A carbon sink intelligent accounting and optimization method for surface source pollution

Abstract

This invention discloses an intelligent carbon sequestration accounting and optimization method for non-point source pollution, addressing the shortcomings of existing methods such as the disconnect between non-point source pollution control and carbon sequestration accounting, large calculation errors in traditional static coefficient methods, inability to meet carbon trading verification requirements, and lack of coordinated optimization measures for pollution reduction and carbon sequestration. This invention first delineates the carbon sequestration accounting boundary coupled with the non-point source pollution migration process, collects multi-source heterogeneous monitoring data, generates net carbon sequestration results that meet carbon trading verification requirements through a dynamic accounting model of mechanism-data fusion, and then outputs a coordinated pollution reduction and carbon sequestration scheme based on a multi-objective optimization model, feeding the optimization parameters back to the accounting module to achieve closed-loop iteration. This invention effectively improves the accuracy of carbon sequestration accounting in non-point source scenarios, balancing pollution control requirements and carbon sequestration benefits, and can be applied to scenarios such as agricultural non-point source pollution control, constructed wetlands, and sponge cities, providing support for coordinated pollution reduction and carbon sequestration management.

Description

Technical Field

[0001] This invention belongs to the technical field of non-point source pollution control, ecological carbon sink measurement and pollution reduction and carbon reduction coordinated management, specifically involving a method for intelligent carbon sink accounting and optimization for non-point source pollution. Background Technology

[0002] With the advancement of the "dual carbon" goals, synergistic efficiency in pollution reduction and carbon sequestration has become the core orientation of ecological and environmental governance. Currently, point source pollution from industry and municipal sources in my country has been effectively controlled. However, non-point source pollution from agriculture and cities has become the core constraint on achieving water quality standards and watershed ecological restoration. Meanwhile, non-point source pollution control facilities such as ecological buffer zones and constructed wetlands have both pollution interception and ecological carbon sequestration functions, making them important carriers for synergistic pollution reduction and carbon sequestration.

[0003] Currently, the technical system in the field of non-point source pollution control still focuses solely on achieving water quality standards. During project design and operation, it only assesses pollution control indicators such as nitrogen and phosphorus retention rates and effluent quality, neglecting the carbon sequestration effect of the treatment process and failing to quantify the benefits of non-point source pollution interception. The project fails to consider the synergistic carbon benefits from greenhouse gas emission reductions such as methane, and also fails to account for additional carbon emissions from unreasonable operation and maintenance methods such as excessive aeration and excessive application of treatment agents. As a result, non-point source pollution control projects generally have high operation and maintenance costs and lack the support of market-based benefits such as carbon trading, leading to insufficient project sustainability.

[0004] In the field of carbon sequestration, existing common methods mostly adopt the fixed factor method published by the IPCC, which is designed only for independent ecosystems such as forests and conventional farmland. They do not consider the coupling relationship between non-point source pollution migration and transformation processes and carbon sequestration, and cannot calculate carbon items specific to non-point source scenarios such as the synergistic carbon reduction caused by non-point source pollution interception and carbon loss caused by pollution overload. The calculation error is relatively high and cannot meet the requirements of measurable, reportable, and verifiable (MRV) carbon trading.

[0005] Furthermore, existing technologies have not formed a closed-loop logic of "accounting-optimization," and cannot automatically adjust engineering design parameters or operation control strategies based on real-time pollution load and dynamic changes in carbon sinks. This makes it difficult to maximize the synergistic benefits of pollution reduction and carbon reduction, resulting in insufficient applicability. Summary of the Invention

[0006] To address the problems existing in the background technology, this invention proposes an intelligent carbon sink accounting and optimization method for non-point source pollution. This method constructs a dedicated carbon sink accounting boundary delineation mechanism coupled with non-point source pollution processes, a mechanism-AI fusion dynamic carbon sink accounting mechanism with uncertainties, and a multi-objective closed-loop optimization mechanism for coordinated pollution reduction and carbon reduction. This addresses core technical issues in existing technologies, such as the disconnect between non-point source pollution control and carbon sink measurement, large errors in fixed-coefficient accounting methods failing to meet carbon trading verification requirements, and the disconnect between accounting and optimization hindering the maximization of synergistic benefits. The method achieves a combined improvement in the accuracy of carbon sink accounting, pollution control compliance, and overall project benefits in non-point source pollution scenarios.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A smart carbon sequestration and optimization method for non-point source pollution includes the following steps: S1. Define the coupled accounting boundary of the non-point source pollution control unit to be accounted for, and initialize the basic parameters and initial uncertainty matrix required for the accounting; S2. Collect multi-source heterogeneous monitoring data of the non-point source pollution control unit to be accounted for, and perform standardization processing on the raw data to obtain three types of standardized datasets: non-point source pollution load sequence, carbon sink monitoring sequence, and operational carbon emission sequence. S3. Based on the aforementioned basic parameters and three types of standardized datasets, the real-time net carbon sink is calculated by using a dynamic accounting model that couples the mechanism-data fusion of the non-point source pollution migration and transformation process, along with quantitative results of uncertainties. S4. Based on real-time net carbon sink and preset non-point source pollution control constraints, the multi-objective optimization model that takes into account both carbon sink benefits and operating costs outputs the corresponding optimization scheme. The operating parameters of the optimization scheme are fed back to the dynamic accounting model to achieve closed-loop iteration.

[0008] Specifically, the coupled accounting boundaries in step S1 include spatial boundaries, process boundaries, and compliance boundaries. The process boundaries include carbon sink and carbon emission items coupled with the non-point source pollution migration and transformation process. The initialization parameters include three categories: carbon sink accounting parameters, non-point source coupling parameters, and optimization constraint parameters. The initial uncertainty matrix satisfies... ,in The initial uncertainty coefficient, It is an identity matrix.

