Numerical prediction method and platform for carbon flux of coastal zone ecosystem and storage medium

Abstract

The invention relates to the technical field of data processing, in particular to a coastal zone ecological system carbon flux numerical prediction method and platform and a storage medium, and the method comprises the steps: obtaining historical carbon flux monitoring data of a coastal zone from a first database, obtaining a land use plan and an emission reduction plan related to the coastal zone, and obtaining a coastal zone ecological system carbon flux numerical prediction result; building future carbon flux monitoring data; creating a first prediction model and a second prediction model based on a neural network, training the first prediction model based on the first carbon flux monitoring data, and training the second prediction model based on the second carbon flux monitoring data; and creating a third prediction model based on the neural network, combining the training data of the first prediction model and the training data of the second prediction model to form a data matrix, training the third prediction model, and outputting a prediction result corresponding to prediction time as a final prediction result of the coastal zone ecosystem carbon flux value by the third prediction model. According to the method, the carbon flux of the coastal zone ecosystem can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically to a numerical prediction method, platform and storage medium for carbon flux of a coastal ecosystem. Background Art

[0002] Coastal ecosystems play an important role in the global carbon cycle, and accurate prediction of their carbon flux is of great significance for assessing global climate change and formulating carbon reduction strategies. However, existing carbon flux prediction methods mostly rely on complex model calculations and field monitoring data, which have problems such as high cost, complex operation, and limited prediction accuracy.

[0003] Similar prior art includes a Chinese patent application with publication number CN118917559A, which relates to a method for assessing the carbon sink enhancement potential in a near-coastal area, comprising the following steps: S1: collecting multi-dimensional marine environmental data to form a preliminary data set;

[0004] S2: Conduct in-depth analysis of the carbon sink increase characteristics and long-term change trends, and identify factors that affect the offshore carbon sink pattern; S3: Construct a static relationship model between the carbon sink pattern and the factors; S4: Construct a dynamic relationship model of the carbon sink pattern; S5: Combine the static relationship model with the dynamic relationship model to conduct a comprehensive assessment to determine the carbon sink spatial carrying capacity of the coastal area; S6: Based on the spatial carrying capacity assessment results obtained in S5, determine the carbon sink increase potential of each region. This invention can provide more accurate carbon sink management and prediction by integrating static and dynamic models to evaluate the carbon sink capacity of the coastal area.

[0005] Similar prior art includes a Chinese patent application with publication number CN118890604A, which discloses a multimodal environment fusion perception method and system, including: based on the multimodal environment fusion perception needs in a wide-area scenario, setting nodes with multiple functional attributes, and dividing strong nodes and weak nodes according to the perception performance of the nodes; creating a hierarchical perception network through the deployment combination of strong nodes and weak nodes, collecting environmental data, and performing data preprocessing, analysis, prediction and storage in sequence based on the node computing power; establishing communication channels between nodes, and optimizing the perception performance and perception results of the hierarchical perception network based on the data transmission and compensation mechanism between strong nodes and weak nodes; making response decisions based on the perception results, and generating and executing control commands. The invention speeds up decision-making by compensating and optimizing data by strong nodes.

[0006] However, neither of the above two patent applications took into account the prediction of carbon flux in coastal ecosystems, let alone the accuracy of carbon flux prediction. Summary of the invention

[0007] The purpose of the present invention is to provide a numerical prediction method based on the carbon flux of coastal ecosystems, which can utilize historical monitoring data and future land use plans, construct a prediction model through a neural network, and realize accurate prediction of the carbon flux of coastal ecosystems.

[0008] In order to achieve the above-mentioned object of the invention, the present invention provides a numerical prediction method of carbon flux of coastal ecosystem, which is implemented by executing the following steps:

[0009] Step S1: Acquire historical carbon flux monitoring data of a coastal zone from a first database as first carbon flux monitoring data, acquire a land use plan and an emission reduction plan related to the coastal zone, and construct future carbon flux monitoring data based on the land use plan and the emission reduction plan as second carbon flux monitoring data, and store the data in a second database;

[0010] Step S2: creating a first prediction model and a second prediction model based on a neural network, normalizing the first carbon flux monitoring data and the second carbon flux monitoring data respectively, training the first prediction model based on the normalized first carbon flux monitoring data, and obtaining the trained first prediction model, training the second prediction model based on the normalized second carbon flux monitoring data, and obtaining the trained second prediction model;

[0011] Step S3: Create a third prediction model based on a neural network, combine the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and use the data in the data matrix as training data to train the third prediction model, calculate the first type of ecological data and the second type of ecological data corresponding to the prediction time and input them into the third prediction model as input data, the third prediction model outputs the prediction result corresponding to the prediction time, and uses the prediction result as the final prediction result of the carbon flux value of the coastal ecosystem.

[0012] As a preferred technical solution of the present invention, the step S3 further includes:

[0013] Step S4: Obtain an emission reduction target and compare it with the prediction result. When the prediction result is less than the emission reduction target, issue a warning message, and formulate a new land use plan in combination with the land use plan corresponding to the prediction result. Calculate new first type ecological data and second type ecological data based on the new land use plan, and input them into the third prediction model again to obtain a new prediction result, determine whether the new prediction result is greater than the emission reduction target, and continue to adjust the new land use plan when it is less than the emission reduction target until the prediction result of the third prediction model is less than or equal to the emission reduction target.

