Carbon flow transferable system, carbon flow transferable method, device, medium and program product

Abstract

The present application belongs to the field of energy-carbon synergy, and relates to an energy-carbon synergy method, device, medium and program product of an energy-carbon circulation system. The present application comprises: obtaining energy-carbon data and device data collected by each core energy equipment in a target energy-carbon circulation system; determining current energy-carbon activity data of each core energy equipment and a comprehensive confidence of the current energy-carbon activity data according to the device data and the energy-carbon data; constructing a dynamic energy-carbon topology graph between each core energy equipment in the target energy-carbon circulation system according to the current energy-carbon activity data and the comprehensive confidence; determining a target scheduling strategy conforming to the current working condition of each core energy equipment according to the dynamic energy-carbon topology graph and the comprehensive confidence; and updating the target scheduling strategy, the dynamic energy-carbon topology graph or the comprehensive confidence according to the execution result. The present application realizes fine and intelligent collaborative control of energy consumption and carbon emission in an industrial park.

Description

Technical Field

[0001] This invention belongs to the technical field of energy-carbon synergy, and specifically relates to an energy-carbon synergy method, device, medium and program product for an energy-carbon circulation system. Background Technology

[0002] As the proportion of new energy sources, mainly wind and solar, in the energy consumption structure of the park continues to increase, the strong intermittency and volatility of new energy output bring new technical challenges to the park's energy and carbon collaborative management: On the one hand, the uncertainty of new energy output leads to dynamic changes in the quality of energy and carbon data, and traditional static data governance methods are difficult to guarantee data confidence; on the other hand, the multi-energy coupling relationship and the spatiotemporal migration of carbon flow transmission paths caused by the fluctuation of new energy cause the energy and carbon model based on fixed topology to fail; furthermore, the prediction-based optimization scheduling scheme cannot adaptively adjust due to the deviation of actual operating conditions from the preset during the execution process, resulting in a significant reduction in optimization effect.

[0003] Current research on energy and carbon management in industrial parks has made some progress: at the data acquisition level, it has achieved data acquisition from multiple energy types and supports mainstream industrial protocols such as Modbus and OPC UA. Some solutions have further introduced spatiotemporal correlation feature extraction and graph neural network encoding technologies to improve the intelligence level of data processing; at the energy and carbon modeling level, methods such as multi-energy coupling, fixed emission factor accounting, and segmented carbon flow tracking have been developed; at the optimization and scheduling level, game theory models, dual-objective (cost + carbon emission) optimization models, and solution methods such as particle swarm optimization and genetic algorithms have emerged; at the intelligent decision-making level, technologies such as blockchain notarization and digital twins have been introduced to improve data credibility and visualization. However, existing technologies mostly focus on optimizing a single link and have not yet formed an integrated solution covering multi-energy coupling, multi-entity collaboration, and a closed-loop process. Summary of the Invention

[0004] This application addresses the shortcomings of existing technologies that focus on optimizing single aspects and lack a comprehensive solution covering multi-energy coupling, multi-entity collaboration, and a closed-loop process. It provides a method, equipment, medium, and program for energy and carbon transfer systems, enabling refined and intelligent collaborative management of energy consumption and carbon emissions in industrial parks.

[0005] To address the aforementioned technical problems, this application proposes four aspects.

[0006] In a first aspect, this application proposes an energy-carbon coordination method for an energy-carbon circulation system, comprising: acquiring energy-carbon data and equipment data collected by each core energy device in a target energy-carbon circulation system; determining current energy-carbon activity data and a comprehensive confidence level of the current energy-carbon activity data for each core energy device based on the equipment data and the energy-carbon data; constructing a dynamic energy-carbon topology diagram among the core energy devices in the target energy-carbon circulation system based on the current energy-carbon activity data and the comprehensive confidence level; determining a target scheduling strategy that conforms to the current operating conditions of each core energy device based on the dynamic energy-carbon topology diagram and the comprehensive confidence level; acquiring the execution result of the target scheduling strategy; and updating the target scheduling strategy, the dynamic energy-carbon topology diagram, or the comprehensive confidence level based on the execution result, ultimately enabling energy-carbon coordination among the core energy devices.

[0007] In some embodiments, determining the current energy and carbon activity data of each core energy device and the comprehensive confidence level of the current energy and carbon activity data based on the device data and the energy and carbon data includes: processing the energy and carbon data and the device data according to preset data processing rules to obtain the current energy and carbon activity data of each core energy device; calculating a time-varying decay factor based on the current energy and carbon activity data of the new energy devices in the target energy and carbon circulation system, wherein the time-varying decay factor is used to characterize the output uncertainty of the new energy devices; calculating a static score of each core energy device based on the current energy and carbon activity data of each core energy device; calculating an initial confidence level of each current energy and carbon activity data based on the static score and the time-varying decay factor; verifying the current energy and carbon activity data based on the current verification threshold at the current time to obtain a verification result; determining a correction coefficient for the initial confidence level based on the verification result; and correcting the initial confidence level based on the correction coefficient to obtain the comprehensive confidence level of the current energy and carbon activity data.

[0008] In some embodiments, constructing a dynamic energy-carbon topology map among the core energy devices in the target energy-carbon flow system based on the current energy-carbon activity data and the comprehensive confidence level includes: constructing an energy-carbon flow topology map based on the target energy-carbon flow system; determining the current equipment efficiency and current carbon flow intensity of each core energy device based on the current energy-carbon activity data of the target energy-carbon flow system and each core energy device; determining the node characteristics of each node in the energy-carbon flow topology map based on the current equipment efficiency, the current carbon flow intensity, and the current energy-carbon activity data; determining the dynamic attention weight of each edge in the energy-carbon flow topology map based on the comprehensive confidence level; and constructing the dynamic energy-carbon topology map based on the node characteristics of each node, the dynamic attention weight of each edge, and the energy-carbon flow topology map.

[0009] In some embodiments, the method further includes: when a first preset condition is met, obtaining the latest comprehensive confidence level and current energy carbon activity data of each of the core energy devices; updating the dynamic energy carbon topology map according to the latest comprehensive confidence level and current energy carbon activity data; calculating the node centrality of each node in the dynamic energy carbon topology map after each update; and updating the current verification threshold when the change in the node centrality of each node meets a second preset condition.

[0010] In some embodiments, determining the target scheduling strategy that conforms to the current operating conditions of each core energy device based on the dynamic energy-carbon topology map and the comprehensive confidence level includes: calculating a global modulation confidence level based on the comprehensive confidence level and the dynamic energy-carbon topology map, wherein the global modulation confidence level is used to characterize the confidence level of all energy-carbon activity data in the target energy-carbon circulation system; determining the current stability distribution of the target energy-carbon circulation system based on the global modulation confidence level; determining the weight coefficients of each parameter and the penalty values ​​of the penalty parameters based on the stability distribution and the dynamic energy-carbon topology map; and adjusting the currently executing scheduling strategy based on the weight coefficients and the penalty values ​​to obtain the target scheduling strategy.

[0011] In some embodiments, obtaining the execution result of the target scheduling strategy and updating the target scheduling strategy, the dynamic energy carbon topology map, or the comprehensive confidence level based on the execution result includes: predicting the prediction results of each of the core energy devices and the target energy carbon transfer system within a future preset time range based on the target scheduling strategy; obtaining the actual execution results generated by the target energy carbon transfer system when executing the target scheduling strategy; determining the execution deviation based on the prediction results and the actual execution results; when the execution deviation meets a third preset condition, determining the main source of deviation based on the execution deviation; determining the target to be adjusted and the adjustment strategy for the target to be adjusted based on the main source of deviation; and adjusting the target to be adjusted based on the adjustment strategy.

[0012] In some embodiments, determining the target to be adjusted and the adjustment strategy for the target to be adjusted based on the main source of deviation includes: when the main source of deviation is a device-level source, determining the adjustment target as a comprehensive confidence level, and determining the adjustment strategy as adjusting the verification threshold used to determine the comprehensive confidence level; when the main source of deviation is a topology-level source, determining the adjustment target as the dynamic energy carbon topology map, and determining the adjustment strategy as adjusting the update conditions of the dynamic energy carbon topology map; when the main source of deviation is a system-level source, determining the adjustment target as the current adjustment strategy, and determining the adjustment strategy as a determination threshold for stability distribution.

[0013] In a second aspect, this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in any of the first aspects.

[0014] Thirdly, this application proposes a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0015] Fourthly, this application proposes a computer program product comprising a computer program that, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0016] This application discloses an energy-carbon coordination method for an energy-carbon circulation system, comprising: acquiring energy-carbon data and equipment data collected by each core energy device in the target energy-carbon circulation system; determining the current energy-carbon activity data of each core energy device and the comprehensive confidence level of the current energy-carbon activity data based on the equipment data and the energy-carbon data; constructing a dynamic energy-carbon topology map among the core energy devices in the target energy-carbon circulation system based on the current energy-carbon activity data and the comprehensive confidence level; determining a target scheduling strategy that conforms to the current operating conditions of each core energy device based on the dynamic energy-carbon topology map and the comprehensive confidence level; acquiring the execution result of the target scheduling strategy; and updating the target scheduling strategy, the dynamic energy-carbon topology map, or the comprehensive confidence level based on the execution result, ultimately enabling energy-carbon coordination among the core energy devices. This method addresses the problem that existing technologies focus on optimizing a single link and have not yet formed an integrated solution covering multi-energy coupling, multi-entity coordination, and a closed-loop process. It achieves refined and intelligent collaborative management and control of energy consumption and carbon emissions in industrial parks. Attached Figure Description

[0017] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0018] Figure 1 This application provides a main flowchart of an energy-carbon synergy method for an energy-carbon transfer system. Figure 2 A flowchart illustrating the calculation of overall confidence level is provided for an embodiment of this application; Figure 3 A schematic diagram illustrating the process of generating a dynamic energy carbon topology map and its output, provided for embodiments of this application; Figure 4 A flowchart illustrating the generation of a target scheduling strategy is provided for an embodiment of this application. Figure 5A schematic diagram of a feedback adjustment process provided for an embodiment of this application; Figure 6 This is a schematic diagram illustrating an adjustment based on the cause of the main deviation, provided as an embodiment of this application. Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0019] As the proportion of new energy sources, mainly wind and solar, in the energy consumption structure of the park continues to increase, the strong intermittency and volatility of new energy output bring new technical challenges to the park's energy and carbon collaborative management: On the one hand, the uncertainty of new energy output leads to dynamic changes in the quality of energy and carbon data, and traditional static data governance methods are difficult to guarantee data confidence; on the other hand, the multi-energy coupling relationship and the spatiotemporal migration of carbon flow transmission paths caused by the fluctuation of new energy cause the energy and carbon model based on fixed topology to fail; furthermore, the prediction-based optimization scheduling scheme cannot adaptively adjust due to the deviation of actual operating conditions from the preset during the execution process, resulting in a significant reduction in optimization effect.

