Greenhouse gas monitoring and management decision method and system based on multi-source fusion perception, digital twin and source analysis

Abstract

The application discloses a farmland greenhouse gas monitoring and management decision method and system based on multi-source fusion perception, digital twinning and source analysis, belongs to the technical field of big data analysis and ecological environment monitoring, and solves the problems of low farmland greenhouse gas emission dynamic monitoring precision and poor farmland management decision availability in the prior art. The application obtains a farmland source analysis model by adopting a CNN-Transformer model and a physical equation for simulating greenhouse gas emission dynamics based on multi-source perception big data of a target farmland area, so that high-precision greenhouse gas emission data is obtained; a farmland management action and gas emission relationship model and a farmland optimization management decision model based on a reinforcement learning algorithm are further constructed, so that a high-availability farmland management strategy is generated. The application improves the farmland greenhouse gas emission dynamic monitoring precision and the availability of farmland management decision, and solves the problems of low farmland greenhouse gas emission dynamic monitoring precision and poor farmland management decision availability.

Description

Technical Field

[0001] This invention belongs to the field of big data analysis and ecological environment monitoring technology, specifically relating to a method and system for monitoring and managing greenhouse gases in farmland based on multi-source fusion sensing, digital twins and source apportionment. Background Technology

[0002] Global climate change is intensifying, and greenhouse gas emissions from agriculture have become a major driver of climate change. With the development of information technology, big data technology can be applied to various fields such as agricultural environmental monitoring, farmland production management, greenhouse gas emission analysis, and precision regulation. By collecting, fusing, analyzing, and mining multi-source, massive, and continuous data, big data technology helps improve agricultural environmental awareness, enhance process prediction accuracy, and provide more targeted decision support for farmland management. However, traditional methods for monitoring greenhouse gases in farmland typically rely on observations at fixed locations and empirical models. These methods generally suffer from drawbacks such as limited data sources, limited spatial coverage, low sampling frequency, delayed dynamic response, and insufficient monitoring accuracy, making it difficult to provide a real-time, comprehensive, and continuous depiction of changes in greenhouse gas emissions in the farmland environment.

[0003] In existing farmland management decision-making technologies, on the one hand, there is a lack of high-precision and continuous dynamic monitoring data on farmland greenhouse gas emissions, making it difficult to accurately reflect the actual emission status and changing patterns of farmland; on the other hand, there is a lack of big data analysis and intelligent closed-loop control mechanisms based on monitoring data, making it impossible to adjust and automatically optimize farmland management measures such as irrigation, fertilization, and tillage in real time based on farmland greenhouse gas emissions, thus causing a discrepancy between farmland management measures and actual farmland needs, resulting in low usability.

[0004] In summary, it is necessary to develop a method for monitoring greenhouse gases in farmland and a corresponding farmland management decision-making method to improve the accuracy of dynamic monitoring of greenhouse gas emissions from farmland and the usability of farmland management decisions. Summary of the Invention

[0005] In view of the above analysis, the embodiments of the present invention aim to provide a method and system for monitoring and managing greenhouse gas emissions in farmland based on multi-source fusion sensing, digital twin and source apportionment, so as to solve one or more of the problems of low dynamic monitoring accuracy of greenhouse gas emissions in farmland and poor availability of farmland management decisions in the prior art.

[0006] The objective of this invention is achieved as follows: A method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment includes: Acquire characteristic big data of the target farmland area, wherein the characteristic big data includes at least two of the following: greenhouse gas emission data, meteorological parameters, soil physicochemical properties, crop canopy physiological parameters, spatial heterogeneity data, and farmland environmental data; The feature big data is synchronized and fused to obtain multi-source sensing big data of the target farmland area; A convolutional neural network is used to extract the spatial features of the multi-source sensing big data, and a Transformer model and attention mechanism are used to collect the temporal features corresponding to the spatial features. Based on the temporal features, the predicted greenhouse gas emission flux of the target farmland area is obtained. Based on the multi-source sensing big data, a physical equation simulating the dynamics of greenhouse gas emissions is constructed. The physical equation is embedded into a physical information neural network model, a total loss function is constructed, and the physical information neural network model is iterated according to the total loss function to obtain a farmland source apportionment model. The spatiotemporal distribution data of greenhouse gas emissions in the target farmland area are obtained based on the predicted greenhouse gas emission flux and the farmland source apportionment model.

[0007] Furthermore, the greenhouse gas emission data, meteorological parameters, and soil physicochemical properties are acquired by a ground-based Internet of Things sensor array, the crop canopy physiological parameters and spatial heterogeneity data are acquired by drones, and the farmland environmental data are acquired by satellite remote sensing equipment.

[0008] Furthermore, the synchronization and fusion of the feature big data to obtain multi-source sensing big data for the target farmland area specifically includes: The feature big data is preprocessed, and the isolated forest algorithm is used to identify and remove outliers in the preprocessed feature big data to obtain the first feature set; Spatial interpolation algorithms are used to fill in the spatial missing data in the first feature set to obtain the second feature set; A long short-term memory network is used to impute the time series data of the second feature set to obtain a third feature set; After applying registration benchmarks and accuracy constraints to the data in the third feature set, a fourth feature set is obtained; Based on the basic accuracy weight, spatiotemporal matching weight, scene representation weight and dynamic stability weight, the fourth feature set is fused to obtain multi-source sensing big data of the target farmland area. The basic accuracy weight is calculated using the following formula: W bas,i = ; In the formula, W bas,i∈(0,1), representing the normalized base accuracy weights, i=1,2,3, corresponding to the three data sources: the ground-based IoT sensor array, the UAV, and the satellite remote sensing equipment, respectively. RRMSE i Let E be the relative root mean square error of the i-th type of data source. cal,i The laboratory calibration relative error of the i-th type of data source; The spatiotemporal matching weights are calculated using the following formula: W st,i = ; In the formula, W st,i ∈(0,1), representing the normalized spatiotemporal matching weights, i=1,2,3 and corresponding to the three types of data sources: the ground-based IoT sensor array, the UAV, and the satellite remote sensing equipment, respectively. T m,i S is the time matching coefficient of the i-th type of data source. m,i The spatial matching degree coefficient of the i-th type of data source; The scene representation weights are calculated using the following formula: W scene,i = ; In the formula, W scene,i ∈(0,1), where MI is the normalized scene representation weight. i K is the mutual information coefficient. stage K is the crop growth period coefficient. event For event-driven coefficients; The dynamic stability weights are calculated using the following formula: W stab,i = ; In the formula, W stab,i ∈(0,1), where Q is the normalized dynamic stable weight. valid,i For the effective data rate of the i-th type of data source, CV i Let be the volatility coefficient of the i-th type of data source, Cloud i The percentage of cloud data sourced from drones or satellite remote sensing equipment.