[0014] As a preferred technical solution of the present invention, the carbon flux monitoring data in step S1 includes:

[0015] Carbon flux monitoring data refers to first-type ecological data, second-type ecological data and carbon flux, and the carbon flux monitoring data includes the first carbon flux monitoring data and the second carbon flux monitoring data, wherein the first-type ecological data refers to mangroves, salt marshes, seagrass beds and large seaweeds that absorb carbon elements and the corresponding coverage areas, and the second-type ecological data refers to the number of animals that release carbon elements and the area covered by human activities;

[0016] Multiple remote sensing images of the coastal zone are periodically acquired within a preset time period, and ecological data and corresponding areas or quantities of different types of coastal zones on each remote sensing image are identified and stored in a first database in chronological order.

[0017] As a preferred technical solution of the present invention, in step S2, after normalizing the first carbon flux monitoring data and before training the first prediction model, the step further includes:

[0018] Based on a preset time length of the first carbon flux monitoring data, a prediction time is determined, and sequence data of historical carbon flux monitoring data based on the preset time length is created for the prediction time, and the sequence data is paired with the prediction time. The sequence data serves as input data of the first prediction model, the prediction time serves as the time corresponding to a prediction target, and the carbon flux corresponding to the prediction time serves as an output result of the first prediction model. The sequence data and the carbon flux corresponding to the prediction time constitute training data for the first prediction model, wherein the sequence data is data arranged in chronological order, and the prediction target refers to the carbon flux corresponding to the prediction time.

[0019] As a preferred technical solution of the present invention, in step S1, constructing future carbon flux monitoring data includes:

[0020] Based on big data or land use planning documents issued by relevant departments, the land use plan related to the coastal area is obtained. Based on the land use plan, the future first type of ecological data and the future second type of ecological data and the corresponding future carbon flux corresponding to each preset time period in the coastal area are calculated, so as to construct future carbon flux monitoring data and store them in the second database.

[0021] As a preferred technical solution of the present invention, in step S3, the training data of the first prediction model and the training data of the second prediction model are combined to form a data matrix, including:

[0022] Acquire the training data of the first prediction model, create a data matrix with time, the first type of ecological data, the second type of ecological data and the carbon flux as titles and the corresponding data as values, acquire the training data of the second prediction model, align the training data of the second prediction model with the training data of the first prediction model according to the time, and add them to the data matrix to form a comprehensive data matrix containing the first carbon flux monitoring data and the second carbon flux monitoring data, use the first type of ecological data, the second type of ecological data, the future first type of ecological data and the future second type of ecological data in the data matrix as input data of the third prediction model, and use the first prediction result and the second prediction result in the data matrix as output data of the third prediction model to train the third prediction model.

[0023] As a preferred technical solution of the present invention, in step S2, training the first prediction model based on the normalized first carbon flux monitoring data includes:

[0024] Acquire the normalized first carbon flux monitoring data and use it as training data for the first prediction model, determine whether the data in the training data is uniform, and when the data in the training data is not uniform, divide the training data into first learning data and second learning data, and generate more new first learning data through data enhancement technology based on the data features of the first learning data and the second learning data;

[0025] The new first learning data is input into a judgment model, the judgment model outputs a judgment result, and when the value of the judgment result is greater than or equal to a first threshold and less than or equal to a second threshold, the new first learning data is used as the third learning data, the similarity between the third learning data is calculated, and the data with the similarity greater than or equal to the third threshold are clustered together to generate a clustering result of the third learning data;

[0026] The weights and parameters of the first prediction model are adjusted based on the clustering result of the third learning data, and the first prediction model is trained using the third learning data and the second learning data.

[0027] As a preferred technical solution of the present invention, judging whether the data in the training data is uniform, and when the data in the training data is not uniform, dividing the training data into first learning data and second learning data, comprises:

[0028] Calculate the number of the training data that contains the first type of ecological data features and the number of the training data that does not contain the first type of ecological data features, and when the number that does not contain the first type of ecological data features is less than N times the number that contains the first type of ecological data features, determine that the data in the training data is uneven, where N represents a positive integer greater than or equal to 2;

[0029] When the data in the training data is uneven, the training data is divided into first learning data and second learning data, wherein the first learning data refers to data that does not contain the first type of ecological data characteristics, and the second learning data refers to data that contains the first type of ecological data characteristics.

[0030] The present invention also provides a numerical prediction platform for carbon flux of coastal ecosystem as described above, comprising the following modules:

[0031] An acquisition unit is used to acquire historical carbon flux monitoring data of a coastal zone from a first database as first carbon flux monitoring data, acquire a land use plan and an emission reduction plan related to the coastal zone, and based on the land use plan and the emission reduction plan, construct future carbon flux monitoring data as second carbon flux monitoring data, and store the data in a second database;

[0032] A training unit, used to create a first prediction model and a second prediction model, normalize the first carbon flux monitoring data and the second carbon flux monitoring data respectively, train the first prediction model based on the normalized first carbon flux monitoring data, and obtain the trained first prediction model, train the second prediction model based on the normalized second carbon flux monitoring data, and obtain the trained second prediction model;

[0033] An integration unit is used to create a third prediction model, combine the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and use the data in the data matrix as training data to train the third prediction model, calculate the first type of ecological data and the second type of ecological data corresponding to the prediction time and input them into the third prediction model as input data, the third prediction model outputs the prediction result corresponding to the prediction time, and uses the prediction result as the final prediction result of the carbon flux value of the coastal ecosystem.