[0009] Specifically, step S2 includes: S21. Collect three types of raw monitoring data, including non-point source pollution monitoring data, carbon sink characteristic monitoring data, and operational energy consumption monitoring data, among which: For non-point source pollution monitoring data, water quality sensors and flow sensors are installed at the inlet and outlet of the non-point source pollution treatment unit to be accounted for, respectively, to collect the influent TN concentration. TP concentration in influent Inlet flow rate TN concentration in effluent effluent TP concentration Outflow rate , Sampling time; For carbon sink characteristic monitoring data, soil organic carbon content is collected using soil carbon sensors, weather stations, and vegetation biomass monitoring equipment. vegetation biomass Photosynthetically active radiation Temperature ; For operational energy consumption monitoring data, the power consumption of equipment is collected through smart meters, chemical metering devices, and maintenance operation statistical terminals. Diesel consumption for maintenance Dosage of pesticides for non-point source pollution control ; S22. Use the Laida criterion to remove outlier data. Data points that satisfy the following formula are identified as outliers and deleted: ; in, For the original monitoring data points, This represents the average value of the same monitoring indicator within a 24-hour statistical window. The standard deviation of the same monitoring indicator within a 24-hour statistical window; S23. Complete the missing data resulting from anomaly removal. The completion formula is: ; in, For moments when data is missing, The most recent valid data moment before the missing moment. It represents the most recent valid data moment after the missing moment. , They are respectively , Valid monitoring values ​​at any given time; S24. Generate three types of standardized datasets. For the non-point source pollution load sequence, calculate the TN and TP pollution loads of the influent and effluent respectively: ; ; in, , They are respectively Constant influent TN and TP pollution load , They are respectively Constantly monitor the TN and TP pollution loads in the effluent; For carbon sink monitoring sequences, the standardized sequence is used. , , , constitute; For operating carbon emission sequences, energy consumption data is converted to... Equivalent, the formula is: ; in, for Constant carbon emissions As a carbon emission factor for electricity, For diesel carbon emission factors, Carbon emission factors of the treatment agents per unit.

[0010] Specifically, the mechanism-data fusion dynamic accounting model for the coupled non-point source pollution migration and transformation process described in step S3 includes a mechanism accounting layer, a dynamic correction layer, and an uncertainty quantification layer connected in sequence, and the execution steps include: S31. Mechanism accounting layer couples the interception, transformation and overload process of non-point source pollution in the treatment unit, calculates the initial net carbon sink based on the basic parameters and three types of standardized datasets, and outputs five core coefficients to be corrected corresponding to the dimension of the initial uncertainty matrix. S32. The dynamic correction layer introduces time-series monitoring features to dynamically correct the five core coefficients to be corrected output by the mechanism accounting layer, and obtains the corrected net carbon sink calculation value and correction residual. S33. The uncertainty quantification layer iteratively calculates the posterior uncertainty covariance matrix at the current time based on the temporal fluctuation characteristics of the initial uncertainty matrix, correction residuals, and core coefficients, and generates the final net carbon sink result with confidence intervals.

[0011] Specifically, the formula for calculating the initial net carbon sink in step S31 is as follows: ; in, The standard carbon sequestration amount is calculated using the following formula: ; In the formula , These are the vegetation carbon sequestration baseline coefficient and the soil carbon sequestration baseline coefficient, respectively, in the core coefficients to be corrected. for The increase in vegetation biomass at the current accounting time compared to the previous accounting time. for The increase in soil organic carbon at the current accounting time compared to the previous accounting time; The formula for calculating the carbon reduction from area sources is as follows: ; in, , These are the unit TN retention synergistic carbon reduction coefficient and the unit TP retention synergistic carbon reduction coefficient in the core coefficients to be corrected. To calculate the amount of TN retained during the accounting period, This refers to the TP retention amount during the accounting period; The carbon loss due to area source overload is calculated using the following formula: ; in, The threshold for non-point source pollution overload. The total area of ​​the non-point source pollution treatment unit to be calculated. To calculate the total influent TN load during the calculation period, The overload loss coefficient to be corrected is... This represents the amount of regular carbon sequestration at the previous accounting point.

[0012] Specifically, step S32 dynamically corrects the core coefficients of the mechanistic accounting layer using variational Bayesian LSTM. The specific process includes: (1) Collect historical monitoring data of the target area within a predetermined time period, and synchronously match the actual carbon sequestration true value at the corresponding measurement time as a label, and divide it into training set, validation set and test set according to a preset ratio; (2) The model input is defined as a feature vector composed of a non-point source pollution load sequence, a carbon sink monitoring sequence, an operational carbon emission sequence, and time features. The hidden layer consists of a 2-layer LSTM network used to extract the temporal features of the non-point source-carbon sink coupling process. The output layer is a 5-dimensional dynamic correction coefficient vector. Each parameter represents a corresponding baseline coefficient. The model outputs the approximate posterior distribution parameters of each dynamic correction coefficient at each time step, including the posterior mean and posterior standard deviation of each dynamic correction coefficient. (3) Uncertainty propagation is achieved by learning the probability distribution of the weights to train the model. The specific training logic is as follows: a. Assuming the model has all trainable weights Follows a mean of 0 and a standard deviation of Independent Gaussian prior distributions: ; In the formula The pre-defined prior standard deviation, An identity matrix that matches the weight dimensions; Introducing a parameterized approximate posterior distribution Fitting the true posterior distribution ,in For the training dataset, These are trainable parameters that approximate the posterior. Follows an independent Gaussian distribution: ; b. Construct the loss function, using the lower bound of evidence as the optimization objective, with the following formula: ; in, For approximate posterior distribution Expectations The feature vector input to the model, The log-likelihood term corresponds to the goodness of fit between the corrected net carbon sink prediction and the measured true value. For the carbon sink regression task, the Gaussian log-likelihood is taken, which is equivalent to the negative mean square error. ; in The net carbon sink forecast calculated using the corrected coefficients. To measure the true value of carbon sequestration, The standard deviation of the observed noise; The KL divergence between the approximate posterior and prior distributions is used to constrain the weight distribution and avoid overfitting. c. Perform gradient optimization, using reparameterization techniques to decouple the random sampling process from the trainable parameters, representing the weight sampling as: ; in, For independent standard Gaussian noise, This is an element-wise product; therefore, it is optimized using the Adam gradient descent method. During training, the average loss is calculated by sampling the approximate posterior distribution a predetermined number of times in each batch, when the validation set... Training is stopped when the number of consecutive predetermined rounds does not increase, and a pre-trained correction model adapted to the target region is obtained. (4) After the model is put into operation, online dynamic calibration is performed, and the model is fine-tuned by synchronizing the latest measured true value of carbon sink at predetermined intervals.

