Low-carbon park comprehensive energy system collaborative operation and dispatching method based on dynamic carbon potential

Abstract

The present application relates to a kind of low-carbon park integrated energy system collaborative operation scheduling method based on dynamic carbon potential, the method first constructs physical information neural network model, and based on historical data etc. Feature output prediction marginal carbon potential;Then it is used as optimization signal, combined with model predictive control to carry out charge-discharge optimization scheduling to electric energy storage, output system operating state data;Then based on the data, using graph neural network extracts power grid topology and power flow distribution characteristics to calculate node electric carbon potential, based on the second law of thermodynamics and entropy increase principle to calculate node thermal carbon potential;Finally, the actual energy consumption of each main body is multiplied by the node electric, thermal carbon potential to obtain the basic carbon responsibility, combined with entropy weight method to obtain the comprehensive collaborative factor of multi-dimensional factor and the incentive compatible adjustment coefficient of high-carbon period response, multiplication output each main body's dynamic carbon responsibility allocation result.Compared with prior art, the present application realizes the fine description of park carbon emission and the deep binding of carbon responsibility allocation and carbon reduction behavior.

Description

Technical Field

[0001] This invention relates to the field of carbon emission management of integrated energy systems, and in particular to a collaborative operation scheduling method for integrated energy systems in low-carbon industrial parks based on dynamic carbon potential. Background Technology

[0002] Low-carbon parks, represented by industrial parks and commercial complexes, have evolved into integrated energy systems that couple electricity and heat. Given the differentiated energy consumption behaviors of heterogeneous entities within the parks and the large-scale integration of low-carbon energy supply equipment, how to achieve a scientific and fair carbon responsibility traceability and incentive mechanism has become a key challenge in the field of carbon management.

[0003] Current research and practice on carbon management in industrial parks have many limitations: In carbon emission accounting, traditional methods rely on fixed or average carbon potential, which cannot reflect the dynamic changes in system operation status and their marginal impact; in carbon flow tracking, the commonly used proportional sharing method is difficult to capture the nonlinear characteristics of power grid topology and power flow distribution, and lacks description of transmission heat loss and entropy increase effects in heating network modeling, leading to carbon flow allocation bias; in terms of carbon responsibility allocation and synergistic benefits, existing methods tend to be based on linear allocation of energy consumption, ignoring behavioral factors such as load time sequence characteristics, green electricity consumption, and energy use flexibility, while the traditional Shapley value model, due to the assumption of symmetrical contributions from subjects, fails to effectively distinguish the actual differences in technical capabilities and coordination among subjects, resulting in allocation results that lack positive incentives for low-carbon behavior.

[0004] Despite some progress made in carbon management in industrial parks in recent years, the following four core issues and challenges remain when considering existing technologies and methods.

[0005] (1) The carbon potential characterization lacks marginal sensitivity, making it difficult to guide the low-carbon response on the load side. Existing average carbon potential cannot respond to changes in system state in real time, nor can it characterize the marginal carbon emission increment caused by unit load variation, making it difficult for regulation measures such as energy storage and load shifting to accurately pinpoint high-carbon periods. Furthermore, purely data-driven prediction models do not incorporate physical constraints of the power grid, and their prediction results are prone to deviating from the laws governing power operation.

[0006] (2) The carbon flow tracing method does not adequately consider network topology and thermal characteristics, resulting in low spatiotemporal resolution. Traditional methods on the power grid side neglect the nonlinear coupling between topology and power flow, making it impossible to reasonably allocate the carbon responsibility for grid losses; methods on the heating network side do not consider the entropy increase and carbon increase effect caused by heat loss and transmission delay, resulting in an underestimation of the carbon potential of the heating network.

[0007] (3) The carbon responsibility allocation method is too simplistic and lacks incentive-compatible design for behavioral responses. The existing "responsibility based on quantity" approach has limitations: first, it cannot distinguish the temporal differences in electricity consumption by entities during different carbon potential periods; second, it lacks an independent incentive channel for proactive peak shaving and green electricity consumption, and has not constructed a multi-dimensional collaborative responsibility factor that covers load resilience and entity interaction.

[0008] (4) The collaborative contribution allocation ignores subject heterogeneity, and the Shapley value lacks adaptive correction. Traditional Shapley values ​​do not take into account the heterogeneity of entities in terms of technological capabilities, load characteristics, and interaction, which can easily lead to "free-riding" phenomena. At the same time, the lack of multi-dimensional decomposition makes it difficult to provide clear guidance for differentiated emission reduction paths for entities.

[0009] Patent CN121457998A relates to a multi-entity collaborative carbon responsibility allocation method for a low-carbon industrial park integrated energy system. The steps include: collecting multi-source data of the park and performing multi-energy flow optimization scheduling with the goal of minimizing the total operating cost of the integrated energy system; constructing a dynamic carbon flow distribution matrix using a two-way carbon flow dynamic correction model that considers carbon flow transfer caused by energy storage charging and discharging processes; decoupling the total load into multiple sub-load categories using a load carbon responsibility decoupling allocation model, and assigning different carbon responsibility allocation coefficients to each sub-load category; taking various stakeholders in the park as game participants, Nash bargaining as a collaborative optimization framework, maximizing the utility gain of all participants as the objective function, and Shapley value as the basis for carbon responsibility allocation, performing multi-entity collaborative optimization, and obtaining the carbon responsibility allocation scheme for each entity by combining the user carbon responsibility allocation coefficient. However, it has shortcomings in high-precision spatiotemporal characterization of carbon potential, low-carbon closed-loop control, and multi-dimensional behavioral incentives. It cannot accurately capture the dynamic spatiotemporal evolution of carbon potential; it has not achieved closed-loop real-time feedback between marginal carbon potential and energy storage scheduling; and the allocation of carbon responsibility relies on static load classification and weighting, failing to achieve deep binding between the actual response behavior of the subject and the incentive compatibility mechanism. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a collaborative operation and scheduling method for a low-carbon park integrated energy system based on dynamic carbon potential, which realizes the refined prediction and source tracing of carbon potential in the electro-thermal coupled park, as well as the deep binding of multi-entity carbon responsibility allocation and carbon reduction behavior incentives.

[0011] The objective of this invention can be achieved through the following technical solutions: A collaborative operation and scheduling method for integrated energy systems in low-carbon industrial parks based on dynamic carbon potential includes the following steps: A physical information neural network model is constructed, and a physical constraint error, including the power load balance of the park, the upper and lower limits of generator output, and the energy storage power constraint, is introduced into the loss function of the physical information neural network model to train the physical information neural network model. The physical information neural network model is trained by inputting historical load, historical marginal carbon potential, weather data, and date characteristics of the park into the input, and outputs the predicted marginal carbon potential. Using the predicted marginal carbon potential as an optimization signal, model predictive control is used to optimize the charging and discharging of the park's energy storage system, and outputs system operation status data including high carbon periods, photovoltaic output periods, actual energy consumption and branch network losses. Based on the system operating status data, a graph neural network is used to extract the topology features and power flow distribution features of the power grid to calculate the node electrical carbon potential, and the node thermal carbon potential is calculated based on the second law of thermodynamics and the principle of entropy increase. The basic carbon responsibility is obtained by multiplying the actual energy consumption by the node electrical carbon potential and the node thermal carbon potential; by calculating the load time series factor, renewable energy absorption factor, load elasticity factor and subject interaction factor of each subject, the entropy weight method is used to assign weights to each factor to obtain the comprehensive synergy factor, and the incentive compatibility adjustment coefficient is calculated in combination with the response contribution degree of high carbon period to obtain the dynamic carbon responsibility of each subject. The dynamic carbon responsibility system generates load response control instructions and sends them to each entity to dynamically adjust the energy load of each entity, thereby realizing the low-carbon collaborative operation and scheduling of the park's integrated energy system.