[0034] The present invention also provides a storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a processor, the numerical prediction method, platform and storage medium for carbon flux of coastal ecosystems described above are implemented.

[0035] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0036] 1. The technical solution of the present invention obtains the historical carbon flux monitoring data of the coastal area from the first database as the first carbon flux monitoring data, and trains the first prediction model; obtains the land use plan and emission reduction plan related to the coastal area, and based on the land use plan and emission reduction plan, constructs the future carbon flux monitoring data as the second carbon flux monitoring data, and trains the second prediction model. The first prediction model and the second prediction model are created based on the neural network, and the first carbon flux monitoring data and the second carbon flux monitoring data are normalized respectively. The first prediction model is trained based on the normalized first carbon flux monitoring data, and the trained first prediction model is obtained to identify the trend of carbon flux over time. The second prediction model is trained based on the normalized second carbon flux monitoring data, and the trained second prediction model is obtained to provide an expected perspective to predict the trend of future carbon flux changes.

[0037] 2. The technical solution of the present invention creates a third prediction model, combines the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and captures the trend of carbon energy changes from different perspectives. The data in the data matrix is ​​used as training data to train the third prediction model to integrate the first prediction model and the second prediction model, consider the combined impact of historical trends and future plans, consider the combined impact of historical trends and future plans, combine the advantages of these two models, and use the different perspectives they provide to improve the accuracy of the prediction. The first type of ecological data and the second type of ecological data corresponding to the prediction time are calculated and input into the third prediction model as input data. The third prediction model outputs the prediction result corresponding to the prediction time, and the prediction result is used as the final prediction result of the carbon flux value of the coastal ecosystem. Since the third prediction model combines the historical carbon flux monitoring data and the carbon flux monitoring data of the future plan, the output result of the third prediction model makes the numerical prediction of the carbon flux of the coastal ecosystem more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0039] Figure 1 It is a flow chart of the steps of the numerical prediction method of carbon flux of coastal ecosystem in the present invention;

[0040] Figure 2 This is a structural diagram of the numerical prediction platform for carbon flux in coastal ecosystems in the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0042] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0043] Coastal ecosystems play an important role in the global carbon cycle, and accurate prediction of their carbon flux is of great significance for assessing global climate change and formulating carbon reduction strategies. However, existing carbon flux prediction methods mostly rely on complex model calculations and field monitoring data, which have problems such as complex operation and limited prediction accuracy.

[0044] In view of the above technical problems, the present invention proposes Figure 1 The numerical prediction method of carbon flux in coastal ecosystems is shown in the figure and is implemented by executing the following steps:

[0045] Step S1: Obtain historical carbon flux monitoring data of the coastal area from the first database as first carbon flux monitoring data, obtain land use plans and emission reduction plans related to the coastal area, and based on the land use plans and emission reduction plans, construct future carbon flux monitoring data as second carbon flux monitoring data, and store them in the second database.

[0046] Specifically, the carbon flux of the coastal ecosystem includes various carbon exchange processes within the ecosystem, such as photosynthesis and respiration of plants, and respiration of soil microorganisms, which will cause carbon elements to be exchanged between the ecosystem and the atmosphere. By predicting the carbon flux of the coastal ecosystem, we can understand the absorption and emission of carbon by the ecosystem, and then evaluate climate change. The carbon flux monitoring data of the coastal area in the past ten or twenty years are collected and stored in the first database as training data for the first prediction model. The data mainly comes from field measurements, satellite remote sensing or historical climate records. In addition, the change of carbon flux in the coastal ecosystem is also affected by many factors, including climate, topography, vegetation type, soil conditions and human activities. Changes in these factors will lead to changes in the carbon exchange process within the ecosystem, thereby affecting the change of carbon flux; at the same time, these factors are affected by the land use plan and emission reduction targets formulated by the local government or environmental management agency. Therefore, it is also necessary to obtain land use plans and emission reduction targets related to the coastal area from the local government or environmental management agency, including urban planning, agricultural expansion, and establishment of protected areas, to calculate future carbon flux monitoring data and store them in the second database as training data for the second prediction model. Among them, future carbon flux can be estimated in combination with emission reduction plans formulated by relevant policies.

[0047] Step S2: Create a first prediction model and a second prediction model based on a neural network, normalize the first carbon flux monitoring data and the second carbon flux monitoring data respectively, train the first prediction model based on the normalized first carbon flux monitoring data, and obtain the trained first prediction model, train the second prediction model based on the normalized second carbon flux monitoring data, and obtain the trained second prediction model.

