A collaborative optimization system for farmland emission reduction and carbon sequestration

Abstract

The present invention relates to the field of agricultural environmental protection technology, and discloses a collaborative optimization system for farmland emission reduction and carbon fixation. The data processing module is used to collect multi-source data of farmland with major grain production, and pre-process the collected data to obtain pre-processed multi-source data; the model construction module is used to screen key factors that significantly affect farmland carbon storage based on the pre-processed multi-source data, and construct a farmland emission reduction and carbon fixation model; the collaborative optimization module is used to use a non-dominated sorting genetic algorithm to search for a Pareto optimal solution set in a parameter space, introduce a fuzzy logic controller to handle target conflicts, and obtain a collaborative optimization scheme for farmland emission reduction and carbon fixation; the real-time monitoring module is used to monitor farmland environmental changes and emission reduction and carbon fixation effects in real time, and regularly adjust and optimize the collaborative optimization scheme; the present invention improves the synergy and effectiveness of the technology, realizes the synergistic improvement of farmland production increase and emission reduction and carbon fixation, and solves the problem of uncoordinated production increase and emission reduction and carbon fixation in the existing technology.

Description

Technical Field

[0001] The present invention relates to the field of agricultural environmental protection technology, and in particular to a farmland emission reduction and carbon fixation collaborative optimization system. Background Art

[0002] Against the backdrop of global climate change, emission reduction and carbon sequestration in the agricultural sector are of vital importance. Currently, there is strong variability in the carbon balance of farmland in major grain-producing areas, uncertainty in carbon sink estimation, and a lack of key technologies for increasing production, reducing emissions, and increasing sinks. The region faces the dual challenges of reducing carbon emissions and ensuring resource supply. Existing farmland carbon sink accounting methods are not accurate enough, and emission reduction and carbon sequestration technologies need to be optimized. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a coordinated optimization system for farmland emission reduction and carbon sequestration.

[0004] The present invention provides a farmland emission reduction and carbon sequestration collaborative optimization system, which includes:

[0005] The data processing module is used to collect multi-source data of farmland with major grain production, and pre-process the collected data, including data cleaning, missing value filling and outlier removal, to obtain pre-processed multi-source data;

[0006] Construct a model module to screen key factors that significantly affect farmland carbon storage based on preprocessed multi-source data and build a farmland emission reduction and carbon sequestration model;

[0007] The collaborative optimization module is used to search for the Pareto optimal solution set in the parameter space for different planting patterns using a non-dominated sorting genetic algorithm. A fuzzy logic controller is introduced to handle target conflicts. The constructed farmland emission reduction and carbon sequestration model is solved to obtain a collaborative optimization solution for farmland emission reduction and carbon sequestration.

[0008] The real-time monitoring module is used to apply the collaborative optimization scheme of farmland emission reduction and carbon fixation to the actual management of farmland, and to monitor farmland environmental changes and emission reduction and carbon fixation effects in real time. Based on the monitoring data, the collaborative optimization scheme is regularly adjusted and optimized to adapt to different farmland conditions and climate changes, thereby realizing dynamic collaborative optimization of farmland emission reduction and carbon fixation.

[0009] Optionally, in the first implementation of the present invention, the multi-source farmland data includes at least farmland carbon storage, carbon flux, soil physical and chemical properties, meteorological data, farmland area, field management data and shelterbelt data.

[0010] Optionally, in a second implementation of the present invention, the data processing module includes:

[0011] The missing value filling submodule is used to organize the preprocessed multi-source data into time series to obtain input feature vectors, and input them into the LSTM network structure. For time points with missing values, the LSTM network structure predicts and fills the missing values, and smoothes the filled data.

[0012] The outlier removal submodule is used to select environmental factors closely related to carbon flux as features, build an isolation forest model, randomly select features and split points, recursively generate a decision tree, calculate the anomaly score of each sample point in the smoothed data, mark the sample points with anomaly scores exceeding the threshold as outliers, and remove the identified outliers;

[0013] A database submodule is constructed to generate a farmland carbon element grid map through a spatiotemporal interpolation algorithm and to construct a spatiotemporal continuous carbon budget basic database, wherein the carbon budget basic database contains the obtained pre-processed multi-source data.

[0014] Optionally, in a third implementation of the present invention, the database construction submodule includes:

[0015] Organize the preprocessed carbon flux data and environmental factor data, construct a spatial point dataset, calculate the semivariogram of the carbon flux data, and fit the semivariogram using a Gaussian model;

[0016] Based on the variogram model, ordinary kriging was used for spatial interpolation. The interpolation results were converted into a raster format to generate a spatiotemporal continuous raster map of farmland carbon elements. The raster map was then post-processed, including boundary correction and smoothing filtering.

[0017] Integrate the generated carbon factor grid map with environmental factor data to build a spatiotemporally continuous carbon budget database, including spatial, temporal, and attribute dimensions;

[0018] The preprocessed multi-source data are stored in a spatiotemporally continuous carbon budget basic database, and spatial and temporal indexes are established.

