Demand side resource and energy storage coordinated regulation and control method and system based on carbon index constraint

Abstract

The invention discloses a demand side resource and energy storage coordinated regulation and control method and system based on carbon index constraint. The method comprises the steps of collecting real-time load data, renewable energy output data and energy storage equipment state data of a regional power grid and a dynamic carbon emission intensity signal issued by a superior power grid; based on the dynamic carbon emission intensity signal, a load side carbon emission intensity spectrum with temporal-spatial resolution is generated through a source-load-storage full-link carbon flow tracking algorithm; according to the load side carbon emission intensity spectrum, a virtual carbon pool with the carbon cost as a weight factor is established, and a flexible load regulation and control priority sequence is generated; and in combination with the flexible load regulation priority sequence and the real-time charging and discharging boundary of the energy storage system, solving a multi-time-scale collaborative optimization model and generating a source-network-load-storage integrated regulation strategy by taking the condition that the carbon intensity of the system is not out of limit as a core constraint. According to the embodiment of the invention, fine metering and active regulation and control of load side carbon emission can be realized, and the response capability of the system to dynamic carbon constraint is improved.

Description

Technical Field

[0001] This invention belongs to the field of energy storage technology, and in particular to a method and system for coordinated regulation of demand-side resources and energy storage based on carbon index constraints. Background Technology

[0002] With the deepening of the global energy transition, the integration of high proportions of renewable energy into the grid poses severe challenges to the real-time balance and coordinated management of carbon emission reduction in power systems. Existing grid control methods primarily focus on maintaining power balance and economic dispatch, neglecting the coordinated utilization of demand-side resources and energy storage, and generally lacking refined carbon management tools that span the entire power generation-grid-load-storage chain. Traditional approaches typically treat carbon emissions as a system-wide or generation-side indicator, making it difficult to accurately track and regulate the spatiotemporal carbon responsibility of the load side. This results in the demand-side regulation potential not being fully activated to serve the low-carbon operation of the grid. Furthermore, facing dynamically changing carbon emission intensity and renewable energy fluctuations, current methods struggle to achieve real-time, accurate, and coordinated control of carbon emission intensity while ensuring system security. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for coordinated regulation of demand-side resources and energy storage based on carbon index constraints, in order to overcome the shortcomings of the prior art, realize the precise measurement and active regulation of load-side carbon emissions, and improve the system's response capability to dynamic carbon constraints.

[0004] One embodiment of this application provides a method for coordinated regulation of demand-side resources and energy storage based on carbon index constraints, the method comprising: Collect real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals issued by the upper-level power grid in the region; Based on the dynamic carbon emission intensity signal, a load-side carbon emission intensity spectrum with spatiotemporal resolution is generated by a source-load-storage full-process carbon flow tracking algorithm. Based on the load-side carbon emission intensity spectrum and combined with the demand-side resource adjustability potential model, a virtual carbon pool with carbon cost as the weighting factor is established and a flexible load regulation priority sequence is generated. By combining the aforementioned flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, and taking the system carbon intensity not exceeding the limit as the core constraint, a multi-timescale collaborative optimization model is solved and an integrated source-grid-load-storage regulation strategy is generated to achieve real-time controllability of carbon emission intensity and collaborative optimization of system operation economy.

[0005] Optionally, the data collected includes real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals from the upper-level power grid. The system collects power data of load nodes in real time by deploying IoT sensors at various nodes of the regional power grid, obtains real-time output data of renewable energy power plants through the energy management system, and collects state of charge and charging / discharging power data of energy storage devices through the battery management system, generating a raw heterogeneous dataset. The original heterogeneous dataset is timestamped and cleaned to unify data with different sampling frequencies to the same time base, and bad data caused by abnormal jump points and communication interruptions is removed to generate a cleaned and aligned time series dataset. The system receives dynamic carbon emission intensity signals from the upper-level power grid through the power dispatch data network. These signals are updated at a minute-level frequency and are associated with the regional power grid gateway metering points to generate dynamic carbon emission intensity time-series signals. The cleaned and aligned time-series dataset is fused with the dynamic carbon emission intensity time-series signal and encapsulated according to a standardized data model to generate a standardized multi-source dataset for subsequent carbon flow tracing analysis.

[0006] Optionally, the step of generating a load-side carbon emission intensity spectrum with spatiotemporal resolution based on the dynamic carbon emission intensity signal using a source-load-storage full-process carbon flow tracing algorithm includes: The dynamic carbon emission intensity time series signal in the standardized multi-source dataset is analyzed, and a carbon flow network topology is established with the regional power grid gateway as the starting point and the load node as the ending point, generating a carbon flow network model. Based on the carbon flow network model and real-time power flow data, and using the proportional sharing principle, the distribution of electrical carbon flow from the generation side to the load side is calculated, and a node carbon flow intensity matrix is ​​generated. Establish a carbon state transition model for energy storage devices, track the carbon emissions corresponding to the electrical energy absorbed during the charging period of energy storage, and perform carbon flow tracing and allocation during the discharging period to generate an energy storage carbon flow distribution matrix; By combining the carbon flow intensity matrix of the integrated nodes and the carbon flow distribution matrix of the energy storage, a mixed-integer linear programming algorithm is used to accurately allocate carbon emissions to each load node and time segment, ultimately generating a load-side carbon emission intensity spectrum with spatiotemporal resolution.

[0007] Optionally, the step of establishing a virtual carbon pool with carbon cost as a weighting factor and generating a flexible load regulation priority sequence based on the load-side carbon emission intensity spectrum and the demand-side resource adjustability potential model includes: Analyze the carbon emission intensity spectrum on the load side, calculate the marginal carbon emission intensity of each flexible load node at different time sections, and convert it into carbon cost per unit of electricity to generate a carbon cost vector. The demand-side resource adjustability potential model is invoked. Based on historical operating data and load characteristics, this model quantifies the power reduction, transferable electricity, and response time constant of various flexible loads, and generates an adjustability potential mapping table. Using the carbon cost vector as the core weighting factor and combining it with the adjustable potential mapping table, a pricing and settlement mechanism for the virtual carbon pool is designed, and a dynamic carbon price curve that correlates carbon cost with adjustment value is established. Based on the dynamic carbon price curve and the adjustment costs of each flexible load, a priority sequence for flexible load regulation is generated, which is optimally ranked in terms of both carbon emission reduction benefits and economic efficiency.

[0008] Optionally, the step of combining the flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, taking the system carbon intensity not exceeding the limit as the core constraint, solving the multi-timescale collaborative optimization model and generating an integrated source-grid-load-storage regulation strategy, to achieve real-time controllable carbon emission intensity and collaborative optimization of system operation economy, includes: By integrating the priority sequence of flexible load regulation, the real-time charging and discharging power boundary of the energy storage system, and the renewable energy output prediction curve, a collaborative optimization model framework with three time scales—day-ahead, intraday, and real-time—is constructed, generating a multi-time-scale optimization model parameter set. In the optimization model, the real-time carbon intensity at the interface between the regional power grid and the upper-level power grid does not exceed a preset threshold as the core constraint, and energy storage operation constraints and load regulation constraints are embedded to generate an optimization problem with carbon constraints. A distributed robust optimization algorithm is used to solve the carbon-constrained optimization problem. The optimal adjustment amount of each flexible load, the optimal charging and discharging plan of the energy storage system, and the renewable energy consumption scheme are calculated under multiple time sections in the future, and a preliminary coordinated control strategy is generated. The initial coordinated control strategy is verified and dynamically adjusted to form the final executable integrated source-grid-load-storage control instruction set, achieving the coordinated goal of full controllable carbon emission intensity and optimal total system operating cost.

[0009] Another embodiment of this application provides a demand-side resource and energy storage coordinated regulation system based on carbon index constraints, the system comprising: The data acquisition module is used to collect real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals issued by the upper-level power grid. The generation module is used to generate a load-side carbon emission intensity spectrum with spatiotemporal resolution based on the dynamic carbon emission intensity signal and through a source-load-storage full-process carbon flow tracing algorithm. A module is established to create a virtual carbon pool with carbon cost as a weighting factor and generate a flexible load regulation priority sequence based on the load-side carbon emission intensity spectrum and the demand-side resource adjustability potential model. The control module is used to combine the flexible load control priority sequence with the real-time charge and discharge boundary of the energy storage system, with the system carbon intensity not exceeding the limit as the core constraint, to solve the multi-timescale collaborative optimization model and generate an integrated source-grid-load-storage control strategy, so as to achieve real-time controllability of carbon emission intensity and collaborative optimization of system operation economy.