[0048] Specifically, a prediction model is constructed based on a neural network, and a long short-term memory network (LSTM) is selected as the first prediction model. The LSTM model can capture the long-term dependencies between data, and historical carbon flux monitoring data usually have time series characteristics. Therefore, the selection of the LSTM model can effectively process such data and identify the temporal trend of carbon flux. However, when relying solely on past carbon flux monitoring data, it may not be possible to fully capture future changes in carbon flux, because past carbon flux monitoring data may not fully reflect future external influences, such as land changes, resulting in a reduction in mangrove area, etc. The reduction in mangrove area directly affects carbon flux. Therefore, considering future land use plans and emission reduction targets is an important factor in predicting future carbon fluxes. The second prediction model provides an expectation-based perspective by considering future land use plans and expected emission reduction targets in the coastal zone to predict future trends in carbon flux changes. Therefore, a multi-layer perceptron (MLP) is selected as the second prediction model. The MLP model is good at capturing nonlinear relationships between input data. Future planned carbon flux monitoring data will contain multiple variables, and the relationship between these variables and future carbon fluxes may mostly be nonlinear. Unlike LSTM, the MLP model does not have inherent time series processing capabilities, so it is more suitable for processing one-time future planned carbon flux monitoring data that is independent of time. Future land use plans, policy changes and other factors may affect changes in carbon flux. The second prediction model incorporates these expected changes to make the prediction results more adaptable to possible future situations. At the same time, choosing the MLP model as the second prediction model usually predicts results faster than the recurrent neural network.

[0049] Step S3: Create a third prediction model based on a neural network, combine the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and use the data in the data matrix as training data to train the third prediction model, calculate the first type of ecological data and the second type of ecological data corresponding to the prediction time and input them into the third prediction model as input data, the third prediction model outputs the prediction result corresponding to the prediction time, and uses the prediction result as the final prediction result of the carbon flux value of the coastal ecosystem.

[0050] Specifically, the first prediction model is based on historical carbon flux monitoring data, and the second prediction model is based on future planned carbon flux monitoring data. They capture the trend of carbon energy changes from different perspectives. At this time, it is also necessary to create a new multi-layer perceptron model, namely the third prediction model, to integrate the first prediction model and the second prediction model, consider the combined impact of historical trends and future plans, combine the advantages of these two models, and use the different perspectives they provide to improve the accuracy of the prediction. Specifically, the training data of the two models are combined into a data matrix, including historical carbon flux monitoring data and future planned carbon flux monitoring data, and the third prediction model is trained using the data matrix. When the carbon flux monitoring data corresponding to the prediction time is input, the third model outputs the carbon flux prediction result corresponding to the prediction time. Since the third prediction model combines historical carbon flux monitoring data and future planned carbon flux monitoring data, the output result of the third prediction model makes the numerical prediction of carbon flux of coastal ecosystems more accurate.

[0051] Therefore, through the coordination of the above steps, combined with historical carbon flux monitoring data and future planned carbon flux monitoring data, the accuracy of numerical prediction of carbon flux in coastal ecosystems can be effectively improved.

[0052] Furthermore, after the above step S3, the following steps are further included:

[0053] Step S4: Obtain the emission reduction target and compare it with the prediction result. When the prediction result is less than the emission reduction target, issue a warning message, and formulate a new land use plan in combination with the land use plan corresponding to the prediction result. Calculate new first type ecological data and second type ecological data based on the new land use plan, and input them into the third prediction model again to obtain a new prediction result, determine whether the new prediction result is greater than the emission reduction target, and continue to adjust the new land use plan when it is less than the emission reduction target until the prediction result of the third prediction model is less than or equal to the emission reduction target.

[0054] Specifically, the prediction results can also be used as a guide for formulating future land planning and planning the ecological environment in advance. Specific emission reduction targets are obtained from government environmental policies or sustainable development goals, and the final prediction results of the carbon flux values ​​of the coastal ecosystem obtained in step S3 are compared with the emission reduction targets. If the prediction results exceed the emission reduction targets, the system automatically issues an early warning message to prompt managers to take measures to reduce carbon emissions. At the same time, a new land use plan or land planning can be formulated in combination with the early warning information and the current land use plan, including increasing the vegetation planting area, improving wetland protection measures, or reducing human activities to reduce carbon emissions. Based on the new land use plan, the first type of ecological data and the second type of ecological data are recalculated. The new ecological data is input into the third prediction model to obtain a new prediction result. If the new prediction result still exceeds the emission reduction target, the land use plan is adjusted continuously, and the prediction process is repeated until the prediction result meets or is lower than the emission reduction target. This method ensures that the carbon flux prediction results of the coastal ecosystem can meet or be lower than the established emission reduction target, which helps to achieve the regional climate change mitigation goals. At the same time, by continuously adjusting the land use plan to meet the emission reduction target, the resilience and carbon sink function of the coastal ecosystem are enhanced.

[0055] Furthermore, the carbon flux monitoring data in the above step S1 includes:

[0056] Carbon flux monitoring data refers to first-type ecological data, second-type ecological data and carbon flux. Carbon flux monitoring data includes first-type carbon flux monitoring data and second-type carbon flux monitoring data. Among them, first-type ecological data refers to mangroves, salt marshes, seagrass beds and large seaweeds that absorb carbon elements and the corresponding coverage area; second-type ecological data refers to the number of animals that release carbon elements and the area covered by human activities.

[0057] Multiple remote sensing images of the coastal zone are periodically acquired within a preset time period, and ecological data and corresponding areas or quantities of different types of coastal zones on each remote sensing image are identified and stored in a first database in chronological order.