[0020] Current research on energy and carbon management in industrial parks has made some progress: at the data acquisition level, it has achieved data acquisition from multiple energy types and supports mainstream industrial protocols such as Modbus and OPC UA. Some solutions have further introduced spatiotemporal correlation feature extraction and graph neural network encoding technologies to improve the intelligence level of data processing; at the energy and carbon modeling level, methods such as multi-energy coupling, fixed emission factor accounting, and segmented carbon flow tracking have been developed; at the optimization and scheduling level, game theory models, dual-objective (cost + carbon emission) optimization models, and solution methods such as particle swarm optimization and genetic algorithms have emerged; at the intelligent decision-making level, technologies such as blockchain notarization and digital twins have been introduced to improve data credibility and visualization. However, existing technologies mostly focus on optimizing a single link and have not yet formed an integrated solution covering multi-energy coupling, multi-entity collaboration, and a closed-loop process.

[0021] To address the aforementioned technical problems, this invention proposes an energy-carbon synergy method for an energy-carbon circulation system. The implementation details of this energy-carbon synergy method for an energy-carbon circulation system are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution.

[0022] Example 1: like Figure 1 As shown, this application provides an energy-carbon synergy method for an energy-carbon circulation system. The method is applicable to electronic devices, such as servers, mobile terminals, computers, and cloud platforms. The data processing functionality of the electronic device provided in this application embodiment can be implemented by the processor of the electronic device calling program code, wherein the program code can be stored in a computer storage medium. The energy-carbon synergy method for the target energy-carbon circulation system includes: Step S1: Obtain energy and carbon data and equipment data collected by each core energy device in the target energy and carbon transfer system.

[0023] In this application, the energy-carbon flow system refers to a complex system formed by the coupling of energy flow networks and carbon flow networks within an industrial park. It not only encompasses the entire process of production, transmission, conversion, and consumption of various energy sources such as electricity, heat, gas, hydrogen, and cooling, but also closely links energy flow with carbon emission measurement through clearly defined physical conversion relationships (such as carbon emissions from electricity consumption). The core energy equipment within this system includes, but is not limited to, generators, photovoltaic panels, energy storage devices, charging piles, and high-energy-consuming industrial equipment; these are the fundamental nodes constituting this complex network.

[0024] This step involves using edge gateways deployed on core energy devices within the energy and carbon transfer system to collect device-level data via industrial protocols such as Modbus, OPCUA, and MQTT. Simultaneously, it connects to enterprise ERP and MES systems via API interfaces to obtain both device-level and system-level data. Device-level data is collected at a frequency of no more than one minute and includes energy data (such as electricity, heat, gas, hydrogen, and cooling), new energy data (such as photovoltaic output and wind power output), carbon emission data, and device status data. Timestamps and spatial tags are added during collection to ensure spatiotemporal consistency. System-level data (collected as needed) includes production data, market data, and management data. Timestamps are added during collection, and spatial mapping is established through associated fields. This multi-source heterogeneous data forms the foundational input for building a high-fidelity energy and carbon model. "Energy and carbon data" refers to operational data characterizing energy supply, conversion, consumption, and carbon emission intensity; "device data" refers to identifying data characterizing the device's type, spatial location, connectivity, and status.

[0025] Step S2: Determine the current energy and carbon activity data of each of the core energy devices and the comprehensive confidence level of the current energy and carbon activity data based on the device data and the energy and carbon data.

[0026] Due to factors such as sensor noise, communication interference, fluctuations in renewable energy output, and equipment aging, the raw collected data contains uncertainties. If directly used for modeling and optimization, data quality issues will be amplified to the entire system's carbon flow accounting and scheduling decisions. This step cleans, standardizes, and manages the confidence level of the raw collected data, quantifying the credibility of each data point at the current moment. "Current energy and carbon activity data" refers to structured data units (CFAU units) organized in a unified standard format after cleaning and semantic unification, carrying timestamps, spatial tags, data sources, and energy and carbon flow information. "Overall confidence level" is a scalar between 0 and 1; a higher value indicates greater reliability of the data source. It is a quantitative indicator of the reliability of the current energy and carbon activity data, representing the credibility of the data at the current moment after double verification and correction. It is a core endogenous variable that runs through the entire chain of subsequent topology learning, optimization scheduling, and closed-loop correction.

[0027] In some embodiments, such as Figure 2 As shown, step S2, "determining the current energy and carbon activity data of each of the core energy devices based on the device data and the energy and carbon data, and the comprehensive confidence level of the current energy and carbon activity data," includes: Step S21: Process the energy carbon data and the equipment data according to the preset data processing rules to obtain the current energy carbon activity data of each of the core energy devices.

[0028] The raw collected data is cleaned by using median filtering to remove sensor noise, aligning multi-source data time based on NTP or GPS, and using threshold fusion and LSTM residual identification for anomaly detection. The cleaned data is then uploaded to the cloud. Based on pre-configured device ledgers and business rules, the cloud uses semantic mapping to uniformly convert data from different sources and formats into the CFAU (Carbon Flow Activity Unit) standard format. Each CFAU unit includes: a unique identifier (ID), a time stamp (T), a spatial label (S), a data source (Source), a carbon flow type (Type), energy data (E_Data), carbon flow data (C_Data), associated data (M_Data), and a blockchain notarization hash (Cert).

[0029] Step S22: Calculate the time-varying decay factor based on the current energy and carbon activity data of the new energy equipment in the target energy and carbon transfer system. The time-varying decay factor is used to characterize the output uncertainty of the new energy equipment.

[0030] In actual energy and carbon transfer systems, the actual output of new energy equipment (such as photovoltaic and wind power generation) is often significantly affected by natural environmental factors such as weather and seasons, exhibiting strong randomness and volatility. This inherent uncertainty can cause abrupt changes in the collected energy and carbon data between adjacent moments. If such unprocessed data is directly used for subsequent scheduling decisions, it is very easy for the system to misjudge the equipment status, thereby causing scheduling imbalances or safety hazards.

[0031] To characterize and quantify the changes in data reliability caused by the integration of new energy sources, this scheme introduces a time-varying decay factor. This factor is a weighting coefficient that dynamically changes over time, and its core purpose is to capture the degree of uncertainty in new energy output. By calculating the time-varying decay factor, this scheme can dynamically assess the reliability of new energy data at the current moment. During periods of drastic data fluctuations (such as a sharp drop in photovoltaic output due to rapid cloud cover), this factor will decrease significantly, thereby reducing the weight of data from that period in subsequent calculations; conversely, during periods of stable sunlight, this factor will remain at a higher level.

[0032] Therefore, introducing the time-varying decay factor into the subsequent comprehensive confidence calculation is to solve the problem that "static weights cannot adapt to dynamic environments" in the existing technology. Its important contribution to the calculation of comprehensive confidence is that by automatically reducing the confidence weight of relevant data when the uncertainty of new energy sources is high, it effectively suppresses decision noise caused by prediction bias or environmental disturbances, ensuring that the system can still output robust and reliable scheduling strategies under complex operating conditions.

[0033] Specifically, the time-varying decay factor in this application The expression is as follows:

[0034] in, The coefficient of variation of the contribution of new energy sources within the sliding window. and These are the standard deviation and mean of the new energy output sequence within the sliding window, which contains 96 time points (corresponding to 24 hours, with each time point lasting 15 minutes). The mean absolute error of the ultra-short-term prediction model within the most recent time window is calculated from the average absolute deviation between the predicted and actual values. The scaling factor is determined by the ratio of the park's new energy installed capacity to the total load.

[0035] Step S23: Calculate the static score of each core energy device based on the current carbon activity data of each core energy device.

[0036] While time-varying decay factors can effectively capture short-term environmental disturbances caused by fluctuations in renewable energy output, they cannot cover the impact of inherent differences in equipment properties on data reliability due to uncertainties arising from changes in external meteorological conditions. In energy and carbon transfer systems, different equipment exhibit inherent differences in hardware quality, operational stability, and health status: some equipment uses sensors with lower accuracy levels, resulting in inherent errors in the collected data; some equipment has poor consistency in historical operating data, exhibiting irregular fluctuations over a long period; and some equipment has been in operation for a long time, with aging core components causing shifts in output characteristics. These differences are all long-term inherent characteristics at the equipment level and are unrelated to short-term renewable energy fluctuations. If only time-varying decay factors are used to evaluate data reliability, a systematic misjudgment may occur, leading to the assumption that "all equipment data is highly reliable when the environment is stable." Therefore, it is necessary to introduce independent quantitative indicators to characterize the inherent reliability of the equipment, i.e., static scoring.

[0037] In this step, the static score is calculated based on three core dimensions: sensor accuracy, historical data consistency, and equipment health status. Sensor accuracy directly determines the hardware reliability of the data acquisition end; the lower the accuracy level, the greater the inherent error in the data. Historical data consistency reflects the long-term operational stability of the equipment; if historical data frequently shows irregular jumps, it indicates a hidden risk of equipment failure. Equipment health status reflects the operational reliability of the equipment itself; the output characteristics of aging or damaged equipment will deviate, causing the measured data to deviate from the actual operating conditions. The weights of the three dimensions are objectively determined using the entropy weighting method, avoiding the subjectivity of manual weighting. The final static score is a quantitative representation of the average reliability of the equipment throughout its entire life cycle, which does not change with short-term environmental fluctuations, providing a stable benchmark for confidence calculation.

[0038] The introduction of static scoring addresses three core issues: First, it provides a benchmark for equipment reliability, preventing significant jumps in equipment reliability evaluation due to short-term environmental changes and ensuring a stable foundation for initial confidence calculations. Second, it enables differentiated equipment under the same operating conditions. Equipment within the same process chain shares the same time-varying decay factor (i.e., faces the same environmental fluctuations), but static scoring can distinguish individual reliability differences among different equipment under the same environment, ensuring that data from high-reliability equipment receives higher weight in subsequent calculations. Third, it provides a basis for differentiated adaptation of subsequent verification rules. For equipment with low static scores (such as those with poor sensor accuracy or severe aging), the verification threshold can be appropriately relaxed to suit its inherent characteristics, avoiding frequent false alarms caused by the equipment's inherent characteristics and improving the rationality of the verification mechanism.

[0039] The initial confidence level is obtained by multiplying the static score by the time-varying decay factor. This combines the inherent reliability of the equipment with short-term environmental interference correction. However, both are predictive indicators based on historical patterns and preset models, and have not been verified by real-time operating data. They cannot identify occasional anomalies such as sudden equipment failures and instantaneous communication interruptions. Therefore, the calculation result at this time can only be used as the initial confidence level. It needs to be double-verified and corrected by subsequent steps S25-S27 to form the final comprehensive confidence level used for end-to-end decision-making.