[0012] Furthermore, when training the physical information neural network model, load forecast data is acquired, and the load forecast data is input into an economic dispatch model that includes unit output ranking to back-calculate the reference output of marginal units; the predicted marginal unit output is obtained through the predicted marginal carbon potential; the physical constraint error is the error between the predicted marginal unit output and the reference output of the marginal unit.

[0013] Furthermore, in the step of using model predictive control to optimize the charging and discharging scheduling of the park's electric energy storage system, the purchased power, natural gas consumption, and natural gas carbon emission factor are first obtained; the objective function is set to minimize system carbon emissions, where the system carbon emissions are the sum of a first product and a second product, the first product being the product of the predicted marginal carbon potential and the purchased power, and the second product being the product of the natural gas consumption and the natural gas carbon emission factor.

[0014] Furthermore, in the step of using a graph neural network to extract grid topology features and power flow distribution features to calculate node electrocarbon potential, the grid bus is used as the node set of the graph neural network, and the transmission line is used as the edge set of the graph neural network; the update rule of the node hidden state includes aggregating the edge features of neighboring nodes through an aggregation function, multiplying them by a learnable weight matrix, and inputting them into an activation function.

[0015] Furthermore, after the step of using a graph neural network to extract grid topology features and power flow distribution features to calculate the nodal electrocarbon potential, the method further includes: Obtain the system's average carbon potential and current sensitivity; The network loss allocation coefficient is calculated based on the power flow sensitivity. The branch network loss is multiplied by the system average carbon potential and the network loss allocation coefficient as an additional load and superimposed on each node in the node set to correct the node's electrical carbon potential.

[0016] Furthermore, the step of calculating the nodal thermal carbon potential based on the second law of thermodynamics and the principle of entropy increase specifically includes: Obtain the carbon compensation coefficient for heat loss and the ambient temperature; Calculate the initial nodal thermal carbon potential, and use the heat loss carbon compensation coefficient and the ambient temperature to calculate entropy increase compensation; The entropy increase compensation is superimposed on the initial node thermal carbon potential to obtain the node thermal carbon potential.

[0017] Furthermore, the calculation of the load time-series factor, renewable energy absorption factor, load resilience factor, and entity interaction factor for each entity specifically includes: Obtain the amount of electricity traded between the power generation carbon potential and the direct energy exchange between the entities, and calculate the entity interaction factor based on the product of the amount of electricity traded between the entities and the power generation carbon potential; The load timing factor is the normalized value of the proportion of each entity's electrical load to its total electrical load during the high-carbon period. The renewable energy absorption factor is the normalized value of the proportion of each entity's electricity load to its total electricity load during the photovoltaic output period; Obtain the elastic coefficients obtained from the regression of the elastic response model, and use the elastic coefficients as the load elasticity factor.

[0018] Furthermore, the calculation of the incentive compatibility adjustment coefficient based on the response contribution during high-carbon periods specifically includes: Obtain reference energy consumption and reference carbon potential; The product of the difference between the actual energy consumption and the reference energy consumption during the high-carbon period and the reference carbon potential is taken as the actual response contribution. The excitation compatibility adjustment coefficient decreases as the actual response contribution increases.

[0019] Furthermore, after outputting the dynamic carbon responsibility allocation results for each entity, the method also includes: Treating each entity as a cooperative game participant, a cooperative game model is constructed with the carbon emission reduction of the cooperative operation relative to independent operation as the characteristic function, and the Shapley value of each entity is calculated using the characteristic function.

[0020] Furthermore, after calculating the Shapley value of each subject using the characteristic function, the method further includes: Obtain the percentage of equipment self-consumption by each entity and the amount of electricity traded directly between entities; The synergistic contribution coefficient is constructed by dividing the comprehensive synergistic factor by the mean of the comprehensive synergistic factors of all subjects. The adaptive Shapley value is obtained by multiplying the synergistic contribution coefficient by the Shapley value. The adaptive Sharpe ratio is decomposed into technical contribution, load characteristic contribution, and collaborative contribution based on the proportion of the equipment's self-consumption, the proportion of the inverse of the load time series factor, and the allocation ratio of the electricity in the direct energy trading between the entities.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. Improve the physical consistency and robustness of carbon potential prediction: Construct a physical information neural network model and add physical constraint error to the loss function for training. Based on historical data and other features, output the predicted marginal carbon potential so that the prediction results can fit the data while satisfying the physical laws of unit output and grid operation.

[0022] 2. Forming a closed-loop low-carbon operation scheduling: Using the predicted marginal carbon potential as an optimization signal, combined with model predictive control, the charging and discharging of the park's energy storage system is optimized and scheduled, dynamically guiding the park's energy storage to operate in a low-carbon manner and shaving off peaks and filling valleys.

[0023] 3. Achieve precise spatiotemporal calculation of carbon emissions from electro-thermal coupled systems: Graph neural networks are used to extract grid topology and power flow distribution features to calculate node electrical carbon potential. At the same time, based on the second law of thermodynamics and the principle of entropy increase, node thermal carbon potential is calculated to accurately characterize the impact of network nonlinear distribution and heat loss on carbon potential.

[0024] 4. Achieve unity between carbon responsibility allocation and carbon reduction incentives: Utilize the entropy weight method to integrate four-dimensional factors, including load time series, renewable energy consumption, load elasticity, and subject interaction, to obtain a comprehensive synergy factor. Combine this with the incentive compatibility adjustment coefficient calculated based on actual energy consumption during high-carbon periods. Finally, multiply the two factors with the basic carbon responsibility to obtain a dynamic allocation result, so that the allocation of responsibility is tilted towards low-carbon, flexible, and collaborative actual behaviors. Attached Figure Description

[0025] Figure 1 This is a general framework diagram of the present invention; Figure 2 This is a diagram of the system topology. Figure 3 This is a flowchart of the optimization algorithm of the present invention; Figure 4 Diagram of the integrated electricity-heat energy system architecture for the park; Figure 5 This is the electrical load curve of the present invention; Figure 6 This is the load statistics table of the present invention; Figure 7 This is the heat load curve of the present invention; Figure 8 This is the electrical load timing distribution table of the present invention; Figure 9 This is the photovoltaic power generation output curve of the present invention; Figure 10 This invention relates to the marginal carbon potential of the power grid. Figure 11 This invention relates to the carbon potential-load time-series correlation. Figure 12 This is the carbon potential distribution table for this invention; Figure 13 This is the power generation component of the present invention; Figure 14 This is the heating component of the present invention; Figure 15 This is the energy storage state of charge of the present invention; Figure 16 The carbon emission composition of this invention; Figure 17 A comparison of the carbon potential at key nodes in this invention; Figure 18 This is an indicator of the dynamic characteristics of carbon potential in this invention; Figure 19 This invention demonstrates the photovoltaic carbon potential suppression effect; Figure 20 The carbon emission timeline of the system of this invention; Figure 21 For comparison of carbon responsibility allocation in this invention; Figure 22 The rate of change of the dynamic method of this invention; Figure 23 This constitutes the basic carbon responsibility component of the present invention; Figure 24 This is the carbon responsibility allocation fairness table for this invention; Figure 25 This refers to the synergy factor and incentive compatibility adjustment coefficient of this invention. Figure 26 This is a comparison table of scenarios for subject B of the present invention; Figure 27 This is a table showing the economic impact of carbon costs in this invention. Detailed Implementation

[0026] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0027] Example 1 This embodiment provides a collaborative operation and scheduling method for a low-carbon industrial park integrated energy system based on dynamic carbon potential, such as... Figure 1 As shown, the specific steps are as follows: Construct a Physical Information Neural Network (PINN) model, and train the Physical Information Neural Network model by adding physical constraint error to the loss function of the Physical Information Neural Network model; The loss function of the physical information neural network consists of two parts: data fitting loss and physical constraint loss.