[0058] Specifically, satellites or drones are used to periodically acquire remote sensing images of coastal areas within preset time periods, such as one year, two years, five years, and other different time periods. By periodically acquiring remote sensing images, changes in coastal ecosystems can be efficiently monitored without a large number of field surveys, reducing the cost and time of data collection. These remote sensing images can cover visible light, near infrared, and other spectral bands, making it easy to identify different types of ecosystems and monitor their changes. Remote sensing images are analyzed using image processing software and machine learning algorithms to identify first-type ecological data such as mangroves, salt marshes, seagrass beds, and large seaweeds in the images, as well as second-type ecological data such as animal activity areas and human activity coverage areas, thereby improving the accuracy of ecological data identification and ensuring the quality of carbon flux monitoring data. Image recognition belongs to the prior art and will not be repeated here. Through image segmentation, object recognition, and classification techniques, the coverage area or quantity of each ecological type is determined, and specific ecological data are extracted from the processed remote sensing images, including the coverage area of ​​mangroves, salt marshes, seagrass beds, and large seaweeds, as well as the number of animals and the area affected by human activities. The extracted ecological data and their corresponding timestamps are stored in the first database in chronological order to provide basic data for subsequent carbon flux prediction. Among them, carbon flux is closely related to the first type of ecological data and the second type of ecological data, and the relationship between them is learned through the prediction model. By calculating the first type of ecological data and the second type of ecological data in the future time, the carbon flux in the future time is predicted based on the prediction model.

[0059] Furthermore, in the above step S2, after normalizing the first carbon flux monitoring data and before training the first prediction model, the following steps are further included:

[0060] Based on a preset time length of the first carbon flux monitoring data, a prediction time is determined, and sequence data of historical carbon flux monitoring data based on the preset time length is created for the prediction time, and the sequence data is paired with the prediction time. The sequence data is used as input data of the first prediction model, the prediction time is used as the time corresponding to the prediction target, and the carbon flux corresponding to the prediction time is used as the output result of the first prediction model. The sequence data and the carbon flux corresponding to the prediction time constitute the training data of the first prediction model, wherein the sequence data is data arranged in chronological order, and the prediction target refers to the carbon flux corresponding to the prediction time.

[0061] Specifically, a preset time length, such as 3 years, is selected as the time window of the model input data, and a prediction time point, such as XX, is also determined as the future time point we want to predict. The historical carbon flux monitoring data for the three years before XX, that is, the data from XX-3 to XX-1, is extracted from the first database. These data include the coverage area of ​​mangroves, salt marshes, seagrass beds and large seaweeds (first type of ecological data), as well as the number of animals and the coverage area of ​​human activities (second type of ecological data). The extracted 3-year sequence data is paired with the prediction time XX, and XX is determined as the time corresponding to the prediction target, and the carbon flux corresponding to XX is used as the output result of the first prediction model. The sequence data (carbon flux monitoring data from XX-3 to XX-1) and the carbon flux corresponding to the prediction time (XX) constitute the training data set of the first prediction model. The sequence data is arranged in chronological order, and the prediction target refers to the carbon flux corresponding to the prediction time (XX). Before training the model, the sequence data in the training data set is normalized to eliminate the influence of different dimensions and magnitudes so that the data can be compared on the same scale. The first prediction model was trained using the normalized training data set, and the parameters of the LSTM model, such as the number of hidden layers and the number of neurons, were adjusted to optimize the model performance. By using the sequence of historical carbon flux monitoring data, the LSTM model can capture the long-term dependencies in the time series data and improve the accuracy of future carbon flux predictions.

[0062] Furthermore, in the above step S1, constructing future carbon flux monitoring data includes:

[0063] Based on big data or land use planning documents released by relevant departments, the land use plan related to the coastal area is obtained. Based on the land use plan, the future first-type ecological data and future second-type ecological data and the corresponding future carbon flux corresponding to each preset time period in the coastal area are calculated, so as to construct future carbon flux monitoring data and store them in the second database.

[0064] Specifically, by analyzing the land use planning documents obtained from the big data platform or from relevant departments, land use plans related to specific coastal areas are collected, including information such as urban expansion, agricultural development, industrial construction, ecological protection area demarcation, and vegetation planting plans. Determine the preset time period, divide the future time period, and predict the corresponding ecological data. For each preset time period, calculate the new area corresponding to the future first type of ecological data and the future second type of ecological data, and also calculate the corresponding future carbon flux based on the first type of ecological data and the future second type of ecological data, including calculating the carbon absorption potential of vegetation and carbon emissions caused by animal activities and human activities. The calculated future first type of ecological data, future second type of ecological data, and corresponding future carbon flux are integrated into future carbon flux monitoring data and stored in the second database to provide data support for subsequent model training and prediction.

[0065] Furthermore, in the above step S3, the training data of the first prediction model and the training data of the second prediction model are combined to form a data matrix, including:

[0066] The training data of the first prediction model is obtained, and a data matrix is ​​created with time, first type ecological data, second type ecological data and carbon flux as titles and the corresponding data as values, and the training data of the second prediction model is obtained, and the training data of the second prediction model is aligned with the training data of the first prediction model in time, and added to the data matrix to form a comprehensive data matrix containing the first carbon flux monitoring data and the second carbon flux monitoring data, and the first type ecological data, the second type ecological data, the future first type ecological data and the future second type ecological data in the data matrix are used as input data of the third prediction model, and the first prediction results and the second prediction results in the data matrix are used as output data of the third prediction model to train the third prediction model.