[0040] The process for calculating the initial confidence level in this application is as follows:

[0041] The mean of the static scores of each node in the process chain is obtained by weighting the scores of sensor accuracy, historical data consistency, and equipment health status at each node. The weights are objectively determined using the entropy weight method. The calculation formula is as follows:

[0042] in The static score of node i in process chain j is dimensionless and is obtained by weighting three dimensions: sensor accuracy, historical consistency, and equipment health status. The weighting weights are objectively determined by the entropy weighting method. Let j be the total number of nodes in the process chain. This represents the set of nodes in process chain j.

[0043] Step S24: Calculate the initial confidence level of each of the current energy and carbon activity data based on the static score and the time-varying decay factor.

[0044] Step S25: Verify the current energy and carbon activity data according to the current verification threshold at the current moment, and obtain the verification result.

[0045] Step S26: Determine the correction coefficient for the initial confidence level based on the verification result.

[0046] Step S27: Correct the initial confidence level according to the correction coefficient to obtain the comprehensive confidence level of the current energy and carbon activity data.

[0047] The calculation result in step S23 can only be called the initial confidence level, not the final comprehensive confidence level used for scheduling decisions. This is because, although the initial confidence level incorporates structural and environmental factors, it is still a predictive evaluation based on models and historical patterns. In actual energy and carbon transfer systems, equipment may generate transient abnormal data due to sudden failures, communication delays, or human error. These anomalies are often difficult to capture in time using static scoring and conventional time-varying decay. If the initial confidence level is used directly to guide scheduling, potential abnormal data may lead the system to misjudge the equipment status, causing scheduling risks. Therefore, it is necessary to introduce a real-time, dynamic verification mechanism for data mutations to correct them in order to generate a final confidence level that can truly withstand sudden disturbances.

[0048] This application employs a dual verification mechanism of energy conservation verification and operating condition consistency verification. The verification basis and the meaning of the results of the two mechanisms belong to different dimensions, forming a complementary reliability verification system: The energy conservation verification is based on the fundamental physical laws of the energy-carbon circulation system—all energy input, conversion, transmission, and consumption processes strictly follow the law of energy conservation, and there is no energy gain without a source or energy loss without a destination. The verification uses the process chain as the smallest accounting unit, verifying the balance between the total input power, total output power, and total transmission loss within the same process chain. The verification tolerance comprehensively considers all acceptable unmodeled errors, including the overall accuracy level of the CT / PT sensor (allowable error ±0.5%), line losses, and transformer losses. If the verification fails, it indicates that there is a systematic deviation in the measurement data of the current process chain: either the input / output sensor synchronization drift (such as batch failure of CT / PT on the same bus), or the transmission loss model is mismatched with the actual operating conditions, or there are unidentified energy leaks, incorrect data acquisition timestamp alignment, etc. Since energy non-conservation often points to regional measurement system deviations rather than single-point failures, the result of the verification failure can be directly used as the core basis for reducing the confidence of a single process chain and lowering the global correction coefficient of the sub-topology region, so as to achieve "regional" risk isolation and avoid omission of single-point correction.

[0049] It should be noted that the process chain is a logical set divided according to the energy and carbon functional logic, while the sub-topology region is a physical set divided according to the physical measurement infrastructure (such as a set of equipment sharing the same CT / PT measurement bus). The two overlap but are not completely equivalent: a process chain may span multiple sub-topology regions (such as the need to pass through multiple buses from the photovoltaic array to the end load), and a sub-topology region may also carry multiple process chains (such as the same bus supplying power to both the production and office process chains at the same time). The synergy between the two achieves dual coverage of "logical function control" and "physical measurement correction".

[0050] Of course, in actual implementation, the correction coefficient can always exist. However, when the verification passes, the correction coefficient is 1, and when the verification fails, the correction coefficient will be changed. Specifically, a separate correction coefficient is given to the initial confidence of the single process chain that failed the verification, while a group correction coefficient is given to other process chains in the topology-level region related to the single process chain that failed the verification.

[0051] Specific confidence corrections for a single process chain: Reduce the overall confidence level of this process chain. ,in This is the energy conservation failure attenuation coefficient, with a default value of 0.8.

[0052] Global correction coefficients for subtopological regions: A sub-topology region refers to a collection of devices that share the same physical measurement infrastructure. Reduce the global correction factor for the sub-topology region where this process chain resides. , The effective confidence levels of all process chains in this region are adjusted synchronously: .

[0053] in, Indicates the initial confidence level. This indicates the overall confidence level, which is the final confidence level output in step S2. In this application, the parameter for "old" represents the old value, while the parameter for "new" represents the new value. For example... This represents the old global correction factor. This represents the new global correction coefficient.

[0054] In addition to providing correction coefficients, the sampling frequency also needs to be dynamically adjusted when energy non-conservation occurs. The sampling frequency of the equipment in this area was temporarily increased from 1 time / min to 1 time / 15s (4 times encryption) for 6 windows (1.5 hours). Encrypted sampling can capture transient anomalies that may lead to energy non-conservation (such as CT phase abrupt changes and harmonic interference), providing richer data support for the next cycle calibration.

[0055] Furthermore, the basis for the operating condition consistency verification is the inherent physical operating boundary of the core energy equipment—the output change rate of each piece of equipment is rigidly limited by its own physical characteristics (such as gradeability, thermal inertia, and mechanical stress), and cannot exceed the maximum allowable change range in the design. Specifically, the operating condition consistency verification logic in this application is as follows: .

[0056] in These are the physical measurements for the current scheduling cycle of the equipment. These are the physical measurement values ​​from the previous scheduling cycle. This refers to the maximum permissible variation between adjacent cycles, determined based on the equipment's physical characteristics. It represents the absolute physical limit determined by the equipment's ramp rate, thermal inertia, and mechanical stress. The tolerance for consistency verification under operating conditions is set to 3% by default, which is a relative relaxation ratio based on physical limits to accommodate sensor noise and fluctuations in normal operating conditions.

[0057] If the verification fails, the anomaly type can be precisely identified based on the failure mode: exceeding the upper limit indicates that the data may have positive sensor drift or control command overshoot; exceeding the lower limit indicates that the data may have negative sensor drift or a hidden internal fault in the equipment; sudden exceedance... This most likely indicates a communication interruption or complete sensor failure, rendering the data completely unreliable. Unlike the link-level verification of energy conservation checks, operating condition consistency checks are node-level time-series verifications. They can capture instantaneous single-point anomalies that the initial confidence level cannot cover—for example, a device may have a high static score and its process chain may be in energy balance, but the output mutation rate of a single device exceeds the physical limit. Such anomalies can only be identified through operating condition checks. Therefore, the result of a failed check can be directly used as the basis for the initial confidence level of the device corresponding to the nonlinear decay, while triggering an adaptive adjustment of the check threshold.

[0058] That is, when the verification fails, the failure mode is recorded: exceeding the upper limit. This could be due to positive sensor drift or control command overshoot or exceeding the lower limit. This could be due to sensor negative drift, equipment malfunction, or sudden exceeding of limits. In cases where communication is likely interrupted or the sensor malfunctions, the data is directly marked as "untrustworthy." A sliding window counter is maintained for each mode. The threshold has been exceeded cumulatively over the last W windows in a certain pattern. When this happens, the verification threshold for this mode will be automatically adjusted.

[0059] in: The threshold adjustment rate coefficient, , The larger the threshold, the more sensitive it is to historical verification failures, but too large a threshold can cause it to oscillate. The cumulative failure count threshold for triggering automatic threshold adjustment is set to 30 by default, based on the average drift period of industrial sensors (approximately 3-6 months) and a sampling frequency of 15 minutes; W is the sliding window length, set to 100 windows (approximately 25 hours) by default, covering a complete workday cycle.

[0060] Due to gradual deviations caused by long-term sensor drift or equipment aging, a fixed threshold will generate a large number of false alarms, diluting the true abnormal signal. After evolution, the threshold will automatically be relaxed to adapt to the current health status of the sensor.

[0061] Threshold changes directly alter the frequency of verification failures, thus affecting confidence level updates. When a single verification fails... When n consecutive failures occur , , This is the operating condition consistency correction factor. Tightening the threshold (e.g., reducing it from 3% to 2.4%) increases the failure frequency and accelerates the confidence decay, making the system more sensitive to anomalies; relaxing the threshold (e.g., increasing aging adaptation from 3% to 4.5%) reduces the failure frequency and slows down the decay, avoiding continuous false alarms caused by aging.

[0062] Although this application uses two verification criteria, each with its own correction calculation method, they are effective simultaneously. That is, the final overall confidence level is the result of the initial confidence level after continuous energy correction and operating condition consistency correction. The total correction coefficient can be expressed as... For ease of representation, the total correction factor is... Includes The possibility that the process chain will fail to verify energy is due to the use of... As a correction factor, other process chains in the topological region are used. As a correction factor.

[0063] The aforementioned dual verification empirically corrects the initial confidence level prediction from two dimensions: "balance of link-level quantities" and "reasonableness of node-level quantities." The energy conservation verification confirms the systematic consistency of the data and eliminates the influence of regional measurement deviations; the operating condition consistency verification confirms the temporal rationality of the data and eliminates the influence of single-point instantaneous anomalies. The confidence level corrected by the dual verification integrates three layers of information: inherent equipment attributes, environmental fluctuations, and physical law verification. Therefore, it is called the comprehensive confidence level, rather than the initial prediction value.

[0064] Furthermore, the consistency verification threshold in this application possesses a bidirectional adaptive adjustment mechanism: the adjustment period for the relaxation direction (sensor aging adaptation) is W windows (approximately 25 hours), while the adjustment period for the tightening direction (triggered by deviation feedback or a decrease in topology centrality; specific tightening trigger conditions are detailed in step S364) is at least 30 complete scheduling cycles (approximately 7.5 hours). The time scale difference between the two directions is approximately one order of magnitude, thus eliminating the possibility of oscillations. If both directions are triggered simultaneously, the relaxation direction is executed first—because sensor aging is an irreversible physical fact, and increased deviation may be an inevitable result of aging; the confirmed aging trend should not be resisted due to increased deviation. This bidirectional adjustment mechanism, combined with the Lyapunov convergence design in subsequent steps, jointly ensures that the system does not oscillate or diverge from both the internal and external levels of data governance. After double verification and correction, the core fields of the CFAU unit, the verification log and the data hash value are assembled into a storage data package and uploaded to the consortium blockchain. After consensus verification, a unique storage hash is generated and filled into the Cert field. The standardized CFAU data stream is published once per minute through the Kafka message queue for downstream modules to subscribe to and use.

[0065] Step S3: Construct a dynamic energy-carbon topology diagram among the core energy devices in the target energy-carbon circulation system based on the current energy-carbon activity data and the comprehensive confidence level.