[0028] Network input: historical load Historical marginal carbon potential Weather data (Temperature, Irradiance), Date Characteristics (Weekdays / Holidays).

[0029] Network output: Predicting marginal carbon potential .

[0030] Data fitting loss: Physical constraint loss: In the formula: The predicted marginal unit output is derived from the load forecast through an economic dispatch model.

[0031] Total loss function: (39) In the formula: The physical constraint weight coefficient is set to 0.1. The network is trained through backpropagation to ensure that the prediction results both fit historical data and satisfy the physical laws of the power system.

[0032] The physical information neural network model is trained by inputting historical load, historical marginal carbon potential, weather data, and date characteristics of the park into the input, and outputs the predicted marginal carbon potential. like Figure 2 As shown, the low-carbon industrial park in this embodiment consists of three energy-consuming entities: a manufacturing enterprise (entity A), a commercial complex (entity B), and a residential community (entity C). The energy supply side of the system includes gas turbines, gas boilers, photovoltaic power generation systems, electric energy storage systems, and thermal energy storage systems. Each entity achieves energy coupling through the park's power grid and heating network, and can purchase and sell electricity from the external power grid.

[0033] Load time-series characteristics are the basis for carbon responsibility allocation. This embodiment establishes a load model that considers weekday patterns, peak periods, and random fluctuations based on the energy consumption behavior characteristics of different entities.

[0034] Industrial load is primarily driven by continuous production, exhibiting distinct day and night shift characteristics. The electrical load model for the industrial load model (subject A) is as follows: In the formula: The electrical load (kW) of subject A at time t; Based on the base load value (kW); This refers to the production load factor. This is a function indicating production time periods. It takes 1 for weekdays 8:00-12:00 and 13:00-17:00, 0.3 for lunch break at 12:00, and 0 for other time periods. This refers to the start-stop impact coefficient; The start-stop impact function has peak values ​​of 0.4 and 0.3 at 8:00 and 13:00, respectively. The term represents random fluctuations, with a mean of 0 and a standard deviation of . It follows a normal distribution.

[0035] Industrial heat load is mainly process heat, and the model is as follows: In the formula: The heat load (kW) of subject A at time t; Based on the base heat load (kW); For production heat load coefficient; This represents the random fluctuation term of the heat load.

[0036] Commercial load is primarily characterized by business hours, including midday and evening peak hours. The electrical load model for the commercial load model (Subject B) is as follows: In the formula: The electrical load (kW) of subject B at time t; Based on the base load value (kW); This is the operating load factor; This is a function indicating business hours; it takes the value 1 for hours from 10:00 to 22:00. , These are the coefficients for the midday peak and the evening peak, respectively. , The peak shape function is fitted using a sine function: Lighting load factor; This is a function indicating the lighting period; the value is 0.3 for 17:00-23:00. It represents a random fluctuation term.

[0037] The commercial heat load model is as follows: In the formula: The heat load (kW) of subject B at time t; Based on the base heat load (kW); , , These are the heat load coefficients for business hours, midday peak, and evening peak, respectively. It represents a random fluctuation term.

[0038] Residential load exhibits a dual-peak characteristic, occurring in the morning and evening. The electrical load model for the residential load model (subject C) is as follows: In the formula: The electrical load (kW) of the main component C at time t; Based on the base load value (kW); , These are the morning peak and evening peak coefficients, respectively. , Peak shape function: This refers to the nighttime base load factor. This is a function indicating the nighttime hours, with a value of 0.2 for 0:00-6:00 and 23:00-24:00. It represents a random fluctuation term.

[0039] The main heat load for residents is domestic hot water: In the formula: The heat load (kW) of the main body C at time t; Based on the base heat load (kW); , These are the heat load coefficients for the morning peak and evening peak, respectively. It represents a random fluctuation term.

[0040] The operating models of the energy supply equipment on the system's energy supply side include gas turbine models, gas boiler models, photovoltaic power generation system models, electric energy storage system models, and thermal energy storage system models. The modeling methods for each model are as follows: Photovoltaic power generation system model: Photovoltaic power output is affected by solar irradiance, ambient temperature, and weather conditions. This embodiment adopts a photovoltaic power generation model based on physical processes. Specifically, the modeling method for distributed photovoltaic power generation is as follows: Calculation of solar altitude angle: In the formula: Let t be the solar altitude angle (rad). The latitude is the local latitude (rad). The solar declination angle (rad); Let t be the solar hour angle (rad). .

[0041] Total solar irradiance consists of direct radiation and diffuse radiation: Direct radiation: Scattered radiation: In the formula: The total solar irradiance at time t (kW / m²) 2 ); , Direct and diffuse irradiance (kW / m²) 2 ); Clear-sky direct irradiance (kW / m²) 2 ), take 1000 W / m 2 ; The weather impact factor is set at 1.0 for sunny days and 0.3 for cloudy days. Basic diffuse irradiance (kW / m 2 Take 100 W / m 2 ; The scattering coefficient is taken as 0.5.

[0042] Solar panel temperature: In the formula: Let t be the temperature of the solar panel at time t (°C). Let t be the ambient temperature (°C). The temperature rise coefficient of the solar panel is taken as 0.03°C·m. 2 / W.

[0043] Photovoltaic power generation: In the formula: Let t be the photovoltaic power generation (kW); The efficiency of the photovoltaic module is taken as 0.18; Photovoltaic array area (m²) 2 ), take 5500 m 2 ; The temperature coefficient is taken as -0.004 / °C; The installed capacity of the photovoltaic system (kW) is taken as 1000kW.

[0044] To account for random fluctuations caused by cloud cover, the photovoltaic output is adjusted as follows: In the formula: The corrected photovoltaic power generation (kW); The cloud layer influence factor follows a normal distribution with a mean of 1 and a standard deviation of 0.1, and only affects the daytime period (6:00-19:00). This refers to the daytime period.

[0045] Gas turbine model: The gas turbine operates in combined heat and power (CHP) mode, using natural gas to generate electricity and heat. The relationship between its electrical power and thermal power is as follows: In the formula: , These represent the power generation (kW) and heating power (kW) of the gas turbine at time t, respectively. , The power generation efficiency and heat recovery efficiency are respectively taken as 0.38 and 0.42; Let t be the natural gas consumption (kW, converted to calorific value).

[0046] Gas turbine operation must meet technical output constraints and ramp-up constraints: In the formula: , The minimum and maximum technical output (kW) are respectively taken as 400kW and 2000kW; The start / stop status variable at time t (1 indicates running, 0 indicates stopping); The gradeability is set at 300 kW / h. The time step is set to 1 hour.

[0047] Gas-fired boilers only provide heat load: In the formula: Let t be the heating power (kW) of the gas-fired boiler at time t; For thermal efficiency, we take 0.88; Natural gas consumption (kW); , The minimum and maximum technical output (kW) are taken as 200 kW and 1500 kW, respectively; These are start / stop state variables.

[0048] Electric energy storage systems achieve energy time shifting through charging and discharging; their dynamic model is as follows: In the formula: The state of charge (kWh) of the electrical energy storage at time t; The self-discharge rate is set to 0.001. , These are the charging and discharging power (kW), respectively. , The values ​​are the charging and discharging efficiencies, both taken as 0.95.

[0049] Energy storage operation constraints: In the formula: , These are the maximum charging and discharging power (kW), both taken as 500kW; , For charging and discharging state variables (0-1 variables); , These represent the minimum and maximum allowable state of charge (kWh), respectively, and are set to 200kWh and 2000kWh.