[0067] Specifically, the training data of the first prediction model are extracted from the first database, including time, first-type ecological data (coverage area of ​​mangroves, salt marshes, seagrass beds and large seaweeds), second-type ecological data (animal number and human activity coverage area) and carbon flux. A data matrix is ​​created with time as the row index, first-type ecological data, second-type ecological data and carbon flux as column headers, and the corresponding values ​​are filled. The training data of the second prediction model are extracted from the second database, which are future first-type ecological data and second-type ecological data estimated based on future land use plans. The training data of the second prediction model are aligned with the training data of the first prediction model in time and added to the data matrix to form a comprehensive data matrix containing historical and future carbon flux monitoring data. In the comprehensive data matrix, the first-type ecological data, the second-type ecological data, the future first-type ecological data and the future second-type ecological data are defined as the input data of the third prediction model, and the first prediction results and the second prediction results are defined as the output data of the third prediction model. The third prediction model is trained using the comprehensive data matrix. By combining historical and future carbon flux monitoring data, the third prediction model can provide a more comprehensive understanding of the changes in carbon flux in coastal ecosystems and improve the accuracy of predictions.

[0068] Furthermore, in the above step S2, training the first prediction model based on the normalized first carbon flux monitoring data includes:

[0069] The first carbon flux monitoring data after normalization is obtained and used as training data for the first prediction model, and whether the data in the training data is uniform is determined. When the data in the training data is not uniform, the training data is divided into first learning data and second learning data, and more new first learning data is generated through data enhancement technology based on the data features of the first learning data and the second learning data;

[0070] The new first learning data is input into the judgment model, and the judgment model outputs a judgment result. When the value of the judgment result is greater than or equal to the first threshold and less than or equal to the second threshold, the new first learning data is used as the third learning data, the similarity between the third learning data is calculated, and the data with similarity greater than or equal to the third threshold are clustered together to generate a clustering result of the third learning data;

[0071] The weights and parameters of the first prediction model are adjusted based on the clustering result of the third learning data, and the first prediction model is trained using the third learning data and the second learning data.

[0072] Specifically, the first carbon flux monitoring data is normalized so that it has a unified data scale so as to be used for training the first prediction model (LSTM). The normalized training data is analyzed to check whether the training data is uniform. The definition of whether the training data is uniform will be explained below. Uneven data may affect the training effect and prediction accuracy of the model. When the training data is uneven, the training data is divided into first learning data (the part with less feature category data) and second learning data (the part with more feature category data). Based on the data features of the first learning data, more new first learning data are generated through data enhancement technology to increase the data volume of the first learning data. The newly generated first learning data is input into the judgment model. The judgment model can evaluate the authenticity of the newly generated data to judge the closeness of the newly generated learning data to the original first learning data. The closer the judgment result output by the judgment model is to the middle value between the first threshold and the second threshold (the first threshold can be set to 0.45 and the second threshold can be set to 0.55), it means that the judgment model cannot determine whether the newly generated first learning data is actual data or generated virtual data. For example, if the judgment result output by the judgment model is 0.499, it means that the judgment model cannot determine whether the newly generated first learning data is actual data or generated virtual data, and the effective new data is selected as the third learning data. The similarity is calculated for the selected new data (third learning data), and the data with similarity greater than or equal to the third threshold are clustered together to generate clustering results, and the similarity between data points is evaluated to identify and reduce the redundancy and diversity deficiencies in the third learning data to represent the data distribution of the real world, and also enhance the prediction model's recognition of different categories and subcategories in the learning data. According to the clustering results of the third learning data, the weights and parameters of the first prediction model are adjusted. The clustering results can help identify patterns and structures in the data, accelerate the model training process, and improve the performance of the model on new data, especially when the new data is similar to the old data. The first prediction model is trained using the adjusted parameters and the third learning data (clustered data) and the second learning data, which improves the efficiency of the prediction model training and the accuracy of the prediction model prediction.

[0073] Further, judging whether the data in the training data is uniform, and when the data in the training data is not uniform, dividing the training data into first learning data and second learning data, including:

[0074] Calculate the number of training data containing the first type of ecological data features and the number of training data not containing the first type of ecological data features, and when the number not containing the first type of ecological data features is less than N times the number containing the first type of ecological data features, determine that the data in the training data is uneven, where N represents a positive integer greater than or equal to 2;

[0075] When the data in the training data is uneven, the training data is divided into first learning data and second learning data, wherein the first learning data refers to data that does not contain the first type of ecological data features, and the second learning data refers to data that contains the first type of ecological data features.

[0076] Specifically, assuming that there are 2000 training data in total, the number of training data containing the first type of ecological data features is 100, and the number of training data that does not contain the first type of ecological data features is 1900. The number of data containing the first type of ecological data features is much smaller than the number of data not containing the first type of ecological data features, and there is unevenness in the two feature categories. When there are two types of feature category data unevenness in the training data, the output results of the prediction model may be biased towards data with a large number of feature categories, because the data with a large number of feature categories has more samples, and the prediction model is more likely to learn features from these samples, thereby ignoring the data with a small number of feature categories. Therefore, when the data in the training data is uneven, the training data is divided into the first learning data and the second learning data. The purpose of dividing the training data is to make the prediction model training process also focus on the data with a small number of feature categories, so as to more effectively deal with the problem of uneven feature categories in the training data and improve the accuracy of the prediction model. This method is also used to process the training data of the second prediction model.