[0019] Optionally, in a fourth implementation of the present invention, the model building module includes:

[0020] The screening submodule is used to screen key factors that significantly affect farmland carbon storage based on preprocessed multi-source data;

[0021] The extraction submodule is used to extract remote sensing images and corresponding soil attribute data from key factors, construct a three-dimensional input tensor, and generate training samples using a 3×3 sliding window;

[0022] A submodule was constructed to train the CNN network and the GWR network based on the training samples, and then the Bayesian fusion method was used to combine the output of the CNN network with the output of the GWR network, dynamically adjust the weights, quantify the uncertainty of the model prediction through Monte Carlo simulation, calculate the confidence interval of the predicted value, and construct a farmland emission reduction and carbon sequestration model.

[0023] Optionally, in a fifth implementation of the present invention, in a CNN network, the input layer receives 13-channel data, connects three convolution blocks in sequence, each convolution block is followed by a batch normalization layer and a ReLU activation function, a 2×2 maximum pooling layer is inserted between the convolution blocks for dimensionality reduction, and finally outputs the plot-scale SOC prediction value through three fully connected layers;

[0024] In the GWR network, plot-scale SOC prediction values ​​were integrated with regional climate and soil type data to construct a spatial dataset containing 500 sample points. A Gaussian weight function with adaptive bandwidth was used to construct a spatial weight matrix. Weighted regression was performed on the neighborhood of each sample point, and local parameters were estimated using the least squares method to obtain a spatial distribution map, which included at least the temperature sensitivity coefficient and the fertilization response coefficient.

[0025] Optionally, in a sixth implementation of the present invention, the screening submodule includes:

[0026] The pre-processed multi-source data are matched and integrated according to spatial location and time stamp to construct a feature matrix including candidate driving factors;

[0027] Initialize the RF algorithm, train the feature matrix, calculate the Gini impurity reduction of each feature, sort the features in descending order of importance, obtain the feature subset with a cumulative importance of 90%, and obtain the key factors that significantly affect farmland carbon storage.

[0028] Optionally, in a seventh implementation of the present invention, the farmland emission reduction and carbon sequestration model outputs a carbon storage map, a carbon sink potential map, and a key driving factor impact intensity map in a raster format.

[0029] Optionally, in an eighth implementation of the present invention, the collaborative optimization module includes:

[0030] The first calculation submodule is used to randomly generate an initial population, each individual represents a set of technical parameter combinations, encode the parameters, input each parameter combination into the farmland emission reduction and carbon sequestration model, and calculate the carbon emission reduction rate, soil carbon increase rate, and crop yield increase rate;

[0031] The second calculation is to divide the population into different levels according to the Pareto dominance relationship, and calculate the Euclidean distance of individuals in the same level in the target space;

[0032] The selection submodule is used to randomly select 5 individuals from the population using the tournament selection method, and select the individual with the highest non-dominated rank and the largest crowding degree;

[0033] The crossover submodule is used to generate offspring individuals using simulated binary crossover with a crossover probability of 0.9;

[0034] The mutation submodule is used to perturb the parameters by ±10% of the biochar application amount using polynomial mutation with a mutation probability of 0.1;

[0035] The iterator module is used to set the maximum number of iterations or terminate the algorithm when there is no significant update of the non-dominated solution set for 20 consecutive generations, and finally obtain the Pareto optimal solution set;

[0036] The solution submodule is used to introduce a fuzzy logic controller to deal with target conflicts, solve the constructed farmland emission reduction and carbon sequestration model, and obtain a collaborative optimization plan for farmland emission reduction and carbon sequestration.

[0037] Optionally, in a ninth implementation of the present invention, the solution submodule includes:

[0038] The normalized values ​​of the three optimization objectives are used as inputs to the fuzzy controller, where the optimization objectives include carbon emission reduction rate, soil carbon increase rate and crop yield increase rate;

[0039] For the Pareto solution set generated by the non-dominated sorting genetic algorithm, the fuzzy output values ​​are calculated one by one, and the high-priority solutions are determined according to the screening threshold, and a collaborative optimization scheme including emission reduction, carbon sequestration and production increase is obtained, where the screening threshold is priority ≥ 0.6.

[0040] In the technical solution provided by the present invention, multi-source data of farmland with major grain production are collected, and the collected data are preprocessed to obtain preprocessed multi-source data; based on the preprocessed multi-source data, key factors that have a significant impact on farmland carbon storage are screened, and a farmland emission reduction and carbon sequestration model is constructed; for different planting patterns, a non-dominated sorting genetic algorithm is used to search for the Pareto optimal solution set in the parameter space, a fuzzy logic controller is introduced to deal with target conflicts, and the constructed farmland emission reduction and carbon sequestration model is solved to obtain a collaborative optimization scheme for farmland emission reduction and carbon sequestration; the collaborative optimization scheme for farmland emission reduction and carbon sequestration is applied to the actual management of farmland, and farmland environmental changes and emission reduction and carbon sequestration effects are monitored in real time. According to the monitoring data, the collaborative optimization scheme is regularly adjusted and optimized to adapt to different farmland conditions and climate changes, so as to realize dynamic collaborative optimization of farmland emission reduction and carbon sequestration; the farmland emission reduction and carbon sequestration model constructed by the present invention can accurately reflect the complex relationship between farmland carbon storage and various influencing factors, improve the accuracy and reliability of carbon sink accounting, and provide reference for the "dual carbon" of agriculture. It provides accurate data support for the realization of the goals; improves the synergy and effectiveness of the technology, achieves the coordinated improvement of farmland production increase and emission reduction and carbon sequestration, and solves the problem of the lack of coordination between production increase and emission reduction and carbon sequestration in existing technologies; it is replicable and scalable, and can be demonstrated and promoted in major grain-producing areas, radiating and driving regional agriculture to achieve green, low-carbon and sustainable development, and provides a new technical approach and demonstration model for agriculture to respond to climate change and achieve the "dual carbon" goals. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Various other advantages and benefits will become apparent to those skilled in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiment and are not to be considered as limiting the present invention.