[0009] Furthermore, the step of extracting spatial features from the multi-source sensing big data using a convolutional neural network, collecting temporal features corresponding to the spatial features using a Transformer model and attention mechanism, and obtaining the predicted greenhouse gas emission flux of the target farmland area based on the temporal features specifically includes: Using a convolutional neural network, the spatial features of the multi-source sensing big data are extracted using the following formula: ; In the formula, For time The extracted spatial features The input features include at least two of the features from the large dataset. For convolution kernel, For convolution bias; Based on the spatial features extracted by the convolutional neural network, temporal modeling is performed using the Transformer model, and the spatial features are linearly transformed using the following formula: ; ; ; In the formula, This is the weight matrix for the science department. The impact of meteorological and crop characteristics on greenhouse gas emissions, For soil-related characteristics, Information for calculating greenhouse gas emissions; The relative importance between each time step is calculated using a sub-attention mechanism, and the attention features are obtained using the following formula: ; In the formula, For the attention feature, The dimension of the key vector; Based on the aforementioned attention features, a multi-head attention mechanism is employed to capture the temporal dependencies of multiple subspaces, and the temporal features are obtained using the following formula: ; In the formula, The time-series feature; Based on the aforementioned time-series characteristics, the predicted greenhouse gas emission flux for the target farmland area is obtained using the following formula: ; In the formula, and The weights and biases of the output layer of the Transformer model. The predicted flux for greenhouse gas emissions.

[0010] Furthermore, the step of constructing a physical equation simulating the dynamics of greenhouse gas emissions based on the multi-source sensing big data, embedding the physical equation into a physical information neural network model, constructing a total loss function, and iterating the physical information neural network model according to the total loss function to obtain a farmland source apportionment model specifically includes: Based on the ADR equations describing changes in gas concentration and gas transport in farmland, the following physical equation for meteorological transport is obtained: ; In the formula, For meteorological conditions and diffusion coefficient, For source terms that include both biological and abiotic sources, For gas concentration, This refers to the wind speed field.

[0011] Using a reaction kinetic model to represent the microbial source term, the following physical equation for soil microbial activity is obtained: ; In the formula, For microbial origin items, and These are the yield coefficients for the nitration and denitrification reaction processes, respectively. and These represent the nitrification and denitrification rates, respectively. Construct the following physical equation for crop growth: ; In the formula, Leaf area index, In order to absorb light and effective radiation, , , These represent the limiting factors for temperature, moisture, and nitrogen, respectively. The physical equations of meteorological transport, soil microbial activity, and crop growth process are embedded into the physical information neural network model to construct the total loss function; The farmland source analysis model is obtained by iterating the physical information neural network model based on the total loss function.

[0012] Furthermore, the step of embedding the physical equations of meteorological transport, soil microbial activity, and crop growth processes into the physical information neural network model to construct the total loss function specifically includes: The physical information neural network model is trained based on the aforementioned meteorological transport physical equations. Given the gas concentration output by the physical information neural network model, the residual loss of the meteorological transport physical equation is obtained as follows: Construct the soil-atmosphere interface boundary conditions as follows: In the formula, The residual loss of the aforementioned meteorological transport physical equation, The gas flux predicted by the aforementioned meteorological transport physics equation. The external normal vector. It is the soil-atmosphere interface; By embedding the physical equation of soil microbial activity into the source terms of the physical information neural network model, the residual loss of the physical equation of soil microbial activity is obtained as follows: In the formula, This represents the residual loss of the physical equation for soil microbial activity; By embedding the constraints of the physical equations governing the crop growth process into the physical information neural network model, the residual loss of the physical equations governing the crop growth process is obtained as follows: In the formula, This represents the residual loss of the physical equations governing the crop growth process. Based on the prior total throughput of the convolutional neural network and the Transformer model, construct the prior consistency loss of the physical information neural network model: ; ; ; in, The prior consistency loss is... Let be the prior total flux. This represents the total output throughput of the physical information neural network model. and These are the biological and non-biological fluxes output by the physical information neural network model, respectively. Based on the residual losses of the meteorological transport physical equation, the soil microbial activity physical equation, the crop growth process physical equation, and the prior consistency loss, the total loss function is obtained as follows: ; Where L is the total loss function, , , , These are the loss weight coefficients corresponding to the residual loss of the meteorological transport physical equation, the residual loss of the soil microbial activity physical equation, the residual loss of the crop growth process physical equation, and the prior consistency loss, respectively.

[0013] A farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution includes: The spatiotemporal distribution data of greenhouse gas emissions in the target farmland area are obtained by the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin and source apportionment as described in the above embodiments of the invention. Based on the spatiotemporal distribution data of greenhouse gas emissions, a model of the relationship between farmland management actions and gas emissions based on graph neural networks is constructed and trained, wherein the model of the relationship between farmland management actions and gas emissions is used to build a virtual decision-making environment; In the virtual decision-making environment, a farmland optimization management decision model based on reinforcement learning algorithm is constructed; Based on the farmland optimization management decision model, a farmland management strategy for the target farmland area is generated.

[0014] Furthermore, the construction of a farmland optimization management decision model based on reinforcement learning algorithms in the virtual decision-making environment specifically includes: The system's state is constructed based on the following formula: ; In the formula, The state of the system. For current farmland management practices, In the virtual decision-making environment, the gas emission flux corresponding to the current farmland management action; The system's actions are based on the following formula: ; In the formula, For the actions of the system, Adjusted farmland management procedures; The reward function is constructed based on the following formula: ; In the formula, For state Take action below value, For learning rate, As a discount factor, For instant rewards.

[0015] A farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twin, and source apportionment includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin, and source apportionment as described in the above-described embodiments of the invention.

[0016] A management decision-making system based on multi-source fusion sensing, digital twin, and source resolution includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the management decision-making method based on multi-source fusion sensing, digital twin, and source resolution described in the above embodiments of the invention.

[0017] A system includes a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin and source apportionment as described in the above embodiments of the invention, or the management decision-making method based on multi-source fusion sensing, digital twin and source apportionment.

