent division by zero when computing the standard deviation.

[0083] The spatial dimension of the feature map is reduced layer by layer through downsampling, and the number of channels is expanded at the same time. In this embodiment, after 4 downsamplings, the feature map size is reduced to 1 / 8 of the original size, and the number of channels is expanded to 512.

[0084] In step S32, the extracted features are aggregated using a pyramid pooling module (PPM) to enhance the ability to perceive global information. First, the feature map is divided into pooling regions of varying sizes (e.g., 1×1, 2×2, 4×4). The global average of each pooling region is calculated to generate a fixed-length vector. These multi-scale features are then concatenated with the original features to enhance the ability to express context.

[0085] ;

[0086] in, is the feature map after splicing; Contact represents the splicing function; 、 、 They are the feature maps obtained by maximum pooling; is the original feature map.

[0087] The surface energy balance equation (such as the SEBAL model) is embedded in the loss function to ensure that the model training conforms to the physical laws of evapotranspiration. The loss function is defined as:

[0088] ;

[0089] in, Represents the physical constraint loss, which is used to constrain the physical consistency of evapotranspiration simulation; Represents the error of the observed data, which is used to minimize the difference between the model prediction and the true data.

[0090] The extracted features are combined with the physical constraint results to generate a spatiotemporal distribution map of evapotranspiration for the study area. This map includes total evapotranspiration as well as components such as canopy evaporation, vegetation transpiration, and soil evaporation. Furthermore, the process of generating the spatiotemporal distribution map involves coupling the evapotranspiration results with dynamic vegetation indices (such as NDVI). This allows the dynamic changes in the vegetation index to further modify or refine the evapotranspiration simulation results, reflecting the impact of vegetation cover on the evapotranspiration process.

[0091] Vegetation indices (such as NDVI) and and Highly correlated, and Low correlation with NDVI, among which, Indicates vegetation transpiration; represents canopy evaporation; Indicates soil evaporation. Therefore, NDVI is used to correct and The simulation results are shown in Figure 2.

[0092] First, the vegetation coverage changes were dynamically monitored by NDVI, and the vegetation coverage rate was calculated. :

[0093] ;

[0094] is the maximum value of NDVI in the study area (high density vegetation); is the minimum NDVI value (bare soil) in the study area.

[0095] based on The function corrects each component, for canopy evaporation:

[0096] ;

[0097] in is the canopy evaporation coefficient, is the total evapotranspiration.

[0098] For vegetation transpiration:

[0099] ;

[0100] in is the canopy evaporation coefficient.

[0101] For soil evaporation:

[0102] ;

[0103] In the dynamic coupling correction, the temporal and spatial distribution of evapotranspiration is corrected by combining the daily time series evapotranspiration results with the dynamic NDVI. The steps include:

[0104] Calculated based on daily NDVI ; Calculate canopy evaporation, vegetation transpiration, and soil evaporation components for each grid pixel; Smooth NDVI dynamic changes through time series interpolation or filtering (such as Savitzky-Golay filter) to reduce observation noise.

[0105] Assess the contribution of vegetation types and changes. Compare evapotranspiration characteristics of different vegetation cover types to quantify the impact of climate change on the regional water cycle. Divide regions by land use type (forest, grassland, shrub, etc.) to refine evapotranspiration estimates.

[0106] The PINN model also employs supervised learning, using observed evapotranspiration data (such as eddy covariance data or remote sensing inversion data) as labels for comparison with model predictions. Techniques such as data augmentation, learning rate adjustment, and batch normalization are employed to improve model performance. Model performance is evaluated using metrics such as mean squared error (MSE) and coefficient of determination (R²) to quantify the accuracy of simulation results.

[0107] Step S4: Visualize the spatiotemporal distribution of evapotranspiration based on GIS technology to obtain an evapotranspiration intensity map and a change trend map of the study area.

[0108] Based on the generated spatiotemporal distribution map of evapotranspiration, the interannual and seasonal trends of evapotranspiration are analyzed, and the driving effects of meteorological conditions and vegetation changes on evapotranspiration are evaluated. The results are visualized using GIS technology, including regional evapotranspiration intensity maps and trend maps, to provide support for water resource management and ecological and environmental protection. The formula for evapotranspiration intensity change is:

[0109] ;

[0110] Where, is the difference in evapotranspiration before and after the change; is the total area of ​​the study area; is the time span.