[0010] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0011] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0012] Compared with existing technologies, this invention provides a demand-side resource and energy storage coordinated regulation method based on carbon index constraints. It collects real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals from the upper-level power grid. Based on the dynamic carbon emission intensity signals, a load-side carbon emission intensity spectrum with spatiotemporal resolution is generated through a source-load-storage full-process carbon flow tracking algorithm. According to the load-side carbon emission intensity spectrum, a virtual carbon pool with carbon cost as a weighting factor is established, and a flexible load regulation priority sequence is generated. Combining the flexible load regulation priority sequence with the real-time charge-discharge boundary of the energy storage system, and taking the system carbon intensity not exceeding the limit as the core constraint, a multi-timescale collaborative optimization model is solved, and an integrated source-grid-load-storage regulation strategy is generated. This enables refined measurement and proactive regulation of load-side carbon emissions, improving the system's response capability to dynamic carbon constraints. Attached Figure Description

[0013] Figure 1 A hardware structure block diagram of a computer terminal for a method of coordinated regulation of demand-side resources and energy storage based on carbon index constraints, provided in an embodiment of the present invention; Figure 2 A flowchart illustrating a method for coordinated regulation of demand-side resources and energy storage based on carbon index constraints, provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a demand-side resource and energy storage coordinated regulation system based on carbon index constraints, provided as an embodiment of the present invention. Detailed Implementation

[0014] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0015] The present invention first provides a method for coordinated regulation of demand-side resources and energy storage based on carbon index constraints. This method can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0016] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for a method of coordinated regulation of demand-side resources and energy storage based on carbon index constraints, provided as an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. This computer program includes program instructions that, when executed, enable the processor to perform any method for the coordinated regulation of demand-side resources and energy storage based on carbon index constraints.

[0018] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0019] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any method of demand-side resource and energy storage coordinated regulation based on carbon index constraints.

[0020] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0021] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0022] See Figure 2 The present invention provides a method for coordinated regulation of demand-side resources and energy storage based on carbon index constraints, which may include the following steps: S201 collects real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals issued by the upper-level power grid. Specifically, power data of load nodes can be collected in real time through IoT sensors deployed at various nodes of the regional power grid, real-time output data of renewable energy power plants can be obtained through the energy management system, and the state of charge and charging / discharging power data of energy storage devices can be collected through the battery management system to generate raw heterogeneous datasets. This step is the foundational data collection stage for coordinated regulation. Its core is to achieve comprehensive perception of all elements of the regional power grid, including "source-load-storage." It involves constructing a raw data foundation covering power flow and carbon flow analysis through multi-channel data collection. The specific implementation method is as follows: The deployment of IoT sensors follows the principle of "full node coverage + key load encryption." A total of 120 power sensors are deployed at the outgoing lines of 110kV substations, branch nodes of 35kV distribution networks, and 10kV user access points in the regional power grid. The sensors utilize the Hall effect principle, with a measurement range of 0-10000kW, a measurement accuracy of ±0.5%FS, and a sampling frequency set to 10Hz (to meet real-time load tracking requirements). Load node power data is divided into active power P_load and reactive power Q_load. The data format is "node number-timestamp-active power-reactive power-data status bit," where the node number is encoded as "substation number-outgoing line number-load number" (e.g., S1-01-L05), the timestamp uses UTC format (accurate to milliseconds), and the data status bit is 0 for normal operation and 1 for abnormal data acquisition. For example, the collected data for a certain load node is “S1-01-L05-2025-10-10T08:30:00.123-4560kW-1280kvar-0”, which reflects the real-time power consumption status of the node at the corresponding time.

[0023] Real-time output data from renewable energy power plants is acquired through the regional grid energy management system (EMS). This system communicates with the monitoring systems of eight photovoltaic power plants and five wind power plants within the jurisdiction via the IEC 61850 protocol, with a data update frequency of 5Hz. The collected data includes active power output (P_pv), irradiance, and module temperature for photovoltaic power plants, and active power output (P_wind), wind speed, and turbine speed for wind power plants. The measurement accuracy of the output data is ±1%FS, and the irradiance measurement range is 0-2000W / m². 2 The wind speed measurement range is 0-70 m / s. For example, the output data of a photovoltaic power station is "PV-03-2025-10-10T08:30:00.000-3200kW-850W / m".2 The data for a certain wind power station, "-42℃-0", is "WT-02-2025-10-10T08:30:00.000-2800kW-6.5m / s-1200rpm-0". These data directly reflect the real-time power supply capacity of renewable energy.

[0024] Data acquisition for the energy storage devices is achieved through a Battery Management System (BMS). Six energy storage power stations (total capacity 50MWh) are deployed within the regional power grid. Each station's BMS collects real-time data on parameters such as the state of charge (SOC), charge / discharge power (P_ess), individual cell voltage, and module temperature of the battery clusters, with a sampling frequency of 5Hz. The SOC measurement range is 0-100% with an accuracy of ±2%, and the charge / discharge power measurement range is -5000kW to 5000kW (negative values ​​indicate charging, positive values ​​indicate discharging). The data format is "Energy Storage Station Number-Battery Cluster Number-Timestamp-SOC-Charge / Discharge Power-Average Temperature". For example, data from a certain energy storage station might be "ESS-01-C02-2025-10-10T08:30:00.246-68%-1800kW-32℃", indicating that the battery cluster is currently charging, with an SOC of 68%, a charging power of 1800kW, and normal operating temperature.

[0025] The original heterogeneous dataset is a collection of the three types of data mentioned above. It exhibits heterogeneous characteristics due to differences in data source, format, and sampling frequency. The load data is sampled at 10Hz, renewable energy data at 5Hz, and energy storage data at 5Hz, with a total data volume of approximately 1.2GB per hour. The dataset is stored using a distributed file system and managed by partitioning according to "date-data type," such as "20251010-Load Data" and "20251010-Renewable Energy Data," ensuring data traceability and access efficiency, and providing complete raw input for subsequent preprocessing stages.

[0026] The original heterogeneous dataset is timestamped and cleaned to unify data with different sampling frequencies to the same time base and remove bad data caused by abnormal jump points and communication interruptions, generating a cleaned and aligned time series dataset. This step is crucial for improving data quality. Through time synchronization and anomaly handling, noise and biases in the raw data are eliminated, transforming heterogeneous data into time-series data in a unified format, providing high-quality input for subsequent carbon flow tracing. The specific implementation method is as follows: Timestamp alignment employs a "Precise Time Protocol (PTP) + interpolation completion" scheme, using the clock of the regional power grid dispatch center as the reference clock (1μs accuracy). Timestamps from all acquisition devices are calibrated to this reference, with calibration errors controlled within ±1ms. For data with different sampling frequencies, the time reference is uniformly set to 10Hz (consistent with the load data sampling frequency), and linear interpolation is used to upsample low-frequency data. For example, if the sampling frequency of energy storage data is 5Hz, and a time sequence is "08:30:00.000-68%-1800kW" and "08:30:00.200-68.2%-1850kW", then at 08:30:00.100, interpolation will generate "08:30:00.100-68.1%-1825kW", ensuring that the time interval for all data is 100ms. The timestamp format is uniformly "YYYY-MM-DDHH:MM:SS.sss", for example, the calibrated timestamp is "2025-10-10 08:30:00.100", to ensure the time consistency of the data.

[0027] Data cleaning primarily targets abnormal data jumps and bad data caused by communication interruptions, employing a dual detection mechanism of "threshold judgment + local trend analysis." The threshold for abnormal data jumps is determined based on historical data statistics. The active power jump threshold is set at 50% / second (i.e., a jump is defined as a power change rate exceeding 50% between two adjacent time steps), and the SOC jump threshold is set at 5% / second. For example, if the power of a load node abruptly changes from 4560kW to 9200kW (a change rate of 101.7% / second), it is considered an abnormal jump. For jump points, a sliding window median filtering method is used for correction. The window length is set to 5 time steps, and the median of the data within the window is used to replace the outlier. For example, if the sequence before and after the outlier data is 4560kW, 4620kW, 9200kW, 4750kW, 4680kW, and the median of the window is 4680kW, then the outlier is corrected to 4680kW.