[0050] Thermal energy storage model: In the formula: Let t be the amount of heat stored in the thermal energy storage (kWh); The heat loss coefficient is taken as 0.005; , These are the heat charging and heat dissipation power (kW), respectively. The heat dissipation efficiency is taken as 0.92.

[0051] Thermal energy storage operation constraints: In the formula: , These are the maximum heat charging and heat dissipation power (kW), both taken as 400kW; , For the state variables of heat release; , These are the minimum and maximum allowable heat storage (kWh), respectively, taken as 50kWh and 1000kWh.

[0052] The physical definition of the predicted marginal carbon potential is as follows: Let the power grid system contain G generator units, and the generation cost function of unit g is... and carbon emission function Both are quadratic functions: In the formula: The carbon emission coefficient of unit g is determined by the unit type and operating characteristics; Power output of the generator unit (kW).

[0053] The system's economic dispatch aims to minimize total power generation cost, with marginal cost determining the order of unit output. Marginal carbon potential is defined as the system's ability to meet load increments. At that time, the incremental carbon emissions generated by marginal units: In the formula: m(t) is the marginal unit number at time t; Output (kW) to marginal units. Marginal units are determined by the system scheduling sequence and are usually coal-fired power units.

[0054] Using the predicted marginal carbon potential as an optimization signal, model predictive control is employed to optimize the charging and discharging scheduling of the park's energy storage system, outputting system operation status data including high-carbon periods, photovoltaic output periods, actual energy consumption, and branch network losses. The specific steps are as follows: using the marginal carbon potential as an optimization signal, an energy storage charging and discharging optimization model is established, and the prediction error is handled by the rolling optimization method of model predictive control (MPC).

[0055] Optimization objective: Minimize system carbon emissions In the formula: H is the prediction time domain, which is taken as 4 hours; The marginal carbon potential predicted by PINN; Power purchased (kW); The carbon emission factor for natural gas (kgCO2 / kWh).

[0056] MPC rolling optimization steps: 1. At time t, predict the load and marginal carbon potential for the next H steps based on the current state; 2. Solve the optimization problem to obtain the control sequence for the next H steps; 3. Perform the first step of control and update the system status; 4. Repeat the above steps.

[0057] Define the elasticity coefficient of the load with respect to the carbon potential: This coefficient represents the percentage change in load i when the carbon potential changes by 1%. The elasticity coefficient is obtained from historical data through linear regression or neural network regression. In the formula: That is, the elasticity coefficient. ; For other control variables (temperature, date, etc.); This is the error term.

[0058] During operation, when a future increase in carbon potential is predicted, the system sends a load reduction signal to the host, and the signal strength is proportional to the rate of change of carbon potential. The main body adjusts the load based on its own elasticity coefficient and signal strength, forming a closed-loop feedback. Main body load adjustment amount: Based on the system operating status data, a graph neural network is used to extract grid topology and power flow distribution features to calculate the nodal electrocarbon potential, and the nodal thermal carbon potential is calculated based on the second law of thermodynamics and the principle of entropy increase. Specific steps include: modeling the power grid as a graph structure. , where the node set Corresponding to the power grid bus, edge set Corresponding to transmission lines. Each node has time-varying characteristics: power injection and carbon potential. Each edge has time-varying characteristics: power flow and carbon flow.

[0059] Graph neural network models are used to learn the nonlinear relationship between node carbon potential and network topology and power flow distribution. Node carbon potential update rule: In the formula: Let v be the hidden state of node v at level l; Let v be the set of neighboring nodes; represents the edge characteristics of edge (u,v); AGGREGATE is an aggregation function (such as summation or averaging). For activation functions; This is a learnable weight matrix.

[0060] The final node carbon potential is obtained from the output of the last layer: Compared with the traditional proportional sharing method, the GNN method can capture nonlinear carbon flow distribution and more accurately reflect the influence of network topology on carbon potential.

[0061] Carbon flow tracking method considering network losses: Introducing equivalent loads for losses, treating network losses as additional loads allocated to each node. A network loss allocation method based on power flow sensitivity is adopted: Network loss sharing coefficient for node i: In the formula: For nodes i Net injected power (kW); For nodes i Associated branch network losses (kW).

[0062] Corrected nodal carbon potential: In the formula: The system average carbon potential is given.

[0063] A carbon flow tracing method for heat networks based on the second law of thermodynamics: Carbon flow tracing in heat networks needs to consider heat loss and transport delay. A nodal-based carbon flow model for heat networks is established, incorporating entropy increase compensation. Thermal carbon potential at node j: In the formula: For the node i Flow to Node j Mass flow rate (kg / s); Specific heat capacity (J / (kg·℃)); Water supply temperature (°C); For nodes j Injected heat power (kW); The carbon potential (kgCO2 / kWh) is the amount of heat injected into the source.

[0064] Heat loss leads to an increase in carbon potential, which is compensated by entropy increase: In the formula: The carbon compensation coefficient for heat loss is obtained by calibration using historical data. The ambient temperature is in °C.

[0065] The basic carbon responsibility is obtained by multiplying the actual energy consumption by the nodal electrical carbon potential and the nodal thermal carbon potential. In the formula: as the main body i Basic carbon responsibility (kgCO2); The load bus's carbon potential; The carbon potential of the heating network.

[0066] The load time-series impact factor, renewable energy consumption contribution factor, load elasticity factor, and subject interaction factor of each entity are calculated. The entropy weight method is used to assign weights to each factor to obtain a comprehensive synergy factor. Load timing influence factor: defining the set of high-carbon periods .main body i The load timing affects the original value: Normalization: Renewable Energy Consumption Contribution Factor: Defining the Set of Photovoltaic Output Periods .main body i Original value of renewable energy consumption contribution: Normalization: Load elasticity factor: Load elasticity factor From the elasticity coefficient: Inter-entity interaction factor: Defines the carbon emission reductions from direct energy trading between inter-entities. In the formula: Let be the amount of electricity (kW) that i purchases from j at time t; The carbon potential for power generation is j (0 for photovoltaics). Subjective interaction factor: Multidimensional factor fusion based on entropy weighting: The entropy weighting method is used to determine the weights of each factor, avoiding subjectivity. Let there be a total... m =4 factors n =3 subjects, constructing an evaluation matrix ,in as the main body i The k Factor values ​​(normalized).

[0067] Calculate the first k The factor under the th i The proportion of each subject: Calculate information entropy: Calculate the weights: Calculate the comprehensive synergy factor: This design ensures that when the factor value is above 0.5 (i.e., above average), This indicates a positive incentive; below 0.5 indicates a positive incentive. This indicates a negative penalty; and the magnitude of the incentive / penalty is directly proportional to the degree of deviation.

[0068] Obtain reference energy consumption and reference carbon potential, calculate the actual response contribution of each subject based on the actual energy consumption, reference energy consumption and reference carbon potential during the high-carbon period, and use the actual response contribution to calculate the incentive compatibility adjustment coefficient; link carbon responsibility allocation with the actual response behavior of the subjects to establish an incentive-compatible carbon responsibility adjustment method.

[0069] Define the subject i Actual response contribution: In the formula: For reference load (e.g., typical daily load); This represents the actual load. Only the contribution of reduced electricity consumption during high-carbon periods is considered; The reference carbon potential (the average carbon potential can be used).

