Urban wetland carbon sink management system and method based on digital twinning

Abstract

The invention discloses an urban wetland carbon sink management system and method based on digital twinning, and belongs to the field of carbon sink management. The data transmission layer transmits the multi-source monitoring data and stores the multi-source monitoring data to the cloud platform; the digital twin modeling layer is combined with multi-source monitoring data, GPP and Reco to construct a management model of the urban wetland, and NEP equation information and temperature-respiration equation information are added in a loss function as physical constraints; constructing a three-dimensional visual monitoring management platform according to the management model; and the intelligent analysis layer generates a coping strategy when the visual monitoring management platform monitors abnormal data, and collects feedback data after implementation of the coping strategy so as to update the strategy generation model. According to the urban wetland carbon sink management system and method based on digital twinning, through deep fusion of a physical mechanism and data driving and in combination with dynamic calibration, the whole process of sensing, decision making, execution and feedback of urban wetland carbon sinks is achieved, and a management positive cycle is formed.

Description

Technical Field

[0001] The present application relates to the field of carbon sink management, and specifically to an urban wetland carbon sink management system and method based on digital twins. Background Art

[0002] As important natural carbon sinks, wetlands have a carbon absorption capacity three to five times greater than that of forests, playing an irreplaceable role in climate change mitigation. However, with the acceleration of urbanization, urban wetlands are facing issues such as shrinking area and degraded functions, seriously threatening their carbon sequestration capacity. Therefore, efficient and precise urban wetland carbon sequestration management is crucial to achieving carbon neutrality goals.

[0003] Traditional monitoring methods are based on manual sampling and static models. Manual sampling involves regularly collecting wetland soil, vegetation and water samples, and analyzing the carbon content in the laboratory. The data update cycle is long (usually monthly or quarterly), which cannot reflect the dynamic changes in wetland carbon sinks and has high manpower and material costs. Static models are carbon sink models constructed based on historical data (such as the IPCC carbon accounting method), which cannot respond to environmental changes (such as rainfall and temperature fluctuations) in real time and have limited prediction accuracy.

[0004] The monitoring method based on satellite remote sensing uses satellites to obtain data such as wetland vegetation coverage and water quality. It has limited spatial resolution (usually 10-30 meters) and low temporal resolution, making real-time monitoring impossible.

[0005] The monitoring method based on Internet of Things technology deploys sensors such as CO2 concentration, temperature and humidity, and light to collect wetland environmental data in real time. However, the data is isolated and has not been systematically managed. In addition, the sensor deployment density is insufficient and the monitoring range is limited. Edge computing nodes are deployed locally in the wetlands to pre-process the environmental data collected by the sensors, but the computing power is limited and it is difficult to support complex models.

[0006] In summary, the current urban wetland carbon sink management faces problems such as low monitoring accuracy, insufficient data real-time performance, and high management costs, making it difficult to meet the needs of precise and intelligent management. Summary of the Invention

[0007] The present application provides an urban wetland carbon sink management system and method based on digital twins, which can solve the technical problems of urban wetland carbon sink management in the existing technology, such as low monitoring accuracy, insufficient data real-time performance, and high management costs.

[0008] In a first aspect, an embodiment of the present application provides an urban wetland carbon sequestration management system based on digital twins, the system comprising:

[0009] The data perception layer includes hardware sensors, drones, satellite receiving stations, and weather stations, which are used to collect multi-source monitoring data;

[0010] The data transmission layer is used to transmit and store multi-source monitoring data to the cloud platform;

[0011] The digital twin modeling layer is used to combine multi-source monitoring data, gross primary productivity (GPP), and ecosystem respiration (R) eco Constructing a management model for urban wetlands, and adding Net Ecosystem Productivity (NEP) equation information and Temperature-Respiration equation information as physical constraints in the loss function; and also constructing a three-dimensional visual monitoring and management platform based on the management model;

[0012] The intelligent analysis layer is used to generate a response strategy based on the strategy generation model when the visual monitoring management platform detects abnormal data, and collect feedback data after the response strategy is implemented to update the strategy generation model.

[0013] In conjunction with the first aspect, in one embodiment, the multi-source monitoring data includes sensor data collected by sensors, multispectral image data collected by drones, remote sensing data collected by satellites, and meteorological data collected by weather stations;

[0014] The sensor data includes CO2 concentration data, temperature and humidity data, light intensity data, water level depth data, water quality data, vegetation image data, soil moisture data, and water pH value data;

[0015] The multispectral image data includes the Normalized Difference Vegetation Index (NDVI) and the Wide Area Vegetation Index (EVI);

[0016] The remote sensing data includes NDVI, EVI, and land surface temperature;

[0017] The meteorological data includes rainfall data, wind speed data, and temperature data.