[0042] Figure 1 A schematic diagram of the structure of a coordinated optimization system for farmland emission reduction and carbon sequestration provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of the structure of a data processing module provided in an embodiment of the present invention;

[0044] Figure 3 A schematic diagram of the structure of the model building module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar items and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that shown or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, apparatus, product, or device that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include steps or elements not expressly listed or inherent to such process, method, product, or device.

[0046] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 A schematic diagram of the structure of a coordinated optimization system for farmland emission reduction and carbon sequestration provided by an embodiment of the present invention includes:

[0047] The data processing module is used to collect multi-source data of farmland with major grain production, and pre-process the collected data, including data cleaning, missing value filling and outlier removal, to obtain pre-processed multi-source data;

[0048] Construct a model module to screen key factors that significantly affect farmland carbon storage based on preprocessed multi-source data and build a farmland emission reduction and carbon sequestration model;

[0049] The collaborative optimization module is used to search for the Pareto optimal solution set in the parameter space for different planting patterns using a non-dominated sorting genetic algorithm. A fuzzy logic controller is introduced to handle target conflicts. The constructed farmland emission reduction and carbon sequestration model is solved to obtain a collaborative optimization solution for farmland emission reduction and carbon sequestration.

[0050] The real-time monitoring module is used to apply the collaborative optimization scheme of farmland emission reduction and carbon fixation to the actual management of farmland, and to monitor farmland environmental changes and emission reduction and carbon fixation effects in real time. Based on the monitoring data, the collaborative optimization scheme is regularly adjusted and optimized to adapt to different farmland conditions and climate changes, thereby realizing dynamic collaborative optimization of farmland emission reduction and carbon fixation.

[0051] In this embodiment, the multi-source data of farmland include at least farmland carbon storage, carbon flux, soil physical and chemical properties, meteorological data, farmland area, field management data and shelterbelt data, among which farmland area refers to the land area directly used for agricultural product production, covering cultivated land, gardens, woodlands, grasslands and other types of land. Information can be collected over farmland through remote sensing satellites and other technical means to measure farmland area data; field management data include fertilization data, irrigation data, pest and disease control data, planting and thinning data, plant growth data, and shelterbelt data include forest belt location data, forest area data, and forest growth data; meteorological data records climate and weather information such as temperature, precipitation, light, wind speed, etc. in the farmland area; fertilization data records the application amount, time and method of fertilizers such as nitrogen, phosphorus, potassium and organic fertilizers, reflecting the matching of soil nutrient replenishment with crop needs. situation; irrigation data includes irrigation time, water volume, frequency and water source type, which are used to optimize water resource utilization and maintain soil moisture balance; pest and disease control data records the types of pests and diseases, occurrence time, prevention and control measures (such as pesticide use) and effects, and assists in the scientific prevention and control of agricultural disasters; planting and thinning data involve crop varieties, sowing density, thinning time and methods, which affect farmland space utilization and crop competition relationships; plant growth data include plant height, leaf area index (LAI), biomass and other indicators, reflecting crop growth status and yield potential; forest belt location data mark the distribution range of shelterbelts and their spatial relationship with farmland; forest area data counts the coverage area of ​​shelterbelts and quantifies their contribution to the ecological services of farmland (such as carbon sequestration and habitat provision); forest growth data records the growth parameters of shelterbelt trees such as tree height, diameter at breast height, crown width, etc.

[0052] In this embodiment, please refer to Figure 2 , the data processing module includes:

[0053] The missing value filling submodule is used to organize the preprocessed multi-source data into time series to obtain input feature vectors, and input them into the LSTM network structure. For time points with missing values, the LSTM network structure predicts and fills the missing values, and smoothes the filled data.

[0054] The outlier removal submodule is used to select environmental factors closely related to carbon flux as features, build an isolation forest model, randomly select features and split points, recursively generate a decision tree, calculate the anomaly score of each sample point in the smoothed data, mark the sample points with anomaly scores exceeding the threshold as outliers, and remove the identified outliers;

[0055] A database submodule is constructed to generate a farmland carbon element grid map through a spatiotemporal interpolation algorithm and to construct a spatiotemporal continuous carbon budget basic database, wherein the carbon budget basic database contains the obtained pre-processed multi-source data.