[0018] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: a) This invention effectively solves the problems of poor fusion consistency and difficulty in eliminating heterogeneity caused by differences in temporal resolution, spatial resolution, and sampling frequency in existing multi-source data by integrating multi-source heterogeneous data from IoT sensors, drones, and satellite remote sensing, and constructing a source analysis model by combining a CNN-Transformer deep learning model with the dynamic physical equations of greenhouse gas emissions. At the same time, it breaks through the limitations of traditional fixed-point monitoring, such as incomplete spatial coverage, low sampling frequency, and weak dynamic change capture capability, and realizes full-area, high-frequency, continuous real-time monitoring of core greenhouse gas fluxes such as CH4 and N2O in target farmland areas. This significantly improves the spatiotemporal accuracy and data reliability of greenhouse gas monitoring data, and provides accurate basic data support for farmland management decisions. b) This invention constructs a deep learning source analysis model under physical mechanism constraints, abandoning the analysis method of existing technologies that rely too much on empirical models. It can accurately distinguish between biological and non-biological sources of greenhouse gases in farmland, quantitatively analyze the contribution ratio and dynamic impact of different emission sources, different farmland management measures, and different environmental factors on greenhouse gas emissions, and truly restore the spatiotemporal dynamic change characteristics of greenhouse gas emissions in farmland. c) This invention constructs a full-element digital twin of the target farmland area through digital twin technology, realizing real-time mapping, dynamic simulation and extrapolation of all-dimensional elements such as farmland meteorology, soil, crop growth, management measures, and greenhouse gas emissions; it can conduct advance simulation and quantitative evaluation of the greenhouse gas emission effects and crop growth impacts of different farmland management schemes, thereby improving the scientific nature of management decisions and their practical usability in the field. d) This invention is adaptable to farmland scenarios with different climate zones, soil types, and crop planting patterns, and has strong scenario adaptability and large-scale promotion capabilities. It can effectively utilize multi-source sensing big data, and can not only provide various planting entities with directly implementable and usable low-carbon farmland management decision-making solutions, but also effectively promote the digital transformation of agriculture and the realization of the "dual carbon" goals in the agricultural field, and comprehensively enhance the sustainable development capabilities and climate resilience of agricultural production.

[0019] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings. Figure 1 This is a flowchart illustrating the method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment provided in Embodiment 1 of the present invention. Figure 2 This is a flowchart illustrating the farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution provided in Embodiment 2 of the present invention. Figure 3 A schematic diagram of the structure of the farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twin and source apportionment provided in Embodiment 3 of the present invention; Figure 4 A schematic diagram of the structure of the farmland management decision-making system based on multi-source fusion sensing, digital twin and source resolution provided in Embodiment 4 of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] To facilitate understanding of the embodiments of this application, further explanation and description will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application. In the drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0023] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that will be recognized by those skilled in the art.

[0024] Example 1 A specific embodiment of the present invention, such as Figure 1 As shown, a method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment is disclosed, including: S101. Obtain characteristic big data of the target farmland area, wherein the characteristic big data includes at least two of the following: greenhouse gas emission data, meteorological parameters, soil physicochemical properties, crop canopy physiological parameters, spatial heterogeneity data, and farmland environmental data. Specifically, characteristic big data can also include soil temperature and humidity, pH value, nitrogen content, ground temperature, etc. After collecting characteristic big data, the data is transmitted in real time through wireless sensor networks, cellular networks, and satellite networks to facilitate the synchronization and fusion of characteristic big data.

[0025] S102. Synchronize and fuse the feature big data to obtain multi-source sensing big data of the target farmland area; Specifically, after the feature big data is obtained from different acquisition devices, it is formatted and spatially registered. The isolated forest algorithm is used to identify and remove outliers, and spatial interpolation algorithms such as Kriging interpolation or inverse distance weighting are used to fill in spatial missing data. Long Short-Term Memory (LSTM) network is used to fill in time series data, and data with different spatiotemporal resolutions are synchronized and fused to ensure the integrity and consistency of each data source.

[0026] S103. Spatial features of multi-source sensing big data are extracted using convolutional neural networks. Temporal features corresponding to spatial features are collected using Transformer model and attention mechanism. Based on the temporal features, the predicted greenhouse gas emission flux of the target farmland area is obtained. Specifically, based on fused multi-source sensing big data, a convolutional neural network (CNN) and a Transformer model are used to predict farmland gas emissions. This model is able to extract features from the input data and learn the complex nonlinear relationship between gas emissions and environmental factors.

[0027] S104. Based on multi-source sensing big data, construct physical equations to simulate the dynamics of greenhouse gas emissions, embed the physical equations into a physical information neural network model, construct a total loss function, and iterate the physical information neural network model according to the total loss function to obtain a farmland source apportionment model. Specifically, consistency constraints and physical constraints are embedded into the Physical Information Neural Network (PINN) model to obtain the total loss function; minimizing the total loss function improves the prediction accuracy of the farmland source apportionment model.

[0028] S105. Obtain the spatiotemporal distribution data of greenhouse gas emissions in the target farmland area based on the predicted greenhouse gas emission flux and the farmland source apportionment model.

[0029] Specifically, the spatiotemporal distribution data of greenhouse gas emissions are visualized to generate spatiotemporal distribution maps of greenhouse gas emissions in the target farmland area, as well as analytical charts of various sources (biological and non-biological), providing intuitive data support for subsequent farmland management decisions.