[0013] Specifically, the execution steps of step S33, the uncertainty quantification layer, include: (1) Based on the posterior uncertainty covariance matrix at the previous calculation time, and combined with the process noise of coefficient time drift, the prior uncertainty covariance matrix at the current time is calculated using the following formula: ; in, for The prior uncertainty covariance matrix at time t has the same dimensions as the initial uncertainty matrix; This is the state transition matrix, initially set as the identity matrix, used to describe the time evolution of the core correction coefficients; The initial uncertainty matrix is ​​used in the first calculation, which is the posterior uncertainty covariance matrix at the previous calculation time. ; The process noise matrix reflects the uncertainty of the core correction coefficients' natural drift over time, satisfying: ; in This represents the time interval between the current moment and the previous calculation moment. For coefficient drift rate, It is an identity matrix, with the same dimensions as the core correction coefficients; (2) Combining the prior uncertainty covariance matrix and the observation noise matrix, calculate the Kalman gain at the current time. The calculation formula is as follows: ; in, for The Kalman gain at time step is used to adjust the weight of the observation data on the uncertainty update; The observation noise diagonal matrix consists of diagonal elements that are the squares of the posterior standard deviations of the dynamic correction coefficients output by the variational Bayesian LSTM, reflecting the observation uncertainty of the correction process. (3) Based on the prior covariance and Kalman gain, the posterior uncertainty covariance matrix at the current time is obtained iteratively, and the calculation formula is as follows: ; in, for The posterior uncertainty covariance matrix of the final output at time step; (4) Based on the corrected net carbon sink value and the posterior uncertainty covariance matrix output by the pre-trained correction model, output the real-time net carbon sink with confidence interval: .

[0014] Specifically, step S4 includes: S41. With the goal of maximizing the net synergistic benefits of pollution reduction and carbon reduction throughout the entire life cycle of the non-point source pollution control unit to be accounted for, an optimization function is constructed: ; in, The set of decision variables to be optimized includes two categories: engineering design parameters and operation control parameters. , To optimize the target weight coefficients and satisfy ; The total net carbon sink over the entire life cycle is obtained by summing up the real-time net carbon sink. The benchmark price for carbon trading; The total operating cost over the entire lifecycle is calculated to satisfy: ; in, This is the benchmark value for annual operating cost per unit area. The total area of ​​the non-point source pollution treatment unit to be calculated. The increase in operating costs resulting from adjusting decision variables; S42. Set constraints for the optimization process, including: (1) The constraints of non-point source pollution control shall meet the following requirements: ; in, TN retention rate Minimum retention rate requirements pre-set for compliance boundaries; The TN concentration in the effluent. The upper limit of TN in effluent as stipulated by local water environment standards; For the total TN load of the influent, This is the overload threshold for non-point source pollution. (2) Project boundary constraints, meeting the project's pre-set usable area and budget requirements: ; in, The actual construction area is [not specified]. Pre-determine the usable area for the project; The total lifecycle cost includes construction and operating costs. This is the total project budget; (3) Carbon sink stability constraints, ensuring that fluctuations in carbon sink volume meet the stability requirements of carbon trading projects, satisfying: ; in, The coefficient of variation is 1. The standard deviation of real-time net carbon sinks within the accounting period. This represents the average real-time net carbon sink over the accounting period. The fluctuation threshold; S43. The objective function is solved using the particle swarm optimization algorithm, and the optimal decision variables are output after iterative convergence. The corresponding optimization scheme feeds back the operating parameters in the optimization scheme to the mechanism-data fusion dynamic accounting model in step S3, updates the input boundary of the next round of accounting, and realizes closed-loop iteration.

[0015] Specifically, step S43 includes: (1) Set the particle swarm size and the maximum number of iterations. The dimension of each particle is consistent with the dimension of the set of decision variables to be optimized. The initial position of the particles is... Randomly generated within the engineering feasibility interval of each decision variable, where For particle serial numbers; (2) Calculate the particle fitness and the position of each particle. For a specific set of decision variables, Substituting the values ​​into the optimization function in step S41, we obtain the fitness value of the particle: ; (3) Iterate through the particles, updating their velocity and position in each iteration according to the following formula: ; ; in, Inertial weights are used to balance global and local search capabilities; , These are learning factors, representing the weights by which a particle learns towards its individual optimum and global optimum, respectively. , for Random numbers within the range are used to increase the randomness of the search; For the first Before each particle The optimal position of an individual in the next iteration corresponds to the combination of decision variables that represent the highest fitness value in the particle's history. For the entire particle swarm The global optimal position in the next iteration corresponds to the combination of decision variables with the highest historical fitness value among all particles; (4) When the maximum number of iterations is reached, or the change in the global optimal fitness value after a preset number of consecutive iterations is less than a preset threshold, the iteration is stopped, and the optimal decision variable corresponding to the global optimal position is output. .

[0016] In summary, the beneficial technical effects of the present invention are as follows: 1. Improved adaptability of accounting dimensions: This invention deeply couples the migration and transformation process of non-point source pollution with carbon sink accounting, and incorporates carbon items specific to non-point source scenarios such as non-point source collaborative carbon reduction and overload carbon loss. This solves the problem of traditional general carbon sink accounting methods omitting carbon items specific to non-point source governance. The accounting logic is in line with the actual operation rules of non-point source governance scenarios, avoiding overestimation or underestimation of carbon sink.

[0017] 2. Improved accounting accuracy and compliance: This invention adopts a three-layer accounting architecture of mechanism model + Bayesian LSTM correction + uncertainty quantification. It retains the interpretability of the mechanism model and adapts to the differentiated characteristics of different regions and seasons through dynamic coefficient correction, thereby improving the accounting accuracy. At the same time, the carbon sink results with confidence intervals output meet the requirements of measurable, reportable and verifiable (MRV) carbon trading.

[0018] 3. Enhanced Synergistic Benefits and Project Sustainability: This invention constructs a closed-loop logic of accounting-optimization-feedback iteration, outputting multi-objective optimization schemes based on real-time carbon sequestration results and pollution control constraints. Under the premise of ensuring compliance with non-point source pollution interception standards and carbon sequestration stability, it achieves increased carbon sequestration revenue and reduced operating costs, expands the carbon trading market revenue channels for non-point source pollution control projects, reduces dependence on fiscal subsidies, and enhances the sustainability of the project throughout its entire life cycle.