[0066] Traditional physical topologies are static and cannot accurately represent the dynamic changes in the energy-carbon coupling relationship of a park under high-proportion renewable energy access over time. Based on the high-quality CFAU data stream and structured comprehensive confidence score output from step S2, this step introduces a Graph Attention Network (GAT) to construct a dynamic energy-carbon topology graph. Here, the comprehensive confidence score is no longer just an isolated data quality scale, but is directly embedded into the GAT attention calculation as a multiplicative factor of the node feature vectors. The feature vectors of low-confidence nodes are compressed as a whole before the attention calculation, causing them to be structurally marginalized in multi-layer GAT propagation (attention residue is less than 1% after 3 layers), fundamentally eliminating the interference of low-quality data on the carbon flow accounting of the entire system. Simultaneously, this step outputs a time-varying energy transfer matrix A(t) and a carbon flow transfer matrix T(t), defining a dynamic constraint space for subsequent optimization, and calculating the rate of change of node topological centrality to feed back mutual signals to step S2, synchronizing topology learning with data quality assessment.

[0067] In some embodiments, such as Figure 3 As shown, step S3, "Constructing a dynamic energy-carbon topology map between core energy devices in the target energy-carbon circulation system based on the current energy-carbon activity data and the comprehensive confidence level," includes: Step S31: Construct an energy-carbon transfer topology diagram based on the target energy-carbon transfer system.

[0068] To construct a dynamic energy-carbon topology, a static energy-carbon flow topology must first be defined. The energy-carbon flow system in this application not only records equipment information but also constructs models for each device and records the energy-carbon flow relationships between them. Therefore, the core energy devices in the energy-carbon flow system can be used as nodes, and the energy-carbon flow relationships between each energy device can be used as edges to construct a static energy-carbon flow topology. Then, based on real-time energy-carbon activity data and comprehensive confidence levels, the parameters of each edge and node are updated, thereby achieving the construction of a dynamic energy-carbon topology.

[0069] Step S32: Determine the current equipment efficiency and current carbon flow intensity of each core energy device based on the current energy and carbon activity data of the target energy carbon transfer system and each core energy device.

[0070] In addition to the energy and carbon activity data output in step S2, the characteristics of the nodes also need to consider the current equipment efficiency and carbon flow intensity of each device. Since the energy and carbon flow system of this application has already built a model for the equipment efficiency and carbon flow intensity of each device in advance, the energy and carbon activity output in step S2 can be substituted into the model to obtain the carbon flow intensity and equipment efficiency of each device under the current energy and carbon activity data.

[0071] The following are the preset models for some devices in this application: Modeling photovoltaic equipment:

[0072] in, Contribute to photovoltaic power Light intensity, For the area of ​​the photovoltaic panel, Photovoltaic conversion efficiency. Photovoltaic carbon flux intensity. (Zero-carbon energy).

[0073] Modeling wind power equipment:

[0074] in, To provide power for wind power, air density, For the area swept by the wind turbine, For wind speed. Efficiency. It varies non-linearly with wind speed and decreases slowly with operating time. Wind power carbon flux intensity =0 (zero carbon energy).

[0075] Modeling the electrolytic cell equipment:

[0076] in, For hydrogen production capacity, For input electrical power, For electrolysis efficiency, the carbon flow intensity of the electrolyzer depends on the proportion of green electricity used.

[0077] in For the proportion of green electricity used, The intensity of carbon flow in the power grid.

[0078] The modeling of equipment such as gas-fired boilers, combined heat and power (CHP) units, hydrogen fuel cells, P2G devices, and CCUS devices follows the same principles. Output power is determined by the product of input energy and dynamic efficiency, which dynamically changes with load rate, operating time, or operating conditions. Carbon flow intensity is calculated based on the energy source and conversion efficiency. Specifically, the carbon flow intensity of P2G and CCUS devices is negative, representing carbon fixation or removal; the carbon flow intensity of hydrogen fuel cells inherits the carbon footprint of upstream hydrogen.

[0079] Step S33: Determine the node characteristics of each node in the energy and carbon flow topology graph based on the current equipment efficiency, the current carbon flow intensity, and the current energy and carbon activity data.

[0080] Based on the equipment efficiency of the above devices Carbon flow intensity Using CFAU data streams, all energy equipment in the park is abstracted as nodes (PV / Wind / Energy Storage / Electrolyzer / CCUS / Load), and the energy-carbon transfer relationships between devices are used as edges (Electricity / Heat / Carbon Flow) to construct an energy-carbon coupled heterogeneous graph. Node feature vectors are used. Includes device identifier, spatial tag, device type, real-time energy data, real-time carbon flow data, and a comprehensive confidence scalar output by S2. The equipment type is used to distinguish whether an equipment is a carbon flow generator, transmitter, consumer, or reducer. The initial construction of edges is based on the physical connections between equipment (such as cables, pipes, heating networks, etc.), with an initial weight of 1. Subsequent steps use GAT to learn the dynamic weights of carbon transfer between equipment.

[0081] Step S34: Determine the dynamic attention weight of each edge in the energy carbon flow topology graph based on the comprehensive confidence level.

[0082] Step S35: Construct the dynamic energy-carbon topology graph based on the node characteristics of each node, the dynamic attention weights of each edge, and the energy-carbon flow topology graph.

[0083] This application employs a graph attention network to train the energy and carbon transfer topology graph, learning the dynamic attention weights for energy and carbon transfer between devices. During training, the attention level of node i to its neighbor node j is calculated. At that time, As node feature vectors Multiplicative factors, embedded in GAT attention computation:

[0084] in, Let be the original attention score of node i to node j. For activation function, Let be the feature vector of node i, and W be the learnable linear transformation matrix. It is the attention weight vector. It's a vector concatenation operation. The comprehensive confidence scalar for node i is obtained from the structured confidence tensor output by S1.

[0085] When a certain node When low, the feature vector The feature vector (i.e., the vector of node features) is compressed as a whole, and the feature representation after linear transformation is systematically weakened. The attention contribution of the node to its neighbors is suppressed at the feature level. Feature vector-level embedding compresses low-quality node features from the source of information aggregation. After 3 layers, the residual is <1%. Low-quality nodes are structurally marginalized, and their interference with the carbon flow accounting of the whole system is fundamentally eliminated.

[0086] After normalization using the Softmax function, the result is obtained Attention scores for intervals are automatically assigned during normalization to connections in high-quality data regions.

[0087]

[0088] in, The normalized attention score is the dynamic attention weight used in this application. This is a dynamic edge weight matrix. A dynamic threshold is set. ,when At that moment, it is assumed that there is no efficient carbon transfer between nodes i and j, and the corresponding edge is set to zero. The above calculation is repeated for all node pairs to obtain the final dynamic edge weight matrix. This provides adaptive topological weight support for subsequent system-level energy-carbon coupling modeling.

[0089] Of course, after obtaining The construction of the dynamic energy-carbon topology graph has been completed. However, subsequent calculations, especially in step S4, require the extraction of a large amount of data from the dynamic energy-carbon topology graph. This data includes the energy transfer matrix A and the carbon flow transfer matrix T. Therefore, it is necessary to separate the energy transfer matrix A(t) and the carbon flow transfer matrix T(t) from the dynamic energy-carbon topology graph. Thus, in this application, the final generated dynamic edge weight matrix is ​​obtained first. Then, based on the device-level dynamic energy conversion efficiency and carbon flow intensity, the energy transfer matrix A(t) and the carbon flow transfer matrix T(t) are constructed.

[0090] Specifically, the process of constructing the energy transfer matrix A(t) and the carbon flux transfer matrix T(t) is as follows: Let the energy flow transfer matrix be defined. Therefore, matrix elements The efficiency of energy transfer from energy p to energy q is obtained by weighted summing of the dynamic weights of the GAT of all device nodes undertaking the transfer task and the device efficiency:

[0091] in, Corresponding to electricity, heat, gas, hydrogen, and cold, respectively. The set of device node pairs responsible for the "energy p→q" transfer; Node pairs output by GAT Dynamic attention weights; For device node pairs The energy conversion efficiency of the equipment it represents.

[0092] Let the carbon flow transfer matrix be defined. Therefore, matrix elements The carbon flux intensity transfer coefficient, representing the transfer of energy from p to q, is obtained by weighted summation of the GAT dynamic weights of all device node pairs undertaking the transfer task and the carbon flux intensity of each device.

[0093] in, For device node pairs Equipment-grade carbon flow intensity.

[0094] In some embodiments, the method further includes: Step S361: When the first preset condition is met, obtain the latest comprehensive confidence level and current energy and carbon activity data of each core energy device.

[0095] Step S362: Update the dynamic energy and carbon topology map based on the latest comprehensive confidence level and the current energy and carbon activity data.

[0096] Although steps S31-S35 explain how to construct a dynamic energy and carbon topology map based on the current energy and carbon activity data and comprehensive confidence level output in step S2, the comprehensive confidence level and current energy and carbon activity data output in step S2 are updated in real time at the minute level. The dynamic energy and carbon topology map should have iterated synchronously with the update of the above data to match the real-time status of the energy and carbon circulation system to the greatest extent. However, in practical engineering, indiscriminate real-time full updates can lead to three core problems: First, the multi-layer training computation of Graph Attention Networks (GAT) is computationally expensive. During periods of stable operation, the topology itself does not undergo substantial changes, and high-frequency repetitive calculations will ineffectively occupy edge computing resources, which does not meet the requirements for lightweight deployment of park-level systems. Second, short-term random fluctuations in renewable energy output and load demand (such as second-level jumps in photovoltaic output caused by cloud cover and minute-level adjustments in flexible loads) can cause instantaneous disturbances in confidence and equipment operation data. If the topology is updated synchronously, it will cause frequent oscillations in the topology edge weights, making it difficult for subsequent optimization scheduling to capture stable energy and carbon transfer patterns. This can easily lead to a "ping-pong effect" of frequent switching of scheduling strategies, and even cause system instability. Third, equipment efficiency decay and seasonal changes in carbon flow intensity are slow time-varying processes. High-frequency updates cannot capture these long-term evolutionary trends and will instead mask the true characteristics of topological changes.

[0097] Therefore, this application does not adopt an indiscriminate real-time synchronous update mechanism, but sets a "trigger-based conditional update" rule: the iterative update of the topology is only triggered when the first preset condition is met (such as the fluctuation of new energy output exceeds the threshold, the rate of change of equipment efficiency exceeds the critical value, or the node topology centrality shows a continuous shift). This ensures the sensitivity of the topology to substantial changes in operating conditions, avoids invalid calculations and strategy oscillations, and also provides logical support for the flexible switching between the basic update mode and the enhanced update mode.

[0098] This application has two update modes: a basic update mode and an enhanced update mode.

[0099] • Basic mode: Iterates every 15 minutes, updating GAT weights and system-level matrices based on the latest CFAU data. It has low computing power requirements and is suitable for parks with stable operating conditions.