[0111] Example 2

[0112] This embodiment discloses an evapotranspiration simulation system based on a PINN neural network;

[0113] like Figure 2 As shown, an evapotranspiration simulation system based on a PINN neural network includes:

[0114] The input dataset acquisition module is configured to: acquire multi-source meteorological data and high-resolution remote sensing image data of the study area and construct the input dataset;

[0115] The data grid generation module is configured to: perform standardization processing on the meteorological data and remote sensing image data in the input data set to generate a remote sensing image grid and a meteorological data grid of the study area respectively;

[0116] The spatiotemporal distribution map acquisition module is configured to: input the remote sensing image grid and the meteorological data grid into the trained PINN model, use the dynamic changes of the vegetation index to correct and refine the evapotranspiration simulation results, and output the spatiotemporal distribution map of the evapotranspiration in the study area;

[0117] The visualization display module is configured to: visualize the spatiotemporal distribution map of evapotranspiration based on GIS technology, and obtain the evapotranspiration intensity map and change trend map of the study area.

[0118] Among them, the PINN model uses multi-layer convolution to extract features from remote sensing image grids and meteorological data grids to generate feature maps representing the evapotranspiration process;

[0119] The surface energy balance equation is embedded into the loss function to implement physical constraints on the model;

[0120] The extracted features are combined with the physical constraint results to generate the spatiotemporal distribution map of evapotranspiration.

[0121] Example 3

[0122] The purpose of this embodiment is to provide a computer-readable storage medium.

[0123] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the evaporation simulation method based on a PINN neural network as described in Example 1.

[0124] Example 4

[0125] The purpose of this embodiment is to provide an electronic device.

[0126] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps of the evaporation simulation method based on a PINN neural network as described in Example 1 are implemented.

[0127] The steps involved in the apparatuses of Examples 2, 3, and 4 above correspond to those of Method Example 1. For detailed implementations, please refer to the relevant description of Example 1. The term "computer-readable storage medium" should be understood to mean a single medium or multiple media containing one or more instruction sets; it should also be understood to include any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and causing the processor to perform any method of the present invention.

[0128] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0129] Although the above describes the specific embodiments of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without any creative work are still within the scope of protection of the present invention.

Claims

1. A PINN neural network-based evapotranspiration simulation method, characterized in that: include: Obtain multi-source meteorological data and high-resolution remote sensing image data of the study area and construct input datasets; Standardize the meteorological data and remote sensing image data in the input dataset to generate remote sensing image grids and meteorological data grids of the study area respectively; The remote sensing image grid and meteorological data grid are input into the trained PINN model, and the dynamic changes of vegetation index are used to correct and refine the evapotranspiration simulation results, and the spatiotemporal distribution map of evapotranspiration in the study area is output; The correction and refinement specifically include: coupling the evapotranspiration results with the dynamic vegetation index to analyze the impact of vegetation type and dynamic changes on evapotranspiration; through regional zoning simulation, achieving a refined estimation of evapotranspiration characteristics under different vegetation cover types; Based on GIS technology, the temporal and spatial distribution of evapotranspiration is visualized to obtain the evapotranspiration intensity map and change trend map of the study area; Among them, the PINN model uses multi-layer convolution to extract features from remote sensing image grids and meteorological data grids to generate feature maps representing the evapotranspiration process; During the training of the PINN model, the surface energy balance equation is embedded in the loss function to implement the physical constraints of the model. Specifically, the key variables of the surface energy balance are extracted and embedded as physical constraints. Among them, the key variables of the surface energy balance include: net radiation flux , sensible heat flux H, latent heat flux LE and soil heat flux G; establish energy conservation constraints: Physical constraints serve as auxiliary training objectives to enhance the physical consistency of model predictions. The trained PINN model extracts features from the input data through multi-layer convolution to generate a feature map representing the evapotranspiration process. During the feature extraction process, a dual-branch network with shared weights is used to extract deep features of the dual-phase input data. The extracted features are combined with the physical constraint results to generate the spatiotemporal distribution map of evapotranspiration.