[0018] In conjunction with the first aspect, in one embodiment, the digital twin modeling layer includes:

[0019] The data preprocessing module is used to clean and extract features from multi-source monitoring data to obtain multi-source modeling data;

[0020] ST-Fusion module, used for spatiotemporal alignment of multi-source modeling data;

[0021] Physics-AI model module for calculating GPP and R based on multi-source modeling data eco , based on multi-source modeling data, GPP, and R eco Training management model;

[0022] The three-dimensional rendering module is used to construct a three-dimensional visual monitoring and management platform according to the management model.

[0023] In combination with the first aspect, in one embodiment, the NEP equation information includes NEP is not greater than GPP, R eco Not less than 0, and GPP not less than 0;

[0024] The temperature-respiration equation information includes the Arrhenius equation.

[0025] In combination with the first aspect, in one embodiment, the digital twin modeling layer is also used to collect multi-source monitoring data for a preset time period before the current moment when the multi-source monitoring data conforms to an environmental mutation or the management model prediction error exceeds a preset threshold, and retain a preset proportion of historical multi-source monitoring data to re-train and update the management model.

[0026] In conjunction with the first aspect, in one embodiment, constructing a three-dimensional visual monitoring and management platform according to the management model specifically includes the following steps:

[0027] A visual monitoring and management platform for urban wetlands is constructed by combining management models, digital elevation models, and multispectral image data.

[0028] In conjunction with the first aspect, in one embodiment, the intelligent analysis layer includes:

[0029] A real-time carbon sink calculation module is used to obtain the integrated carbon sink data output by the management model and process it to obtain regional carbon sink capacity assessment results;

[0030] An anomaly detection module, used to compare the fused carbon sink data with historical baseline data, and mark the abnormal area and alarm level when the comparison result is abnormal;

[0031] The strategy optimization service module is used to process abnormal areas and alarm levels based on the strategy generation model to generate a response strategy; it is also used to collect feedback data after the implementation of the response strategy to update the strategy generation model.

[0032] In conjunction with the first aspect, in one embodiment, generating a response strategy based on a strategy generation model and collecting feedback data after the response strategy is implemented to update the strategy generation model specifically includes the following steps:

[0033] generating a plurality of response strategies based on a strategy generation model, each response strategy including at least one regulatory measure;

[0034] Cost-effectiveness calculations were conducted for multiple preliminary response strategies, including incremental carbon sinks, real-time costs, and ecological risks. Cost-effectiveness was positively correlated with incremental carbon sinks and negatively correlated with real-time costs and ecological risks.

[0035] sorting the multiple preliminary response strategies according to cost-effectiveness, and collecting the carbon sequestration improvement rate of each preliminary response strategy from high to low in terms of cost-effectiveness;

[0036] The action value of the regulatory measures is evaluated according to the carbon sink improvement rate, and the strategy generation model is updated according to the updated action value.

[0037] In combination with the first aspect, in one embodiment, the strategy generation model includes a preset rule database and a reinforcement learning DQN model, and the preset rule database contains multiple response strategies associated with abnormal data.

[0038] In a second aspect, an embodiment of the present application provides a method for urban wetland carbon sequestration management based on digital twins, the method comprising:

[0039] Configure hardware sensors, drones, satellite receiving stations, and weather stations to collect multi-source monitoring data;

[0040] Transmit and store multi-source monitoring data to the cloud platform;

[0041] Combining multi-source monitoring data, gross primary productivity (GPP), and ecosystem respiration (R) eco Constructing an urban wetland management model and adding Net Ecosystem Productivity (NEP) equation information and Temperature-Respiration equation information as physical constraints to the loss function; building a three-dimensional visualization monitoring and management platform based on the management model;

[0042] When the visual monitoring management platform detects abnormal data, a response strategy is generated based on the strategy generation model, and feedback data after the response strategy is implemented is collected to update the strategy generation model.

[0043] The beneficial effects of the technical solutions provided in the embodiments of the present application include:

[0044] The digital twin-based urban wetland carbon sink management system and method deeply integrates physical mechanisms with data-driven approaches, combined with dynamic calibration, to achieve the full process of urban wetland carbon sink perception, decision-making, execution, and feedback, forming a positive management cycle. This also reduces manual inspection costs and achieves rational resource allocation through dynamic adjustment strategies based on carbon sink prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a schematic diagram of the architecture of an embodiment of an urban wetland carbon sequestration management system based on digital twins of this application;

[0046] Figure 2 This application is a schematic diagram of the overall process of the digital twin modeling layer;

[0047] Figure 3 Recommend an overall process diagram for this application strategy;

[0048] Figure 4 This is a flow chart of an embodiment of the urban wetland carbon sequestration management method based on digital twins of this application. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0051] In a first aspect, an embodiment of the present application provides an urban wetland carbon sequestration management system based on digital twins.