[0028] Bad data caused by communication interruption (data status bit is 1) is handled according to the interruption duration: When the interruption duration is ≤ 3 time steps (300ms), linear interpolation is used for repair, with the formula x_restore=x_prev+(x_next-x_prev)×(t-t_prev) / (t_next-t_prev), where x_prev and x_next are the normal data before and after the interruption, and t is the timestamp of the bad data; when the interruption duration is > 3 time steps, historical similar operating condition data from the same period of the previous 24 hours of the node is used for replacement. For example, if a load node experiences a communication interruption from 08:30:00.300 to 08:30:00.600, the power data of 4580kW, 4610kW, 4590kW, and 4630kW from the previous day from 08:30:00.300 to 08:30:00.600 are used for replacement.

[0029] The cleaned and aligned time-series dataset adopts a "timestamp-global state vector" format, with each timestamp corresponding to a global state vector. The vector dimension is 3×N_load + 2×N_re + 3×N_ess (N_load is the number of load nodes, N_re is the number of renewable energy power plants, and N_ess is the number of energy storage battery clusters). It contains core parameters such as the active / reactive power of all loads, the active power output of renewable energy, and the SOC / charge / discharge power / temperature of energy storage. For example, in the global state vector of a certain timestamp, the parameters of load S1-01-L05 are "4680kW-1290kvar", the parameters of photovoltaic PV-03 are "3250kW", and the parameters of energy storage ESS-01-C02 are "68.1%-1825kW-32℃". All data are stored continuously in chronological order to form a well-organized time-series dataset.

[0030] The system receives dynamic carbon emission intensity signals from the upper-level power grid through the power dispatch data network. These signals are updated at a minute-level frequency and are associated with the regional power grid gateway metering points to generate dynamic carbon emission intensity time-series signals. This step is crucial for obtaining the basis for carbon constraints. By receiving dynamic carbon emission intensity signals from the upstream power grid, a correlation between carbon intensity and the regional power grid is established, providing core input for subsequent carbon flow tracking. The specific implementation method is as follows: The power dispatch data network adopts a dual-link redundancy architecture (primary link + backup link), uses TCP / IP as the communication protocol, has a transmission rate of 1Gbps, and an end-to-end latency of ≤50ms to ensure the real-time and reliable transmission of dynamic carbon emission intensity signals. Signals issued by the upper-level power grid are generated based on carbon emission monitoring data from the provincial power market, reflecting the overall carbon emission level of the current power grid. The signal update frequency is 1 minute / time (updated every 60 seconds), and the signal format is "timestamp-carbon emission intensity value-confidence level-update cycle". The unit of the carbon emission intensity value is kgCO2 / kWh (the amount of carbon dioxide emitted per kilowatt-hour of electricity), with a value range of 0.2-0.8 kgCO2 / kWh, and a confidence level ≥90% (reflecting data reliability). For example, a signal "2025-10-10 08:30:00-0.52kgCO2 / kWh-95%-60s" indicates that the dynamic carbon emission intensity of the upstream power grid at that moment is 0.52kgCO2 / kWh, the data confidence level is 95%, and the next update time is 08:31:00.

[0031] The association between dynamic carbon emission intensity signals and regional power grid gateway metering points adopts a "geographical zoning + power supply range" mapping rule. The regional power grid has three upstream power grid gateway metering points (G1, G2, and G3), each corresponding to a different power supply area: G1 covers the western industrial load area, G2 covers the central commercial and residential load area, and G3 covers the eastern renewable energy consumption area. The real-time power supply range and load distribution of each gateway are queried through the EMS system to establish a correlation table between signals and gateways. For example, when a received signal is labeled "Coverage Area: G1, G2", this signal only applies to the load nodes corresponding to G1 and G2, while the nodes corresponding to G3 still use the previously received signal (if not updated). During the correlation process, if the signal is not updated in time for a certain period (exceeding the update cycle by 10 seconds), the current carbon intensity value is predicted by linear extrapolation. The prediction formula is e_pred=e_prev+(e_prev-e_prev2)×Δt / T, where e_prev is the previous signal value, e_prev2 is the signal value two years before last, Δt is the time difference between the current time and the previous update, and T is the update cycle (60 seconds), ensuring the continuity of the carbon intensity signal.

[0032] The generated dynamic carbon emission intensity time-series signal adopts the format of "timestamp-gate number-carbon emission intensity-applicable scope" and is interpolated and extended according to a 10Hz time base (to keep synchronized with the cleaned time-series dataset). For example, the signal value at 08:30:00 is 0.52kgCO2 / kWh, and the interpolated signal from 08:30:00.100 to 08:30:00.900 is also 0.52kgCO2 / kWh, until it is updated to a new signal value at 08:31:00. The time-series signal is completely aligned with the timestamps of the cleaned load, renewable energy, and energy storage data, ensuring that the electricity data at each time step can match the corresponding carbon intensity data during subsequent carbon flow tracking.

[0033] The cleaned and aligned time-series dataset is fused with the dynamic carbon emission intensity time-series signal and encapsulated according to a standardized data model to generate a standardized multi-source dataset for subsequent carbon flow tracing analysis.

[0034] This step is the final stage of data integration. Through data fusion and standardized encapsulation, the scattered electricity data and carbon intensity data are integrated into a dataset with a unified format, providing directly usable input for carbon flow tracing algorithms. The specific implementation method is as follows: Data fusion employs a "timestamp primary key association + field supplementation" approach, using a 10Hz timestamp as the unique primary key to associate the cleaned and aligned time-series dataset (containing source-load-storage power parameters) with the dynamic carbon emission intensity time-series signal (containing carbon intensity parameters). Each timestamp corresponds to a fused data record, which includes a "power parameter subset + carbon intensity parameter subset." The power parameter subset contains 286 fields, including P_load / Q_load for load nodes, P_pv / P_wind for renewable energy power plants, and SOC / P_ess / temperature for energy storage devices. The carbon intensity parameter subset contains three fields: carbon emission intensity value, confidence level, and applicable scope at each threshold. The total number of fields in the fused data record is 289, ensuring that all parameters related to carbon flow tracking are included.

[0035] The standardized data model is an extension of the Common Information Model (CIM) of the IEC 61970 standard. It defines data naming conventions, data types, units, and value ranges. For example, the field for active power load is named "Load_P_Node Number," with a floating-point data type, unit of kW, and a value range of 0-10000kW; the field for carbon emission intensity is named "Carbon_Intensity_Border Number," with a floating-point data type, unit of kgCO2 / kWh, and a value range of 0-1.0 kgCO2 / kWh. The model also defines data quality standards, requiring a missing rate of ≤0.1% for all numerical data and an outlier correction rate of ≥99% to ensure data availability.

[0036] The encapsulation process uses JSON format to serialize the fused data. Each data record corresponds to a JSON object, where the key is a standardized field name and the value is the corresponding data content. For example, the encapsulated data for a timestamp is: "{"TimeStamp":"2025-10-10 08:30:00.100","Load_P_S1-01-L05":4680.0,"Load_Q_S1-01-L05":1290.0,"PV_P_PV-03":3250.0,"WT_P_WT-02":2800.0,"ESS_SOC_ESS-01-C02":68.1,"ESS_P_ESS-01-C02":1825.0,"Carbon_Intensity_G1":0.52,"Carbon_Intensity_G2":0.52,"Carbon_Intensity_G3":0.48,...}". The encapsulated dataset is stored as a file per hour, with the file name format "Standard_Data_YYYYMMDD_HH.json", for example "Standard_Data_20251010_08.json". The files are stored in a compressed format to save space, with a compression ratio of approximately 3:1.

[0037] The standardized multi-source dataset also includes a data dictionary and a quality report. The data dictionary details the meaning, source, data type, and unit of each field, while the quality report records indicators such as data collection success rate, anomaly correction rate, and missing rate. For example, "Data collection success rate was 99.8% on 2025-10-10 08, anomaly correction rate was 99.2%, and missing rate was 0.08%", providing a reference for setting parameters for the subsequent carbon flow tracing algorithm and ensuring the accuracy of the analysis results.