[0019] 4. Strong versatility: It can be widely adapted to various non-point source pollution control scenarios such as agricultural non-point source ecological buffer zones, non-point source pollution control artificial wetlands, high-standard farmland, and sponge city rainwater management facilities. It can provide scheme optimization support in the engineering design stage and dynamic control basis in the operation stage. Its application scope covers the current mainstream non-point source pollution control scenarios. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0021] To make the technical means, creative features, objectives and effects of this invention clearer and easier to understand, the invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0022] Example like Figure 1 As shown, the present invention provides a smart carbon sequestration accounting and optimization method for non-point source pollution, which specifically includes the following steps: S1. Define the coupled accounting boundary of the non-point source pollution control unit to be accounted for, and initialize the basic parameters and initial uncertainty matrix required for the accounting; Specifically, the coupled accounting boundary includes spatial boundary, process boundary and compliance boundary, where the process boundary includes carbon sink items and carbon emission items coupled with the migration and transformation process of non-point source pollution; Carbon sequestration includes conventional carbon sequestration and non-point source co-reduction. Conventional carbon sequestration is the total natural carbon sequestration by vegetation and soil within the non-point source pollution control unit, directly related to vegetation growth gain and soil organic matter accumulation after non-point source pollutant interception. Non-point source co-reduction is the carbon reduction achieved by the control unit after intercepting nitrogen and phosphorus non-point source pollutants, avoiding carbon sequestration and degradation. Greenhouse gas emissions such as methane are converted Equivalent to the additional carbon sequestration benefits brought about by the treatment of non-point source pollution; Carbon emissions include operational carbon emissions and carbon loss from non-point source overload. Operational carbon emissions are calculated based on the carbon emissions generated during equipment operation, reagent application, and maintenance work during the non-point source pollution control process of the treatment unit. Equivalent; non-point source overload carbon loss is the carbon released from vegetation death and microbial decomposition when the non-point source pollution load input to the treatment unit exceeds the carrying capacity threshold. equivalent; The spatial boundary is the geographical boundaries and total area of ​​the non-point source pollution treatment unit to be calculated, which is obtained from the project survey data; the compliance boundary is the preset non-point source pollution interception rate and effluent water quality compliance control requirements. The basic initialization parameters include three categories: carbon sink accounting parameters, area source coupling parameters, and optimization constraint parameters, specifically including: (1) Carbon sequestration accounting parameters: Vegetation carbon sequestration baseline coefficient Soil carbon sequestration baseline coefficient This is used for the calculation of conventional carbon fixation in the subsequent step S3; (2) Surface source coupling parameters: unit TN interception synergistic carbon reduction coefficient Unit TP Retention Coefficient for Carbon Reduction Non-point source pollution overload threshold This is used for calculating the area source co-source carbon reduction and overload carbon loss in subsequent step S3; (3) Optimize constraint parameters: carbon trading benchmark price Benchmark value of annual operating cost per unit area Optimize target weight coefficients , This is used for the multi-objective optimization calculation in the subsequent step S4; The initial uncertainty matrix satisfies ,in The initial uncertainty coefficient, It is an identity matrix.

[0023] S2. Collect multi-source heterogeneous monitoring data of the non-point source pollution control units to be accounted for, and perform standardization processing on the raw data to obtain three types of standardized datasets: non-point source pollution load sequence, carbon sink monitoring sequence, and operational carbon emission sequence. The specific steps include: S21. Collect three types of raw monitoring data, including non-point source pollution monitoring data, carbon sink characteristic monitoring data, and operational energy consumption monitoring data, among which: For non-point source pollution monitoring data, water quality sensors and flow sensors are respectively installed at the inlet and outlet of the non-point source pollution treatment unit to be accounted for. The sampling frequency is preferably set to once every 15 minutes to collect the influent TN concentration. TP concentration in influent Inlet flow rate TN concentration in effluent effluent TP concentration Outflow rate , Sampling time; For carbon sink characteristic monitoring data, soil organic carbon content is collected using soil carbon sensors, weather stations, and vegetation biomass monitoring equipment. vegetation biomass Photosynthetically active radiation Temperature ; For operational energy consumption monitoring data, the power consumption of equipment is collected through smart meters, chemical metering devices, and maintenance operation statistical terminals. Diesel consumption for maintenance Dosage of pesticides for non-point source pollution control The sampling frequency is preferably set to 1 time / 1h; S22. Use the Laida criterion to remove outlier data. Data points that satisfy the following formula are identified as outliers and deleted: ; in, For the original monitoring data points, This represents the average value of the same monitoring indicator within a 24-hour statistical window. The standard deviation of the same monitoring indicator within a 24-hour statistical window; S23. Complete the missing data resulting from anomaly removal. The completion formula is: ; in, For moments when data is missing, The most recent valid data moment before the missing moment. It represents the most recent valid data moment after the missing moment. , They are respectively , Valid monitoring values ​​at any given time; S24. Generate three types of standardized datasets. For the non-point source pollution load sequence, calculate the TN and TP pollution loads of the influent and effluent respectively: ; ; in, , They are respectively Constant influent TN and TP pollution load , They are respectively Constantly monitor the TN and TP pollution loads in the effluent; For carbon sink monitoring sequences, the standardized sequence is used. , , , constitute; For operating carbon emission sequences, energy consumption data is converted to... Equivalent, the formula is: ; in, for Constant carbon emissions As a carbon emission factor for electricity, For diesel carbon emission factors, Carbon emission factors of the treatment agents per unit.

[0024] S3. Based on the aforementioned basic parameters and three types of standardized datasets, the real-time net carbon sink, along with uncertain quantification results, is calculated using a mechanism-data fusion dynamic accounting model that couples the migration and transformation process of non-point source pollution. The mechanism-data fusion dynamic accounting model includes a sequentially connected mechanism accounting layer, dynamic correction layer, and uncertain quantification layer. The execution steps include: S31. Mechanism accounting layer couples the interception, transformation and overload process of non-point source pollution in the treatment unit, calculates the initial net carbon sink based on the basic parameters and three types of standardized datasets, and outputs five core coefficients to be corrected corresponding to the dimension of the initial uncertainty matrix. The formula for calculating the initial net carbon sink is: ; in, Conventional carbon sequestration, characterizing the total natural carbon sequestration of vegetation and soil in the treatment unit during the accounting period, is calculated using the following formula: ; In the formula , These are the vegetation carbon sequestration baseline coefficient and the soil carbon sequestration baseline coefficient, respectively, in the core coefficients to be corrected. for The increase in vegetation biomass at the current accounting time compared to the previous accounting time. for The increase in soil organic carbon at the current accounting time compared to the previous accounting time; The carbon reduction from non-point sources represents the additional carbon reduction benefit from the interception of nitrogen and phosphorus non-point source pollutants by the treatment unit. The calculation formula is as follows: ; in, , These are the unit TN retention synergistic carbon reduction coefficient and the unit TP retention synergistic carbon reduction coefficient in the core coefficients to be corrected. To calculate the amount of TN retained during the accounting period, This refers to the TP retention amount during the accounting period; Carbon loss due to non-point source overload characterizes the amount of carbon loss when the pollution load exceeds the carrying capacity of the treatment unit. The calculation formula is as follows: ; in, The threshold for non-point source pollution overload. The total area of ​​the non-point source pollution treatment unit to be calculated. To calculate the total influent TN load during the calculation period, The overload loss coefficient to be corrected is... This represents the amount of regular carbon sequestration at the previous accounting point.