2. The evapotranspiration simulation method based on a PINN neural network according to claim 1, characterized in that: The multi-source meteorological data includes temperature, precipitation, relative humidity, wind speed and radiation data in the study area.

3. The evapotranspiration simulation method based on a PINN neural network according to claim 1, characterized in that: The process of standardizing the meteorological data and remote sensing image data in the input data set to generate the remote sensing image raster and meteorological data raster of the study area includes: unifying the data format of the meteorological data and the remote sensing image data, eliminating outliers, performing projection transformation, spatial resampling, time series interpolation and missing value filling; and performing atmospheric correction and radiation correction on the remote sensing image data; and finally generating the remote sensing image raster and meteorological data raster of the study area that are consistent in time and space.

4. The evapotranspiration simulation method based on a PINN neural network according to claim 1, characterized in that: During the training process of the PINN model, a supervised learning method is used to compare the observed evapotranspiration data with the model prediction results using them as labels; Improve model performance using data augmentation, learning rate adjustment, and batch normalization; The mean square error and coefficient of determination were used as evaluation indicators of model performance to quantify the accuracy of the simulation results.

5. The evapotranspiration simulation method based on a PINN neural network according to claim 1, characterized in that: The loss function is: ； in, Represents the physical constraint loss, which is used to constrain the physical consistency of evapotranspiration simulation; Represents the error of the observed data, which is used to minimize the difference between the model prediction and the true data.

6. The evapotranspiration simulation method based on a PINN neural network according to claim 1, characterized in that: The process of visualizing the spatiotemporal distribution of evapotranspiration based on GIS technology and obtaining the evapotranspiration intensity map and change trend map of the study area is as follows: Based on the generated spatiotemporal distribution maps of evapotranspiration, the interannual and seasonal trends of evapotranspiration were analyzed, and the driving effects of meteorological conditions and vegetation changes on evapotranspiration were evaluated. The results were visualized using GIS technology, including regional evapotranspiration intensity maps and trend maps. The formula for evapotranspiration intensity change is: ； Where, is the difference in evapotranspiration before and after the change; is the total area of ​​the study area; is the time span.

7. An evapotranspiration simulation system based on a PINN neural network, characterized by: include: The input dataset acquisition module is configured to: acquire multi-source meteorological data and high-resolution remote sensing image data of the study area and construct the input dataset; The data grid generation module is configured to: perform standardization processing on the meteorological data and remote sensing image data in the input data set to generate a remote sensing image grid and a meteorological data grid of the study area respectively; The spatiotemporal distribution map acquisition module is configured to: input the remote sensing image grid and the meteorological data grid into the trained PINN model, use the dynamic changes of the vegetation index to correct and refine the evapotranspiration simulation results, and output the spatiotemporal distribution map of the evapotranspiration in the study area; The correction and refinement specifically include: coupling the evapotranspiration results with the dynamic vegetation index to analyze the impact of vegetation type and dynamic changes on evapotranspiration; through regional zoning simulation, achieving a refined estimation of evapotranspiration characteristics under different vegetation cover types; The visualization display module is configured to: visualize the spatiotemporal distribution of evapotranspiration based on GIS technology, and obtain the evapotranspiration intensity map and change trend map of the study area; Among them, the PINN model uses multi-layer convolution to extract features from remote sensing image grids and meteorological data grids to generate feature maps representing the evapotranspiration process; During the training of the PINN model, the surface energy balance equation is embedded in the loss function to implement the physical constraints of the model. Specifically, the key variables of the surface energy balance are extracted and embedded as physical constraints. Among them, the key variables of the surface energy balance include: net radiation flux , sensible heat flux H, latent heat flux LE and soil heat flux G; establish energy conservation constraints: Physical constraints serve as auxiliary training objectives to enhance the physical consistency of model predictions. The trained PINN model extracts features from the input data through multi-layer convolution to generate a feature map representing the evapotranspiration process. During the feature extraction process, a dual-branch network with shared weights is used to extract deep features of the dual-phase input data. The extracted features are combined with the physical constraint results to generate the spatiotemporal distribution map of evapotranspiration.

8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the evaporation simulation method based on the PINN neural network as described in any one of claims 1 to 6 are implemented.

9. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the evapotranspiration simulation method based on the PINN neural network are implemented as described in any one of claims 1 to 6.

Images