[0038] S202, Based on the dynamic carbon emission intensity signal, a load-side carbon emission intensity spectrum with spatiotemporal resolution is generated by a source-load-storage full-process carbon flow tracking algorithm; Specifically, it can analyze the time series signal of dynamic carbon emission intensity in a standardized multi-source dataset, establish a carbon flow network topology with the regional power grid gateway as the starting point and the load node as the ending point, and generate a carbon flow network model. This step is the foundational framework building stage for carbon flow tracing. Its core is to extract a carbon intensity benchmark through data analysis and construct a network carrier for carbon flow transmission by combining it with the physical structure of the power grid. This clarifies the transmission path of carbon flow from its source to the load, providing structural support for subsequent quantitative calculations. The specific implementation method is as follows: The parsing of standardized multi-source datasets focuses on extracting key parameters from the time-series signal of dynamic carbon emission intensity. The parsing process adopts a workflow of "field matching - validity verification - spatiotemporal correlation". First, it matches three core fields, "Carbon_Intensity_gate number", "TimeStamp", and "confidence level", from the standardized data in JSON format. For example, the parsing yields that the carbon emission intensity of gate G1 on 2025-10-10 08:30:00.100 is 0.52 kgCO2 / kWh with a confidence level of 95%; the intensity of gate G2 during the same period is 0.52 kgCO2 / kWh, and that of G3 is 0.48 kgCO2 / kWh. Validity verification is achieved by comparing a confidence threshold (preset to 90%), eliminating signals with confidence levels below the threshold. If a signal at a certain threshold is invalid, it is replaced by the weighted average of signals from adjacent thresholds (the weights are calculated based on the load share of each threshold). For example, if signal G2 has a confidence level of 88%, and the load shares of G1 and G3 are 60% and 40% respectively, then the replacement intensity of G2 = 0.52 × 0.6 + 0.48 × 0.4 = 0.504 kgCO2 / kWh. Spatiotemporal correlation binds the parsed carbon intensity signal with the grid topology data of the corresponding time segment, ensuring that the carbon flow calculation matches the real-time grid structure.

[0039] The carbon flow network topology is established based on the physical connections of the regional power grid and is constructed using a node-edge undirected graph model. The node types include four core elements: First, the carbon flow initiation nodes, namely the metering points connecting the regional power grid and the upper-level power grid (G1, G2, G3), which serve as the source of external carbon flow input; Second, the power generation side nodes, including photovoltaic power stations (PV-01 to PV-08) and wind power stations (WT-01 to WT-05) within the region. The carbon intensity of these nodes is set according to the energy type (photovoltaic and wind power are zero-carbon energy, with an intensity value of 0 kgCO2 / kWh); Third, the energy storage nodes, namely the battery clusters of each energy storage power station (ESS-01-C01 to ESS-06-C12), which serve as temporary storage and transfer nodes for carbon flow; Fourth, the carbon flow endpoint nodes, namely all flexible load and rigid load nodes (L01 to L120), which are the final consumption ends of carbon flow. Edges correspond to the transmission and distribution lines of the power grid. The attributes of each edge include line impedance (in Ω), rated transmission power (in kW), and real-time power flow direction (from high-potential nodes to low-potential nodes). For example, line L1-01 connecting gate G1 and photovoltaic power station PV-03 has an impedance of 0.05Ω, a rated power of 10000kW, and a real-time power flow from G1 to PV-03 with a power of 2000kW.

[0040] The carbon flow network model is a digital representation of its topology and parameters, stored using a combination of adjacency matrix and attribute dictionary. The adjacency matrix has an N×N dimension (N is the total number of nodes, N=152 here). Matrix elements A[i][j] have values ​​of 1 indicating that nodes i and j are directly connected, and 0 indicating no connection. For example, A[G1 number][PV-03 number]=1, A[PV-03 number][L05 number]=1. The attribute dictionary records detailed parameters of each node and edge. The node dictionary contains "node number-type-real-time power-carbon intensity benchmark". For example, the dictionary entry for G1 is "G1-gate-8000kW-0.52kgCO2 / kWh", and the entry for PV-03 is "PV-03-photovoltaic-3250kW-0kgCO2 / kWh". The edge dictionary contains "edge number-start point-end point-impedance-transmission power". For example, the entry for L1-01 is "L1-01-G1-PV-03-0.05Ω-2000kW". The model fully presents the transmission path and constraints of carbon flow in the regional power grid, ensuring that subsequent carbon flow tracking can accurately match the actual operating status of the power grid.

[0041] Based on the carbon flow network model and real-time power flow data, and using the proportional sharing principle, the distribution of electrical carbon flow from the generation side to the load side is calculated, and a node carbon flow intensity matrix is ​​generated. This step is the core of carbon flow quantification. It uses a proportional sharing principle to allocate carbon emission intensity to each node according to the proportion of electricity flow, realizing the step-by-step calculation of carbon flow from source to load, ultimately forming a quantified carbon intensity result at the node level. The specific implementation method is as follows: Real-time power flow data is extracted from a standardized multi-source dataset. Core parameters include the real-time active power transmission power P_line of each line and the injected power P_inject of each node (positive for generator nodes, negative for load nodes). The data update frequency is consistent with the carbon flow network model (10Hz). For example, the power flow data for a certain time segment is as follows: 8000kW injected power at the G1 junction, 3250kW injected by PV-03, and 2800kW injected by the WT-02 wind farm; 4680kW absorbed by load L05 and 3200kW by load L06; 2000kW transmitted by line L1-01 (G1-PV-03), 3500kW by line L1-02 (PV-03-L05), and 3200kW by line L2-01 (G2-L06). The power flow data must satisfy power balance constraints, i.e., the algebraic sum of the injected power at all nodes must be 0. If an imbalance exists (error ≤ 0.5%), it is corrected through power allocation at the load nodes.

[0042] The core logic of the proportional sharing principle is "carbon flow and electrical energy flow are distributed proportionally." That is, the carbon intensity of a node is obtained by weighting all upstream nodes supplying power to it according to their power ratios. The calculation formula is e_i = ∑(P_ij / P_i_in) × e_j, where e_i is the carbon intensity of node i, P_ij is the power transmitted from node j to node i, P_i_in is the total input power to node i, and e_j is the carbon intensity of node j. This principle is applicable to radial distribution networks, accurately reflecting the impact of electrical energy sources on load carbon intensity and avoiding errors caused by traditional average distribution methods.

[0043] The carbon flow distribution calculation proceeds step-by-step in the order of "source node → intermediate node → load node". First, the carbon intensity benchmark for the source node is determined: the dynamic carbon emission intensity of the gate node is adopted after analysis (e.g., G1=0.52, G2=0.504, G3=0.48), the renewable energy node is fixed at 0, and the initial state of the energy storage node (when not charging) is 0. Taking the photovoltaic node PV-03 as an example, its input power comes from G1 (2000kW) and its own power generation (3250kW), and the total input power P_in=2000+3250=5250kW. Therefore, the carbon intensity of PV-03 is e_PV03=(2000 / 5250)×0.52+(3250 / 5250)×0≈0.198kgCO2 / kWh.

[0044] The calculation of carbon intensity at the load node is crucial. Taking L05 as an example, its input power comes from PV-03 (3500kW) and energy storage ESS-01 (1180kW), with a total input P_in = 3500 + 1180 = 4680kW. Given that e = 0.198 for PV-03 and ESS-01 is currently not discharging (calculated as 0), then e_L05 = (3500 / 4680) × 0.198 + (1180 / 4680) × 0 ≈ 0.149 kg CO2 / kWh. For a node that is powered by multiple gateways, such as L10 simultaneously receiving power from G1 (1500kW, e=0.52) and G3 (1000kW, e=0.48), with a total input of 2500kW, then e_L10=(1500 / 2500)×0.52+(1000 / 2500)×0.48=0.312+0.192=0.504kgCO2 / kWh.