[0052] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of an embodiment of the urban wetland carbon sequestration management system based on digital twins. Figure 1 As shown in the figure, the urban wetland carbon sequestration management system based on digital twin includes:

[0053] The data perception layer, including hardware sensors, drones, satellite receiving stations, and weather stations, is used to collect multi-source monitoring data.

[0054] The data transmission layer is used to transmit and store multi-source monitoring data to the cloud platform.

[0055] The digital twin modeling layer is used to combine multi-source monitoring data, GPP (Gross Primary Productivity), and ecosystem respiration R eco A management model for urban wetlands was constructed, and the NEP (Net Ecosystem Productivity) equation and the temperature-respiration equation were added as physical constraints to the loss function. This was also used to construct a 3D visualization monitoring and management platform based on the management model.

[0056] The intelligent analysis layer is used to generate response strategies based on the strategy generation model when the visual monitoring management platform detects abnormal data, and collect feedback data after the implementation of the response strategy to update the strategy generation model.

[0057] In this embodiment, the digital twin-based urban wetland carbon sink management system and method deeply integrates physical mechanisms with data-driven approaches, combined with dynamic calibration, to achieve the full process of urban wetland carbon sink perception, decision-making, execution, and feedback, forming a positive management cycle. This also reduces manual inspection costs and achieves rational resource allocation through dynamic carbon sink prediction and adjustment strategies.

[0058] Furthermore, in one embodiment, the multi-source monitoring data includes sensor data collected by sensors, multispectral image data collected by drones, remote sensing data collected by satellites, and meteorological data collected by weather stations.

[0059] The sensor data includes CO2 concentration data, temperature and humidity data, light intensity data, water level depth data, water quality data, vegetation image data, soil moisture data, and water pH value data.

[0060] Multispectral image data includes NDVI (Normalized Difference Vegetation Index) and EVI (Enhanced Vegetation Index).

[0061] Remote sensing data include NDVI, EVI, and land surface temperature.

[0062] Meteorological data includes rainfall data, wind speed data, and temperature data.

[0063] In this embodiment, the data perception layer includes hardware sensors and drones. Extended data sources include satellite receiving stations and weather station data sources. Hardware sensors include environmental sensors, water quality sensors, vegetation monitoring equipment, and soil sensors. Extended data sources include satellite remote sensing data and meteorological data.

[0064] Environmental sensors include: a CO2 concentration sensor with an accuracy of approximately ±70 ppm, deployed at a density of 1 per hectare; a temperature and humidity sensor covering the core wetland area; and a light sensor with a spectral range of 400-1100 nm to monitor photosynthesis intensity.

[0065] Water quality sensors include: dissolved oxygen sensors with an accuracy of approximately ±0.3mg / L, deployed at key points in the water body; and pH sensors with a measurement range of 0-14, which monitor water acidification in real time.

[0066] Vegetation monitoring equipment includes: multispectral drone equipped with a multispectral camera (5 bands: blue, green, red, red edge, near infrared), a resolution of 5cm, and daily inspections.

[0067] Soil sensors include: soil moisture sensor, with a range of 0-100%, an accuracy of approximately ±5%, and a deployment density of 3 per hectare.

[0068] Expanded data sources include: Satellite remote sensing data from Landsat-8 (30m resolution) and Sentinel-2 (10m resolution), updated daily. Meteorological data, accessed from local weather stations (rainfall, wind speed, and temperature), updated every 5 minutes.

[0069] Furthermore, in one embodiment, the data transmission layer is divided into two parts: communication technology and cloud platform. The communication technology includes: LPWAN (Low-Power Wide-Area Network), 5G private network, and edge computing nodes. The cloud platform includes a data storage module and a real-time computing module.

[0070] The low-power wide area network utilizes the LoRaWAN protocol (868MHz frequency band), has a coverage radius of 3km, and supports 10,000 nodes. The 5G private network is used for real-time drone video streaming, with latency less than 50ms. Edge computing nodes are deployed to run data preprocessing algorithms (Kalman filter denoising and data compression) to pre-process the data.

[0071] The data storage module uses cloud object storage services to store multi-source heterogeneous data. The real-time computing module is based on the Apache Flink stream processing framework and supports real-time processing of tens of thousands of data items per second.

[0072] Furthermore, in one embodiment, referring to Figure 2 , the digital twin modeling layer includes:

[0073] Data preprocessing module. It is used to clean and extract features from multi-source monitoring data to obtain multi-source modeling data.

[0074] The ST-Fusion module is used to perform spatiotemporal alignment of multi-source modeling data.

[0075] The Physics-AI Model module is used to calculate GPP and R based on multi-source modeling data. eco , based on multi-source modeling data, GPP, and R eco Training management model.