[0077] The present invention also provides a Figure 2 The numerical prediction platform for carbon flux in coastal ecosystems shown in the figure includes the following modules:

[0078] An acquisition unit is used to acquire historical carbon flux monitoring data of the coastal zone from the first database as first carbon flux monitoring data, acquire a land use plan and an emission reduction plan related to the coastal zone, and construct future carbon flux monitoring data based on the land use plan and the emission reduction plan as second carbon flux monitoring data, and store the data in a second database;

[0079] A training unit, used to create a first prediction model and a second prediction model, normalize the first carbon flux monitoring data and the second carbon flux monitoring data respectively, train the first prediction model based on the normalized first carbon flux monitoring data, and obtain the trained first prediction model, train the second prediction model based on the normalized second carbon flux monitoring data, and obtain the trained second prediction model;

[0080] The integration unit is used to create a third prediction model, combine the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and use the data in the data matrix as training data to train the third prediction model, calculate the first type of ecological data and the second type of ecological data corresponding to the prediction time and input them into the third prediction model as input data, the third prediction model outputs the prediction result corresponding to the prediction time, and uses the prediction result as the final prediction result of the carbon flux value of the coastal ecosystem.

[0081] The present invention also provides a computer storage medium, which stores program instructions. When the program instructions are executed, the device where the computer storage medium is located is controlled to execute the numerical prediction method for the carbon flux of the coastal ecosystem.

[0082] In summary, the technical solution of the present invention obtains the historical carbon flux monitoring data of the coastal area from the first database as the first carbon flux monitoring data to train the first prediction model; obtains the land use plan and emission reduction plan related to the coastal area, and based on the land use plan and emission reduction plan, constructs the future carbon flux monitoring data as the second carbon flux monitoring data to train the second prediction model. The first prediction model and the second prediction model are created based on the neural network, and the first carbon flux monitoring data and the second carbon flux monitoring data are normalized respectively. The first prediction model is trained based on the normalized first carbon flux monitoring data, and the trained first prediction model is obtained to identify the trend of carbon flux over time. The second prediction model is trained based on the normalized second carbon flux monitoring data, and the trained second prediction model is obtained to provide an expected perspective to predict the trend of future carbon flux changes.

[0083] The third prediction model is created based on the neural network, and the training data of the first prediction model and the training data of the second prediction model are combined to form a data matrix to capture the trend of carbon energy changes from different perspectives. The data in the data matrix is ​​used as training data to train the third prediction model to integrate the first prediction model and the second prediction model, consider the comprehensive impact of historical trends and future plans, consider the comprehensive impact of historical trends and future plans, combine the advantages of the two models, and use the different perspectives they provide to improve the accuracy of the prediction. The first type of ecological data and the second type of ecological data corresponding to the prediction time are calculated and input into the third prediction model as input data. The third prediction model outputs the prediction result corresponding to the prediction time, and the prediction result is used as the final prediction result of the carbon flux value of the coastal ecosystem. Since the third prediction model combines the historical carbon flux monitoring data and the carbon flux monitoring data of the future plan, the output result of the third prediction model makes the numerical prediction of the carbon flux of the coastal ecosystem more accurate.

[0084] Through the coordination of the above steps, combined with historical carbon flux monitoring data and future planned carbon flux monitoring data, the accuracy of numerical prediction of carbon flux in coastal ecosystems can be effectively improved.

[0085] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0086] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The above-mentioned program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0087] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

[0089] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A numerical prediction method for carbon flux in coastal ecosystems, characterized in that: The method comprises: Step S1: Acquire historical carbon flux monitoring data of a coastal zone from a first database as first carbon flux monitoring data, acquire a land use plan and an emission reduction plan related to the coastal zone, and construct future carbon flux monitoring data based on the land use plan and the emission reduction plan as second carbon flux monitoring data, and store the data in a second database; Step S2: creating a first prediction model and a second prediction model based on a neural network, normalizing the first carbon flux monitoring data and the second carbon flux monitoring data respectively, training the first prediction model based on the normalized first carbon flux monitoring data, and obtaining the trained first prediction model, training the second prediction model based on the normalized second carbon flux monitoring data, and obtaining the trained second prediction model; Step S3: Create a third prediction model based on a neural network, combine the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and use the data in the data matrix as training data to train the third prediction model, calculate the first type of ecological data and the second type of ecological data corresponding to the prediction time and input them into the third prediction model as input data, the third prediction model outputs the prediction result corresponding to the prediction time, and uses the prediction result as the final prediction result of the carbon flux value of the coastal ecosystem.

2. The method according to claim 1, characterized in that After step S3, the following steps are also included: Step S4: Obtain an emission reduction target and compare it with the prediction result. When the prediction result is less than the emission reduction target, issue a warning message, and formulate a new land use plan in combination with the land use plan corresponding to the prediction result. Calculate new first type ecological data and second type ecological data based on the new land use plan, and input them into the third prediction model again to obtain a new prediction result, determine whether the new prediction result is greater than the emission reduction target, and continue to adjust the new land use plan when it is less than the emission reduction target until the prediction result of the third prediction model is less than or equal to the emission reduction target.