[0070] Incentive compatibility adjustment factor: In the formula: The first term is an adjustment coefficient, set to 0.2; the second term is the reward for increased electricity consumption during low-carbon periods, and the third term is the reward for reduced electricity consumption during high-carbon periods (already reflected in the first term). This adjustment coefficient allows proactively responding entities to receive less carbon responsibility, forming incentive compatibility. Multiplying the basic carbon responsibility, the comprehensive synergy factor, and the incentive compatibility adjustment coefficient yields the dynamic carbon responsibility allocation results for each entity.

[0071] Ultimate dynamic carbon responsibility: Traditional carbon liability allocation methods use a fixed average emission factor: In the formula: , ; Take 0.55 kg CO2 / kWh, Take 0.225 kg CO2 / kWh.

[0072] Deviation rates of the two methods: After completing the dynamic carbon responsibility allocation for each entity according to the aforementioned steps, in order to further address the issue of fair distribution of emission reduction benefits in multi-entity collaborative operation, this embodiment treats each entity in the park as a participant in a cooperative game and constructs a collaborative contribution allocation mechanism based on alliance game theory and adaptive Shapley value. The specific implementation is as follows: Alliance Game Model Construction: The three entities are considered as participants in a cooperative game. Define the characteristic function v(S) as the coalition... The carbon emission reduction benefit is the amount of carbon emission reduction achieved by the alliance relative to the baseline scenario.

[0073] Baseline scenario: Each entity operates independently with no energy sharing.

[0074] Alliance operation: Considering energy trading and coordinated scheduling among members.

[0075] The characteristic function satisfies superadditivity: ,when .

[0076] Basic contribution allocation based on Shapley values: Traditional Shapley value calculation of the average marginal contribution of each entity to the coalition: In the formula: as the main body i The Shapley value is the share of carbon emission reduction benefits that a company is entitled to.

[0077] An adaptive Shapley value improvement method based on synergy factors: Introducing synergy factors to adaptively correct the Shapley value, so that the contribution allocation matches the actual synergistic contribution of each subject.

[0078] Define the subject i Synergistic contribution coefficient: Adaptive Shapley value: This correction guarantees That is, the total emission reduction benefits remain unchanged, but the allocation ratio is adjusted according to the synergistic contribution.

[0079] Multidimensional decomposition method of contribution: Decompose the modified Shapley value into three parts: Technological contribution: Technological contribution stems from the emission reduction benefits generated by the low-carbon equipment invested in or operated by the entity. Define the entity. i Technology contribution coefficient: In the formula: as the main bodyi Self-use photovoltaic power generation (kW); as the main body i Self-used energy storage discharge power (kW). Technical contribution: Load characteristic contribution: The load characteristic contribution originates from the optimized energy consumption behavior of the entity itself. Define the load characteristic contribution coefficient: In the formula: This is the normalized time-series influence factor. The contribution of load characteristics is then: Synergistic contribution: The synergistic contribution is a margin, reflecting the additional emission reduction benefits generated by the interaction and cooperation among stakeholders. This breakdown can provide targeted carbon reduction guidelines for each entity: technological contributions encourage investment in low-carbon equipment, load characteristic contributions encourage optimized energy use behavior, and collaborative contributions encourage participation in system interaction.

[0080] After obtaining the dynamic carbon responsibility allocation results and multi-dimensional collaborative contributions of each entity, to ensure the transparent execution and immutability of the allocation results and reward / penalty incentives, this embodiment constructs an internal carbon trading and settlement system for the park based on blockchain smart contracts. The specific execution process is as follows: A carbon trading market will be established within the park based on dynamic carbon responsibility. Initial carbon allowances for each entity will be allocated using traditional methods. In the formula: Assuming a target carbon potential of 0.4 kg CO2 / kWh; Take 0.15 kg CO2 / kWh.

[0081] The difference between actual carbon liability and allowance is the trading volume: like Carbon credits must be purchased; if Carbon allowances can be sold.

[0082] A blockchain-based carbon trading settlement method: This method utilizes blockchain smart contracts to automate carbon trading settlements. The smart contracts are deployed on a consortium blockchain, with all stakeholders and park managers acting as nodes.

[0083] Smart contract functions include: 1. Automatic matching: Based on the transaction needs of each party, the system automatically matches buyers and sellers and uses a two-sided auction mechanism to determine the transaction price.

[0084] 2. Transaction Execution: After the transaction is confirmed, the fee will be automatically deducted from the buyer's account and the carbon allowance will be transferred to the buyer.

[0085] 3. Ledger Records: All transaction records are recorded on the blockchain, are immutable, and are traceable.

[0086] 4. Settlement and Clearing: The net amount of carbon trading for each entity is automatically settled daily / monthly.

[0087] Carbon trading prices are based on a marginal cost-based pricing method. In the formula: Based on the carbon price (yuan / kg CO2); This is the price elasticity coefficient.

[0088] Incentive-compatible reward and punishment methods: Design reward and punishment functions to link carbon trading revenues to the entity's response behavior. In the formula: Basic reward and punishment coefficient; In response to reward and punishment coefficients; The actual response contribution of subject i; This represents the average response contribution. If... This indicates that the entity receives a subsidy; if This indicates that a fine must be paid.

[0089] Investment payback period analysis method: Considering the entity's investment in photovoltaic, energy storage, and other equipment, an investment decision analysis model is established. Entity i Total benefits of investing in low-carbon equipment: In the formula: For carbon trading revenue; For energy-saving benefits; To subsidize revenue.

[0090] Dynamic investment payback period satisfy: In the formula: r The discount rate; The initial investment cost is (in yuan). The internal rate of return (IRR) is calculated by: To support efficient computation throughout the entire process of physical modeling, dynamic carbon potential prediction, carbon flow tracing, responsibility allocation, game decomposition, and on-chain settlement, this embodiment constructs a modular, hierarchical solution architecture, such as... Figure 3 As shown, the overall algorithm's solution and calculation process is as follows: The proposed method employs a modular, hierarchical solution architecture, divided into five logical layers. The bottom layer is the data generation module, responsible for generating time-series data such as load, photovoltaic output, and grid marginal carbon potential. The second layer is the operation optimization and carbon flow tracking module, which solves the system's operational trajectory based on predicted data and calculates the carbon potential of each node. The third layer is the responsibility allocation and contribution analysis module, completing the calculation of collaborative factors, carbon responsibility allocation, alliance game theory, and contribution decomposition. The fourth layer is the transaction settlement module, which completes carbon trading and reward / penalty settlement based on blockchain smart contracts. The top layer is the result output and visualization module. Information is exchanged between modules through a standardized data format, supporting independent updates and expansions of each module.

[0091] Data Generation and Scenario Construction: Load and photovoltaic (PV) data generation employs Monte Carlo sampling. For each main electrical and thermal load, a baseline time-series curve is first generated using a load model. Multiple sample sets are then generated by overlaying a normally distributed random fluctuation term, with each set containing 24 hours of time-series data. PV output is calculated using a physical model, incorporating a cloud shading factor based on the baseline irradiance to generate a PV output sequence with random fluctuation characteristics. The grid marginal carbon potential is generated using the PINN prediction model, which takes historical load, historical carbon potential, and weather forecast characteristics as inputs and outputs a predicted marginal carbon potential value for the next 24 hours.

[0092] To reduce the computational complexity of subsequent optimization calculations, the K-means clustering algorithm is used to reduce the number of generated scenes. Each scene is treated as a high-dimensional vector (dimension 1). Similar scenes are merged through iterative clustering, and cluster centers are selected as typical scenes, with each typical scene assigned a corresponding weight. The number of typical scenes retained after scene reduction is determined based on the variance explained rate, and is usually controlled within 5-10, which significantly reduces the computational burden while ensuring computational accuracy.