[0030] In this embodiment, preferably, in S101, greenhouse gas emission data, meteorological parameters, and soil physicochemical properties are acquired by a ground-based Internet of Things sensor array, crop canopy physiological parameters and spatial heterogeneity data are acquired by a drone, and farmland environmental data are acquired by a satellite remote sensing device. S102. Synchronize and fuse the feature big data to obtain multi-source sensing big data for the target farmland area, specifically including: The feature big data is preprocessed, and the isolated forest algorithm is used to identify and remove outliers in the preprocessed feature big data to obtain the first feature set. Spatial interpolation algorithms are used to fill in the spatially missing data in the first feature set to obtain the second feature set; A long short-term memory network is used to impute the time series data of the second feature set to obtain the third feature set; The fourth feature set is obtained by applying registration benchmarks and accuracy constraints to the data in the third feature set; Based on the basic accuracy weight, spatiotemporal matching weight, scene representation weight and dynamic stability weight, the fourth feature set is fused to obtain multi-source sensing big data of the target farmland area. The basic precision weight is calculated using the following formula: W bas,i = ; In the formula, W bas,i ∈(0,1), representing the normalized base precision weights, i=1,2,3, corresponding to three data sources: ground-based IoT sensor arrays, drones, and satellite remote sensing equipment, respectively. RRMSE i Let E be the relative root mean square error of the i-th type of data source. cal,i The laboratory calibration relative error of the i-th type of data source; The spatiotemporal matching weights are calculated using the following formula: W st,i = ; In the formula, W st,i ∈(0,1), representing the normalized spatiotemporal matching weights, i=1,2,3, corresponding to three types of data sources: ground-based IoT sensor arrays, drones, and satellite remote sensing equipment, respectively. T m,i S is the time matching coefficient of the i-th type of data source. m,i The spatial matching degree coefficient of the i-th type of data source; The scene representation weights are calculated using the following formula: W scene,i = ; In the formula, W scene,i ∈(0,1), where MI is the normalized scene representation weight. i K is the mutual information coefficient. stage K is the crop growth period coefficient. event For event-driven coefficients; The dynamic stability weights are calculated using the following formula: W stab,i = ; In the formula, W stab,i ∈(0,1), where Q is the normalized dynamic stable weight. valid,i For the effective data rate of the i-th type of data source, CV i Let be the volatility coefficient of the i-th type of data source, Cloud i The percentage of cloud data sourced from drones or satellite remote sensing equipment.

[0031] Specifically, preprocessing of feature big data includes geographic coordinate system normalization, time normalization, and unit and field normalization.

[0032] Specifically, the Isolation Forest algorithm is used to identify and remove outliers from the preprocessed feature dataset to obtain the first feature set, which includes: For data acquired by ground-based IoT sensor arrays, the 3σ principle is used to eliminate hardware anomalies. Values ​​exceeding [μ-3σ, μ+3σ] are identified as the first abnormal value obtained from sensor hardware failure. In the formula, μ is the sliding window mean of the sensor within a preset time period, and σ is the standard deviation. For example, the preset time period can be 7 days, 30 days, or other custom values.

[0033] The Pettit mutation test was used to identify time-series mutation points and to identify the differences in mean before and after the mutation point that conformed to the emission pattern. If a mutation point exists and there is no corresponding agricultural management or meteorological phenomenon, the mutation point is regarded as a non-driving mutation value in the second outlier. If a mutation point exists and there is a corresponding agricultural management or meteorological phenomenon, the mutation point is regarded as a driving mutation value in the second outlier. The outlier score of an isolated forest is calculated using the following formula: ; In the formula, For feature big data samples The corresponding anomaly score ranges from (0,1), with scores closer to 1 indicating a higher degree of anomaly. The first correction coefficient corresponds to the first outlier. This is the second correction coefficient corresponding to the second outlier. For feature big data samples Average path length across all isolated trees For sampling scale The baseline value for average path length.

[0034] for It can be considered This is a hardware anomaly correction factor, when When a value is marked as the first outlier, it corresponds to a hardware malfunction in the IoT sensor array. At this point, [the appropriate action is taken]. = ( >1), otherwise take =1.

[0035] for It can be considered For mutation anomaly correction coefficient, when When a value is marked as the second outlier and corresponds to a driverless mutation value identified by the Pettitt mutation test, take... = ( >1), for example, A value of 1.5 can be used to amplify outliers; when If a value is marked as the second outlier and corresponds to a driver mutation value identified by the Pettitt mutation test, then take... = (0 < <1), for example, A value of 0.5 can be used to suppress outliers and avoid misjudgments; otherwise, a value of 0.5 can be used. =1.

[0036] Finally, based on the outlier scores of the isolated forest, outliers in the above samples are identified and removed to obtain the first feature set.

[0037] Specifically, the relative root mean square error (RRMSE) of the i-th type of data source i This can be obtained by prior measurement in farmland using high-precision standard instruments; furthermore, RRMSE i =RMSE i / In the formula, RMSE i Let be the root mean square error of the i-th type of data source. This is the measured average.

[0038] Specifically, the crop growth period coefficient K stage It can assign values ​​to the main emission factors at different growth stages of crops.

[0039] In a specific implementation, the crop growth period can be divided into three stages based on the crop type of the target farmland area: sowing-tillering, jointing-heading, and grain filling-maturity. During the sowing-tillering stage, the main greenhouse gas emission factors are soil temperature, humidity, base fertilizer, and organic matter. Therefore, the sampling impact of the ground-based IoT sensor array should be increased during this period. stage >K of drone data sources stage >K of satellite remote sensing equipment data sources stage During the jointing-heading stage, the main greenhouse gas emission factors are crop canopy (LAI), topdressing, and root activity. Therefore, the sampling impact of the hyperspectral sensors and lidar carried by UAVs should be enhanced during this period. stage >K of the data source for terrestrial IoT sensor arrays stage >K of satellite remote sensing equipment data sources stage During the grain-filling to maturity stage, the main greenhouse gas emission factors are regional hydrothermal activity, crop senescence, and large-scale environmental factors. Therefore, the sampling impact of satellite remote sensing equipment should be increased during this period. stage >K of drone data sources stage >K of the data source for terrestrial IoT sensor arrays stage .

[0040] Specifically, the event-driven coefficient Kevent Values ​​can be assigned to driving events, which include fertilization, irrigation, heavy rainfall, etc.

[0041] In a specific implementation, when in a stable period without driver events, the event-driven coefficient K corresponding to different data source categories is... event All values ​​are assigned to 1; when it is within the first preset time period after fertilization or irrigation, the K value of the ground IoT sensor array data source is... event >K of drone data sources event >K of satellite remote sensing equipment data sources event The first preset time period can be the peak period of greenhouse gas emissions, for example, 7 days; when it is within the second preset time period after heavy rainfall, the K data source of the ground IoT sensor array... event >K of drone data sources event =K of satellite remote sensing equipment data source event And at this time, the K data source of the ground IoT sensor array event K with drone data source event The difference is greater than the difference between the two within the first preset time period after fertilization or irrigation, where the second preset time period can be the peak period of rainfall impact, for example, 3 days.