[0045] The nodal carbon flow intensity matrix is ​​a spatiotemporal integration of the carbon intensity calculation results for all nodes. The matrix has a dimension of T×N (T is the number of time segments, calculated over one hour, T=36000, N is the total number of nodes, 152). Matrix element M[t][n] represents the carbon intensity value (unit: kgCO2 / kWh) of the nth node at the t-th time segment. For example, the element value in the L05 column of row 3601 (corresponding to 08:30:00.100) is 0.149, and the element value in the G1 column is 0.52. The matrix uses column-major storage, facilitating the querying of carbon intensity changes at different times by node dimension, providing fundamental data support for subsequent energy storage carbon flow calculations and intensity spectrum generation.

[0046] Establish a carbon state transition model for energy storage devices, track the carbon emissions corresponding to the electrical energy absorbed during the charging period of energy storage, and trace and allocate carbon flow during the discharging period to generate an energy storage carbon flow distribution matrix; This step is crucial for addressing the "carbon storage-carbon release" characteristics of energy storage. By constructing a carbon state transition model, it enables full lifecycle tracking of carbon flow during energy storage charging and discharging, avoiding mismatches in carbon flow over time and ensuring the accuracy of load-side carbon intensity calculations. The specific implementation method is as follows: The core of the carbon state transition model for energy storage devices is to link the changes in the SOC of energy storage with carbon flow. The "Carbon State of Charge (C-SOC)" is defined as the core state variable, representing the current carbon emissions stored in the energy storage, expressed in kgCO2. The calculation formula is C-SOC_t = C-SOC_{t-1} + P_ess_t × Δt × e_charge_t, where C-SOC_{t-1} is the carbon charge at the previous time step, P_ess_t is the charging / discharging power at time t (positive for charging, negative for discharging), Δt is the time step (0.1 seconds), and e_charge_t is the carbon intensity corresponding to the charging energy at time t (this value is 0 during discharging; carbon flow release is based on historical charging carbon intensity tracing). The model also includes a carbon flow tracing mechanism, that is, during discharging, historical charging carbon intensity is allocated according to the "first-in, first-out (FIFO)" principle to ensure that the carbon source of each kilowatt-hour of discharged energy is traceable.

[0047] Carbon flow tracking during charging periods requires precise matching of the source of charging energy with the corresponding carbon intensity. Taking the energy storage ESS-01-C02 as an example, during the period from 08:30:00.000 to 08:30:01.000, continuous charging occurs with a stable charging power of 1800kW and a time step of 0.1 seconds. The charging amount at each time step is ΔE = 1800 × 0.1 / 3600 = 0.05kWh. The charging energy during this period comes from the gate G1 (1000kW, e = 0.52) and the photovoltaic PV-03 (800kW, e = 0.198). Therefore, the carbon intensity at each time step is e_charge = (1000 / 1800) × 0.52 + (800 / 1800) × 0.198 ≈ 0.289 + 0.088 = 0.377kgCO2 / kWh. Therefore, the carbon charge increment ΔC-SOC at each time step is approximately 0.05 × 0.377 ≈ 0.01885 kg CO2. After 10 time steps (1 second), C-SOC increases from the initial 1200 kg CO2 to 1200 + 0.01885 × 10 ≈ 1200.1885 kg CO2. At the same time, the carbon intensity log for charging during this period is recorded: time interval, charging amount, and average carbon intensity, forming a traceable carbon flow record.

[0048] Carbon flow allocation during discharge periods is based on the FIFO principle, tracing historical charging carbon intensity. Assuming ESS-01-CO2 begins discharging at 08:35:00.000 with a discharge power of 2000kW, ΔE = 2000 × 0.1 / 3600 ≈ 0.0556 kWh / time step. The carbon intensity log is queried, prioritizing the earliest charging record: 0.5 kWh charging (10 time steps × 0.05 kWh) from 08:30:00.000 to 08:30:01.000, with a carbon intensity of 0.377 kg CO2 / kWh; and 2 kWh charging from 08:30:01.000 to 08:30:05.000, with a carbon intensity of 0.35 kg CO2 / kWh. During discharge, the carbon flow corresponding to 0.5 kWh is released first. The carbon intensity of the discharge at each time step is calculated based on the traced charging intensity. For example, in the first 9 time steps (0.9 seconds), 0.5 kWh of electrical energy is released, and the carbon intensity is 0.377 kg CO2 / kWh. From the 10th time step onwards, the electrical energy recorded in subsequent charging is released, and the carbon intensity switches to 0.35 kg CO2 / kWh. At this time, the formula for the change of C-SOC is C-SOC_t = C-SOC_{t-1} - ΔE × e_discharge_t, where e_discharge_t is the traced historical charging carbon intensity, ensuring that the change in carbon charge is completely matched with the release of carbon flow.

[0049] The energy storage carbon flow distribution matrix is ​​a quantitative representation of the carbon flow during energy storage charging and discharging. The matrix dimension is T×M (T is the number of time segments, M is the number of energy storage battery clusters, here M=72). The matrix element C[t][m] represents the carbon intensity released to the load by the m-th energy storage battery cluster at the t-th time segment (the traced carbon intensity value during discharge, 0 during charging, and -1 during standby). For example, ESS-01-C02 is in a discharging state at 08:35:00.100 (the 3901st time step), and the traced carbon intensity is 0.377, so C

[3901] [ESS-01-C02 number]=0.377; it is in a charging state at 08:30:00.100 (the 3601st time step), so C

[3601] [ESS-01-C02 number]=0. The time dimension of this matrix is ​​perfectly aligned with that of the node carbon flow intensity matrix. When the energy storage discharges, the released carbon intensity is included in the calculation of the carbon intensity of the downstream load node through power transmission, thus achieving precise matching of carbon flow in time and space.

[0050] By combining the carbon flow intensity matrix of the integrated nodes and the carbon flow distribution matrix of the energy storage, a mixed-integer linear programming algorithm is used to accurately allocate carbon emissions to each load node and time segment, ultimately generating a load-side carbon emission intensity spectrum with spatiotemporal resolution.

[0051] This step is the final integration stage of carbon flow tracing. It solves the problem of accurate allocation of multi-source carbon flows by using a mixed-integer linear programming algorithm, eliminating ambiguity in carbon flow calculation, and accurately locating carbon emissions to each time segment of each load node, forming a complete load-side carbon emission intensity spectrum. The specific implementation method is as follows: The integration of the node carbon flow intensity matrix and the energy storage carbon flow distribution matrix requires addressing the allocation problem of "dual carbon flow inputs"—the carbon flow of load nodes comes from both the upstream grid and renewable energy (already calculated in the node matrix) and from energy storage discharge (provided by the energy storage matrix). When both exist simultaneously, the carbon intensity weights need to be accurately allocated according to the power ratio. Taking load L05 as an example, during the period 08:35:00.100, the input power comes from PV-03 (3000kW, e=0.20) and ESS-01-C02 (1680kW, e=0.377), with a total input of 4680kW. At this time, the carbon intensity weights of the two need to be determined through an algorithm to ensure that the comprehensive carbon intensity calculation of L05 conforms to both power balance and the conservation of total carbon emissions.

[0052] The core of the mixed-integer linear programming algorithm is to construct an optimization model of "objective function-constraints" to achieve optimal carbon emission allocation. The objective function is set as "minimizing the calculation error of carbon intensity at load nodes", i.e., Minimize∑|e_load_actual-e_load_calc|, where e_load_actual is the actual carbon intensity inferred from the total carbon emissions, and e_load_calc is the carbon intensity calculated by the model. The constraints include three core types of constraints: first, power balance constraints, i.e., the total input power of the load node is equal to the sum of the output power of each upstream node, P_load=∑P_source; second, carbon flow conservation constraints, i.e., the total carbon emissions of the load node are equal to the sum of the carbon emissions transmitted by each upstream node, e_load_calc×P_load=∑(e_source×P_source); and third, integer constraints, used to mark the energy storage discharge state (0=not discharged, 1=discharged), ensuring that the carbon flow of the energy storage only participates in allocation during discharge.