[0076] The three-dimensional rendering module is used to build a three-dimensional visualization monitoring and management platform based on the management model.

[0077] In this embodiment, the digital twin modeling layer constructs a high-precision virtual mapping of urban wetland carbon sinks through physical models, data-driven models, model fusion technology, and dynamic calibration mechanisms.

[0078] The digital twin modeling layer includes model construction and 3D visualization. Model construction includes the physical-AI model and the ST-Fusion module, which is used to align multi-source data. 3D visualization includes platform construction and data overlay.

[0079] The overall process of the digital twin modeling layer is as follows: Figure 2 As shown, it includes data preprocessing, ST-Fusion spatiotemporal alignment, physical-AI model construction, adding physical constraints, building a new loss function, and 3D visualization rendering.

[0080] The input data for data preprocessing include sensor data (CO2 concentration, temperature and humidity, light intensity, water level depth, soil moisture content, pH value, with a time resolution of 10 minutes), UAV multispectral image data (used to calculate NDVI and EVI vegetation index), satellite remote sensing data (Landsat-8 NDVI, Sentinel-2 surface temperature), and meteorological data (precipitation, wind speed, evaporation, with a time resolution of 10 minutes).

[0081] Data preprocessing includes data cleaning and feature extraction. Data cleaning includes outlier detection and missing value filling. Outlier detection includes using the 3σ principle (three times the standard deviation) to eliminate outliers. Missing value filling includes using linear interpolation or ARIMA model prediction for time series data, and using Kriging interpolation for spatial data (images). Feature extraction includes calculating the NDVI and EVI indices using multispectral images. The calculation formulas are shown in the following formulas (1) and (2):

[0082]

[0083] Among them, NIR represents the near-infrared band, Red represents the red light band, and Blue represents the blue light band.

[0084] Furthermore, due to the mismatch in data sources involved in wetland carbon sequestration management, spatiotemporal alignment is required.

[0085] Specifically, the spatial scale differences include the resolution difference between satellites (10m) and drones (5cm).

[0086] The time frequency differences include: different update cycles of satellites (daily), drones (daily), and sensors (every minute).

[0087] Coordinate system differences include: drones use a local coordinate system, satellites use WGS84, and sensors have no spatial coordinates.

[0088] Therefore, spatiotemporal alignment is necessary to address the spatiotemporal gaps in multi-source data. First, the datum is unified: all data is converted to UTC (Universal Time Coordinated) timestamps. Spatially, the data is unified to the WGS84 coordinate system, and the grid projections are aligned.

[0089] Spatial alignment involves unifying the 5cm drone data and 10m satellite data into a 1m grid using a GCN algorithm. This reduces errors at the target resolution. This includes downsampling the multispectral imagery data to 1m as nodes (100×100). Upsampling the satellite data to 1m (50×50→100×100 after interpolation, with some nodes left empty). After three layers of GCN, the output is a 1m fused NDVI. The clarity of vegetation boundaries is close to that of the multispectral imagery data, and large water areas are consistent with the satellite data.

[0090] Time alignment involves using the DTW (Dynamic Time Warping) algorithm to match data streams with different sampling frequencies. Daily satellite data is synchronized with sensor minute-by-minute and daily multispectral imagery data to a 10-minute granularity.

[0091] Furthermore, the construction of the physical-AI model includes the construction of the physical model and the construction of the AI ​​model (also known as the management model). The model construction involves the NEP formula. NEP (Net Ecosystem Productivity) is the core indicator for measuring the carbon sequestration capacity of an ecosystem, which represents the net difference between the carbon fixed by the ecosystem through photosynthesis and the carbon released through respiration per unit time. NEP>0 means that the ecosystem is a carbon sink (absorbing more CO2 than releasing, such as healthy forests and wetlands). NEP<0 means that the ecosystem is a carbon source (releasing more CO2 than absorbing, such as forest land after a fire). NEP=0 means that the carbon budget is balanced. The definition formula of NEP is shown in the following formula (3):

[0092]

[0093] Where ∈ represents the light energy utilization rate (default value is 0.5gC / MJ). PAR represents photosynthetically active radiation (obtained by the light sensor). f(T) represents the temperature response function (Logistic curve, optimum temperature is 25°C). f(W) represents the humidity response function (based on soil moisture sensor data). R h Indicates heterotrophic respiration (soil microorganisms decomposing organic matter). a Indicates autotrophic respiration (plant respiration). 10 =2.0 is the temperature sensitivity coefficient (the respiration rate doubles for every 10°C increase in temperature). hIndicates the heterotrophic respiration coefficient value (0.01-0.05). soil Indicates the organic carbon content of the surface soil (2-10). T indicates the actual temperature of the surface soil.