[0025] S32. The dynamic correction layer introduces time-series monitoring features to dynamically correct the five core coefficients to be corrected output by the mechanism accounting layer, eliminating regional and seasonal biases of the general benchmark coefficients, and obtaining the corrected net carbon sink calculation value and correction residual. Specifically, the core coefficients of the mechanistic accounting layer are dynamically corrected using variational Bayesian LSTM. The process includes: (1) Collect historical monitoring data of the target area within a predetermined time period, and synchronously match the actual carbon sequestration true value at the corresponding measurement time as a label, and divide it into training set, validation set and test set according to a preset ratio; (2) The model input is defined as a feature vector composed of a non-point source pollution load sequence, a carbon sink monitoring sequence, an operational carbon emission sequence, and time features. The hidden layer consists of a 2-layer LSTM network used to extract the temporal features of the non-point source-carbon sink coupling process. The output layer is a 5-dimensional dynamic correction coefficient vector. Each parameter represents a corresponding baseline coefficient. The model outputs the approximate posterior distribution parameters of each dynamic correction coefficient at each time step. Specifically, the approximate posterior distribution parameters include the posterior mean and posterior standard deviation of each dynamic correction coefficient. (3) Uncertainty propagation is achieved by learning the probability distribution of the weights to train the model. The specific training logic is as follows: a. Assuming the model has all trainable weights Follows a mean of 0 and a standard deviation of Independent Gaussian prior distributions: ; In the formula The preset prior standard deviation is preferably 0.1; An identity matrix that matches the weight dimensions; Introducing a parameterized approximate posterior distribution Fitting the true posterior distribution ,in For the training dataset, These are trainable parameters that approximate the posterior. Follows an independent Gaussian distribution: ; b. Construct the loss function, using the lower bound of evidence as the optimization objective, with the following formula: ; in, For approximate posterior distribution Expectations The feature vector input to the model, The log-likelihood term corresponds to the goodness of fit between the corrected net carbon sink prediction and the measured true value. For the carbon sink regression task, the Gaussian log-likelihood is taken, which is equivalent to the negative mean square error. ; in The net carbon sink forecast calculated using the corrected coefficients. To measure the true value of carbon sequestration, To observe the standard deviation of noise, a value of 0.05 is preferred; The KL divergence between the approximate posterior and prior distributions is used to constrain the weight distribution and avoid overfitting. c. Perform gradient optimization, using reparameterization techniques to decouple the random sampling process from the trainable parameters, representing the weight sampling as: ; in, For independent standard Gaussian noise, This is an element-wise product; therefore, it is optimized using the Adam gradient descent method. During training, the average loss is calculated by sampling the approximate posterior distribution a predetermined number of times in each batch, when the validation set... Training is stopped when the number of consecutive predetermined rounds does not increase, and a pre-trained correction model adapted to the target region is obtained. (4) After the model is put into operation, online dynamic correction is performed. The latest measured true value of carbon sink is synchronized at predetermined intervals to fine-tune the model in order to adapt to the coefficient drift caused by seasonal and pollution load fluctuations.

[0026] S33. The uncertainty quantification layer iteratively calculates the posterior uncertainty covariance matrix at the current time based on the initial uncertainty matrix, the correction residuals, and the time-series fluctuation characteristics of the core coefficients, generating the final net carbon sink result with confidence intervals. Specific execution steps include: (1) Based on the posterior uncertainty covariance matrix at the previous calculation time, and combined with the process noise of coefficient time drift, the prior uncertainty covariance matrix at the current time is calculated using the following formula: ; in, for The prior uncertainty covariance matrix at time t has the same dimensions as the initial uncertainty matrix; This is the state transition matrix, initially set as the identity matrix, used to describe the time evolution of the core correction coefficients; The initial uncertainty matrix is ​​used in the first calculation, which is the posterior uncertainty covariance matrix at the previous calculation time. ; The process noise matrix reflects the uncertainty of the core correction coefficients' natural drift over time, satisfying: ; in This represents the time interval between the current moment and the previous calculation moment. For coefficient drift rate, It is an identity matrix, with the same dimensions as the core correction coefficients; (2) Combining the prior uncertainty covariance matrix and the observation noise matrix, calculate the Kalman gain at the current time. The calculation formula is as follows: ; in, for The Kalman gain at time step is used to adjust the weight of the observation data on the uncertainty update; The observation noise diagonal matrix consists of diagonal elements that are the squares of the posterior standard deviations of the dynamic correction coefficients output by the variational Bayesian LSTM, reflecting the observation uncertainty of the correction process. (3) Based on the prior covariance and Kalman gain, the posterior uncertainty covariance matrix at the current time is obtained iteratively, and the calculation formula is as follows: ; in, for The posterior uncertainty covariance matrix of the final output at time step; (4) Based on the corrected net carbon sink value and the posterior uncertainty covariance matrix output by the pre-trained correction model, output the real-time net carbon sink with confidence interval: .