[0093] Operational optimization algorithm: The operational optimization employs a model predictive control framework to solve a mixed-integer linear programming problem within a finite time domain at each time step. Optimization variables include gas turbine start-up and shutdown status and output power, gas boiler output, electric energy storage charging and discharging power, thermal energy storage charging and discharging power, and purchased and sold electricity power, totaling approximately 300 variables, of which approximately 100 are integer variables. Constraints cover equipment output upper and lower limits, ramp-up constraints, energy storage SOC constraints, power balance constraints, etc., totaling approximately 500 linear constraints.

[0094] The objective function of the optimization problem is to minimize the carbon emissions of the system, and its mathematical form is: In the formula: t The current moment; H To predict the length of the time domain (take 4 hours); Carbon emission factor of natural gas; , These are the natural gas consumption figures for gas turbines and gas boilers, respectively. For the power purchased; This represents the marginal carbon potential predicted by PINN. The objective function unifies the dimensions of direct and indirect emissions, aiming to minimize system carbon emissions.

[0095] The solver chosen is Gurobi, which utilizes its branch and bound algorithm to handle mixed integer variables. The MPC rolling optimization process is as follows: at the current time... t Obtain system status (energy storage SOC, current equipment output, measured load), and call the PINN prediction model to output future... H Based on the load and marginal carbon potential at each time period, the above MILP optimization problem is constructed and solved using a solver to obtain the future... H The optimal control sequence for each time period is determined by executing the control instructions for the first time period only. After the system state is updated, the system moves to the next time period and repeats the above process until all 24 hours of optimization are completed.

[0096] Carbon Flow Tracing Algorithm: Grid carbon flow tracing employs a graph neural network method. First, the grid topology is modeled as a graph structure. Node set Corresponding to the power grid bus, edge set The corresponding transmission line. The feature vector of each node includes attributes such as injected power and node type (power source, load, tie node), and the features of each edge include line impedance, power flow direction, and transmitted power.

[0097] The GNN model aggregates neighbor node information through a message passing mechanism, and the update rule for the hidden state of a node is as follows: In the formula: For nodes v In the hidden state of layer l; For nodes v The set of neighboring nodes; Let be the edge features of edge (u,v); AGGREGATE is the aggregation function (using summation). It is the ReLU activation function; The weight matrix is ​​learnable. The model contains three layers of graph convolutions, with each layer having a hidden state dimension of 64. The final node carbon potential is output by a fully connected layer. The model is trained using power grid simulation data, with the node carbon potential obtained from power flow calculations as the label. The loss function includes both prediction error and physical constraints. The number of training iterations is set to 500 epochs, and an adaptive moment estimation optimizer is used for the learning rate.

[0098] Carbon flow tracking in the heating network employs a nodal method, considering heat loss and transmission delay. After obtaining the temperature and flow distribution at each node through thermo-hydraulic calculations, these values ​​are substituted into the heating network carbon flow equation for iterative solution. The solution is applied when the carbon potential change between two consecutive iterations is less than a set threshold (e.g., ...). Convergence is determined when the time is right.

[0099] Synergy Factor Calculation and Carbon Responsibility Allocation: The synergy factor calculation includes four dimensions: load time-series impact factor, renewable energy absorption factor, load resilience factor, and subject interaction factor. After each factor is calculated, the entropy weight method is used to determine the weight. The core of the entropy weight method is to use information entropy to measure the degree of indicator variation; the greater the degree of indicator variation, the smaller the information entropy, and the greater the weight. The formula for calculating the entropy value of the k-th indicator is: In the formula: n The main quantity; as the main body i The kth factor value (normalized); For the first i The subject in the first k The proportion under each factor. When At that time, take The weighting coefficients are derived from the entropy value: In the formula: m The total number of factors is 4. The comprehensive synergistic factor is a weighted synthesis of each factor and mapped to the interval [0.8, 1.2] to ensure that the physical meaning of the synergistic factor is clear.

[0100] The incentive compatibility adjustment coefficient is calculated based on the actual response contribution of the subject. By comparing the actual load with the reference load, the higher the contribution of reducing electricity consumption during high-carbon periods, the smaller the adjustment coefficient and the lower the carbon responsibility. The dynamic carbon responsibility is ultimately expressed as the product of the basic carbon responsibility, the comprehensive synergy factor, and the incentive compatibility adjustment coefficient.

[0101] Coalition Game Solution and Contribution Decomposition: Solving the coalition game requires calculating the characteristic function values ​​of all sub-coalitions. The characteristic function is defined as the carbon emission reduction of the coalition operation relative to the independent operation scenario. Since the number of sub-coalitions increases exponentially with the number of subjects (7 sub-coalitions for 3 subjects), they can be directly enumerated and calculated without the need for sampling approximation. The characteristic function of each sub-coalition is obtained by simulating the collaborative operation of subjects within that coalition; subjects outside the coalition are treated as operating independently.

[0102] The Shapley value is calculated using an analytical formula: In the formula: S For without a main body i The sub-alliance; |S| is the size of the sub-alliance; This is the characteristic function. For three entities, the formula expands to have four terms summed, which can be calculated directly.

[0103] The adaptive Shapley value introduces a synergy factor correction on the basis of the traditional Shapley value. The correction coefficient is composed of the comprehensive synergy factor. The process involves exporting the contribution allocation to match the actual collaborative contribution. The contribution decomposition breaks down the corrected Shapley value into three parts: technical contribution, load characteristic contribution, and collaborative contribution. The technical contribution is decomposed according to the proportion of the system's total self-consumption of low-carbon equipment (photovoltaics, energy storage); the load characteristic contribution is decomposed according to the proportion of the inverse of the time-series influence factor to the sum of the inverses of all main components; and the collaborative contribution is the remaining amount after subtracting the first two items from the corrected Shapley value.

[0104] Blockchain smart contract settlement algorithm: Carbon trading settlement is implemented based on consortium blockchain smart contracts. The smart contract contains three core functions: a matching function that automatically matches buyers and sellers based on the surplus or shortage of carbon allowances for each entity, and uses a bilateral auction mechanism to determine the transaction price; an execution function that completes the transfer of carbon allowances and funds, and writes the transaction record into the blockchain; and a settlement function that settles the net amount at the end of each day / month and generates a settlement report.

[0105] Smart contracts are deployed on Hyperledger Fabric or the Ethereum consortium blockchain, with all principal nodes jointly maintaining the ledger to ensure distributed consensus and immutability of data. Transaction triggers occur automatically after each entity reports its actual carbon liability, comparing it with its allowance to generate transaction requests. The reward / penalty function is designed as follows: In the formula: Basic reward and punishment coefficient; In response to reward and punishment coefficients; as the main body i The actual response contribution; Contribution to average response; This represents the carbon trading price. This function reflects both the differences in basic carbon responsibility and the additional contribution from responses, forming a complete incentive and constraint mechanism.

[0106] The complete solution process, based on the above modules, is as follows: First, 24-hour time-series data is generated based on the load model, photovoltaic model, and PINN prediction model, and typical scenarios are obtained through scenario reduction. Then, the system operation trajectory is solved in a rolling manner under the MPC framework, and the operation optimization results such as the output of each device, energy storage SOC, and purchased and sold power are output. Next, the operation results are input into the GNN model and the heating network carbon flow model to calculate the electrical carbon potential and thermal carbon potential of each node, and the basic carbon responsibility is obtained by multiplying the load of each subject with the corresponding carbon potential. On this basis, the synergy factors of four dimensions are calculated and the comprehensive synergy factor is obtained by fusion using the entropy weight method. The dynamic carbon responsibility is calculated by combining the incentive compatibility adjustment coefficient. Then, the Shapley value is calculated by enumerating the characteristic functions of each sub-alliance, and the adaptive Shapley value is obtained after correction by the synergy factor, and the contribution decomposition is completed. Finally, based on the difference between the dynamic carbon responsibility and the quota, the smart contract is called to complete the carbon trading matching, execution, and reward and punishment, and the carbon responsibility, contribution decomposition, and net carbon trading revenue of each subject are output for chart analysis. The algorithm described above is implemented in a Python environment. Core optimizations utilize the Gurobi solver, the GNN model is implemented using PyTorch, and the blockchain smart contract is deployed on a test network. Modules exchange data via JSON format, ensuring modularity and scalability, and supporting subsequent example analysis and method verification.