[0094] The AI ​​model uses machine learning to capture complex nonlinear relationships (such as extreme weather). LSTM (Long Short-Term Memory) is used to train the AI ​​model. The input is multi-source monitoring data for the past 72 hours, and the output is the carbon sink capacity forecast for the next 24 hours (GPP, R eco , and feature parameters), 128 hidden layer nodes, and a learning rate of 0.001.

[0095] LSTM implements time series modeling through three gating units (input gate, forget gate, output gate) and cell states. The model loss function is shown in the following formula (4):

[0096]

[0097] Among them, N represents the total number of samples, y i represents the true value of the i-th sample, represents the predicted value of the i-th sample, represents the square of the prediction error of the i-th sample.

[0098] Furthermore, in one embodiment, the NEP equation information includes NEP not greater than GPP, R eco is not less than 0, and GPP is not less than 0.

[0099] Temperature-respiration equation information includes the Arrhenius equation.

[0100] In this example, when using LSTM to predict wetland carbon sinks, physical constraints are added to the loss function to improve the accuracy of the prediction. The physical constraints and their meanings are as follows:

[0101] NEP≤GPP: Prevent the model from overestimating carbon sequestration capacity and predicting NEP exceeding the actual upper limit of photosynthesis.

[0102] R eco ≥0: Respiration always releases CO2 and cannot be negative.

[0103] GPP ≥ 0: The amount of carbon fixed by photosynthesis is non-negative.

[0104] Temperature-respiration relationship: The respiration rate increases exponentially with increasing temperature, ensuring that the respiration rate conforms to the Arrhenius equation and avoids violating the laws of biochemistry.

[0105] According to the above constraints, a new loss function is reconstructed. This loss function ensures that the LSTM prediction not only fits the observed data but also conforms to the basic biogeochemical laws of the ecosystem through four physical constraints. The new loss function is shown in the following formula (5):

[0106]

[0107] Indicates adding the constraint type NEP≤GPP: a linear penalty is imposed when the NEP prediction value exceeds the GPP. Among them, NEP i represents the model-predicted net ecosystem productivity, GPP i represents the total primary productivity of the mechanistic model.

[0108] Indicates adding constraint type Reco≥0: using the ReLU function to truncate negative values. Among them, Reco i represents the ecosystem respiration predicted by the model.

[0109] Indicates adding a constraint type temperature-respiration relationship: making the predicted value close to the Q10 theoretical curve. Among them, R0 represents the basic respiratory rate, Q 10 represents the temperature sensitivity coefficient, t i Represents the temperature observation value, T ref Indicates the reference temperature.

[0110] Indicates adding constraint type GPP≥0: using the ReLU function to truncate negative values. Among them, GPP i represents the total primary productivity predicted by the model.

[0111] Furthermore, in one embodiment, the digital twin modeling layer is also used to collect multi-source monitoring data for a preset time period before the current moment when the multi-source monitoring data conforms to environmental mutations or the management model prediction error exceeds a preset threshold, and retain a preset proportion of historical multi-source monitoring data to re-train and update the management model.

[0112] In this embodiment, the trigger conditions are sudden environmental changes (such as heavy rain or high temperatures) or prediction errors exceeding a threshold (MSE > 2.0). When retraining and updating the management model, the latest 10 minutes of sensor data are collected, the LSTM model is retrained (incremental learning, retaining 80% of historical data), and the physical model parameters are updated (adjusting the Q10 value).

[0113] Furthermore, in one embodiment, a three-dimensional visual monitoring management platform is constructed based on the management model, specifically including the following steps:

[0114] A visual monitoring and management platform for urban wetlands is constructed by combining management models, digital elevation models, and multispectral image data.

[0115] In this embodiment, 3D visualization includes platform construction and data overlay. The platform construction is based on the CesiumJS engine and integrates DEM (Digital Elevation Model) and drone imagery. Data overlay includes real-time display of CO2 concentration heat maps (color gradient: blue - low carbon sinks, red - high carbon sinks), vegetation cover, and water quality.

[0116] Furthermore, in one embodiment, the intelligent analysis layer includes:

[0117] The real-time carbon sink calculation module is used to obtain the integrated carbon sink data output by the management model and process it to obtain the regional carbon sink capacity assessment results.

[0118] The anomaly detection module is used to compare the integrated carbon sink data with the historical baseline data, and mark the abnormal area and alarm level when the comparison result is abnormal.

[0119] The strategy optimization service module processes abnormal areas and alarm levels based on the strategy generation model to generate response strategies. It also collects feedback data after the response strategy is implemented to update the strategy generation model.

[0120] In this embodiment, the intelligent analysis layer is the decision-making core of the digital twin system, which realizes the precise management of wetland carbon sinks through a four-step closed loop of real-time calculation, anomaly detection, strategy recommendation, and execution feedback.