[0053] The algorithm employs a branch and bound method to determine the power allocation coefficients of each upstream node through iterative optimization, thereby calculating the precise carbon intensity of the load nodes. Taking the optimization calculation of L05 as an example, the model inputs are P_load=4680kW, P_PV03=3000kW, e_PV03=0.20, P_ESS01=1680kW, e_ESS01=0.377. The target value for total carbon emissions, calculated based on the upstream power grid's outlet carbon intensity and transmission power, is 0.52×(8000-2000)+0.48×1000=3120+480=3600kgCO2 / hour. After 10 iterations using the branch-and-bound method, the optimal power allocation coefficient is obtained: PV-03 undertakes 3000kW (64.1%), and ESS-01 undertakes 1680kW (35.9%). Then, e_load_calc=(3000×0.20+1680×0.377) / 4680=(600+633.36) / 4680≈1233.36 / 4680≈0.264kgCO2 / kWh. The error between this value and the actual carbon intensity is only 0.002kgCO2 / kWh, which meets the accuracy requirement (error ≤0.5%).

[0054] The load-side carbon emission intensity spectrum is the spatiotemporal representation of the algorithm optimization results. The "spatiotemporal resolution" is specifically reflected in the following: the temporal resolution is consistent with the data sampling frequency, which is 10Hz (i.e., one data point every 0.1 seconds); the spatial resolution covers all 120 load nodes within the regional power grid, with each node corresponding to an independent carbon intensity time-series curve. The core content of the intensity spectrum is a three-dimensional data set of "load node - time section - carbon intensity." For example, the carbon intensity of L05 at 08:30:00.100 is 0.149 kgCO2 / kWh, at 08:35:00.100 it is 0.264 kgCO2 / kWh, and at 08:40:00.100 it decreases to 0.12 kgCO2 / kWh due to increased photovoltaic output. For ease of application, the intensity spectrum also provides information on statistical dimensions, including the average carbon intensity, peak carbon intensity, and carbon intensity volatility (standard deviation / mean) of each load node over 15 minutes, 1 hour, and 24 hours. For example, the average carbon intensity of L05 from 08:00 to 09:00 is 0.18 kg CO2 / kWh, the peak is 0.264 kg CO2 / kWh, and the volatility is 25%.

[0055] The intensity spectrum is presented using a combination of a numerical matrix and a visual heatmap. The numerical matrix is ​​stored in CSV format for easy access by subsequent algorithms. The heatmap uses time as the horizontal axis and load nodes as the vertical axis, with color depth representing carbon intensity (dark blue for low intensity and dark red for high intensity), visually demonstrating the spatiotemporal distribution characteristics of carbon intensity. For example, in the heatmap, from 08:35 to 08:40, the L05-L10 region appears light red (carbon intensity 0.2-0.3), while the L50-L60 region, being entirely powered by photovoltaics, appears dark blue (carbon intensity 0-0.1), clearly reflecting the differences in carbon emissions from different load nodes and providing direct carbon intensity data for subsequent flexible load control prioritization.

[0056] S203, Based on the load-side carbon emission intensity spectrum and combined with the demand-side resource adjustability potential model, establish a virtual carbon pool with carbon cost as the weighting factor and generate a flexible load regulation priority sequence. Specifically, the load-side carbon emission intensity spectrum can be analyzed, the marginal carbon emission intensity of each flexible load node at different time sections can be calculated, and it can be converted into carbon cost per unit of electricity to generate a carbon cost vector. This step is fundamental to building a carbon cost-linking mechanism. Its core is to extract marginal carbon intensity indicators from the load-side carbon emission intensity spectrum and convert them into quantifiable economic costs. This provides a crucial basis for subsequent virtual carbon pool pricing and regulatory prioritization. The specific implementation method is as follows: The analysis of the load-side carbon emission intensity spectrum follows a core process of "flexible load node screening - time-series carbon intensity extraction - outlier smoothing". First, from the 120 load nodes covered by the intensity spectrum, 45 flexible load nodes with adjustment capabilities were screened out. The screening criteria were load type (industrial cooling load, commercial air conditioning load, interruptible residential load, etc.) and historical adjustment records (at least 10 effective adjustment responses in the past 3 months). For example, industrial load L12 (cooling system of food processing plant), commercial load L35 (air conditioning system of shopping mall), and residential load L88 (centralized electric heating) were all included in the flexible load set.

[0057] For each flexible load node, carbon intensity data for 24 consecutive hours is extracted at a time resolution of 10Hz, forming a node-level carbon intensity time series. The sequence format is "timestamp t - carbon intensity e_load(t)", where the interval of t is 0.1 seconds, and the unit of e_load(t) is kgCO2 / kWh. Taking industrial load L12 as an example, some extracted time series data are: t1 (08:30:00.000)e=0.52kgCO2 / kWh, t2 (08:30:00.100)e=0.518kgCO2 / kWh, t3 (08:30:00.200)e=0.522kgCO2 / kWh, t4 (08:30:00.300)e=0.525kgCO2 / kWh. After extraction, the data needs to be smoothed. The moving average filtering method is used to eliminate high-frequency noise. The filter window length is set to 5 time steps. For example, the smoothing value of t2 is e_smooth(t2)=(e(t1)+e(t2)+e(t3)+e(t4)+e(t5)) / 5 to ensure the continuity and stability of the carbon intensity sequence.

[0058] Marginal Carbon Intensity (MCI) is calculated using the first-order difference method, reflecting the incremental change in carbon intensity when a load node increases its electricity consumption by one unit per unit time. The formula is MCI(t) = [e_smooth(t) - e_smooth(t-1)] / [P_load(t) - P_load(t-1)] × P_base, where P_load(t) is the load power at time t, and P_base is the base power (the 24-hour average power of the load, for example, P_base = 2000kW for L12). The core logic of this formula is to correlate carbon intensity changes with power changes, eliminating the influence of simple power fluctuations on carbon intensity, and accurately capturing the characteristics of "marginal" carbon cost. Taking L12 as an example, at time t3, P_load(t3) = 2050kW, at time t2, P_load(t2) = 2000kW, e_smooth(t3) = 0.522, e_smooth(t2) ​​= 0.519, then MCI(t3) = (0.522-0.519) / (2050-2000)×2000 = 0.003 / 50×2000 = 0.12kgCO2 / kWh, indicating that for every 1kWh increase in electricity consumption at L12 at this time, the marginal carbon emission increases by 0.12kgCO2.

[0059] The conversion of carbon cost per unit of electricity needs to combine the regional carbon market trading price and carbon emission reduction policy subsidies. The conversion formula is C_carbon(t) = MCI(t) × (P_market + S_subsidy), where P_market is the regional carbon market spot price (unit: yuan / kgCO2, taking the recent average of 0.8 yuan / kgCO2), and S_subsidy is the carbon emission reduction policy subsidy (unit: yuan / kgCO2, with a subsidy of 0.2 yuan / kgCO2 for flexible load adjustment to encourage carbon emission reduction behavior). Therefore, the conversion factor is 0.8 + 0.2 = 1.0 yuan / kgCO2. Taking L12's MCI(t3) = 0.12 kgCO2 / kWh as an example, C_carbon(t3) = 0.12 × 1.0 = 0.12 yuan / kWh, that is, the carbon cost per unit of electricity of L12 at this moment is 0.12 yuan. If MCI(t) is negative (indicating a decrease in marginal carbon intensity when electricity consumption increases, such as during periods of sudden increase in photovoltaic output), then the carbon cost is set to 0, encouraging increased electricity consumption to absorb renewable energy during such periods.

[0060] The carbon cost vector is generated by constructing a two-dimensional vector matrix with "time segments as rows and flexible load nodes as columns". The vector dimension is T×N (T is the number of time segments corresponding to 24 hours, T=864000; N is the number of flexible load nodes, N=45). The vector element C[t][n] represents the carbon cost per unit of electricity at the t-th time segment and the n-th flexible load node. For example, in the matrix, the element value of L12 at t3 (08:30:00.200) is 0.12 yuan / kWh, and the element value of L35, the load with a high proportion of photovoltaic output, is 0 yuan / kWh. To facilitate subsequent calculations, the carbon cost data of each time segment is integrated into a one-dimensional row vector. For example, the carbon cost vector at time t3 is [0.12,0.08,0,...,0.15] (45 elements in total), forming a carbon cost vector sequence arranged in chronological order.