[0056] In this embodiment, constructing a database submodule includes: arranging preprocessed carbon flux data and environmental factor data, constructing a spatial point data set, calculating the semivariogram of the carbon flux data, and fitting the semivariogram through a Gaussian model; based on the variation function model, using the ordinary Kriging method for spatial interpolation, converting the interpolation results into a raster format, generating a spatiotemporal continuous raster map of farmland carbon elements, and post-processing the raster map, including boundary correction and smoothing filtering; integrating the generated carbon element raster map with the environmental factor data to construct a spatiotemporal continuous carbon budget basic database, including spatial dimension, time dimension and attribute dimension; storing the preprocessed multi-source data in the spatiotemporal continuous carbon budget basic database, and establishing a spatial index and a time index.

[0057] In this embodiment, please refer to Figure 3 , the model building modules include:

[0058] The screening submodule is used to screen key factors that significantly affect farmland carbon storage based on preprocessed multi-source data;

[0059] The extraction submodule is used to extract remote sensing images and corresponding soil attribute data from key factors, construct a three-dimensional input tensor, and generate training samples using a 3×3 sliding window;

[0060] A submodule was constructed to train the CNN network and the GWR network based on the training samples, and then the Bayesian fusion method was used to combine the output of the CNN network with the output of the GWR network, dynamically adjust the weights, quantify the uncertainty of the model prediction through Monte Carlo simulation, calculate the confidence interval of the predicted value, and construct a farmland emission reduction and carbon sequestration model.

[0061] In this embodiment, in the CNN network, the input layer receives 13-channel data and is connected to three convolution blocks in sequence. Each convolution block is followed by a batch normalization layer and a ReLU activation function. A 2×2 maximum pooling layer is inserted between the convolution blocks for dimensionality reduction. Finally, three fully connected layers are used to output the plot-scale SOC prediction value. In the GWR network, the plot-scale SOC prediction value is integrated with the regional climate and soil type data to construct a spatial data set containing 500 sample points. A Gaussian weight function with adaptive bandwidth is used to construct a spatial weight matrix. Weighted regression is performed on the neighborhood of each sample point. The local parameters are estimated by the least squares method to obtain a spatial distribution map, wherein the spatial distribution map includes at least a temperature sensitivity coefficient and a fertilization response coefficient.

[0062] In this embodiment, the screening submodule includes: matching and integrating the preprocessed multi-source data according to spatial location and timestamp, and constructing a feature matrix including candidate driving factors; initializing the RF algorithm, training the feature matrix, calculating the Gini impurity reduction of each feature, arranging the features in descending order of importance, obtaining a feature subset with a cumulative importance of 90%, and obtaining the key factors that significantly affect farmland carbon storage.

[0063] In this embodiment, the farmland emission reduction and carbon sequestration model outputs a carbon storage map, a carbon sink potential map, and a key driving factor impact intensity map in raster format.

[0064] In this embodiment, the collaborative optimization module includes:

[0065] The first calculation submodule is used to randomly generate an initial population, each individual represents a set of technical parameter combinations, encode the parameters, input each parameter combination into the farmland emission reduction and carbon sequestration model, and calculate the carbon emission reduction rate, soil carbon increase rate, and crop yield increase rate;

[0066] The second calculation is to divide the population into different levels according to the Pareto dominance relationship, and calculate the Euclidean distance of individuals in the same level in the target space;

[0067] The selection submodule is used to randomly select 5 individuals from the population using the tournament selection method, and select the individual with the highest non-dominated rank and the largest crowding degree;

[0068] The crossover submodule is used to generate offspring individuals using simulated binary crossover with a crossover probability of 0.9;

[0069] The mutation submodule is used to perturb the parameters by ±10% of the biochar application amount using polynomial mutation with a mutation probability of 0.1;

[0070] The iterator module is used to set the maximum number of iterations or terminate the algorithm when there is no significant update of the non-dominated solution set for 20 consecutive generations, and finally obtain the Pareto optimal solution set;

[0071] The solution submodule is used to introduce a fuzzy logic controller to deal with target conflicts, solve the constructed farmland emission reduction and carbon sequestration model, and obtain a collaborative optimization plan for farmland emission reduction and carbon sequestration.

[0072] In this embodiment, the solution submodule includes: taking the normalized values ​​of three optimization objectives as the input of the fuzzy controller, where the optimization objectives include carbon emission reduction rate, soil carbon increase rate and crop yield increase rate; calculating the fuzzy output values ​​of the Pareto solution set generated by the non-dominated sorting genetic algorithm one by one, and determining the high-priority solution according to the screening threshold to obtain a collaborative optimization scheme including emission reduction, carbon sequestration and yield increase, where the screening threshold is priority ≥ 0.6.