[0121] The real-time carbon sink capacity calculation input includes the fused carbon sink data stream (with a 10-minute temporal resolution) output by the digital twin modeling layer, and the output includes a regional carbon sink capacity assessment (kg CO2 / ha / day). The real-time carbon sink capacity calculation steps include spatial gridding, dynamic aggregation of carbon fluxes, and functional area carbon sink summary.

[0122] Spatial gridding involves dividing the wetland into multiple 10m×10m grids (based on drone DEM data). Each grid is assigned a unique ID and associated with its functional zone (e.g., emergent plant area, open water area).

[0123] The dynamic aggregation of carbon flux includes performing IDW (Inverse Distance Weighted interpolation) on the CO2 concentration and NDVI data within the grid. The IDW formula is shown in the following formula (6):

[0124]

[0125] in, represents the estimated value of the interpolation i, d ij It represents the Euclidean distance between grid i and sensor j, and the power of 2 is an empirical value.

[0126] The functional area carbon sink summary includes the generation of a visual heat map to render the grid data into a heat map, with the color mapping rule: blue (<1kg / ha / day) - green (1-3) - yellow (3-5) - red (>5).

[0127] Furthermore, the input for anomaly detection includes real-time carbon sink data streams and historical baseline data (mean ± 2σ (standard deviation) for the same period over the past 30 days). The output includes abnormal area markings and alarm levels (yellow / orange / red). The anomaly detection process includes sliding window statistics and multi-level alarm triggering.

[0128] Sliding window statistics include calculating the carbon sink value of each grid within the current window (within 1 hour) every 10 minutes and calculating its Z-Score value (the Z-score uses the standard deviation as a ruler to measure the distance of a raw score from the mean). The Z-Score value formula is shown in the following formula (7):

[0129]

[0130] Among them, μ t ,σ t Indicates the historical mean and standard deviation of the current period (such as 14:00-14:10).

[0131] The multi-level alarm trigger includes detecting abnormal carbon sink values ​​(set below the historical average), triggering three levels of warning (yellow - warning, orange - warning, red - emergency). Among them, the alarm level is yellow when -2σ≤Z≤-1σ, the alarm level is orange when -3σ≤Z≤-2σ, and the alarm level is red when Z<-3σ or Z<-2σ three times in a row.

[0132] Furthermore, the root cause analysis of the abnormal data is carried out, including correlation analysis of the environmental parameters of the abnormal grid (water level, temperature, pH value and other multi-dimensional features), and the contribution of each factor is calculated using the Bayesian network.

[0133] Furthermore, the overall process of strategy recommendation is as follows Figure 3 As shown, it includes rule engine matching, reinforcement learning dynamic optimization, policy conflict resolution, and policy execution and feedback.

[0134] The rule engine matches based on a preset policy library (policy actions implemented when conditions are met). The input is anomaly detection results, real-time environment data, and the management rule library. The output is a list of optimized policies (sorted by priority). The rule engine matches by loading and matching rules in the preset policy library.

[0135] Reinforcement learning dynamic optimization includes environmental modeling. Specifically, the wetland is considered as an MDP (Markov Decision Process). The state space includes temperature, humidity, CO2 concentration, NDVI, EVI, light intensity, water depth, water pH, GPP, R eco The action space includes human intervention measures, such as water level regulation: opening a gate to release water (-0.1m) or closing a gate to store water (+0.1m). Another example is vegetation management: replanting carbon-tolerant plants or harvesting withered vegetation. Another example is nutritional intervention: adding slow-release fertilizers or spraying microbial agents. Another example is artificial oxygenation.

[0136] The DQN (Deep Q-Network, reinforcement learning) algorithm is used to train a dynamic optimization strategy. The reward function is shown in the following formula (8):

[0137] R=α·ΔC seq -β·Cost (8)

[0138] Where, ΔC seq represents the carbon sink increment, Cost represents the implementation cost, α=0.7, β=0.3 are the correlation coefficients.

[0139] Strategy conflict resolution involves using MAUT (Multiattribute Utility Theory) to make decisions when the rule engine and DQN recommendations conflict (e.g., the rule recommends watering but DQN recommends fertilizing). The decision formula is shown in the following formula (9):

[0140] U(A i )=0.6·u carbon (A i )+0.3·u cost (A i )+0.1·u risk (A i ) (9)

[0141] Among them, u carbon represents the carbon sink gain utility (normalized to 0-1), u cost represents cost-effectiveness (inversely proportional to budget), u risk Represents the ecological risk utility (based on the historical event database).

[0142] The input of strategy execution and feedback is the optimal strategy (e.g., "replenish water to 0.5m"), and the output is the execution effect evaluation and model iteration. When the strategy is executed, a command is sent to the IoT device. Within 2 hours after the strategy is implemented, data from the target area is collected every 20 minutes and the carbon sink value is calculated. The improvement rate is calculated based on the average carbon sink value. The formula for the improvement rate is shown in the following formula (10):

[0143]

[0144] Where η represents the improvement rate, C t represents the average carbon sink value 2 hours before the implementation of the strategy, C t+2h It represents the average carbon sequestration value 2 hours after the implementation of the strategy.