[0061] The demand-side resource adjustability potential model is invoked. Based on historical operating data and load characteristics, this model quantifies the power reduction, transferable electricity, and response time constant of various flexible loads, and generates an adjustability potential mapping table. This step is crucial for clarifying the flexible load adjustment capability. It quantifies the adjustment boundary and response characteristics of each load through an adjustable potential model, providing key data in the "adjustment capability" dimension for subsequent priority ranking of control measures. The specific implementation method is as follows: The demand-side resource adjustability potential model adopts a hybrid modeling method of "load type clustering + historical data training". The model input includes three types of data: first, historical operating data of flexible loads (power time series data and adjustment response records for the past 12 months); second, load characteristic parameters (load type, type of electrical equipment, operating cycle, temperature sensitivity, etc.); and third, external environmental data (ambient temperature, electricity price period, holiday information, etc.). The model output consists of three core quantitative indicators: power reduction potential P_cut, transferable power E_shift, and response time constant τ_response, which comprehensively reflect the load's adjustment potential and response speed.

[0062] The core calculation process of the model consists of three steps: The first step is load type clustering. The K-means clustering algorithm is used to divide the 45 flexible loads into three categories: industrial loads (15), commercial loads (20), and residential loads (10). The clustering characteristics include power fluctuation amplitude, daily load curve shape, and adjustment response frequency. For example, the cluster center of industrial loads is "high power (1000-5000kW), small fluctuation, slow response", commercial loads are "medium power (500-2000kW), large fluctuation, fast response", and residential loads are "low power (100-500kW), random fluctuation, medium response".

[0063] The second step is to quantify the reducible power P_cut, defined as the maximum power consumption that can be reduced per unit time without affecting production and living needs. The calculation is based on the "historical minimum power method" and "equipment redundancy analysis". For industrial loads (such as the cooling system of the L12 food processing plant), P_cut = P_avg - P_min, where P_avg is the average power of the load during the production period (2000kW), and P_min is the minimum operating power under the historical production conditions (1500kW). Therefore, P_cut = 2000 - 1500 = 500kW. At the same time, the equipment redundancy constraint must be met (the minimum power of the cooling system must not be lower than 40% of the rated power; the rated power of L12 is 3000kW, 40% is 1200kW, and 1500kW meets the constraint). For commercial loads (such as the L35 shopping mall air conditioning system), P_cut is related to the ambient temperature and is calculated using a linear regression model: P_cut = a × (T_actual - T_set) + b, where T_actual is the actual ambient temperature (32℃), T_set is the upper limit of the comfortable temperature (28℃), a is the temperature coefficient (50kW / ℃), and b is the basic reduction amount (100kW). Therefore, P_cut = 50 × (32 - 28) + 100 = 300kW. For residential loads (such as the L88 centralized electric heating system), P_cut is taken as 30% of the rated power (rated power 400kW, P_cut = 120kW) to ensure that the indoor temperature is not lower than 18℃.

[0064] The third step is the quantification of transferable electricity E_shift, which refers to the total amount of electricity that can be transferred from high-carbon-intensity periods to low-carbon-intensity periods. The calculation is based on "load operation cycle analysis," and the formula is E_shift = P_avg × ΔT_shift, where ΔT_shift is the transferable time window (unit: hours). For industrial load L12, the cooling system has a 24-hour operating cycle, and the transferable time window is the 6-hour period between off-peak hours (00:00-06:00) and peak hours (08:00-12:00). Therefore, E_shift = 2000 × 6 = 12000 kWh. For commercial load L35, the air conditioning system's transferable time window is from 12:00-14:00 (high-carbon period) to 18:00-20:00 (low-carbon period), ΔT_shift = 2 hours, and E_shift = 1500 × 2 = 3000 kWh. The time window for transferring electric heating for residential load L88 is from 07:00-09:00 (high carbon period) to 22:00-24:00 (low carbon period), ΔT_shift=3 hours, E_shift=300×3=900kWh.

[0065] The response time constant τ_response quantifies the time from when the load receives the control command to when the adjustment is completed. It is obtained using "historical response curve fitting" and is defined as the time required for the load power to reach 63.2% of the adjustment target. Industrial loads have a larger τ_response due to the large inertia of the equipment. The historical response curve of L12 shows that it takes 120 seconds from the issuance of the command to the power reduction to 63.2% of the target value (316kW, 500kW × 63.2%), therefore τ_response = 120s. Commercial loads have a fast response; the response time constant of L35 is 30s. Residential loads have a moderate response speed; the τ_response of L88 is 60s.

[0066] The adjustable potential mapping table is a structured representation of the model output. The core fields in the table are "Flexible Load Number - Load Type - Cuttable Power P_cut - Transferable Power E_shift - Response Time Constant τ_response". For example, the mapping table entry for L12 is "L12-Industrial Load-500kW-12000kWh-120s", for L35 it is "L35-Commercial Load-300kW-3000kWh-30s", and for L88 it is "L88-Residential Load-120kW-900kWh-60s". The mapping table is stored categorized by load type and also labels the confidence level of each indicator (based on historical data fitting accuracy, industrial load confidence level 92%, commercial load 95%, residential load 88%), providing a reliable reference for subsequent virtual carbon pool pricing.

[0067] Using the carbon cost vector as the core weighting factor and combining it with the adjustable potential mapping table, a pricing and settlement mechanism for the virtual carbon pool is designed, and a dynamic carbon price curve that correlates carbon cost with adjustment value is established. This step is crucial in connecting carbon costs with regulatory incentives. Through the market-based mechanism of the virtual carbon pool, carbon costs are transformed into dynamic carbon prices, achieving a regulatory orientation of "high incentives during high-carbon periods and low incentives during low-carbon periods." The specific implementation method is as follows: The virtual carbon pool is a market-based virtual trading platform based on credit settlement. Its core functions are "carbon cost collection - dynamic pricing - adjustment incentive settlement". Participants include regional power grid control centers (carbon pool operators) and flexible load users (adjustment implementers). The carbon pool's funding comes from carbon emission reduction subsidies from the upper-level power grid and carbon cost payments from high-carbon loads. The funds are used to incentivize flexible load adjustment behavior, forming a closed-loop mechanism of "high carbon pays, low carbon rewards".

[0068] The pricing mechanism adopts a composite pricing model of "base carbon price + adjustment potential premium". The core formula is P_cpool(t) = C_carbon(t) × W_carbon + (P_cut / P_avg) × W_potential × P_base_price, where P_cpool(t) is the dynamic carbon price of the virtual carbon pool at time t (unit: yuan / kWh), W_carbon is the carbon cost weighting coefficient (taken as 0.7, reflecting the core position of carbon cost), (P_cut / P_avg) is the load adjustment potential coefficient (reflecting the strength of adjustment capability), W_potential is the potential weighting coefficient (taken as 0.3), and P_base_price is the base incentive price (taken as 0.3 yuan / kWh, ensuring the minimum incentive threshold). The core logic of this formula is: the higher the carbon cost and the greater the adjustment potential of the time period and load, the higher the dynamic carbon price and the stronger the incentive.

[0069] Taking the parameters of L12 at time t3 as an example, C_carbon(t3) = 0.12 yuan / kWh, P_cut = 500kW, P_avg = 2000kW, and the adjustable potential coefficient = 500 / 2000 = 0.25, then P_cpool(t3) = 0.12 × 0.7 + 0.25 × 0.3 × 0.3 = 0.084 + 0.0225 = 0.1065 yuan / kWh. At the same time, for L35, C_carbon(t3) = 0.08 yuan / kWh, P_cut = 300kW, P_avg = 1500kW, and the potential coefficient = 0.2, then P_cpool(t3) = 0.08 × 0.7 + 0.2 × 0.3 × 0.3 = 0.056 + 0.018 = 0.074 yuan / kWh. During the low-carbon period t10 (09:00:00.000), L12's C_carbon(t10) = 0 yuan / kWh, with a potential coefficient of 0.25. Therefore, P_cpool(t10) = 0 × 0.7 + 0.25 × 0.3 × 0.3 = 0.0225 yuan / kWh, which is significantly lower than that during the high-carbon period, thus achieving the guidance of "high carbon, high incentive".