[0100] • Enhanced Mode: Automatically switches when renewable energy output fluctuates drastically or equipment efficiency changes beyond a set range. Building upon the basic mode, it introduces a Transformer time-series forecasting model to predict renewable energy output, load demand, and energy storage SOC for the next 1-4 hours based on historical CFAU data. A confidence gating mechanism is incorporated, and the weights of the predicted data are:

[0101] in The confidence level of the prediction model itself (calculated by a sliding window of historical prediction accuracy). This represents the weight of the predicted data, with values ​​between (0,1). A larger value indicates greater confidence in the predicted data. This represents the confidence level of the actual (measured) data, corresponding to the actual data collected in the system (such as sensor readings). It is typically provided by the "overall confidence level" calculated in the preceding step (S2), reflecting the quality and reliability of the actual data at the current moment.

[0102] This gating ensures that low-quality prediction data does not contaminate topology learning. rate of change Automatically switches to enhanced mode when the trigger threshold is exceeded. Dramatic changes indicate that the operating conditions of new energy sources are undergoing rapid shifts, and real-time data alone cannot reflect the forward-looking migration of carbon flow paths. When the indicator falls below the threshold for two consecutive update cycles, the system switches back to the basic mode. The two modes can be flexibly switched via a configuration switch, balancing the practicality and advanced nature of the solution.

[0103] Step S363: After each update, calculate the node centrality of each node in the dynamic energy carbon topology graph.

[0104] Step S364: Update the current verification threshold when the change in the node centrality of each node satisfies the second preset condition.

[0105] Node centrality is the core topological feature output by Graph Attention Network (GAT). Essentially, it is an aggregated statistical analysis of the attention weights of nodes in the global topology, which quantifies the "voice" of a single device in the energy and carbon flow network—that is, the influence weight of the node's operating data and energy and carbon conversion behavior on the full-link carbon flow accounting and scheduling decisions. This application introduces node centrality calculation, the core of which is to address the limitation that "single-point confidence evaluation cannot cover topological evolution characteristics": the comprehensive confidence output in step S2 is a data quality prediction at the single device level, which can only reflect the inherent reliability and short-term fluctuations of single-point data, and cannot reflect the changes in the functional importance of the device in the global energy and carbon network; while node centrality integrates three pieces of information: "data quality (comprehensive confidence from S2) + topological location + energy and carbon transfer contribution", which is a post-hoc verification of device reliability at the topological level. If the high-confidence node output by S2 maintains high centrality in the topology, it indicates that the confidence evaluation is accurate and the data quality is reliable; if the centrality of a high-confidence node continues to decline, it indicates that the existing verification rules have failed to capture the hidden anomalies of the node (such as slow sensor drift, loose internal connections, etc., which are not identified by single-point verification), or that the functional positioning of the node in the energy and carbon network has undergone substantial changes, and the upstream verification rules need to be adjusted through a feedback mechanism.

[0106] Therefore, in this application, the node centrality of each node will be calculated after the dynamic energy carbon topology graph is updated. The accuracy of the confidence evaluation will be determined based on the change in the centrality of each node after each update. For example, when the centrality of a node drops below the threshold for three consecutive windows, which is the second preset condition of this application, it indicates that there is a problem with the overall confidence of the node. Therefore, it is necessary to update the verification threshold to tighten the verification and improve the accuracy of the overall confidence.

[0107] Step S4: Determine the target scheduling strategy that conforms to the current operating conditions of each core energy device based on the dynamic energy carbon topology map and the comprehensive confidence level.

[0108] Traditional optimization methods often use fixed parameters, making it difficult to cope with data quality fluctuations and multiple constraint competitions. This step defines a dynamic constraint space using A(t) and T(t) output from step S3, constructs a two-level optimization model based on Stackelberg game, and uses comprehensive confidence to perform nonlinear piecewise modulation of the optimization parameters, while also combining constraints with topological spectral characteristics. Under specific operating conditions, the modulation direction can be conditionally reversed, and the contributions of different topological role devices to global modulation are weighted according to their topological roles, thus enabling the output of a scheduling scheme with high executability even when data quality is uncertain.

[0109] In some embodiments, such as Figure 4 As shown, step S4, "Determine the target scheduling strategy that conforms to the current operating conditions of each core energy device based on the dynamic energy carbon topology map and the comprehensive confidence level," includes: Step S41: Calculate the global modulation confidence level based on the comprehensive confidence level and the dynamic energy-carbon topology map. The global modulation confidence level is used to characterize the confidence level of all energy-carbon activity data in the target energy-carbon flow system.

[0110] Step S42: Determine the current stability distribution of the target energy carbon transfer system based on the global modulation confidence level.

[0111] Step S43: Determine the weighting coefficients of each parameter and the penalty values ​​of the penalty parameters based on the stability distribution and the dynamic energy carbon topology diagram.

[0112] Step S44: Adjust the current scheduling policy being executed according to the weight coefficient and the penalty value to obtain the target scheduling policy.

[0113] This application does not generate a scheduling strategy out of thin air, but rather adjusts an existing scheduling strategy—the current scheduling strategy in this application—so that the equipment's operating state can be adjusted to conform to the current operating conditions. Therefore, the current scheduling strategy in this application refers to the instruction set generated based on the previous scheduling cycle, representing the equipment's current expected operating mode, and is a planned output from the control side; the target scheduling strategy refers to the new instruction set, after optimization and adjustment in this step, ready to be issued to the equipment for execution. It should be noted that the scheduling strategy is not equivalent to the actual operating state of the equipment (i.e., the real-time operating condition on the physical side). The former is the "expected instruction," and the latter is the "execution result." The two only perfectly match under ideal, unbiased operating conditions; in actual operation, minor fluctuations conforming to physical laws are allowed. The design of obtaining the target scheduling strategy based on the current scheduling strategy, rather than directly generating a completely new target strategy, aims to ensure the smoothness and stability of the carbon transfer system's operation and avoid system oscillations caused by abrupt changes in control instructions.

[0114] In this application, to adjust the current dispatch strategy to a target dispatch strategy that aligns with current operating conditions, the park operator takes the lead, owning all proprietary equipment such as energy storage, photovoltaic / wind power, and CCHP, and formulates rigid rules for time-of-use pricing, carbon quotas, and energy storage dispatch, coordinating equipment operation and green electricity allocation. Its optimization objectives are to maximize overall revenue, maximize green electricity absorption rate, and minimize total carbon emissions, constructing a multi-objective function, i.e., an optimization model.

[0115] A collaborative solution strategy combining the upper-level NSGA-III algorithm and the lower-level distributed alternating direction multiplier method (ADMM) is employed, iterating to Stackelberg equilibrium. The upper-level algorithm is a three-objective optimization problem (economic benefits, green energy consumption, and carbon emissions, i.e., the optimization model). NSGA-III, through its reference point mechanism, exhibits good Pareto front distribution when handling multiple objectives. The lower-level algorithm is a multi-agent distributed optimization problem, where the data privacy of each follower needs to be protected. ADMM, through its decomposition-coordination mechanism, allows each agent to independently solve local problems, achieving collaborative convergence by exchanging dual variables.

[0116] The solution process is as follows: The upper layer uses time-of-use pricing, carbon quota allocation, and energy storage scheduling rules as decision variables, with a population size of 100 and 50 iterations. It substitutes the responses of each agent in the lower layer into the calculation of multi-objective fitness, generating a Pareto optimal solution set through non-dominated sorting and reference point selection. The lower layer decomposes the multi-agent optimization problem into four sub-problems, with each agent independently solving for local optima. Information is exchanged through dual variables, iterating until convergence (penalty parameter ρ=10, convergence threshold ε=0.0001). The upper layer distributes the new electricity price and quota schemes to the lower layer, which re-solves and provides feedback. This process is repeated until the upper layer's overall revenue fluctuation is less than 1% and none of the followers have a unilateral motivation to improve, i.e., Stackelberg equilibrium is reached. The optimal scheduling strategy, i.e., the target scheduling strategy, is output for subsequent module execution.

[0117] The specific expression for the optimization model is as follows:

[0118] in, For the overall benefit of the park; Economic benefits (in yuan) include revenue from energy sales and carbon trading, minus external energy purchase costs, equipment operation and maintenance costs, and carbon over-emission penalty costs. The revenue from green electricity consumption (dimensionless) includes the unit green electricity consumption subsidy × green electricity consumption volume + consumption rate incentive coefficient × consumption rate; The cost of carbon emissions (in yuan) includes the cost of purchasing carbon allowances and the cost of penalties for exceeding carbon emission limits. , , Let be the weighting coefficient, satisfying

[0119] Green electricity absorption rate constraints: , Policy requirements for green electricity consumption thresholds Operating constraints of self-owned energy equipment: , , The upper and lower limits of equipment output.

[0120] The lower layer comprises four stakeholders, each independently optimizing their own interests under the prices and rules set by the operator.

[0121] Energy suppliers, including grid suppliers and natural gas suppliers, are responsible for supplying external energy to the park to supplement its energy gap. The goal is to maximize net profit, and constraints include supply capacity constraints, price constraints, and supply gap constraints.

[0122] Energy storage operators: As energy storage operators, they undertake the scheduling of energy storage equipment and execute charge-discharge arbitrage strategies within the park's rules to achieve the spatial and temporal transfer of energy and the consumption of surplus green electricity. The goal is to maximize charge-discharge arbitrage profits, with constraints including energy storage capacity constraints, charge-discharge power constraints, cycle balance constraints, and charge-discharge mutual exclusion constraints.

[0123] CCUS operators, as core players in carbon reduction, capture emissions and process waste gas from equipment such as CCHP and gas-fired boilers, prioritizing their supply to P2G for CO2 resource recovery, with the remainder stored to obtain emission reduction subsidies and sales revenue. The goal is to maximize overall benefits, constrained by both capture capacity constraints and total emission conservation constraints.

[0124] End-user load providers: Absorb multiple types of energy (electricity / heat / gas / hydrogen / cooling), adhere to green electricity usage ratio requirements, and reduce energy costs through load shifting and energy substitution. The goal is to minimize overall energy costs, constrained by energy consumption constraints, green electricity usage constraints, and load shifting constraints.

[0125] To use an optimization model for collaborative solving, it is necessary to define the weight coefficients and constraint penalty parameters of each item in the optimization model.

[0126] Since the optimization model in this step involves system-level decision-making, it is necessary to aggregate equipment-level reliability information into a system-level reliability representation. However, the impact of different devices on the overall carbon transfer system varies significantly. For example, confidence fluctuations at primary energy inlets such as photovoltaics propagate along the carbon transfer matrix, affecting the entire carbon flow calculation, while confidence fluctuations at the terminal load only affect local consumption metering. If a simple arithmetic average is applied to the confidence levels of each device, low-impact load data fluctuations will frequently trigger unnecessary strategy switching, while high-impact photovoltaic data anomalies will be "diluted" by the average, leading to misjudgments in the strategy.