[0145] The evaluation criteria are: when η ≥ 15%, the strategy is successful; when 5% ≤ η < 15%, it is partially effective; when η < 5%, the strategy fails.

[0146] When the strategy succeeds, the state-action-reward tuple is stored in the DQN experience replay pool. When the strategy fails, the rule engine is triggered to summarize the logic and generate new rules.

[0147] In one specific example, the carbon sink value in a certain area fell below the baseline by 2σ for three consecutive times, resulting in an orange alert. Correlation analysis revealed that the water level had dropped to 0.2m (normally 0.5m). The rule engine recommended "replenish water to 0.5m." The DQN suggested "replenish water to 0.4m and apply slow-release fertilizer." The MAUT decision opted for pure watering (due to the high risk weight of fertilization). Six hours after the watering, the carbon sink rebounded by 18%, updating the DQN's "fertilize" action value.

[0148] Furthermore, in one embodiment, the interactive application layer is developed based on the Vue.js framework and includes features such as a real-time dashboard displaying carbon sink totals, anomaly alerts, and strategy recommendations. Historical data backtracking allows for viewing carbon sink trends over time.

[0149] On the second aspect, the embodiments of the present application also provide an urban wetland carbon sink management method based on digital twins.

[0150] In one embodiment, referring to Figure 4 , Figure 4 This is a flow chart of an embodiment of the urban wetland carbon sequestration management method based on digital twins of this application. Figure 4 As shown in Figure 2, urban wetland carbon sequestration management methods based on digital twins include:

[0151] Step S1: Configure hardware sensors, drones, satellite receiving stations, and weather stations to collect multi-source monitoring data.

[0152] Step S2: Transmit and store the multi-source monitoring data to the cloud platform.

[0153] Step S3: Combine multi-source monitoring data, gross primary productivity (GPP), and ecosystem respiration (R) eco A management model for urban wetlands was constructed, and the Net Ecosystem Productivity (NEP) equation and the temperature-respiration equation were added as physical constraints to the loss function. A 3D visualization monitoring and management platform was constructed based on the management model.

[0154] Step S4: When the visual monitoring management platform detects abnormal data, a response strategy is generated based on the strategy generation model, and feedback data after the implementation of the response strategy is collected to update the strategy generation model.

[0155] Among them, the above-mentioned urban wetland carbon sink management method based on digital twin is used to realize the role of each functional unit in the urban wetland carbon sink management system based on digital twin, and its functions and implementation process will not be described here one by one.

[0156] In summary, a digital twin-based urban wetland carbon sink management system and method, built on the integration of multi-source sensor networks, AI algorithms, and digital twin technology, aims to address current challenges in urban wetland carbon sink management, such as low monitoring accuracy, data fragmentation, and inefficient management, by constructing a virtual mapping of the physical world. This approach provides efficient and reliable decision support for wetland carbon sink management, improving its accuracy, real-time nature, and intelligence, ultimately contributing to the achievement of carbon neutrality goals.

[0157] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0158] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit the "first", "second" and "third" to different types.

[0159] In the description of the embodiments of this application, the words "exemplary," "for example," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary," "for example," or "for example" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary," "for example," or "for example" is intended to present the relevant concepts in a concrete manner.

[0160] In the description of the embodiments of the present application, unless otherwise specified, " / " means or. For example, A / B can mean A or B. The "and / or" in the text is merely a description of the association relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, "plurality" means two or more than two.

[0161] In some processes described in the embodiments of the present application, multiple operations or steps are included that appear in a specific order. However, it should be understood that these operations or steps may not be performed in the order in which they appear in the embodiments of the present application or may be performed in parallel. The sequence numbers of the operations are only used to distinguish between different operations, and the sequence numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations or steps may be performed in sequence or in parallel, and these operations or steps may be combined.

[0162] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course, by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for enabling a terminal device to execute the methods of each embodiment of the present application.

[0163] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An urban wetland carbon sequestration management system based on digital twins, characterized by: The system comprises: The data perception layer includes hardware sensors, drones, satellite receiving stations, and weather stations, which are used to collect multi-source monitoring data; The data transmission layer is used to transmit and store multi-source monitoring data to the cloud platform; The digital twin modeling layer is used to combine multi-source monitoring data, gross primary productivity (GPP), and ecosystem respiration (R) eco Constructing a management model for urban wetlands, and adding Net Ecosystem Productivity (NEP) equation information and Temperature-Respiration equation information as physical constraints in the loss function; and also constructing a three-dimensional visualization monitoring and management platform based on the management model; The intelligent analysis layer is used to generate a response strategy based on the strategy generation model when the visual monitoring management platform detects abnormal data, and collect feedback data after the response strategy is implemented to update the strategy generation model.