[0042] Specifically, by constructing a three-dimensional monitoring network composed of ground-based IoT sensor arrays, drones, and satellite remote sensing, comprehensive perception of farmland greenhouse gases, meteorological, soil, and crop growth parameters was achieved, thus obtaining multi-source big data. Based on this, the isolated forest algorithm, Pettitt mutation test, and long short-term memory network were comprehensively applied to identify anomalies, fill in missing data, and standardize the raw data, effectively eliminating noise interference caused by hardware failures and undriven mutations, significantly improving the continuity and integrity of the data. Furthermore, by constructing a multi-dimensional weighting system integrating basic accuracy, spatiotemporal matching degree, scene representation capability, and dynamic stability, and introducing crop growth period coefficients and farmland event-driven coefficients, adaptive dynamic adjustment of the contribution of different data sources was achieved, thereby solving the problem of accurate fusion of multi-source heterogeneous data in terms of spatiotemporal scale and representation.

[0043] In this embodiment, preferably, step S103 involves using a convolutional neural network to extract spatial features from multi-source sensing big data, employing a Transformer model and attention mechanism to collect temporal features corresponding to the spatial features, and obtaining the predicted greenhouse gas emission flux for the target farmland area based on the temporal features. Specifically, this includes: Using a convolutional neural network, spatial features of multi-source sensing big data are extracted using the following formula: ; In the formula, For time The extracted spatial features The input features include at least two from the large dataset of features. For convolution kernel, For convolution bias; Based on the spatial features extracted by the convolutional neural network, temporal modeling is performed using the Transformer model, and the spatial features are linearly transformed using the following formula: ; ; ; In the formula, This is the weight matrix for the science department. The impact of meteorological and crop characteristics on greenhouse gas emissions, Soil-related characteristics, Information for calculating greenhouse gas emissions; The relative importance between each time step is calculated using a sub-attention mechanism, and the attention features are obtained using the following formula: ; In the formula, For attention features, The dimension of the key vector; Based on attention features, a multi-head attention mechanism is adopted to capture the temporal dependencies of multiple subspaces, and the temporal features are obtained by the following formula: ; In the formula, It is a time-series feature; Based on time-series characteristics, the predicted greenhouse gas emission flux for the target farmland area is obtained using the following formula: ; In the formula, and The weights and biases of the output layer of the Transformer model. Predicting fluxes for greenhouse gas emissions.

[0044] Specifically, by employing convolutional neural networks to extract spatial features from multi-source sensing big data, and combining the Transformer model with an attention mechanism to perform temporal modeling of the spatial features, the predicted greenhouse gas emission fluxes of the target farmland area are finally output. The CNN convolution operation effectively eliminates the fusion barrier of multi-source heterogeneous data in the spatial dimension. By utilizing the Transformer and attention mechanism to deeply mine the dynamic change patterns of greenhouse gas emissions over time, high-precision spatiotemporal feature extraction and quantitative prediction of greenhouse gas fluxes such as CH4 and N2O in farmland are achieved, thereby significantly improving the consistency of multi-source data fusion and the spatiotemporal resolution and data reliability of greenhouse gas emission monitoring.

[0045] In this embodiment, preferably, step S104 involves constructing a physical equation simulating the dynamics of greenhouse gas emissions based on multi-source sensing big data, embedding the physical equation into a physical information neural network model, constructing a total loss function, and iterating the physical information neural network model according to the total loss function to obtain a farmland source apportionment model. Specifically, this includes: Based on the ADR equations describing changes in gas concentration and gas transport in farmland, the following physical equation for meteorological transport is obtained: ; In the formula, For meteorological conditions and diffusion coefficient, For source terms that include both biological and abiotic sources, For gas concentration, This refers to the wind speed field.

[0046] Nitrification and denitrification are the main biological sources of non-carbon dioxide greenhouse gas emissions. Therefore, a reaction kinetic model is used to represent the microbial source term, resulting in the following physical equation for soil microbial activity: ; In the formula, For microbial origin items, and These are the yield coefficients for the nitration and denitrification reaction processes, respectively. and These represent the nitrification and denitrification rates, respectively. Construct the following physical equation for crop growth: ; In the formula, Leaf area index, In order to absorb light and effective radiation, , , These represent the limiting factors for temperature, moisture, and nitrogen, respectively. The physical equations of meteorological transport, soil microbial activity, and crop growth process are embedded into a physical information neural network model to construct a total loss function; The farmland source apportionment model is obtained by iterating the physical information neural network model based on the total loss function.

[0047] Specifically, the ADR equation, or convection-diffusion-reaction equation, is used to describe changes in gas concentration and gas transport in farmland.

[0048] Specifically, nitrification rate and denitrification rate It depends on soil nitrogen concentration, temperature, pH value, etc., and the calculation process is as follows: ; ; in, and Soil nitrogen source concentration, , , This is a function of the effects of temperature, humidity, and pH.

[0049] In this embodiment, preferably, the physical equations of meteorological transport, soil microbial activity, and crop growth processes are embedded into a Physical Information Neural Network (PINN) model to construct a total loss function, specifically including: A physical information neural network model is trained based on the physical equations of meteorological transport. Given the gas concentration output by the physical information neural network model, the residual loss of the meteorological transport physics equation is obtained as follows: ; Construct the soil-atmosphere interface boundary conditions as follows: ; In the formula, For the residual loss of the physical equations of meteorological transport, The gas flux predicted by the physical equations of meteorological transport. For the external normal vector, It is the soil-atmosphere interface; By embedding the physical equations of soil microbial activity into the source terms of the physical information neural network model, the residual loss of the physical equations of soil microbial activity is obtained as follows: ; In the formula, This represents the residual loss in the physical equations of soil microbial activity; By embedding the constraints of the physical equations governing crop growth into a physical information neural network model, the residual loss of the physical equations governing crop growth is obtained as follows: ; In the formula, This represents the residual loss in the physical equations governing crop growth. Based on the prior total throughput of the convolutional neural network and the Transformer model, a prior consistency loss for the physical information neural network model is constructed: ; ; ; in, For prior consistency loss, For the prior total flux, This represents the total output flux of the physical information neural network model. and These represent the biological and non-biological fluxes output by the physical information neural network model, respectively. Based on the residual losses from the physical equations of meteorological transport, soil microbial activity, and crop growth processes, as well as the prior consistency loss, the total loss function is obtained as follows: ; Where L is the total loss function, , , , These are the residual losses of the meteorological transport physical equation, the soil microbial activity physical equation, the crop growth process physical equation, and the loss weight coefficients corresponding to the prior consistency loss.

[0050] Specifically, during training, the parameters of the PINN model are continuously adjusted using the backpropagation algorithm to minimize the total loss function L. In this way, the PINN model not only conforms to data-driven prior predictions but also strictly follows the physical laws governing meteorological transport, microbial activity, and crop growth, thereby achieving precise decoupling of biogenic and abiotic fluxes and improving prediction accuracy.