[0127] Therefore, in this application, the device node is based on the energy transfer matrix. or carbon flow transport matrix Based on the functional location within the network, the equipment in the park is divided into four topological roles, and a differentiated weighting coefficient k is introduced for each topological role. This allows for the generation of a differentiated weighting coefficient lookup table based on the dynamic energy carbon topology map. The global modulation confidence level is then calculated based on this lookup table and the overall confidence level of each device. The lookup table between the differentiated weighting coefficient k and the topological roles is as follows:

[0128] Based on the aforementioned topological role weights, the global modulation confidence is... Perform differential weighted calculation:

[0129] in, The differential weight coefficients are obtained by querying the pre-configured differential weight coefficient lookup table based on the equipment type. The differential weight coefficient lookup table can be manually adjusted according to the actual topology of the park or automatically learned and updated based on PageRank centrality.

[0130] For example: photovoltaic nodes During a sudden drop A significant decrease (k=1.0) was observed, and the system promptly strengthened constraints to ensure performance; only load nodes... During descent, Only a slight decrease (k=0.3) is made to avoid overreaction affecting scheduling stability.

[0131] After obtaining the global modulation confidence level, the weight coefficients of each parameter in the optimization model and the constraint penalty parameters can be determined based on the global modulation confidence level. Under ideal data quality, the three weights can be fixed according to policy requirements, and the penalty parameters can be taken as basic values. However, in actual energy and carbon transfer systems, data uncertainty can significantly change the optimal decision-making process. Studies have shown that increased uncertainty in renewable energy output leads to decreased operator revenue and increased carbon emissions and energy consumption; measurement uncertainty can even change the renewable utilization rate assessment results by more than 22.4%. Therefore, the global modulation confidence level must be used as a modulation signal to dynamically adjust the weight coefficients and constraint penalty parameters.

[0132] Therefore, this application also establishes a stability assessment system for energy and carbon transfer systems. It evaluates the stability of the energy and carbon transfer system through global modulation confidence levels, assigning different weighting coefficients and constraint penalty parameters for different levels of stability. Then, based on the pre-given penalty methods, it determines the weighting coefficients and penalty values ​​for each parameter. Specifically, this application divides stability into three zones: a safe zone, a transition zone, and a risk zone. Each zone is defined by a confidence threshold. and The process is broken down into different segments, each with its own handling method. The specific segments, thresholds, and handling methods are related as shown in the table below:

[0133] in, Within the sliding window The 75th percentile is the boundary between the safe zone and the transition zone. Within the sliding window The 25th percentile marks the boundary between the transition zone and the risk zone. Safe zone: , Transition zone: , Risk Zone: , ,in, This refers to a jump to at least one order of magnitude above the final value of the transition zone, where These are the weight coefficients of the optimization model described above, since this step can determine... ,so , Then according to Perform normal calculations to obtain the weight coefficients of each item in the optimization model. This represents the base value of the penalty parameter.

[0134] At the same time, the penalty parameter also needs to be related to the energy transfer matrix. The joint constraint on the spectral properties of the Laplace matrix is ​​expressed as follows:

[0135] in: For effective penalty parameters; for The largest eigenvalue of the Laplacian matrix reflects the sensitivity to topological connectivity. , It is a degree matrix; The historical steady-state baseline value is taken from the steady-state operating conditions of the past 30 days. The mean; This is the spectral sensitivity coefficient, determined by the network topology complexity and device coupling density of the campus, with a default value of 0.3. The larger the value, the more significant the amplification effect of changes in topological connectivity on the penalty parameter.

[0136] When topological connectivity is enhanced ( When the confidence level increases, the speed and range of local data disturbances propagate through the network increase, and even if the confidence level of the local data is acceptable, its potential impact on the entire system is greater. This "structural amplification effect" cannot be captured by confidence scalars alone; topological spectrum characteristics must be introduced for joint constraints, so that the penalty parameter is driven by both "data quality" and "structural sensitivity".

[0137] Theoretically, the more unreliable the data (the lower the confidence level), the tighter the carbon constraints should be (increasing the confidence level). Increase the constraint and penalty parameters The numerical value is used, but this logic assumes that the carbon emission data is correct. However, in reality, carbon emission accounting relies on carbon flow intensity, which in turn depends on the overall confidence level of S2. If the data itself is unreliable (e.g., carbon emission sensor drift, resulting in higher calculated carbon emissions than actual emissions; or inaccurate renewable energy output data leading to incorrect calculation of the green electricity ratio), tightening carbon constraints in this case is equivalent to using false data to set a stricter, more rigid rule. In addition, when the margin of total carbon emission constraints is insufficient, the system faces multiple competing constraints: there is a conflict between the hard constraint of total carbon emissions (exceeding the limit will result in penalties for production stoppages and restrictions) and the strengthening of carbon emission constraints based on unreliable data (soft constraints, which may be distorted to be too strict by low-quality data). If carbon emission constraints are still tightened according to conventional logic, the optimizer may not be able to find a feasible solution under the overly strict constraints constructed by unreliable data.

[0138] Therefore, this application establishes a conditional modulation direction reversal mechanism. This means that under specific technical conditions such as insufficient carbon emission control margin (remaining margin below the critical margin) or periods of tight grid supply and demand (grid dispatch instructions requiring reduced electricity consumption or increased output), the modulation direction can be conditionally reversed. Even in these situations... In the low value range, the weight of carbon emission constraints has decreased instead of increased, while the weight of economic targets has increased instead of decreased.

[0139] The direction reversal mechanism can prioritize satisfying the hard constraints on total carbon emissions and equipment safety based on unreliable data, enabling the optimizer to recover a feasible solution. Essentially, direction reversal uses reliable physical boundary signals to correct spurious constraints constructed from unreliable data, rather than parameter sweeping at the Solver level.

[0140] The direction reversal mechanism is triggered by a ternary joint judgment:

[0141] Modulation direction reversed:

[0142] in: This represents the carbon emission margin. The threshold value for carbon emission constraint margin is set at 5% by default, and is determined comprehensively based on the park’s historical carbon quota usage and policy requirements. This is a variable indicating the power grid supply and demand status, with values ​​of {TIGHT, NORMAL, LOOSE}, determined by power grid dispatch instructions or spot electricity price peak signals. As carbon emission constraint weights; As the basic weight for carbon emissions; This is the inversion gain factor, with a default value of 0.6. It is set to the minimum value to ensure that the executable rate after inversion is >80% and to avoid over-inversion. This is the boundary value between the safe zone and the transition zone.

[0143] Step S5: Obtain the execution result of the target scheduling strategy, and update the target scheduling strategy, the dynamic energy-carbon topology map, or the comprehensive confidence level based on the execution result, so as to achieve energy-carbon synergy among the core energy devices.

[0144] Step S4 yields the target scheduling strategy based on the current operating conditions, but whether the target scheduling strategy can be executed correctly and whether the target scheduling strategy is correct still needs to be verified and analyzed.

[0145] In some embodiments, such as Figure 5 As shown, step S5, "obtaining the execution result of the target scheduling strategy, and updating the target scheduling strategy, the dynamic energy-carbon topology map, or the comprehensive confidence level based on the execution result," includes: Step S51: Predict the prediction results of each of the core energy devices and the target energy carbon transfer system within a future preset time range according to the target scheduling strategy.

[0146] Step S52: Obtain the actual execution results generated by the target energy carbon transfer system when it executes according to the target scheduling strategy.

[0147] Step S53: Determine the execution deviation based on the prediction result and the actual execution result.

[0148] This application first integrates a high-fidelity virtual image of the physical world of the park with a deep integration of equipment-level models, energy-carbon coupling models, and target scheduling strategies. Using the target scheduling instructions as the strategy to be verified, real-time CFAU data streams as the initial values ​​of the system's current state, and the predicted renewable energy output and load data generated in enhanced mode for the next 1-4 hours as boundary conditions, the simulation is run in a twin to verify the safety, economy, and carbon reduction effect of the scheme (default safety margin 5%-10%, operating cost increase ≤2%, carbon emission reduction ≥3%, thresholds support parameterized configuration). If the simulation passes, the optimization instructions are marked as executable; if the simulation fails, if it is due to model parameter mismatch, the upstream model parameters are corrected and the simulation is re-verified; if the optimization strategy is infeasible, feedback is sent to S4 to adjust the optimization direction and re-solve until verification is successful.

[0149] Optimization commands that have passed simulation verification are sent to the physical equipment in the park to execute operations such as equipment start-up and shutdown, output adjustment, energy storage charging and discharging, and load transfer. CFAU data streams (including equipment status, energy data, and carbon emission data) during the actual execution process are collected simultaneously and stored together with the simulation expectation data in the twin database to form an "expected-actual" comparison dataset for subsequent intelligent diagnostics.

[0150] Step S54: When the execution deviation meets the third preset condition, determine the main source of deviation based on the execution deviation.

[0151] Then, based on the expected-actual dataset, it is determined whether the target scheduling strategy is executed correctly. If it is not executed correctly, the cause of the deviation is analyzed. In determining whether it is executed correctly, it is necessary to determine the execution deviation based on the expected-actual dataset. When the execution deviation exceeds a preset threshold, it is considered that the current execution deviation needs to be corrected. Before correction, it is necessary to determine the source of the deviation, or the root cause of the deviation.

[0152] In energy and carbon coordinated scheduling systems, the corresponding correction logic differs fundamentally depending on the stage at which deviations occur: if the deviation originates from sensor drift or control failure in a single device, incorrectly adjusting the topology learning parameters will lead to misjudgment of the topology structure; if the deviation originates from errors in topology path characterization, incorrectly tightening the verification threshold of a single device will overlook regional measurement risks; if the deviation originates from parameter mismatch in the optimization strategy, incorrectly correcting the device confidence level will lead to distorted data quality assessment. Traditional closed-loop corrections often employ a "one-size-fits-all" approach to global parameter adjustments, which is highly prone to causing system oscillations or even divergence. Therefore, it is essential to accurately identify the root cause of the deviation in order to achieve targeted correction actions. Combined with Lyapunov convergence design, this ensures that the closed-loop system achieves stable convergence based on operational experience.

[0153] Therefore, this application adopts a three-level decomposition architecture of equipment level, topology level, and system level, which perfectly matches the hierarchical characteristics of the energy and carbon transfer system: the equipment level corresponds to the execution deviation of a single device and is the smallest unit of deviation; the topology level corresponds to the characterization deviation of the energy and carbon transfer path and is the intermediate level connecting the device and the system; the system level corresponds to the systematic offset of the global optimization strategy and is the highest level of scheduling decision. The three-level architecture covers all possible sources of deviation without omission or overlap, and corresponds one-to-one with the comprehensive confidence level in step S2, the dynamic energy and carbon topology map in step S3, and the segment division of the optimized scheduling in step S4, enabling precise matching of deviation attribution and corrective actions. Compared with the two-level decomposition method that only distinguishes between "equipment deviation" and "non-equipment deviation", the three-level decomposition can further distinguish between topology path error and optimization strategy error, avoiding misalignment of corrective actions.