2. The urban wetland carbon sequestration management system based on digital twin according to claim 1, characterized in that: The multi-source monitoring data includes sensor data collected by sensors, multispectral image data collected by drones, remote sensing data collected by satellites, and meteorological data collected by weather stations; The sensor data includes CO2 concentration data, temperature and humidity data, light intensity data, water level depth data, water quality data, vegetation image data, soil moisture data, and water pH value data; The multispectral image data includes the Normalized Difference Vegetation Index (NDVI) and the Wide Area Vegetation Index (EVI); The remote sensing data includes NDVI, EVI, and land surface temperature; The meteorological data includes rainfall data, wind speed data, and temperature data.

3. The urban wetland carbon sequestration management system based on digital twin according to claim 1, characterized in that: The digital twin modeling layer includes: The data preprocessing module is used to clean and extract features from multi-source monitoring data to obtain multi-source modeling data; ST-Fusion module, used for spatiotemporal alignment of multi-source modeling data; Physics-AI model module for calculating GPP and R based on multi-source modeling data eco , based on multi-source modeling data, GPP, and R eco Training management model; The three-dimensional rendering module is used to construct a three-dimensional visual monitoring and management platform according to the management model.

4. The urban wetland carbon sequestration management system based on digital twin according to claim 1, characterized in that: The NEP equation information includes NEP not greater than GPP, R eco Not less than 0, and GPP not less than 0; The temperature-respiration equation information includes the Arrhenius equation.

5. The urban wetland carbon sequestration management system based on digital twin according to claim 1, characterized in that: The digital twin modeling layer is also used to collect multi-source monitoring data for a preset time period before the current moment when the multi-source monitoring data meets the environmental mutation or the management model prediction error exceeds a preset threshold, and retain a preset proportion of historical multi-source monitoring data to re-train and update the management model.

6. The urban wetland carbon sequestration management system based on digital twin according to claim 1, characterized in that: The construction of the three-dimensional visual monitoring and management platform according to the management model specifically includes the following steps: A visual monitoring and management platform for urban wetlands is constructed by combining management models, digital elevation models, and multispectral image data.

7. The urban wetland carbon sequestration management system based on digital twin according to claim 1, characterized in that: The intelligent analysis layer includes: A real-time carbon sink calculation module is used to obtain the integrated carbon sink data output by the management model and process it to obtain regional carbon sink capacity assessment results; An anomaly detection module, used to compare the fused carbon sink data with historical baseline data, and mark the abnormal area and alarm level when the comparison result is abnormal; The strategy optimization service module is used to process abnormal areas and alarm levels based on the strategy generation model to generate a response strategy; it is also used to collect feedback data after the implementation of the response strategy to update the strategy generation model.

8. The urban wetland carbon sequestration management system based on digital twin according to claim 7, characterized in that: Generating a response strategy based on a strategy generation model and collecting feedback data after the response strategy is implemented to update the strategy generation model specifically includes the following steps: generating a plurality of response strategies based on a strategy generation model, each response strategy including at least one regulatory measure; Cost-effectiveness calculations were conducted for multiple preliminary response strategies, including incremental carbon sinks, real-time costs, and ecological risks. Cost-effectiveness was positively correlated with incremental carbon sinks and negatively correlated with real-time costs and ecological risks. sorting the multiple preliminary response strategies according to cost-effectiveness, and collecting the carbon sequestration improvement rate of each preliminary response strategy from high to low in terms of cost-effectiveness; The action value of the regulatory measures is evaluated according to the carbon sink improvement rate, and the strategy generation model is updated according to the updated action value.

9. The urban wetland carbon sequestration management system based on digital twin according to claim 7, characterized in that: The strategy generation model includes a preset rule database and a reinforcement learning DQN model, wherein the preset rule database contains multiple response strategies associated with abnormal data.

10. A method for urban wetland carbon sequestration management based on digital twins, characterized in that: The method comprises: Configure hardware sensors, drones, satellite receiving stations, and weather stations to collect multi-source monitoring data; Transmit and store multi-source monitoring data to the cloud platform; Combining multi-source monitoring data, gross primary productivity (GPP), and ecosystem respiration (R) eco Constructing an urban wetland management model and adding Net Ecosystem Productivity (NEP) equation information and Temperature-Respiration equation information as physical constraints to the loss function; building a three-dimensional visualization monitoring and management platform based on the management model; When the visual monitoring management platform detects abnormal data, a response strategy is generated based on the strategy generation model, and feedback data after the response strategy is implemented is collected to update the strategy generation model.

Images