3. The method according to claim 1, characterized in that The carbon flux monitoring data in step S1 includes: Carbon flux monitoring data refers to first-type ecological data, second-type ecological data and carbon flux, and the carbon flux monitoring data includes the first carbon flux monitoring data and the second carbon flux monitoring data, wherein the first-type ecological data refers to mangroves, salt marshes, seagrass beds and large seaweeds that absorb carbon elements and the corresponding coverage areas, and the second-type ecological data refers to the number of animals that release carbon elements and the area covered by human activities; Multiple remote sensing images of the coastal zone are periodically acquired within a preset time period, and ecological data and corresponding areas or quantities of different types of coastal zones on each remote sensing image are identified and stored in a first database in chronological order.

4. The method according to claim 1, characterized in that: In the step S2, after normalizing the first carbon flux monitoring data and before training the first prediction model, the step further includes: Based on a preset time length of the first carbon flux monitoring data, a prediction time is determined, and sequence data of historical carbon flux monitoring data based on the preset time length is created for the prediction time, and the sequence data is paired with the prediction time. The sequence data serves as input data of the first prediction model, the prediction time serves as the time corresponding to a prediction target, and the carbon flux corresponding to the prediction time serves as an output result of the first prediction model. The sequence data and the carbon flux corresponding to the prediction time constitute training data for the first prediction model, wherein the sequence data is data arranged in chronological order, and the prediction target refers to the carbon flux corresponding to the prediction time.

5. The method according to claim 1, characterized in that In step S1, constructing future carbon flux monitoring data includes: Based on big data or land use planning documents issued by relevant departments, the land use plan related to the coastal area is obtained. Based on the land use plan, the future first type of ecological data and the future second type of ecological data and the corresponding future carbon flux corresponding to each preset time period in the coastal area are calculated, so as to construct future carbon flux monitoring data and store them in the second database.

6. The method according to claim 5, characterized in that In step S3, the training data of the first prediction model and the training data of the second prediction model are combined to form a data matrix, including: Acquire the training data of the first prediction model, create a data matrix with time, the first type of ecological data, the second type of ecological data and the carbon flux as titles and the corresponding data as values, acquire the training data of the second prediction model, align the training data of the second prediction model with the training data of the first prediction model according to the time, and add them to the data matrix to form a comprehensive data matrix containing the first carbon flux monitoring data and the second carbon flux monitoring data, use the first type of ecological data, the second type of ecological data, the future first type of ecological data and the future second type of ecological data in the data matrix as input data of the third prediction model, and use the first prediction result and the second prediction result in the data matrix as output data of the third prediction model to train the third prediction model.

7. The method according to claim 1, characterized in that In the step S2, training the first prediction model based on the normalized first carbon flux monitoring data includes: Acquire the normalized first carbon flux monitoring data and use it as training data for the first prediction model, determine whether the data in the training data is uniform, and when the data in the training data is not uniform, divide the training data into first learning data and second learning data, and generate more new first learning data through data enhancement technology based on the data features of the first learning data and the second learning data; The new first learning data is input into a judgment model, the judgment model outputs a judgment result, and when the value of the judgment result is greater than or equal to a first threshold and less than or equal to a second threshold, the new first learning data is used as the third learning data, the similarity between the third learning data is calculated, and the data with the similarity greater than or equal to the third threshold are clustered together to generate a clustering result of the third learning data; The weights and parameters of the first prediction model are adjusted based on the clustering result of the third learning data, and the first prediction model is trained using the third learning data and the second learning data.

8. The method according to claim 7, characterized in that Determining whether the data in the training data is uniform, and when the data in the training data is not uniform, dividing the training data into first learning data and second learning data, comprises: Calculate the number of the training data that contains the first type of ecological data features and the number of the training data that does not contain the first type of ecological data features, and when the number that does not contain the first type of ecological data features is less than N times the number that contains the first type of ecological data features, determine that the data in the training data is uneven, where N represents a positive integer greater than or equal to 2; When the data in the training data is uneven, the training data is divided into first learning data and second learning data, wherein the first learning data refers to data that does not contain the first type of ecological data characteristics, and the second learning data refers to data that contains the first type of ecological data characteristics.

9. A numerical prediction platform for carbon flux of coastal ecosystems, used to implement the method according to any one of claims 1 to 8, characterized in that: The platform includes the following modules: An acquisition unit is used to acquire historical carbon flux monitoring data of a coastal zone from a first database as first carbon flux monitoring data, acquire a land use plan and an emission reduction plan related to the coastal zone, and based on the land use plan and the emission reduction plan, construct future carbon flux monitoring data as second carbon flux monitoring data, and store the data in a second database; A training unit, used to create a first prediction model and a second prediction model, normalize the first carbon flux monitoring data and the second carbon flux monitoring data respectively, train the first prediction model based on the normalized first carbon flux monitoring data, and obtain the trained first prediction model, train the second prediction model based on the normalized second carbon flux monitoring data, and obtain the trained second prediction model; An integration unit is used to create a third prediction model, combine the training data of the first prediction model and the training data of the second prediction model to form a data matrix, and use the data in the data matrix as training data to train the third prediction model, calculate the first type of ecological data and the second type of ecological data corresponding to the prediction time and input them into the third prediction model as input data, the third prediction model outputs the prediction result corresponding to the prediction time, and uses the prediction result as the final prediction result of the carbon flux value of the coastal ecosystem.

10. A computer storage medium, characterized in that: The storage medium stores program instructions, wherein when the program instructions are executed, the device where the storage medium is located is controlled to execute the numerical prediction method for carbon flux of coastal ecosystems according to any one of claims 1-8.

Images