[0107] Example 2 This embodiment applies a collaborative operation scheduling method for a low-carbon park integrated energy system based on dynamic carbon potential to a typical low-carbon park integrated energy system for simulation and application verification. For example... Figure 4 As shown, the low-carbon industrial park's integrated energy system comprises three typical energy-consuming entities: a manufacturing enterprise (Entity A), a commercial complex (Entity B), and a residential community (Entity C). The system configuration is as follows: a 1000kWp photovoltaic power generation system installed on the roof of Entity B; a gas turbine with a rated power output of 2000kW, a power generation efficiency of 38%, and a heat recovery efficiency of 42%; a gas boiler with a rated heating power output of 1500kW and a thermal efficiency of 88%; an electric energy storage system with a capacity of 2000kWh and a maximum charge / discharge power of 500kW; and a thermal energy storage system with a capacity of 1000kWh and a maximum charge / discharge power of 400kW. The park is connected to the external power grid via a 10kV / 380V substation, enabling bidirectional power purchase and sale.

[0108] The simulation uses typical spring and autumn weekday data with a time resolution of 1 hour, totaling 24 hours. Load data, photovoltaic output, and grid carbon potential are all generated by a simulation program. The generation model is based on the actual energy consumption behavior characteristics of each entity: industrial load is mainly during production periods (8:00-12:00, 13:00-17:00), with a brief decline at midday; commercial load operates from 10:00-22:00, with double peaks at midday and in the evening; residential load shows two peaks in the morning and evening. Photovoltaic output is generated based on solar altitude angle, weather factors (lightly cloudy, taken as 0.9), and random fluctuations in cloud cover. Grid carbon potential is constructed by superimposing a segmented time-series model with random fluctuations, reflecting the actual characteristics of high carbon potential during morning and evening peaks and low carbon potential after photovoltaic substitution at midday.

[0109] Representative moments of the basic data are shown in Table 1. The total electricity load exhibits three characteristic periods: a morning peak, a midday mini-peak, and an evening peak. The maximum load occurs around 19:00, reaching 1275.5 kW. Photovoltaic output begins to rise from 6:00, reaching a peak of 756.8 kW at 13:00. The grid carbon potential is lower at night (0.46-0.62), rising above 0.62 during the morning peak, dropping to around 0.50 at midday due to high photovoltaic power generation, and then climbing again to a peak of 0.87 during the evening peak. This basic data provides reliable input for subsequent operational optimization, carbon flow tracking, and responsibility allocation.

[0110] Table 1. Representative moments of basic data In the analysis of park load characteristics and low-carbon resource operation, such as Figure 5 and Figure 6 As shown, industrial load (main load A) experiences production peaks at 8:00 and 13:00, with peak values ​​of 512kW and 558kW respectively, totaling approximately 8820kWh per day; commercial load (main load B) operates from 10:00 to 22:00, with double peaks at 13:00 and 19:00, peaking at 599kW, totaling approximately 8640kWh per day; residential load (main load C) exhibits double peaks in the morning and evening, from 6:00-7:00 and from 19:00-20:00, with a peak value of 278kW, totaling approximately 5160kWh per day; the peak and valley curves of the three main load types are interspersed and highly complementary. Figure 7 As shown, the heat load of the three types of entities is highly synchronized with the energy consumption rhythm, with peak values ​​of 380kW for industry, 320kW for commerce, and 210kW for residential use. Figure 8 This indicates that industrial load accounts for the highest proportion (over 45%) during the daytime (8:00-17:00), commercial load accounts for 47% at 19:00, and residential load accounts for 37% at 6:00. For example... Figure 9 and Figure 10As shown, photovoltaic power generation output exhibits a single-peak characteristic, reaching its peak between 12:00 and 14:00 (710~757kW), with a daily power generation of approximately 5360kWh; the marginal carbon potential of the power grid exhibits a double-peak characteristic (approximately 0.68kgCO2 / kWh in the morning and approximately 0.87kgCO2 / kWh in the evening), decreasing to 0.50~0.55kgCO2 / kWh at midday due to photovoltaic power replacing thermal power, with a daily average carbon potential of 0.63kgCO2 / kWh. Figure 11 and Figure 12 Correlation analysis showed that the correlation coefficient between total load and grid carbon potential reached 0.72, and the carbon potential remained in the high carbon range of 0.80~0.85 kgCO2 / kWh for 4 hours, proving the necessity of introducing time-series factors for differentiated allocation.

[0111] System operating status as follows Figure 13 , 14 As shown in Figures 15 and 16, photovoltaic, gas turbine, and grid-purchased electricity accounted for 17.2%, 14.5%, and 68.3% of the total electricity, respectively; the thermal storage system had a maximum heat release power of 380kW, reducing carbon emissions from the heating system by approximately 8%; the electric energy storage system charged during the low-carbon period from 0:00 to 6:00 (increasing the State of Charge (SOC) to 1680kWh) and discharged during the high-carbon period from 17:00 to 21:00 (reducing the SOC to 520kWh), reducing additional carbon emissions during the high-carbon period by approximately 12%. The dynamic carbon potential and spatiotemporal tracking effect of the electro-thermal coupling are shown in Figures 15 and 16. Figure 17 and Figure 18 As shown, the overall carbon potential of the busbar is lower than that of the grid, especially during the peak photovoltaic power generation period at midday, when it drops to 0.38~0.45 kgCO2 / kWh (0.10~0.15 kgCO2 / kWh lower than the grid). The carbon potential of the heating network remains between 0.48~0.65 kgCO2 / kWh, and the busbar carbon potential fluctuation coefficient is 0.21, with a maximum carbon potential difference of 0.49 kgCO2 / kWh, demonstrating dramatic time-series dynamic characteristics. The photovoltaic suppression effect is as follows: Figure 19 As shown, photovoltaic grid connection reduces the average daily carbon potential by 0.07 kg CO2 / kWh, resulting in a daily carbon reduction of 1575 kg CO2 (accounting for 21.7% of total emissions). Figure 20 The system's carbon emission peak occurred between 17:00 and 20:00 (a maximum of 485 kg CO2 / h, with a total daily emission of 5680 kg CO2), and its correlation coefficient with the carbon potential curve was as high as 0.94, verifying the effectiveness of marginal carbon potential as a characterization indicator of carbon emissions. Regarding the verification of dynamic carbon responsibility allocation and incentive compatibility, Figure 21 and Figure 22This indicates that, assuming a total carbon liability of 13,620 kg CO2 for all entities, the traditional average emission factor method results in liabilities of 5,430 kg CO2, 5,020 kg CO2, and 3,160 kg CO2 for entities A, B, and C, respectively. However, under the dynamic source allocation method of this invention, these liabilities are 5,680 kg CO2, 4,680 kg CO2, and 3,340 kg CO2, respectively. Specifically, entity B's carbon liability decreased by 6.8% due to effective absorption of photovoltaic power during midday high load; entity A's carbon liability increased by 4.6% due to concentrated electricity consumption during the morning peak and sustained high load during midday; and entity C's carbon liability increased by 5.7% due to a higher proportion of electricity consumption during the evening peak high-carbon period. Figure 23 As shown, electricity responsibility accounts for 65% to 75% of basic carbon responsibility, while heat responsibility accounts for 25% to 35%.