[0073] In this embodiment, an integrated "air-space-ground" monitoring system is used to collect multi-source farmland data under typical planting patterns in major grain-producing areas (such as "wheat-corn" rotation and "wheat-rice" rotation), including but not limited to farmland carbon storage, carbon flux, soil physical and chemical properties (such as soil organic matter content, pH value, water content, soil bulk density), meteorological data (such as temperature, rainfall, light intensity), crop growth indicators (such as crop height, leaf area index, yield), farmland inputs (such as fertilizers, commercial organic fertilizers, livestock and poultry manure, pesticides, biochar), agricultural machinery inputs, etc.; air-based: using drones equipped with multispectral sensors to obtain high-frequency remote sensing data such as vegetation coverage and crop stress index; space-based: accessing long-term time series data such as soil moisture and surface temperature from satellite remote sensing (such as Sentinel-2); ground-based: using the Internet of Things sensor network (such as EC flux observation system, soil three-parameter sensor) to collect carbon flux (CO2 / N2O / CH4), soil physical and chemical properties (organic matter, pH value, soil moisture content, soil bulk density, etc.) in real time. , electrical conductivity) and micrometeorological parameters (temperature, humidity, light, rainfall); pre-process the collected data, including data cleaning, missing value filling, and outlier removal, to improve data quality; construct a farmland emission reduction and carbon sequestration model, and establish a nonlinear relationship model between farmland carbon storage and various influencing factors (such as soil organic matter content, crop type, fertilizer application rate, tillage method, etc.); use historical monitoring data to train and verify the model, and improve the prediction accuracy of the model by adjusting the model parameters. For example, the optimal number of trees and feature selection number of the random forest model are determined by cross-validation method, so that the model's prediction error for farmland carbon storage is less than 5%; design a variety of emission reduction and carbon sequestration coordinated regulation technology solutions based on the characteristics of farmland with different planting patterns, such as stable carbon-based fertilizer equipped with functional fertilizer technology, carbon-nitrogen coupling inhibitor technology, straw rapid composting and carbon supplementation and biochar addition technology, etc.; take farmland carbon emission reduction and soil organic carbon increase as optimization goals, establish a multi-objective optimization model, and determine the optimal combination of technical parameters. For example, for wheat-corn rotation farmland, a particle swarm optimization algorithm was used to determine parameters such as the application rate of stable carbon-based fertilizer, the type of functional fertilizer, and the application time and frequency. This resulted in a carbon emission reduction of over 20% and an increase in soil organic carbon of over 15%. The optimized synergistic regulation technology for emission reduction and carbon sequestration was integrated to construct a model for emission reduction and carbon sequestration in typical farmland. The model was locally adjusted and optimized based on the climate, soil, and planting conditions in different regions of major grain-producing areas. Demonstration bases were established in typical areas to demonstrate and promote the model. Real-time monitoring of the carbon balance of farmland and crop growth in these demonstration bases allowed for timely adjustments to the technical plan to ensure its effectiveness and replicability.

[0074] In this embodiment, the preprocessed data is organized into time series to construct an input feature vector, including carbon flux data and its related environmental factors (temperature, humidity, light, etc.), divided into training sets and test sets, and a sliding window method is used to generate training samples; an LSTM network structure is constructed, including an input layer, a hidden layer (2-3 LSTM layers), and an output layer. The input layer dimension is the number of features, the output layer dimension is 1 (the carbon flux value to be filled), the number of hidden layer neurons is determined by cross-validation, and the activation function is tanh; for time points with missing values, historical data (the first n time steps) are used as input. The missing values ​​are predicted by the trained LSTM model, and the filled data are smoothed to ensure the continuity of the time series; environmental factors closely related to carbon flux (such as temperature, humidity, wind speed, etc.) are selected as features, an isolation forest model is constructed, features and split points are randomly selected, a decision tree is recursively generated, and the anomaly score of each sample point is calculated. The higher the anomaly score, the more likely the point is an outlier; an anomaly score threshold is set (such as 0.7), and sample points exceeding the threshold are marked as outliers. The identified outliers are corrected or eliminated. The correction method can be linear interpolation of the previous and next time steps; the pre-processing is sorted out. The processed carbon flux data and environmental factor data are used to construct a spatial point dataset, calculate the semivariogram of the carbon flux data, analyze its spatial autocorrelation, select an appropriate theoretical model (such as Gaussian model, exponential model) to fit the semivariogram, and based on the variogram model, use ordinary kriging or co-kriging method for spatial interpolation, considering the influence of the time dimension, interpolate the data at different time points separately, or use the space-time kriging method to consider the space-time correlation at the same time, convert the interpolation results into raster format, generate a spatiotemporal continuous raster map of farmland carbon elements (such as soil organic carbon content, carbon flux), and perform spatial and temporal interpolation on the raster map. Post-processing, including boundary correction and smoothing filtering, integrates the generated carbon element raster map with other environmental factor data (meteorological, soil type, etc.), designs the spatiotemporal database structure, including spatial dimensions (latitude and longitude, altitude), time dimensions (year, month, day) and attribute dimensions (carbon flux, soil organic carbon, etc.), stores the integrated data in the spatiotemporal database, establishes spatial and temporal indexes, supports efficient query and analysis, performs quality checks on the data in the database to ensure data integrity, consistency and accuracy, establishes a data update mechanism, regularly adds newly collected data, and ensures the timeliness of the database.