[0154] Specifically, this application calculates the equipment output deviation. and carbon emission deviation Perform a three-level decomposition: Equipment-level deviation The output deviation of a single device is attributed to equipment malfunction or control error, which is the overall confidence level in step S2. For device-level execution results, This is the prediction result at the device level; Topological deviation The deviation in the carbon flow path is attributed to the error in depicting the actual transport path by the dynamic carbon topology diagram in step S3. This is the execution result at the topology level. These are topology-level prediction results; System-level deviation The overall carbon emission deviation is attributed to the systematic shift of the S4 optimization strategy under uncertainty, i.e., the stability range, in which... For system-level execution results, This represents the system-level prediction results.

[0155] Identifying the source of dominant bias through variance decomposition: If If the variance contribution is >60%, then equipment-level bias dominates; if If the variance contribution is greater than 60%, then topological bias dominates; otherwise, system-level bias dominates.

[0156] Step S55: Determine the target to be adjusted and the adjustment strategy for the target to be adjusted based on the main sources of deviation.

[0157] Step S56: Adjust the target to be adjusted according to the adjustment strategy.

[0158] When the primary source of deviation is at the device level, the adjustment target is determined to be the overall confidence level, and the adjustment strategy is to adjust the verification threshold used to determine the overall confidence level. When the primary source of deviation is at the topology level, the adjustment target is determined to be the dynamic energy-carbon topology map, and the adjustment strategy is to adjust the update conditions of the dynamic energy-carbon topology map. When the primary source of deviation is at the system level, the adjustment target is determined to be the current adjustment strategy, and the adjustment strategy is the determination threshold for stability distribution.

[0159] Once the root cause of the deviation has been identified, the output of the corresponding step can be adjusted, and the adjustment strategy can be determined based on the dominant source of the deviation, such as... Figure 6 As shown, perform differential correction: • When equipment-level deviation dominates:

[0160] in, The device's attenuation correction coefficient, defaulting to 0.5, must satisfy the Lyapunov convergence condition. When the actual execution result of the device deviates from the S4 scheduling strategy, the device's confidence level is nonlinearly suppressed. The larger the value, the faster the confidence level drops and the stronger the correction force under the same deviation; at the same time, a "threshold tightening command" is sent to S2, which temporarily tightens the verification threshold for the consistency of the device's operating conditions, and the tightening magnitude is the same as... Proportional.

[0161] • When topological level deviation dominates:

[0162] in, To correct the attenuation coefficient of the topology, the Lyapunov convergence condition must be met; at the same time, an update speed adjustment instruction is sent to S3, which temporarily increases the learning rate of the GAT attention weights corresponding to this process chain, accelerating the adaptation of the topology to the actual path.

[0163] • When system-level bias dominates: Correct the interval boundaries in S4 The adaptive statistical window length is adjusted, and the window is shortened by 50% to improve the sensitivity of the interval boundaries to changes in operating conditions.

[0164] This hierarchical linkage accurately corrects the parameters of the corresponding links based on the source of deviation, avoiding system oscillations caused by "one-size-fits-all" corrections, and making the closed-loop feedback directional.

[0165] Here, the Lyapunov function of the closed-loop system is defined. The total deviation of the system from steady state is quantified, and the global convergence of the system is ensured by constraining and correcting the gain parameter.

[0166] + +

[0167] in, Characterizes the total error cost of the system's overall deviation from the optimal steady state. The larger the value, the more severe the deviation of the system from steady state; as long as The value continuously decreases monotonically and eventually approaches 0, indicating stable convergence of the system. , , These are the steady-state optimal values ​​of the comprehensive confidence level, energy transfer matrix, and effective penalty parameter, respectively, which are determined through offline training using historical data. These are the weight coefficients of the Lyapunov function, representing the relative importance of confidence convergence, topological convergence, and penalty parameter convergence, respectively, with default values ​​of 1.0, 0.5, and 0.3. Let L2 norm be the vector. Let f be the Frobenius norm of the matrix.

[0168] By designing a modified gain , satisfy:

[0169] It can be proven This ensures that the closed-loop system converges to a steady state based on operational experience, rather than oscillating continuously.

[0170] The collaborative method provided by this invention, through a closed-loop technical link in steps S1 to S5, solves industry problems such as unreliable data quality, static and rigid topology, fixed and rigid optimization parameters, and system oscillations caused by "one-size-fits-all" corrections in energy and carbon collaborative scheduling of industrial parks with high proportions of renewable energy access. Each step forms a self-regulating loop based on confidence levels, and is technically inseparable, generating an emergent effect where "the whole is greater than the sum of its parts." Specific beneficial effects are as follows: This invention abandons the traditional evaluation method of a single scalar confidence level, and introduces a new energy fluctuation-prediction joint metric factor λ(t) to quantify the time-varying uncertainty of wind and solar power output, effectively identifying the hidden risk of "high overall reliability but internal abnormal nodes". Simultaneously, it establishes a dual verification mechanism of energy conservation verification and operating condition consistency verification, and empirically corrects the initial confidence level from two dimensions: "balance of link-level quantities" and "reasonableness of node-level changes". The verification threshold has bidirectional adaptive adjustment capability (relaxing for sensor aging and tightening for topology feedback), eliminating the possibility of threshold oscillation from a mechanism perspective. Furthermore, through topology-confidence mutual feedback signals, data quality assessment is synchronized with topology evolution, avoiding the lag problem of "topology sensing anomalies but data verification not yet triggered".

[0171] This invention directly embeds the comprehensive confidence level as a multiplicative factor into the attention calculation of a Graph Attention Network (GAT). The feature vectors of low-confidence nodes are systematically compressed during multi-layer propagation (attention residue is less than 1% after 3 layers), fundamentally eliminating the interference of low-quality data on the carbon flow calculation of the entire system, without requiring additional data cleaning steps. Based on the energy transfer matrix A(t) and carbon flow transfer matrix T(t) learned from the GAT, dynamic coupling modeling of five-dimensional energy (electricity, heat, gas, hydrogen, and cooling) with carbon flow is achieved, accurately matching the dynamic changes in equipment efficiency with operating conditions. Simultaneously, through continuous calculation and mutual feedback mechanisms of node topological centrality, a self-regulating closed loop of "topology awareness - verification and adaptation - confidence evolution" is realized, enabling the topology to proactively adapt to sudden changes in new energy operating conditions.

[0172] This invention constructs a two-layer optimization model based on Stackelberg game theory. The upper layer uses global modulation confidence as the basis for system-level reliability and adjusts the objective function weights and constraint penalty parameters in real time through a nonlinear piecewise modulation function: prioritizing economic efficiency in the safe zone, reducing the economic weight to zero and nonlinearly increasing the penalty parameter to more than 10 times the final value of the transition zone in the risk zone, forcing the optimizer to prioritize the physically feasible solution of the energy and carbon constraints. Simultaneously, the spectral characteristics of the A(t) Laplace matrix are introduced for joint constraints to capture the "structural amplification effect" brought about by enhanced topological connectivity, making the penalty parameter simultaneously driven by both "data quality" and "structural sensitivity". Crucially, under specific operating conditions such as insufficient margin in total carbon emission constraints or tight grid supply and demand, a modulation direction reversal logic is designed. This uses reliable physical boundary signals to correct spurious constraints constructed from unreliable data, temporarily reducing the carbon cost weight rather than tightening the constraints. This avoids the possibility of no solution due to spurious data compressing the feasible region and ensures the rigidity of hard constraints by maintaining a high penalty parameter. This design of "soft constraints with concessions and hard penalties without leniency" ensures that the scheduling scheme maintains an execution rate of over 80% under extreme conditions.

[0173] This invention ensures security before command issuance through digital twin simulation verification and decomposes deviations into three levels—device-level, topology-level, and system-level—after execution. Variance decomposition (>60% variance contribution rate) accurately identifies the dominant deviation source, avoiding system oscillations caused by traditional "one-size-fits-all" corrections. Differentiated corrections are applied to different deviation sources: when device-level deviations dominate, the confidence baseline of the corresponding device is corrected and the verification threshold is tightened; when topology-level deviations dominate, the GAT learning rate is increased to accelerate topology adaptation; when system-level deviations dominate, the statistical window of the modulation function is shortened to improve sensitivity. Through Lyapunov function design, the correction gain is constrained to meet convergence conditions, theoretically guaranteeing global convergence of the closed-loop system rather than continuous oscillations.

[0174] Without S2, S3-S5 lack structured confidence input, and the mutual feedback mechanism of S2 requires the topological centrality of S3 to function; without S4, the constraint space and spectral joint constraints of S5 cannot be generated, and the mutual feedback of S2 has no signal source; without S4, the execution deviation of S5 cannot be generated, and the topology learning results of S3 have nowhere to be applied; without S5, the structured confidence cannot evolve from operational experience, and the system loses its adaptive capability. These four steps, with confidence as the sole structural endogenous variable, run through the entire chain, forming a complete inverse closed loop of "physical execution → deviation decomposition → hierarchical correction → confidence update → topology relearning → optimization re-solution → physical re-execution," enabling the park's energy and carbon system to self-evolve from operational experience and achieving refined, intelligent, and collaborative management of energy consumption and carbon emissions.

[0175] Example 2: The following simulation, built using typical daily operating data from an industrial park, illustrates the complete operation of this method under a scenario of a sudden drop in photovoltaic output. Since some steps have no significant effect in the stable phase, the four stages T0-T4 below mainly explain and calculate the core content, omitting processes that do not contribute to the actual operation of these stages. The park has 30MW of photovoltaic capacity, 20MWh of energy storage, 15MW of electrolyzers, 5MW of CCUS, and 25MW of various loads. The simulation is based on a Python 3.10 + PyTorch 2.0 environment. The park model includes 28 core energy devices (8 photovoltaic units, 4 wind power units, 2 energy storage units, 2 electrolyzers, 2 CCHP units, 2 gas boilers, 1 P2G unit, 1 CCUS unit, and 6 load nodes), with new energy capacity accounting for 42%.

[0176] Numerical comparison under different operating conditions (simulation example):

[0177] • At time t0 (sunny day steady state): Photovoltaic output is 100%. = 0.998, The static scores of each node within the process chain are stable, and the photovoltaic inverter... Electrolytic cell Energy storage CCUS mean Internal consistency measure The level is extremely low, indicating a high degree of consistency in data quality within the process chain. GAT learning... Attention weight for photovoltaic cells → electrolyzer: 0.85. Landing in the safe zone The optimization paradigm seeks economic optimality. The system is in an ideal steady state, all confidence indicators are optimal, and the scheduling plan is issued and executed normally.