[0051] Specifically, by constructing a deep learning source analysis model under the constraints of physical mechanisms, we can accurately distinguish between biological and non-biological sources of greenhouse gases from farmland, quantitatively analyze the contribution ratio and dynamic impact of different emission sources, farmland management measures, and environmental factors on greenhouse gas emissions, and truly restore the spatiotemporal dynamic changes of greenhouse gas emissions from farmland.

[0052] In specific implementations, the spatiotemporal distribution data of greenhouse gas emissions in the target farmland area obtained from the greenhouse gas emission prediction flux and the farmland source apportionment model can be visualized to generate maps and charts of non-CO2 greenhouse gas emissions. Specifically, a spatiotemporal distribution map of greenhouse gas emissions in the target farmland area and analytical charts of various sources (including biological and abiotic sources) are generated to provide intuitive data support for subsequent farmland management decisions.

[0053] Compared with existing technologies, the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin, and source apportionment provided in this embodiment can achieve at least one of the following beneficial effects: a) This invention effectively solves the problems of poor fusion consistency and difficulty in eliminating heterogeneity caused by differences in temporal resolution, spatial resolution, and sampling frequency in existing multi-source data by integrating multi-source heterogeneous data from IoT sensors, drones, and satellite remote sensing, and constructing a source analysis model by combining a CNN-Transformer deep learning model with the dynamic physical equations of greenhouse gas emissions. At the same time, it breaks through the limitations of traditional fixed-point monitoring, such as incomplete spatial coverage, low sampling frequency, and weak dynamic change capture capability, and realizes full-area, high-frequency, continuous real-time monitoring of core greenhouse gas fluxes such as CH4 and N2O in target farmland areas. This significantly improves the spatiotemporal accuracy and data reliability of greenhouse gas monitoring data, and provides accurate basic data support for farmland management decisions. b) This invention constructs a deep learning source analysis model under physical mechanism constraints, abandoning the analysis method of existing technologies that rely too much on empirical models. It can accurately distinguish between biological and non-biological sources of greenhouse gases in farmland, quantitatively analyze the contribution ratio and dynamic impact of different emission sources, different farmland management measures, and different environmental factors on greenhouse gas emissions, and truly restore the spatiotemporal dynamic change characteristics of greenhouse gas emissions in farmland. c) This invention constructs a full-element digital twin of the target farmland area through digital twin technology, realizing real-time mapping, dynamic simulation and extrapolation of all-dimensional elements such as farmland meteorology, soil, crop growth, management measures, and greenhouse gas emissions; it can conduct advance simulation and quantitative assessment of the greenhouse gas emission effects and crop growth impacts of different farmland management schemes, thereby improving the scientific nature of subsequent farmland management decisions and their practical usability in the field.

[0054] Example 2 A specific embodiment of the present invention, such as Figure 2 As shown, a farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution is disclosed, including: S201. Obtain the spatiotemporal distribution data of greenhouse gas emissions in the target farmland area obtained by the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin and source analysis in Embodiment 1 of the present invention; S202. Based on the spatiotemporal distribution data of greenhouse gas emissions, construct and train a model of the relationship between farmland management actions and gas emissions based on graph neural networks. The model of the relationship between farmland management actions and gas emissions is used to build a virtual decision-making environment. S203. In a virtual decision-making environment, construct a farmland optimization management decision model based on reinforcement learning algorithm; S204. Based on the farmland optimization management decision model, generate farmland management strategies for the target farmland area.

[0055] Specifically, through continuous iteration, the optimal strategy is solved based on the relationship between gas emission predictions and agricultural management measures, ultimately generating an optimized agricultural management strategy to achieve the goal of minimizing greenhouse gas emissions and improving agricultural production efficiency.

[0056] Specifically, farmland management strategies include specific recommendations on fertilization, irrigation, and tillage, aiming to balance environmental and economic benefits.

[0057] In this embodiment, preferably, in the farmland management decision-making method, a graph neural network (GNN) is used to model the complex nonlinear relationship between farmland management measures (such as fertilization, irrigation, and tillage) and non-carbon dioxide gas emissions. Specifically, farmland management measures and gas emissions are modeled using a graph structure, where each node represents a management measure or gas emission state, and each edge represents the relationship between the management measure and gas emissions. Each node The feature vectors are: ; in, For nodes In time The feature vector at time step, For nodes The input features include meteorological, soil, and crop data, etc. and For the weights and biases of the GNN layer, This is the activation function.

[0058] During the propagation of a graph neural network, information between nodes is transmitted through the adjacency matrix. Update: ; in, For nodes in the graph and nodes Adjacency matrix elements between them For nodes The set of adjacent nodes.

[0059] In this embodiment, preferably, in a virtual decision-making environment, a farmland optimization management decision model based on reinforcement learning algorithms is constructed, specifically including: The system state is constructed based on the following formula: ; In the formula, For the state of the system, For current farmland management practices, To determine the gas emission flux corresponding to the current farmland management actions in a virtual decision-making environment; The system's actions are based on the following formula: ; In the formula, For the system's actions, Adjusted farmland management procedures; The reward function is constructed based on the following formula: ; In the formula, For state Take action below value, For learning rate, As a discount factor, For instant rewards.

[0060] Specifically, by training a Generative Neural Network (GNN), the deep-seated relationship between agricultural management practices and gas emissions is learned. The GNN encodes this complex relationship as node features and edge weights. Through training, the GNN can automatically learn the impact of agricultural management practices on gas emissions and predict future emissions.

[0061] Compared with existing technologies, the farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution provided in this embodiment can achieve at least the following beneficial effects: This invention is adaptable to farmland scenarios with different climate zones, soil types, and crop planting patterns, and has strong scenario adaptability and large-scale promotion capabilities. It can effectively utilize multi-source sensing big data, and can not only provide various planting entities with directly implementable and usable low-carbon farmland management decision-making solutions, but also effectively promote the digital transformation of agriculture and the realization of the "dual carbon" goals in the agricultural field, and comprehensively enhance the sustainable development capabilities and climate resilience of agricultural production.

[0062] Example 3 A specific embodiment of the present invention, such as Figure 3As shown, a farmland greenhouse gas monitoring system 1 based on multi-source fusion sensing, digital twin, and source apportionment includes: a processor 11, a memory 12, and a computer program stored in the memory 12 and executable on the processor, such as a farmland greenhouse gas monitoring program based on multi-source fusion sensing, digital twin, and source apportionment. When the processor 11 executes the computer program, it implements the steps in the various embodiments of the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin, and source apportionment described above. Alternatively, when the processor 11 executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0063] For example, a computer program can be divided into one or more modules / units, one or more of which are stored in memory and executed by a processor to complete the present invention. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in a farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twins, and source apportionment.