[0112] Figure 24 The fairness analysis shows that the responsibility ratio of subject B under the dynamic method (30%) is closer to its actual low-carbon contribution than that under the traditional method (32%), reflecting the fairness principle of "more emissions, more responsibility; more reductions, more compensation." Regarding the synergistic effect and economic incentive effect, such as... Figure 25 As shown, the synergy factors of subjects A, B, and C are 1.05, 0.92, and 1.03, respectively, and the adjustment coefficients are 0.98, 0.96, and 1.02, respectively. Among them, the synergy factor of subject B is less than 1, which means that it has made a positive contribution to the low-carbon operation of the system. Figure 26 A comparison of scenario B shows that after installing photovoltaics and energy storage, the carbon liability of Entity B decreased from 4680 kg CO2 to 4120 kg CO2 (a reduction of 12.0%), and the synergy factor was optimized to 0.85, forming a positive feedback loop of equipment investment - synergy optimization - liability reduction. The carbon cost economics analysis based on a carbon price of 0.05 yuan / kg CO2 is as follows... Figure 27 As shown, after adopting the dynamic method of this invention, subject B saves 17.0 yuan in carbon costs (estimated at approximately 4250 yuan per year based on 250 operating days), while subjects A and C incur additional expenses of 12.5 yuan (annual increase of approximately 3125 yuan) and 9.0 yuan (annual increase of approximately 2250 yuan), respectively. This economic difference forms an effective behavioral incentive mechanism. Based on the results of the above embodiments, this invention has clear technical advantages and practical value in terms of carbon potential characterization accuracy, spatiotemporal tracking resolution, allocation fairness, and incentive compatibility.

[0113] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A collaborative operation and scheduling method for a low-carbon industrial park integrated energy system based on dynamic carbon potential, characterized in that, Includes the following steps: A physical information neural network model is constructed, and a physical constraint error, including the power load balance of the park, the upper and lower limits of generator output, and the energy storage power constraint, is introduced into the loss function of the physical information neural network model to train the physical information neural network model. The physical information neural network model is trained by inputting historical load, historical marginal carbon potential, weather data, and date characteristics of the park into the input, and outputs the predicted marginal carbon potential. Using the predicted marginal carbon potential as an optimization signal, model predictive control is used to optimize the charging and discharging of the park's energy storage system, and output system operation status data including high carbon periods, photovoltaic output periods, actual energy consumption and branch network losses. Based on the system operating status data, a graph neural network is used to extract the topology features and power flow distribution features of the power grid to calculate the node electrical carbon potential, and the node thermal carbon potential is calculated based on the second law of thermodynamics and the principle of entropy increase. The basic carbon responsibility is obtained by multiplying the actual energy consumption by the nodal electrical carbon potential and the nodal thermal carbon potential. By calculating the load time series factor, renewable energy absorption factor, load elasticity factor and subject interaction factor of each subject, the entropy weight method is used to assign weights to each factor to obtain the comprehensive synergy factor. Combined with the response contribution degree of high carbon period, the incentive compatibility adjustment coefficient is calculated to obtain the dynamic carbon responsibility of each subject. The dynamic carbon responsibility system generates load response control instructions and sends them to each entity to dynamically adjust the energy load of each entity, thereby realizing the low-carbon collaborative operation and scheduling of the park's integrated energy system.

2. The method according to claim 1, characterized in that, When training the physical information neural network model, load forecast data is acquired, and the load forecast data is input into an economic dispatch model that includes unit output ranking to back-calculate the reference output of marginal units; the predicted marginal unit output is obtained through the predicted marginal carbon potential; the physical constraint error is the error between the predicted marginal unit output and the reference output of the marginal unit.

3. The method according to claim 1, characterized in that, In the step of using model predictive control to optimize the charging and discharging scheduling of the park's electric energy storage system, the purchased power, natural gas consumption, and natural gas carbon emission factor are first obtained; the objective function is set to minimize system carbon emissions, where the system carbon emissions are the sum of a first product and a second product, the first product being the product of the predicted marginal carbon potential and the purchased power, and the second product being the product of the natural gas consumption and the natural gas carbon emission factor.

4. The method according to claim 1, characterized in that, In the step of using a graph neural network to extract grid topology features and power flow distribution features to calculate node electrocarbon potential, the grid bus is used as the node set of the graph neural network, and the transmission line is used as the edge set of the graph neural network; the update rule of the node hidden state includes aggregating the edge features of neighboring nodes through an aggregation function, multiplying them by a learnable weight matrix, and inputting them into an activation function.

5. The method according to claim 4, characterized in that, After the step of using a graph neural network to extract grid topology features and power flow distribution features to calculate nodal electrocarbon potential, the method further includes: Obtain the system's average carbon potential and current sensitivity; The network loss allocation coefficient is calculated based on the power flow sensitivity. The branch network loss is multiplied by the system average carbon potential and the network loss allocation coefficient as an additional load and superimposed on each node in the node set to correct the node's electrical carbon potential.

6. The method according to claim 1, characterized in that, The steps for calculating the nodal thermal carbon potential based on the second law of thermodynamics and the principle of entropy increase specifically include: Obtain the carbon compensation coefficient for heat loss and the ambient temperature; Calculate the initial nodal thermal carbon potential, and use the heat loss carbon compensation coefficient and the ambient temperature to calculate entropy increase compensation; The entropy increase compensation is superimposed on the initial node thermal carbon potential to obtain the node thermal carbon potential.

7. The method according to claim 1, characterized in that, The calculation of the load time-series factor, renewable energy absorption factor, load resilience factor, and entity interaction factor for each entity specifically includes: Obtain the amount of electricity traded between the power generation carbon potential and the direct energy exchange between the entities, and calculate the entity interaction factor based on the product of the amount of electricity traded between the entities and the power generation carbon potential; The load timing factor is the normalized value of the proportion of each entity's electrical load to its total electrical load during the high-carbon period. The renewable energy absorption factor is the normalized value of the proportion of each entity's electricity load to its total electricity load during the photovoltaic output period; Obtain the elastic coefficients obtained from the regression of the elastic response model, and use the elastic coefficients as the load elasticity factor.

8. The method according to claim 1, characterized in that, The calculation of the incentive compatibility adjustment coefficient based on the response contribution during high-carbon periods specifically includes: Obtain reference energy consumption and reference carbon potential; The product of the difference between the actual energy consumption and the reference energy consumption during the high-carbon period and the reference carbon potential is taken as the actual response contribution. The excitation compatibility adjustment coefficient decreases as the actual response contribution increases.

9. The method according to claim 1, characterized in that, After outputting the dynamic carbon responsibility allocation results for each entity, the following is also included: Treating each entity as a cooperative game participant, a cooperative game model is constructed with the carbon emission reduction of the cooperative operation relative to independent operation as the characteristic function, and the Shapley value of each entity is calculated using the characteristic function.

10. The method according to claim 9, characterized in that, After calculating the Shapley value of each subject using the characteristic function, the method further includes: Obtain the percentage of equipment self-consumption by each entity and the amount of electricity traded directly between entities; The synergistic contribution coefficient is constructed by dividing the comprehensive synergistic factor by the mean of the comprehensive synergistic factors of all subjects. The adaptive Shapley value is obtained by multiplying the synergistic contribution coefficient by the Shapley value. The adaptive Sharpe ratio is decomposed into technical contribution, load characteristic contribution, and collaborative contribution based on the proportion of the equipment's self-consumption, the proportion of the reciprocal of the load time series factor, and the allocation ratio of the electricity in the direct energy trading between the entities.

Images