[0027] S4. Based on real-time net carbon sink and preset constraints for non-point source pollution control, an optimization scheme for the corresponding scenario is output through a multi-objective optimization model that balances carbon sink benefits and operating costs. The operating parameters of the optimization scheme are then fed back to the dynamic accounting model to achieve closed-loop iteration. The specific steps include: S41. With the goal of maximizing the net synergistic benefits of pollution reduction and carbon reduction throughout the entire life cycle of the non-point source pollution control unit to be accounted for, an optimization function is constructed: ; in, The set of decision variables to be optimized includes two categories: engineering design parameters and operation control parameters. , To optimize the target weight coefficients and satisfy It can be adjusted according to the need to prioritize carbon sink benefits or operating costs; The total net carbon sink over the entire life cycle is obtained by summing up the real-time net carbon sink. The benchmark price for carbon trading; The total operating cost over the entire lifecycle is calculated to satisfy: ; in, This is the benchmark value for annual operating cost per unit area. The total area of ​​the non-point source pollution treatment unit to be calculated. The increase in operating costs resulting from adjusting decision variables; S42. Set constraints for the optimization process, including: (1) The constraints of non-point source pollution control shall meet the following requirements: ; in, TN retention rate Minimum retention rate requirements pre-set for compliance boundaries; The TN concentration in the effluent. The upper limit of TN in effluent as stipulated by local water environment standards; For the total TN load of the influent, This is the overload threshold for non-point source pollution. (2) Project boundary constraints, meeting the project's pre-set usable area and budget requirements: ; in, The actual construction area is [not specified]. Pre-determine the usable area for the project; The total lifecycle cost includes construction and operating costs. This is the total project budget; (3) Carbon sink stability constraints, ensuring that fluctuations in carbon sink volume meet the stability requirements of carbon trading projects, satisfying: ; in, The coefficient of variation is 1. The standard deviation of real-time net carbon sinks within the accounting period. This represents the average real-time net carbon sink over the accounting period. The fluctuation threshold; S43. The objective function is solved using the particle swarm optimization algorithm, and the optimal decision variables are output after iterative convergence. The corresponding optimization scheme feeds back the operating parameters from the optimization scheme to the mechanism-data fusion dynamic accounting model in step S3, updating the input boundary for the next round of accounting and achieving closed-loop iteration. The specific steps for calculating the optimal decision variables include: (1) Set the particle swarm size and the maximum number of iterations. The dimension of each particle is consistent with the dimension of the set of decision variables to be optimized. The initial position of the particles is... Randomly generated within the engineering feasibility interval of each decision variable, where For particle serial numbers; (2) Calculate the particle fitness and the position of each particle. For a specific set of decision variables, Substituting the values ​​into the optimization function in step S41, we obtain the fitness value of the particle: ; Since the optimization objective is to maximize net profit, the higher the fitness value, the better the decision variable scheme corresponding to that particle is. (3) Iterate through the particles, updating their velocity and position in each iteration according to the following formula: ; ; in, Inertial weights are used to balance global and local search capabilities; , These are learning factors, representing the weights by which a particle learns towards its individual optimum and global optimum, respectively. , for Random numbers within the range are used to increase the randomness of the search; For the first Before each particle The optimal position of an individual in the next iteration corresponds to the combination of decision variables that represent the highest fitness value in the particle's history. For the entire particle swarm The global optimal position in the next iteration corresponds to the combination of decision variables with the highest historical fitness value among all particles; (4) When the maximum number of iterations is reached, or the change in the global optimal fitness value after a preset number of consecutive iterations is less than a preset threshold, the iteration is stopped, and the optimal decision variable corresponding to the global optimal position is output. .

[0028] Furthermore, this method adapts to different scenarios based on the different stages of the non-point source pollution control project, adapting to two types of optimization scenarios: engineering design stage optimization and operation stage optimization. The decision variables of the two types of scenarios belong to a subset of the set of decision variables to be optimized in S41, corresponding to engineering design parameters and operation control parameters, respectively. For optimization during the engineering design phase, the triggering time is before the preliminary design of the non-point source pollution control project. At this time, there is no real-time monitoring data, and the historical average monitoring data of the target area over the past three years is used as the calculation input. For different types of treatment units, corresponding engineering design parameters can be selected as decision variables. Specifically, the decision parameters for ecological buffer zone projects include buffer zone width and the proportion of trees, shrubs, and herbs in the vegetation community; for constructed wetland projects, hydraulic retention time and filler type can be added; for high-standard farmland projects, no-till ratio and straw return rate can be added; for sponge city non-point source pollution control facilities projects, the proportion of permeable pavement and rain garden depth can be added. After optimization convergence, the parameter combination corresponding to the optimal decision variables is output as a reference for project design. For dynamic optimization during the operational phase, the triggering time is set automatically once after the project is completed and put into operation. Different operational control parameters are selected as decision variables for different types of governance units. Specifically, general parameters include vegetation harvesting frequency and non-point source pollution control pesticide dosage; for constructed wetland projects, additional parameters can be added, such as aeration power and hydraulic retention time adjustment coefficient; for high-standard farmland projects, additional parameters can be added, such as topdressing amount adjustment coefficient and irrigation water volume adjustment coefficient; and for sponge city facility projects, additional parameters can be added, such as the rainfall threshold for starting and stopping rainwater storage tanks. The optimal operational parameters output after optimization convergence are sent to the governance unit for execution, and simultaneously fed back to the dynamic accounting model to update the carbon sequestration coefficient and operational carbon emission baseline value for the next round of accounting, achieving a closed-loop iteration of accounting and optimization.

[0029] Therefore, this invention provides a smart carbon sequestration accounting and optimization method for non-point source pollution. By defining a dedicated carbon sequestration accounting boundary coupled with the migration process of non-point source pollution, it solves the adaptability problem of traditional general carbon sequestration methods that miss carbon items specific to non-point source scenarios, and establishes an accounting foundation that fits the laws of non-point source pollution control. Utilizing a mechanism-AI fusion dynamic accounting architecture with uncertain quantification, it eliminates the regional and seasonal biases of the general fixed coefficient method, improves accounting accuracy, and ensures that the carbon sequestration results with attached confidence intervals meet the MRV verification requirements for carbon trading. Finally, it constructs a collaborative closed loop of accounting-optimization-feedback iteration for pollution reduction and carbon reduction, achieving increased carbon sequestration benefits and reduced operating costs while ensuring that non-point source pollution interception meets standards. This solves the core pain points of existing technologies, such as the disconnect between non-point source pollution control and carbon sequestration measurement, insufficient accounting accuracy that cannot meet carbon trading requirements, and lack of collaborative optimization paths. It provides a full life-cycle solution that takes into account pollution control, carbon sequestration value-added, and cost control for various non-point source pollution control scenarios, such as agricultural non-point source buffer zones, remedial artificial wetlands, and sponge city facilities.

[0030] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent carbon sequestration accounting and optimization for non-point source pollution, applied to the coordinated management and control of pollution reduction and carbon reduction in non-point source pollution treatment units, characterized in that, The method includes the following steps: S1. Define the coupled accounting boundary of the non-point source pollution control unit to be accounted for, and initialize the basic parameters and initial uncertainty matrix required for the accounting; S2. Collect multi-source heterogeneous monitoring data of the non-point source pollution control unit to be accounted for, and perform standardization processing on the raw data to obtain three types of standardized datasets: non-point source pollution load sequence, carbon sink monitoring sequence, and operational carbon emission sequence. S3. Based on the aforementioned basic parameters and three types of standardized datasets, the real-time net carbon sink is calculated by using a dynamic accounting model that couples the mechanism-data fusion of the non-point source pollution migration and transformation process, along with quantitative results of uncertainties. S4. Based on real-time net carbon sink and preset non-point source pollution control constraints, the multi-objective optimization model that takes into account both carbon sink benefits and operating costs outputs the corresponding optimization scheme. The operating parameters of the optimization scheme are fed back to the dynamic accounting model to achieve closed-loop iteration.