[0064] A farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twins, and source apportionment may include, but is not limited to, processors and memory. Those skilled in the art will understand that the schematic diagram is merely an example of a farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twins, and source apportionment, and does not constitute a limitation on such a system. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, a farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twins, and source apportionment may also include input / output devices, network access devices, CAN buses, etc.

[0065] Example 4 A specific embodiment of the present invention, such as Figure 4 As shown, a farmland management decision-making system 2 based on multi-source fusion sensing, digital twin, and source resolution includes: a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor, such as a farmland management decision-making program based on multi-source fusion sensing, digital twin, and source resolution. When the processor 21 executes the computer program, it implements the steps in the various embodiments of the farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution described above. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module / unit in the various device embodiments described above.

[0066] For example, a computer program can be divided into one or more modules / units, one or more of which are stored in memory and executed by a processor to complete the present invention. One or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in a farmland management decision-making system based on multi-source fusion sensing, digital twins, and source resolution.

[0067] A farmland management decision-making system based on multi-source fusion sensing, digital twins, and source resolution may include, but is not limited to, processors and memory. Those skilled in the art will understand that the schematic diagram is merely an example of a farmland management decision-making system based on multi-source fusion sensing, digital twins, and source resolution, and does not constitute a limitation on such a system. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, a farmland management decision-making system based on multi-source fusion sensing, digital twins, and source resolution may also include input / output devices, network access devices, CAN buses, etc.

[0068] Example 5 The present invention provides a computer-readable storage medium, characterized in that the computer-readable storage medium includes a stored computer program, wherein, when the computer program is running, it controls the device where the computer-readable storage medium is located to execute, as in Embodiment 1 of the present invention, a method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin and source analysis, or Embodiment 2 of the present invention, a method for farmland management decision-making based on multi-source fusion sensing, digital twin and source analysis.

[0069] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application-Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor serves as the control center for a farmland greenhouse gas monitoring system or a farmland management decision-making system based on multi-source fusion sensing, digital twins, and source apportionment. It connects various parts of the entire system via various interfaces and lines.

[0070] The memory can be used to store computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory, realizes various functions of a farmland greenhouse gas monitoring system or a farmland management decision-making system based on multi-source fusion sensing, digital twins, and source apportionment. The memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0071] Modules / units integrated into farmland greenhouse gas monitoring systems based on multi-source fusion sensing, digital twins, and source apportionment, or farmland management decision-making systems based on multi-source fusion sensing, digital twins, and source apportionment, can be stored in a computer-readable storage medium if implemented as software functional units and sold or used as independent products. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in computer-readable media may be appropriately added to or subtracted from the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, computer-readable media may not include electrical carrier signals and telecommunication signals, in accordance with legislation and patent practice.

[0072] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they 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. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0073] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment, characterized in that, include: Acquire characteristic big data of the target farmland area, wherein the characteristic big data includes at least two of the following: greenhouse gas emission data, meteorological parameters, soil physicochemical properties, crop canopy physiological parameters, spatial heterogeneity data, and farmland environmental data; The feature big data is synchronized and fused to obtain multi-source sensing big data of the target farmland area; A convolutional neural network is used to extract the spatial features of the multi-source sensing big data, and a Transformer model and attention mechanism are used to collect the temporal features corresponding to the spatial features. Based on the temporal features, the predicted greenhouse gas emission flux of the target farmland area is obtained. Based on the multi-source sensing big data, a physical equation simulating the dynamics of greenhouse gas emissions is constructed. The physical equation is embedded into a physical information neural network model, a total loss function is constructed, and the physical information neural network model is iterated according to the total loss function to obtain a farmland source apportionment model. The spatiotemporal distribution data of greenhouse gas emissions in the target farmland area are obtained based on the predicted greenhouse gas emission flux and the farmland source apportionment model.

2. The method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment as described in claim 1, characterized in that, The greenhouse gas emission data, meteorological parameters, and soil physicochemical properties are acquired by a ground-based Internet of Things sensor array; the crop canopy physiological parameters and spatial heterogeneity data are acquired by drones; and the farmland environmental data are acquired by satellite remote sensing equipment.

3. The method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment as described in claim 1, characterized in that, The process of synchronizing and fusing the feature big data to obtain multi-source sensing big data for the target farmland area specifically includes: The feature big data is preprocessed, and the isolated forest algorithm is used to identify and remove outliers in the preprocessed feature big data to obtain the first feature set; Spatial interpolation algorithms are used to fill in the spatial missing data in the first feature set to obtain the second feature set; A long short-term memory network is used to impute the time series data of the second feature set to obtain a third feature set; After applying registration benchmarks and accuracy constraints to the data in the third feature set, a fourth feature set is obtained; Based on the basic accuracy weight, spatiotemporal matching weight, scene representation weight and dynamic stability weight, the fourth feature set is fused to obtain multi-source sensing big data of the target farmland area. Preferably, the basic accuracy weight is calculated using the following formula: W bas,i = ； In the formula, W bas,i ∈(0,1), representing the normalized base accuracy weights, i=1,2,3, corresponding to the three data sources: the ground-based IoT sensor array, the UAV, and the satellite remote sensing equipment, respectively. RRMSE i Let E be the relative root mean square error of the i-th type of data source. cal,i The laboratory calibration relative error of the i-th type of data source; The spatiotemporal matching weights are calculated using the following formula: W st,i = ； In the formula, W st,i ∈(0,1), representing the normalized spatiotemporal matching weights, i=1,2,3 and corresponding to the three types of data sources: the ground-based IoT sensor array, the UAV, and the satellite remote sensing equipment, respectively. T m,i S is the time matching coefficient of the i-th type of data source. m,i The spatial matching degree coefficient of the i-th type of data source; The scene representation weights are calculated using the following formula: W scene,i = ； In the formula, W scene,i ∈(0,1), where MI is the normalized scene representation weight. i K is the mutual information coefficient. stage K is the crop growth period coefficient. event For event-driven coefficients; The dynamic stability weights are calculated using the following formula: W stab,i = ； In the formula, W stab,i ∈(0,1), where Q is the normalized dynamic stable weight. valid,i For the effective data rate of the i-th type of data source, CV i Let be the volatility coefficient of the i-th type of data source, Cloud i The percentage of cloud data sourced from drones or satellite remote sensing equipment.