[0178] • At time t1 (PV power drops sharply to 30%): Rapid cloud movement causes PV output to drop from 30MW to 9MW. =0.80, , = 0.565, The photovoltaic nodes within the process chain experience sampling jitter due to sensor fluctuations during sudden drops. It dropped to 0.78. The value increased approximately 8 times compared to time t0, triggering a process chain-level quality warning. GAT was updated... Upon relearning, the photovoltaic → electrolyzer attention plummeted to 0.08 (photovoltaic features were compressed to approximately 38% before being included in the calculation), while the electrolyzer → energy storage attention rose to 0.71 (carbon flow paths were automatically reconfigured to energy storage for power supply). After topological differentiation weighting, it falls into the transition region. , Modulated to approximately 2.1 At this time, S2 outputs Largest eigenvalue of the Laplace matrix (benchmark value) = 6.0 (due to a sudden drop in photovoltaic power leading to localized network sparsity, but with enhanced energy storage → electrolyzer link, overall connectivity remains higher than steady state), spectral sensitivity coefficient. Joint constraints based on topological spectrum: Topological spectrum constraints increase the effective penalty parameter by an additional 36%, ensuring that even if the confidence of a single node is in the transition zone, the risk of disturbance propagation caused by changes in topological connectivity is still sufficiently suppressed. The optimization paradigm shifts to a compromise-based approach.

[0179] • At time t2 (execution deviation triggers S5 correction): After the scheduling scheme generated in S4 is executed, the following occurs. Output deviation (actual adjustment rate of the electrolyzer is lower than expected). S4 calculates three levels of deviation: equipment level contributes 55%, topology level 35%, and system level 10%, meaning equipment level deviation is dominant. Correction is then performed. After correction Triggering end-to-end recalculation, S2 GAT Relearning further weakens the photovoltaic → electrolyzer link; S3 detection Falling into the risk zone It jumped to more than an order of magnitude above the end of the transition zone. (16.7 times), the optimization paradigm structurally shifted to prioritize safety constraints, abandoning economic optimization on low-quality data and only satisfying hard safety constraints. The executability of the revised solution recovered to 95%+ (compared to 61% for the t1-t3 period of the open-loop solution without closed loop).

[0180] Simultaneously triggering the S2→S1 topology-confidence mutual feedback: S3 calculates the topology centrality of the photovoltaic inverter during the relearning process. The value dropped from 0.72 to 0.35 over three consecutive windows (a decrease of 51%), exceeding the threshold. This triggers a topology-confidence mutual feedback signal to S1. In the next verification cycle, S1 temporarily tightens the photovoltaic inverter's operating condition consistency verification threshold from 3% to 2.4% (a 20% tightening), and increases the energy conservation verification weight from 1.0 to 1.3 (a 30% increase). After tightening for six consecutive windows, this verification mode captures a real power sensor drift (drift amplitude of 4.2%, undetectable at the original threshold of 3%). After correction, the topology centrality recovers to 0.68, and the mutual feedback is released. This mutual feedback process synchronizes data quality assessment with topology evolution, avoiding the lag where "the topology has detected an anomaly, but data verification has not yet been triggered."

[0181] • At time t3 (carbon quota constraint competition, triggering a reversal of direction): the remaining margin of the park's total carbon emission constraint drops to 3%, below the critical value. . 25 remains in the risk zone. Although conventional modulation should increase the weight of carbon emission constraints, the ternary joint decision-making triggers a reversal in direction. The efficiency was reduced from 0.92 to 0.28, relaxing the overly stringent carbon emission constraints distorted by low-quality data. Before the reversal, the optimizer had no feasible solutions (the distorted constraints made the feasible region empty); after the reversal, the feasibility of the solution recovered to 88%.

[0182]

[0183] • At time t4 (PV power restored to 90%): The clouds dissipate and PV output recovers to 27MW. =0.60, (Prediction has been restored) = 0.895, GAT learning restored normal connection strength (photovoltaic → electrolyzer attention recovered to 0.83). Recovered to the safe zone Restore to The recovery of photovoltaic power has led to a rebound in confidence across the entire value chain, and the system has smoothly switched from a safety-constraint-first mode back to an economic optimization mode.

[0184] End-to-end comparison summary table at each time point

[0185] End-to-end performance comparison (cumulative across all time periods from t1 to t4)

[0186] This embodiment demonstrates the end-to-end adaptive process from steady state at t0 → disturbance at t1 → closed-loop correction at t2 → constraint competition at t3 → recovery at t4. Confidence level, as the only structural endogenous variable, drives the real-time linkage of topology learning, optimized modulation, and closed-loop correction, enabling the system to maintain a 95%+ feasibility rate even under extreme conditions, while keeping carbon emission deviation below 4.2%.

[0187] Example 3: Thirdly, this application provides an electronic device, such as Figure 7 As shown, it includes: at least one processor 100; and a memory 200 communicatively connected to the at least one processor 100; wherein the memory 200 stores instructions executable by the at least one processor 100, the instructions being executed by the at least one processor 100 to enable the at least one processor 100 to execute an energy-carbon synergy method for an energy-carbon transfer system according to the above embodiments.

[0188] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0189] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0190] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).

[0191] The processor can communicate with external devices via the I / O bus through wired or wireless networks.

[0192] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0193] Example 4: Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0194] Computer-readable storage media can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Computer-readable storage media may include, but are not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, and computer storage media (e.g., hard disk, floppy disk, solid-state drive, removable disk, CD-ROM, DVD-ROM, Blu-ray disc, etc.).

[0195] Computer-readable storage media may also store at least one computer-executable program / instruction, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0196] Example 5: Fifthly, this application proposes a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method described in any of the first aspects.

[0197] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0198] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0199] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0200] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0201] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0202] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a controller to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0203] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein.

[0204] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for energy-carbon synergy in an energy-carbon transfer system, characterized in that, include: Acquire energy and carbon data and equipment data collected by each core energy device in the target energy and carbon transfer system; Based on the equipment data and the energy carbon data, determine the current energy carbon activity data of each of the core energy devices, and the comprehensive confidence level of the current energy carbon activity data; Based on the current energy and carbon activity data and the comprehensive confidence level, a dynamic energy and carbon topology diagram is constructed among the core energy devices in the target energy and carbon circulation system. Based on the dynamic energy carbon topology diagram and the comprehensive confidence level, a target scheduling strategy that conforms to the current operating conditions of each core energy device is determined. The execution result of the target scheduling strategy is obtained, and the target scheduling strategy, the dynamic energy and carbon topology map, or the comprehensive confidence level is updated based on the execution result, so as to achieve energy and carbon synergy among the core energy devices.

2. The method according to claim 1, characterized in that, The step of determining the current energy and carbon activity data of each of the core energy devices based on the device data and the energy and carbon data, and the comprehensive confidence level of the current energy and carbon activity data, includes: The energy and carbon data and the equipment data are processed according to preset data processing rules to obtain the current energy and carbon activity data of each of the core energy devices; A time-varying attenuation factor is calculated based on the current energy and carbon activity data of the new energy equipment in the target energy and carbon transfer system. The time-varying attenuation factor is used to characterize the output uncertainty of the new energy equipment. Calculate a static score for each of the core energy devices based on their current energy and carbon activity data. Calculate the initial confidence level of each of the current energy and carbon activity data based on the static score and the time-varying decay factor; The current energy and carbon activity data is verified based on the current verification threshold at the current moment to obtain the verification result; Based on the verification results, a correction coefficient for the initial confidence level is determined; The initial confidence level is corrected based on the correction coefficient to obtain the overall confidence level of the current energy and carbon activity data.

3. The method according to claim 2, characterized in that, The construction of a dynamic energy-carbon topology map among the core energy devices in the target energy-carbon circulation system based on the current energy-carbon activity data and the comprehensive confidence level includes: Construct an energy-carbon transfer topology diagram based on the target energy-carbon transfer system; The current equipment efficiency and current carbon flow intensity of each core energy device are determined based on the current energy and carbon activity data of the target energy carbon transfer system and each core energy device. The node characteristics of each node in the energy and carbon flow topology are determined based on the current equipment efficiency, the current carbon flow intensity, and the current energy and carbon activity data. The dynamic attention weights of each edge in the energy-carbon transfer topology graph are determined based on the comprehensive confidence level. The dynamic energy-carbon topology graph is constructed based on the node characteristics of each node, the dynamic attention weights of each edge, and the energy-carbon flow topology graph.

4. The method according to claim 3, characterized in that, The method further includes: When the first preset condition is met, obtain the latest comprehensive confidence level and current energy and carbon activity data of each core energy device; The dynamic energy and carbon topology map is updated based on the latest comprehensive confidence level and the current energy and carbon activity data; After each update, the node centrality of each node in the dynamic energy carbon topology graph is calculated; The current verification threshold is updated when the change in the node centrality of each node satisfies the second preset condition.

5. The method according to claim 1, characterized in that, The step of determining the target scheduling strategy that conforms to the current operating conditions of each core energy device based on the dynamic energy carbon topology map and the comprehensive confidence level includes: The global modulation confidence level is calculated based on the comprehensive confidence level and the dynamic energy-carbon topology map. The global modulation confidence level is used to characterize the confidence level of all energy-carbon activity data in the target energy-carbon flow system. The current stability distribution of the target energy carbon transfer system is determined based on the global modulation confidence level. The weighting coefficients of each parameter and the penalty values ​​of the penalty parameters are determined based on the stability distribution and the dynamic energy-carbon topology diagram. The target scheduling policy is obtained by adjusting the current scheduling policy based on the weighting coefficient and the penalty value.

6. The method according to claim 1, characterized in that, The step of obtaining the execution result of the target scheduling strategy and updating the target scheduling strategy, the dynamic energy-carbon topology map, or the comprehensive confidence level based on the execution result includes: Based on the target scheduling strategy, predict the prediction results of each of the core energy devices and the target energy carbon transfer system within a future preset time range; Obtain the actual execution results generated by the target energy carbon transfer system when it executes according to the target scheduling strategy; The execution deviation is determined based on the prediction results and the actual execution results; When the execution deviation meets the third preset condition, the main source of the deviation is determined based on the execution deviation; Based on the main sources of deviation, determine the target to be adjusted and the adjustment strategy for the target to be adjusted; The target to be adjusted is adjusted according to the adjustment strategy.

7. The method according to claim 6, characterized in that, The step of determining the target to be adjusted and the adjustment strategy for the target to be adjusted based on the main sources of deviation includes: When the main source of deviation is a device-level source, the adjustment target is determined to be the overall confidence level, and the adjustment strategy is determined to be to adjust the verification threshold when determining the overall confidence level. When the main source of deviation is a topology-level source, the adjustment target is determined to be the dynamic energy-carbon topology map, and the adjustment strategy is determined to be to adjust the update conditions of the dynamic energy-carbon topology map; When the main source of deviation is a system-level source, the adjustment target is determined as the current adjustment strategy, and the adjustment strategy is determined as the threshold for determining the stability distribution.

8. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

Images