[0075] In this example, remote sensing images (such as Sentinel-2 multispectral data) are cropped and resampled, spatially aligned with soil sampling points, and an input tensor is constructed. This includes multi-band remote sensing data (such as blue, green, red, near-infrared, and short-wave infrared bands) and auxiliary soil data (such as pH, texture, and previous SOC measurements). A sliding window method is used to generate training samples, where each sample contains a central pixel and its neighborhood (such as 3×3 or 5×5 pixels). The CNN model architecture is designed as follows: the input layer receives a multi-channel input tensor (such as 10 bands + 3 soil attributes); the convolutional layer: Set up 3-4 convolution blocks, each of which contains a convolution layer (such as 32 / 64 / 128 filters), a batch normalization layer, and a ReLU activation function; pooling layer: insert a maximum pooling layer (such as 2×2 pooling) between convolution blocks to reduce the feature dimension; fully connected layer: flatten the convolution features and connect 2-3 fully connected layers, and finally output the SOC prediction value; model training and optimization, divide the training set (70%), validation set (15%), and test set (15%), use mean square error (MSE) as the loss function, and adjust the model parameters with the Adam optimizer. Early Stopping prevents overfitting and saves the model parameters that perform best on the validation set. The plot-scale SOC inversion results are integrated with regional environmental variables (climate data, soil type maps, and land use data). A spatial point dataset is constructed, with each point containing a dependent variable (SOC or carbon flux) and an independent variable (key driver). Euclidean or geographic distances are calculated between sample points. A spatial weight function (e.g., Gaussian or quadratic) is selected, and bandwidth parameters (e.g., adaptive bandwidth) are set to generate a spatial weight matrix that reflects the spatial dependencies between sample points. For GWR network fitting, for each sample point, a weight matrix is ​​used to weight neighboring points, and a local regression model is constructed. The least squares method is used to estimate the parameters of each local model, obtain the spatial distribution of the parameters, analyze the spatial heterogeneity of the parameters, and identify the dominant driving factors of carbon sequestration in different regions. For model validation and dynamic correction, the dataset is divided into K non-overlapping subsets (e.g., K=5), of which K-1 subsets are used as training sets and the remaining subset is used as the validation set. The average prediction error (e.g., RMSE, MAE) and coefficient of determination of the K validations are calculated. The contribution of each feature to the model prediction is calculated using the Kernel SHAP or Tree SHAP method, and a SHAP value distribution map is generated to intuitively display the positive / negative impact of each factor on carbon storage. The explanatory power of the model is verified to ensure that the SHAP values ​​of key driving factors are consistent with ecological significance. Annual field trial data are collected (e.g., changes in SOC under different fertilization treatments), and the online gradient descent algorithm is used to dynamically adjust the model parameters based on the new data. The model is retrained regularly (e.g., annually) to incorporate new environmental and management measures data.

[0076] In this embodiment, the input variables are: the normalized values ​​of the three optimization objectives (range 0-1) are used as the input of the fuzzy controller, and the linguistic variables are defined as:

[0077] Carbon emission reduction rate: low (≤0.2), medium (0.2-0.4), high (≥0.4);

[0078] Soil carbon accumulation rate: low (≤0.15), medium (0.15-0.3), high (≥0.3);

[0079] Crop yield increase rate: low (≤0.08), medium (0.08-0.15), high (≥0.15).

[0080] Output variable: comprehensive priority (0-1), defined language value: low, medium, high.

[0081] Design conflict resolution rules, such as:

[0082] When {f1=high, f2=medium, f3=low}, give priority to improving crop yield rate and output priority "medium";

[0083] When {f1=medium, f2=high, f3=medium}, the emphasis is on soil carbon sequestration and the output priority is “high”;

[0084] A total of 3×3×3=27 rules are defined, and the rule weights are calibrated through expert experience or historical data.

[0085] In this embodiment, the fuzzy output value (comprehensive priority) of the Pareto solution set generated by NSGA-II is calculated one by one;

[0086] Set a screening threshold (such as priority ≥ 0.6), retain high-priority solutions, and ultimately obtain 5-10 collaborative optimization solutions that take into account emission reduction, carbon sequestration and production increase.

[0087] In this embodiment, the selected optimization solutions are analyzed for parameters, for example:

[0088] Wheat-corn rotation plan: biochar 250kg / mu (particle size 1mm) + full straw return + nitrification inhibitor 2.5%, predicted v=22%, f2=16%, f3=9%;

[0089] "Wheat-rice" rotation plan: intermittent irrigation + organic fertilizer replacing 30% of chemical fertilizer + biochar covering 150kg / mu, predicted f1=21%, f2=17%, f3=7%.

[0090] In this example, a field experiment was set up in a typical region, with each scenario replicated three times, and the control was local routine management. Monitoring indicators included soil carbon flux (static chamber method), SOC content (potassium dichromate oxidation method), and crop yield (actual yield measurement). Simulated values ​​were compared with measured values. If the error was greater than 10%, the fuzzy controller rules or NSGA-II parameters (such as crossover probability) were adjusted, and the algorithm was rerun.

[0091] In this example, based on the farmland type (e.g., dryland, paddy field) and regional climate characteristics (e.g., temperate monsoon zone, subtropical humid zone), an appropriate collaborative optimization scheme (e.g., "straw carbonization and return to the field + biochar-based fertilizer" combination) is selected, and a detailed farming operation guide is formulated with clear parameter implementation standards. For example, in the "wheat-corn rotation zone" scheme, the biochar application rate is 200 kg / mu, which is evenly mixed into the 0-20 cm soil layer by plowing before corn sowing; nitrification inhibitors are mixed with urea at a ratio of 2% as a base application.