4. The method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment as described in claim 1, characterized in that, The process involves using a convolutional neural network to extract spatial features from the multi-source sensing big data, employing a Transformer model and attention mechanism to collect temporal features corresponding to the spatial features, and obtaining the predicted greenhouse gas emission flux for the target farmland area based on the temporal features. Specifically, this includes: Using a convolutional neural network, the spatial features of the multi-source sensing big data are extracted using the following formula: ； In the formula, For time The extracted spatial features The input features include at least two of the features from the large dataset. For convolution kernel, For convolution bias; Based on the spatial features extracted by the convolutional neural network, temporal modeling is performed using the Transformer model, and the spatial features are linearly transformed using the following formula: ； ； ； In the formula, This is the weight matrix for the science department. The impact of meteorological and crop characteristics on greenhouse gas emissions, Soil-related characteristics, Information for calculating greenhouse gas emissions; The relative importance between each time step is calculated using a sub-attention mechanism, and the attention features are obtained using the following formula: ； In the formula, For the attention feature, The dimension of the key vector; Based on the aforementioned attention features, a multi-head attention mechanism is employed to capture the temporal dependencies of multiple subspaces, and the temporal features are obtained using the following formula: ； In the formula, The time-series feature; Based on the aforementioned time-series characteristics, the predicted greenhouse gas emission flux for the target farmland area is obtained using the following formula: ； In the formula, and The weights and biases of the output layer of the Transformer model. The predicted flux for greenhouse gas emissions.

5. The method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment as described in claim 1, characterized in that, The process involves constructing physical equations to simulate the dynamics of greenhouse gas emissions based on the multi-source sensing big data, embedding these physical equations into a physical information neural network model, constructing a total loss function, and iterating the physical information neural network model according to the total loss function to obtain a farmland source apportionment model. Specifically, this includes: Based on the ADR equations describing changes in gas concentration and gas transport in farmland, the following physical equation for meteorological transport is obtained: ； In the formula, For meteorological conditions and diffusion coefficient, For source terms that include both biological and abiotic sources, For gas concentration, This refers to the wind speed field.

6. Using a reaction kinetic model to represent the microbial source term, the following physical equation for soil microbial activity is obtained: ； In the formula, For microbial origin items, and These are the yield coefficients for the nitration and denitrification reaction processes, respectively. and These represent the nitrification and denitrification rates, respectively. Construct the following physical equation for crop growth: ； In the formula, Leaf area index, In order to absorb light and effective radiation, , , These represent the limiting factors for temperature, moisture, and nitrogen, respectively. The physical equations of meteorological transport, soil microbial activity, and crop growth process are embedded into the physical information neural network model to construct the total loss function; The farmland source analysis model is obtained by iterating the physical information neural network model based on the total loss function.

7. The method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source apportionment as described in claim 5, characterized in that, The process of embedding the physical equations of meteorological transport, soil microbial activity, and crop growth into the physical information neural network model to construct the total loss function specifically includes: The physical information neural network model is trained based on the aforementioned meteorological transport physical equations. Given the gas concentration output by the physical information neural network model, the residual loss of the meteorological transport physical equation is obtained as follows: ； Construct the soil-atmosphere interface boundary conditions as follows: ； In the formula, The residual loss of the aforementioned meteorological transport physical equation, The gas flux predicted by the aforementioned meteorological transport physics equation. The external normal vector. It is the soil-atmosphere interface; By embedding the physical equation of soil microbial activity into the source terms of the physical information neural network model, the residual loss of the physical equation of soil microbial activity is obtained as follows: ； In the formula, This represents the residual loss of the physical equation for soil microbial activity; By embedding the constraints of the physical equations governing the crop growth process into the physical information neural network model, the residual loss of the physical equations governing the crop growth process is obtained as follows: ； In the formula, This represents the residual loss of the physical equations governing the crop growth process. Based on the prior total throughput of the convolutional neural network and the Transformer model, construct the prior consistency loss of the physical information neural network model: ； ； ； in, The prior consistency loss is... Let be the prior total flux. This represents the total output throughput of the physical information neural network model. and These are the biological and non-biological fluxes output by the physical information neural network model, respectively. Based on the residual losses of the meteorological transport physical equation, the soil microbial activity physical equation, the crop growth process physical equation, and the prior consistency loss, the total loss function is obtained as follows: ； Where L is the total loss function, , , , These are the loss weight coefficients corresponding to the residual loss of the meteorological transport physical equation, the residual loss of the soil microbial activity physical equation, the residual loss of the crop growth process physical equation, and the prior consistency loss, respectively.

8. A farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution, characterized in that, include: Obtain the spatiotemporal distribution data of greenhouse gas emissions in the target farmland area obtained by the farmland greenhouse gas monitoring method based on multi-source fusion sensing, digital twin and source apportionment as described in any one of claims 1-6; Based on the spatiotemporal distribution data of greenhouse gas emissions, a model of the relationship between farmland management actions and gas emissions based on graph neural networks is constructed and trained, wherein the model of the relationship between farmland management actions and gas emissions is used to build a virtual decision-making environment; In the virtual decision-making environment, a farmland optimization management decision model based on reinforcement learning algorithm is constructed; Based on the farmland optimization management decision model, a farmland management strategy for the target farmland area is generated.

9. The farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution according to claim 7, characterized in that, The construction of a farmland optimization management decision model based on reinforcement learning algorithms within the virtual decision-making environment specifically includes: The system state is constructed based on the following formula: ； In the formula, The state of the system. For current farmland management practices, In the virtual decision-making environment, the gas emission flux corresponding to the current farmland management action; The system's actions are based on the following formula: ； In the formula, For the actions of the system, Adjusted farmland management procedures; The reward function is constructed based on the following formula: ； In the formula, For state Take action below value, For learning rate, As a discount factor, For instant rewards.

10. A farmland greenhouse gas monitoring system based on multi-source fusion sensing, digital twin, and source apportionment, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement a method for monitoring greenhouse gases in farmland based on multi-source fusion sensing, digital twin, and source resolution as described in any one of claims 1 to 6.

11. A management decision-making system based on multi-source fusion sensing, digital twin, and source resolution, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements a farmland management decision-making method based on multi-source fusion sensing, digital twin, and source resolution as described in any one of claims 7 to 8.

Images