2. The intelligent carbon sequestration accounting and optimization method for non-point source pollution according to claim 1, characterized in that, The coupled accounting boundaries in step S1 include spatial boundaries, process boundaries, and compliance boundaries. The process boundaries include carbon sink and carbon emission items coupled with the non-point source pollution migration and transformation process. The initialization parameters include three categories: carbon sink accounting parameters, non-point source coupling parameters, and optimization constraint parameters. The initial uncertainty matrix satisfies... ,in The initial uncertainty coefficient, It is an identity matrix.

3. The intelligent carbon sequestration accounting and optimization method for non-point source pollution according to claim 2, characterized in that, Step S2 specifically includes: S21. Collect three types of raw monitoring data, including non-point source pollution monitoring data, carbon sink characteristic monitoring data, and operational energy consumption monitoring data, among which: For non-point source pollution monitoring data, water quality sensors and flow sensors are installed at the inlet and outlet of the non-point source pollution treatment unit to be accounted for, respectively, to collect the influent TN concentration. TP concentration in influent Inlet flow rate TN concentration in effluent effluent TP concentration Outflow rate , Sampling time; For carbon sink characteristic monitoring data, soil organic carbon content is collected using soil carbon sensors, weather stations, and vegetation biomass monitoring equipment. vegetation biomass Photosynthetically active radiation Temperature ; For operational energy consumption monitoring data, the power consumption of equipment is collected through smart meters, chemical metering devices, and maintenance operation statistical terminals. Diesel consumption for maintenance Dosage of pesticides for non-point source pollution control ; S22. Use the Laida criterion to remove outlier data. Data points that satisfy the following formula are identified as outliers and deleted: ； in, For the original monitoring data points, This represents the average value of the same monitoring indicator within a 24-hour statistical window. The standard deviation of the same monitoring indicator within a 24-hour statistical window; S23. Complete the missing data resulting from anomaly removal. The completion formula is: ； in, For moments when data is missing, The most recent valid data moment before the missing moment. It represents the most recent valid data moment after the missing moment. , They are respectively , Valid monitoring values ​​at any given time; S24. Generate three types of standardized datasets. For the non-point source pollution load sequence, calculate the TN and TP pollution loads of the influent and effluent respectively: ； ； in, , They are respectively Constant influent TN and TP pollution load , They are respectively Constantly monitor the TN and TP pollution loads in the effluent; For carbon sink monitoring sequences, the standardized sequence is used. , , , constitute; For operating carbon emission sequences, energy consumption data is converted to... Equivalent, the formula is: ； in, for Constant carbon emissions As a carbon emission factor for electricity, For diesel carbon emission factors, Carbon emission factors of the treatment agents per unit.

4. The intelligent carbon sequestration accounting and optimization method for non-point source pollution according to claim 3, characterized in that, The mechanism-data fusion dynamic accounting model for the coupled non-point source pollution migration and transformation process described in step S3 includes a mechanism accounting layer, a dynamic correction layer, and an uncertainty quantification layer connected in sequence. The execution steps include: S31. Mechanism accounting layer couples the interception, transformation and overload process of non-point source pollution in the treatment unit, calculates the initial net carbon sink based on the basic parameters and three types of standardized datasets, and outputs five core coefficients to be corrected corresponding to the dimension of the initial uncertainty matrix. S32. The dynamic correction layer introduces time-series monitoring features to dynamically correct the five core coefficients to be corrected output by the mechanism accounting layer, and obtains the corrected net carbon sink calculation value and correction residual. S33. The uncertainty quantification layer iteratively calculates the posterior uncertainty covariance matrix at the current time based on the temporal fluctuation characteristics of the initial uncertainty matrix, correction residuals, and core coefficients, and generates the final net carbon sink result with confidence intervals.

5. The intelligent carbon sequestration and optimization method for non-point source pollution according to claim 4, characterized in that, The formula for calculating the initial net carbon sink in step S31 is: ； in, The standard carbon sequestration amount is calculated using the following formula: ； In the formula , These are the vegetation carbon sequestration baseline coefficient and the soil carbon sequestration baseline coefficient, respectively, in the core coefficients to be corrected. for The increase in vegetation biomass at the current accounting time compared to the previous accounting time. for The increase in soil organic carbon at the current accounting time compared to the previous accounting time; The formula for calculating the carbon reduction from area sources is as follows: ； in, , These are the unit TN retention synergistic carbon reduction coefficient and the unit TP retention synergistic carbon reduction coefficient in the core coefficients to be corrected. To calculate the amount of TN retained during the accounting period, This refers to the TP retention amount during the accounting period; The carbon loss due to area source overload is calculated using the following formula: ； in, The threshold for non-point source pollution overload. The total area of ​​the non-point source pollution treatment unit to be calculated. To calculate the total influent TN load during the calculation period, The overload loss coefficient to be corrected is... This represents the amount of regular carbon sequestration at the previous accounting point.