[0092] In this embodiment, the Internet of Things monitoring system is deployed:

[0093] Carbon flux monitoring: Install an automatic gas sampling box (such as LI-8100A) to collect CO2, CH4, and N2O concentration data every hour;

[0094] Soil environment: Bury multiple layers of soil sensors (such as Decagon 5TE) to monitor temperature, humidity, conductivity, and redox potential in real time;

[0095] Crop growth: Use a drone for remote sensing monitoring (such as the DJI M300 RTK) to obtain vegetation indices such as NDVI and LAI monthly.

[0096] Establish a data transmission network to upload field data to the cloud server in real time via 4G / 5G.

[0097] In this embodiment, real-time monitoring and data collection:

[0098] Environmental parameter collection includes:

[0099] Meteorological data: Temperature, precipitation, and light data are obtained through field weather stations (such as WatchDog 2000) with a resolution of 10 minutes per time.

[0100] Agricultural data: Real-time data on soil moisture, seedling conditions, pests, weeds, and disasters can be obtained through field weather stations;

[0101] Soil data: Automatically collects temperature, humidity, and SOC content in the 0-10 cm, 10-20 cm, and 20-30 cm soil layers (measured in situ using a near-infrared spectroscopy sensor);

[0102] Crop data: Combine UAV remote sensing with ground-based spectrometers (such as ASD FieldSpec) to monitor leaf area index, biomass accumulation, and nitrogen nutrition status.

[0103] The assessment of emission reduction and carbon sequestration effects includes:

[0104] Carbon sink accounting: Based on monitoring data, use carbon sink models (such as DNDC and RothC) to calculate soil carbon storage changes in real time, with a frequency of once a month;

[0105] Calculation of emission reductions: Compare greenhouse gas emission fluxes before and after the implementation of the plan, and convert them into CO2 equivalents (CO2-eq) based on the IPCC default value;

[0106] Crop response: Evaluate the impact of the program on crop production through yield monitoring (such as intelligent yield measurement systems) and quality analysis (such as protein content and starch content).

[0107] In this embodiment, dynamic optimization and solution adjustment:

[0108] 1. Threshold trigger adjustment: Set the threshold of key indicators and start the adjustment process when the monitoring data exceeds the threshold range:

[0109] For example: if soil moisture is less than 40% of field capacity for three consecutive months, irrigation strategy optimization is triggered;

[0110] If the CH4 emission flux suddenly increases by 20%, initiate adjustments to rice field water management.

[0111] Seasonal correction: Dynamically adjust management measures based on crop growth period, such as:

[0112] Apply biochar-based topdressing during the jointing stage of wheat, and add nitrification inhibitors during the flaring stage of corn;

[0113] 2. Reinforcement learning model: Monitoring data is input into a pre-trained deep Q network (DQN) to optimize management parameters through reward functions (such as carbon sink increment and economic benefits);

[0114] Adaptive adjustment: Update the program parameters every quarter, for example:

[0115] If the SOC growth rate is less than 1.5% for two consecutive years, the biochar application rate will be increased by 50 kg / mu in the next year;

[0116] If the N2O emission intensity continues to be high, increase the inhibitor ratio from 2% to 2.5%.

[0117] In this embodiment, the multi-dimensional evaluation system:

[0118] Environmental benefits: Calculate the annual carbon sequestration per unit area (tCO2-eq / ha·yr) and the global warming potential (GWP) reduction rate;

[0119] Economic benefits: Calculate the input-output ratio (e.g., biochar costs and increased production benefits) and assess the potential benefits of carbon trading;

[0120] Ecological benefits: monitoring soil quality indices (such as aggregate stability and microbial diversity) and changes in farmland biodiversity.

[0121] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A farmland emission reduction and carbon sequestration collaborative optimization system, characterized in that: The system includes: The data processing module is used to collect multi-source data of farmland with major grain production, and pre-process the collected data, including data cleaning, missing value filling and outlier removal, to obtain pre-processed multi-source data; Construct a model module to screen key factors that significantly affect farmland carbon storage based on preprocessed multi-source data and build a farmland emission reduction and carbon sequestration model; The collaborative optimization module is used to search for the Pareto optimal solution set in the parameter space for different planting patterns using a non-dominated sorting genetic algorithm. A fuzzy logic controller is introduced to handle target conflicts. The constructed farmland emission reduction and carbon sequestration model is solved to obtain a collaborative optimization solution for farmland emission reduction and carbon sequestration. A real-time monitoring module is used to apply the collaborative optimization scheme for farmland emission reduction and carbon sequestration to actual farmland management, and to monitor farmland environmental changes and emission reduction and carbon sequestration effects in real time. Based on the monitoring data, the collaborative optimization scheme is regularly adjusted and optimized to adapt to different farmland conditions and climate changes, thereby achieving dynamic collaborative optimization of farmland emission reduction and carbon sequestration. The model building module includes: The screening submodule is used to screen key factors that significantly affect farmland carbon storage based on preprocessed multi-source data; The extraction submodule is used to extract remote sensing images and corresponding soil attribute data from key factors, construct a three-dimensional input tensor, and generate training samples using a 3×3 sliding window; Construct a submodule for training CNN and GWR networks based on training samples, then combine the output of the CNN network with the output of the GWR network using a Bayesian fusion method, dynamically adjust the weights, quantify the uncertainty of the model prediction through Monte Carlo simulation, calculate the confidence interval of the predicted value, and construct a farmland carbon sequestration model. In the CNN network, the input layer receives 13-channel data and is sequentially connected to three convolutional blocks. Each convolutional block is followed by a batch normalization layer and a ReLU activation function. A 2×2 maximum pooling layer is inserted between convolutional blocks for dimensionality reduction. Finally, three fully connected layers are used to output the plot-scale SOC prediction value. In the GWR network, plot-scale SOC prediction values ​​were integrated with regional climate and soil type data to construct a spatial dataset containing 500 sample points. A Gaussian weight function with adaptive bandwidth was used to construct a spatial weight matrix. Weighted regression was performed on the neighborhood of each sample point, and local parameters were estimated using the least squares method to obtain a spatial distribution map, which included at least the temperature sensitivity coefficient and the fertilization response coefficient.

2. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 1, characterized in that: The multi-source farmland data include at least farmland carbon storage, carbon flux, soil physical and chemical properties, meteorological data, farmland area, field management data and shelterbelt data.

3. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 1, characterized in that: The data processing module includes: The missing value filling submodule is used to organize the preprocessed multi-source data into time series to obtain input feature vectors, and input them into the LSTM network structure. For time points with missing values, the LSTM network structure predicts and fills the missing values, and smoothes the filled data. The outlier removal submodule is used to select environmental factors closely related to carbon flux as features, build an isolation forest model, randomly select features and split points, recursively generate a decision tree, calculate the anomaly score of each sample point in the smoothed data, mark the sample points with anomaly scores exceeding the threshold as outliers, and remove the identified outliers; A database submodule is constructed to generate a farmland carbon element grid map through a spatiotemporal interpolation algorithm and to construct a spatiotemporal continuous carbon budget basic database, wherein the carbon budget basic database contains the obtained pre-processed multi-source data.

4. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 3, characterized in that: The database construction submodule includes: Organize the preprocessed carbon flux data and environmental factor data, construct a spatial point dataset, calculate the semivariogram of the carbon flux data, and fit the semivariogram using a Gaussian model; Based on the variogram model, ordinary kriging was used for spatial interpolation. The interpolation results were converted into a raster format to generate a spatiotemporal continuous raster map of farmland carbon elements. The raster map was then post-processed, including boundary correction and smoothing filtering. Integrate the generated carbon factor grid map with environmental factor data to build a spatiotemporally continuous carbon budget database, including spatial, temporal, and attribute dimensions; The preprocessed multi-source data are stored in a spatiotemporally continuous carbon budget basic database, and spatial and temporal indexes are established.

5. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 1, characterized in that: The screening submodule includes: The pre-processed multi-source data are matched and integrated according to spatial location and time stamp to construct a feature matrix including candidate driving factors; Initialize the RF algorithm, train the feature matrix, calculate the Gini impurity reduction of each feature, sort the features in descending order of importance, obtain the feature subset with a cumulative importance of 90%, and obtain the key factors that significantly affect farmland carbon storage.

6. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 1, characterized in that: The farmland emission reduction and carbon sequestration model outputs a carbon storage map, a carbon sink potential map, and a key driving factor impact intensity map in raster format.

7. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 1, characterized in that: The collaborative optimization module includes: The first calculation submodule is used to randomly generate an initial population, each individual represents a set of technical parameter combinations, encode the parameters, input each parameter combination into the farmland emission reduction and carbon sequestration model, and calculate the carbon emission reduction rate, soil carbon increase rate, and crop yield increase rate; The second calculation is to divide the population into different levels according to the Pareto dominance relationship, and calculate the Euclidean distance of individuals in the same level in the target space; The selection submodule is used to randomly select 5 individuals from the population using the tournament selection method, and select the individual with the highest non-dominated rank and the largest crowding degree; The crossover submodule is used to generate offspring individuals using simulated binary crossover with a crossover probability of 0.9; The mutation submodule is used to perturb the parameters by ±10% of the biochar application amount using polynomial mutation with a mutation probability of 0.1; The iterator module is used to set the maximum number of iterations or terminate the algorithm when there is no significant update of the non-dominated solution set for 20 consecutive generations, and finally obtain the Pareto optimal solution set; The solution submodule is used to introduce a fuzzy logic controller to deal with target conflicts, solve the constructed farmland emission reduction and carbon sequestration model, and obtain a collaborative optimization plan for farmland emission reduction and carbon sequestration.

8. The farmland emission reduction and carbon sequestration collaborative optimization system according to claim 7, characterized in that: The solution submodule includes: The normalized values ​​of the three optimization objectives are used as inputs to the fuzzy controller, where the optimization objectives include carbon emission reduction rate, soil carbon increase rate and crop yield increase rate; For the Pareto solution set generated by the non-dominated sorting genetic algorithm, the fuzzy output values ​​are calculated one by one, and the high-priority solutions are determined according to the screening threshold, and a collaborative optimization scheme including emission reduction, carbon sequestration and production increase is obtained, where the screening threshold is priority ≥ 0.6.

Images