6. The intelligent carbon sequestration and optimization method for non-point source pollution according to claim 5, characterized in that, Step S32 dynamically corrects the core coefficients of the mechanistic accounting layer using variational Bayesian LSTM. The specific process includes: (1) Collect historical monitoring data of the target area within a predetermined time period, and synchronously match the actual carbon sequestration true value at the corresponding measurement time as a label, and divide it into training set, validation set and test set according to a preset ratio; (2) The model input is defined as a feature vector composed of a non-point source pollution load sequence, a carbon sink monitoring sequence, an operational carbon emission sequence, and time features. The hidden layer consists of a 2-layer LSTM network used to extract the temporal features of the non-point source-carbon sink coupling process. The output layer is a 5-dimensional dynamic correction coefficient vector. Each parameter represents a corresponding baseline coefficient. The model outputs the approximate posterior distribution parameters of each dynamic correction coefficient at each time step, including the posterior mean and posterior standard deviation of each dynamic correction coefficient. (3) Uncertainty propagation is achieved by learning the probability distribution of the weights to train the model. The specific training logic is as follows: a. Assuming the model has all trainable weights Follows a mean of 0 and a standard deviation of Independent Gaussian prior distributions: ； In the formula The pre-defined prior standard deviation, An identity matrix that matches the weight dimensions; Introducing a parameterized approximate posterior distribution Fitting the true posterior distribution ,in For the training dataset, These are trainable parameters that approximate the posterior. Follows an independent Gaussian distribution: ； b. Construct the loss function, using the lower bound of evidence as the optimization objective, with the following formula: ； in, For approximate posterior distribution Expectations The feature vector input to the model, The log-likelihood term corresponds to the degree of fit between the corrected net carbon sink predicted value and the measured true value. Its expression is: ； in The net carbon sink forecast calculated using the corrected coefficients. To measure the true value of carbon sequestration, The standard deviation of the observed noise; The KL divergence between the approximate posterior and prior distributions is used to constrain the weight distribution and avoid overfitting. c. Perform gradient optimization, using reparameterization techniques to decouple the random sampling process from the trainable parameters, representing the weight sampling as: ； in, For independent standard Gaussian noise, This is an element-wise product; therefore, it is optimized using the Adam gradient descent method. During training, the average loss is calculated by sampling the approximate posterior distribution a predetermined number of times in each batch, when the validation set... Training is stopped when the number of consecutive predetermined rounds does not increase, and a pre-trained correction model adapted to the target region is obtained. (4) After the model is put into operation, online dynamic calibration is performed, and the model is fine-tuned by synchronizing the latest measured true value of carbon sink at predetermined intervals.

7. The intelligent carbon sequestration and optimization method for non-point source pollution according to claim 6, characterized in that, The specific execution steps of step S33, the uncertainty quantification layer, include: (1) Based on the posterior uncertainty covariance matrix at the previous calculation time, and combined with the process noise of coefficient time drift, the prior uncertainty covariance matrix at the current time is calculated using the following formula: ； in, for The prior uncertainty covariance matrix at time t has the same dimensions as the initial uncertainty matrix; This is the state transition matrix, initially set as the identity matrix, used to describe the time evolution of the core correction coefficients; The initial uncertainty matrix is ​​used in the first calculation, which is the posterior uncertainty covariance matrix at the previous calculation time. ; The process noise matrix reflects the uncertainty of the core correction coefficients' natural drift over time, satisfying: ； in This represents the time interval between the current moment and the previous calculation moment. For coefficient drift rate, It is an identity matrix, with the same dimensions as the core correction coefficients; (2) Combining the prior uncertainty covariance matrix and the observation noise matrix, calculate the Kalman gain at the current time. The calculation formula is as follows: ； in, for The Kalman gain at time step is used to adjust the weight of the observation data on the uncertainty update; The observation noise diagonal matrix consists of diagonal elements that are the squares of the posterior standard deviations of the dynamic correction coefficients output by the variational Bayesian LSTM, reflecting the observation uncertainty of the correction process. (3) Based on the prior covariance and Kalman gain, the posterior uncertainty covariance matrix at the current time is obtained iteratively, and the calculation formula is as follows: ； in, for The posterior uncertainty covariance matrix of the final output at time step; (4) Based on the corrected net carbon sink value and the posterior uncertainty covariance matrix output by the pre-trained correction model, output the real-time net carbon sink with confidence interval: .

8. The intelligent carbon sequestration and optimization method for non-point source pollution according to claim 6, characterized in that, Step S4 specifically includes: S41. With the goal of maximizing the net synergistic benefits of pollution reduction and carbon reduction throughout the entire life cycle of the non-point source pollution control unit to be accounted for, an optimization function is constructed: ； in, The set of decision variables to be optimized includes two categories: engineering design parameters and operation control parameters. , To optimize the target weight coefficients and satisfy ; The total net carbon sink over the entire life cycle is obtained by summing up the real-time net carbon sink. The benchmark price for carbon trading; The total operating cost over the entire lifecycle is calculated to satisfy: ； in, This is the benchmark value for annual operating cost per unit area. The total area of ​​the non-point source pollution treatment unit to be calculated. The increase in operating costs resulting from adjusting decision variables; S42. Set constraints for the optimization process, including: (1) The constraints of non-point source pollution control shall meet the following requirements: ； in, TN retention rate Minimum retention rate requirements pre-set for compliance boundaries; The TN concentration in the effluent. The upper limit of TN in effluent as stipulated by local water environment standards; For the total TN load of the influent, This is the overload threshold for non-point source pollution. (2) Project boundary constraints, meeting the project's pre-set usable area and budget requirements: ； in, The actual construction area is [not specified]. Pre-determine the usable area for the project; The total lifecycle cost includes construction and operating costs. This is the total project budget; (3) Carbon sink stability constraints, ensuring that fluctuations in carbon sink volume meet the stability requirements of carbon trading projects, satisfying: ； in, The coefficient of variation is 1. The standard deviation of real-time net carbon sinks within the accounting period. This represents the average real-time net carbon sink over the accounting period. The fluctuation threshold; S43. The objective function is solved using the particle swarm optimization algorithm, and the optimal decision variables are output after iterative convergence. The corresponding optimization scheme feeds back the operating parameters in the optimization scheme to the mechanism-data fusion dynamic accounting model in step S3, updates the input boundary of the next round of accounting, and realizes closed-loop iteration.

9. The intelligent carbon sequestration and optimization method for non-point source pollution according to claim 8, characterized in that, Step S43 specifically includes: (1) Set the particle swarm size and the maximum number of iterations. The dimension of each particle is consistent with the dimension of the set of decision variables to be optimized. The initial position of the particles is... Randomly generated within the engineering feasibility interval of each decision variable, where For particle serial numbers; (2) Calculate the particle fitness and the position of each particle. For a specific set of decision variables, Substituting the values ​​into the optimization function in step S41, we obtain the fitness value of the particle: ； (3) Iterate through the particles, updating their velocity and position in each iteration according to the following formula: ； ； in, Inertial weights are used to balance global and local search capabilities; , These are learning factors, representing the weights by which a particle learns towards its individual optimum and global optimum, respectively. , for Random numbers within the range are used to increase the randomness of the search; For the first Before each particle The optimal position of an individual in the next iteration corresponds to the combination of decision variables that represent the highest fitness value in the particle's history. For the entire particle swarm The global optimal position in the next iteration corresponds to the combination of decision variables with the highest historical fitness value among all particles; (4) When the maximum number of iterations is reached, or the change in the global optimal fitness value after a preset number of consecutive iterations is less than a preset threshold, the iteration is stopped, and the optimal decision variable corresponding to the global optimal position is output. .

Images