[0070] The settlement mechanism adopts a two-stage model of "pre-settlement + final settlement" to ensure timely incentive payment and accurate accounting. Pre-settlement is carried out before the execution of the control order. The pre-settlement amount is calculated based on the predicted adjustment amount and the current dynamic carbon price. The pre-settlement ratio is 70%. For example, if L12 receives a control order to reduce 400kW for 10 minutes at time t3, the predicted adjustment amount is 400×(10 / 60)=66.67kWh, and the pre-settlement amount is 66.67×0.1065×70%≈4.94 yuan. Final settlement is conducted within 24 hours after the implementation of the regulation. It is recalculated based on the actual regulated volume and the actual carbon price, using the formula: Final_payment = Actual_E × P_cpool_actual - Pre_payment. Actual_E is the actual regulated electricity (if the actual reduction is 400kW for 10 minutes, Actual_E = 66.67kWh), and P_cpool_actual is the average carbon price during the actual implementation period (if the average carbon price for t3-t3+10 minutes is 0.108 yuan / kWh). Therefore, Final_payment = 66.67 × 0.108 - 4.94 ≈ 7.20 - 4.94 ≈ 2.26 yuan, and the total incentive amount after final settlement is 7.20 yuan. If the actual regulated volume does not reach the target (e.g., only a 300kW reduction), the calculation is based on the actual regulated volume, and the shortfall is deducted from the pre-settlement amount, ensuring that the incentive is linked to the regulation effect.

[0071] The dynamic carbon price curve is constructed with time on the horizontal axis and dynamic carbon price on the vertical axis. It integrates the carbon price data of all flexible load nodes at each time segment, and takes the average value as the carbon pool carbon price at that moment, forming a continuous carbon price curve. The time resolution of the curve is 10Hz, consistent with the carbon cost vector. For example, from 08:29:00 to 08:31:00, the carbon price rises from 0.09 yuan / kWh to 0.1065 yuan / kWh (time t3), and then falls back to 0.10 yuan / kWh, reflecting the fluctuation trend of carbon intensity during this period. To facilitate control decisions, the curve is marked with both "high carbon price range" (P_cpool(t) ≥ 0.1 yuan / kWh, such as 08:30:00-08:30:30) and "low carbon price range" (P_cpool(t) ≤ 0.03 yuan / kWh, such as 00:10:00-00:20:00). The high carbon price range is the key period for flexible load control, while the low carbon price range encourages loads to consume more electricity from renewable energy sources.

[0072] Based on the dynamic carbon price curve and the adjustment costs of each flexible load, a priority sequence for flexible load regulation is generated, which is optimally ranked in terms of both carbon emission reduction benefits and economic efficiency.

[0073] This step is the core output of the regulatory decision-making process. By comprehensively weighing the benefits of carbon emission reduction with the economic efficiency of regulation, the priority of flexible load regulation is determined to ensure that limited regulatory resources are first invested in loads with "good emission reduction effect and low cost". The specific implementation method is as follows: The core evaluation index system for comprehensive ranking includes two primary indicators: carbon emission reduction benefit (B_carbon) and adjustment economy (C_adjust). The weights of the two indicators are 0.6 and 0.4, respectively. The weights are determined based on the control objective of "prioritizing carbon emission reduction while taking into account economic costs" and are calculated using the analytic hierarchy process (AHP). The consistency test coefficient CR = 0.08 ≤ 0.1, ensuring the rationality of the weight allocation.

[0074] The carbon emission reduction benefit B_carbon reflects the carbon emission reduction and economic value brought about by load regulation behavior. The formula is B_carbon=(MCI(t)×ΔE)×(P_market+S_subsidy), where ΔE is the maximum adjustable load capacity (the larger of the adjustable capacity and the transferable capacity, for example, ΔE=12000kWh for L12, ΔE=3000kWh for L35, and ΔE=900kWh for L88). Taking L12 at time t3 as an example, MCI(t3)=0.12kgCO2 / kWh, ΔE=12000kWh, then B_carbon=(0.12×12000)×1.0=1440 yuan, that is, L12 can achieve a carbon emission reduction benefit of 1440 yuan when it completes the maximum regulation. The MCI(t3) of L35 is 0.08 kg CO2 / kWh, ΔE is 3000 kWh, and B_carbon is (0.08 × 3000) × 1.0 = 240 yuan. The MCI(t3) of L88 is 0.15 kg CO2 / kWh, ΔE is 900 kWh, and B_carbon is (0.15 × 900) × 1.0 = 135 yuan.

[0075] The calculation of the economic efficiency of regulation, C_adjust, reflects the cost required for the load to execute regulation commands. Cost types include equipment operating costs, production interruption losses (industrial load only), and user comfort losses (commercial and residential loads). The calculation formula is C_adjust = C_operation + C_interrupt + C_comfort. For industrial load L12, C_operation is the equipment energy consumption cost during the regulation process (0.05 yuan / kWh, C_operation = 12000 × 0.05 = 600 yuan), C_interrupt is the marginal loss from production interruption (loss coefficient for food processing interruption: 0.02 yuan / kWh, C_interrupt = 12000 × 0.02 = 240 yuan), and C_comfort = 0. Therefore, C_adjust = 600 + 240 = 840 yuan. For commercial load L35, C_operation = 3000 × 0.04 = 120 yuan, C_interrupt = 0 (air conditioning adjustment does not affect commercial operation), and C_comfort is the cost of comfort loss (the loss coefficient for temperature deviation from the comfortable range is 0.03 yuan / kWh, C_comfort = 3000 × 0.03 = 90 yuan). Therefore, C_adjust = 120 + 90 = 210 yuan. For residential load L88, C_operation = 900 × 0.03 = 27 yuan, C_interrupt = 0, and C_comfort = 900 × 0.05 = 45 yuan (comfort loss due to reduced heating temperature). Therefore, C_adjust = 27 + 45 = 72 yuan.

[0076] To achieve a comprehensive ranking, the two indicators need to be standardized to the same scale (0-100 points). The extreme value standardization method is used, with the formula: Score = (X - X_min) / (X_max - X_min) × 100, where X is the original value of the indicator, X_min is the minimum value of the indicator, and X_max is the maximum value. For carbon emission reduction benefits, X_max = 1440 yuan (L12) and X_min = 135 yuan (L88). Therefore, the B_score for L12 is (1440-135) / (1440-135) × 100 = 100 points, the B_score for L35 is (240-135) / (1440-135) × 100 ≈ (105 / 1305) × 100 ≈ 8.05 points, and the B_score for L88 is 0 points. The evaluation of economic efficiency adopts the logic of "the lower the cost, the higher the score". Therefore, the standardized formula is Score=(X_max-X) / (X_max-X_min)×100, X_max=840 yuan (L12), X_min=72 yuan (L88). Then, the C_score of L12 is (840-840) / (840-72)×100=0 points, the C_score of L35 is (840-210) / (840-72)×100≈(630 / 768)×100≈81.9 yuan≈81.9 points, and the C_score of L88 is 100 points.

[0077] The formula for calculating the overall score is Total_score = 0.6 × B_score + 0.4 × C_score. The overall scores for each load are calculated as follows: L12's Total_score = 0.6 × 100 + 0.4 × 0 = 60 points; L35's Total_score = 0.6 × 8.05 + 0.4 × 81.9 ≈ 4.83 + 32.76 ≈ 37.59 points; L88's Total_score = 0.6 × 0 + 0.4 × 100 = 40 points. Furthermore, the response time constant τ_response is used as a ranking correction factor. When the difference in overall scores between two loads is ≤ 5 points, the load with the shorter response time has higher priority. For example, between residential loads L89 (overall score 40 points, τ_response = 45s) and L88 (overall score 40 points, τ_response = 60s), L89 has a higher priority than L88.

[0078] The priority sequence for flexible load control is arranged from high to low based on the comprehensive score. If the scores are the same, they are sorted from shortest to longest response time. The sequence format is "priority ranking-load number-load type-comprehensive score-core control parameter". Based on the calculation results of 45 flexible loads, the priority sequence of the top 10 is as follows: 1-L12-Industrial Load-60 points-P_cut=500kW,E_shift=12000kWh; 2-L9-Industrial Load-58 points-P_cut=450kW,E_shift=10000kWh; 3-L89-Residential Load-40 points-P_cut=150kW,E_shift=1200kWh; 4-L88-Residential Load-40 points-P_cut=120kW,E_shift=900kWh; 5-L35-Commercial Load-37.59 points-P_cut=300kW,E_shift=3000kWh… This sequence clarifies the order of control implementation. High-priority loads are prioritized during periods of high carbon prices to ensure maximum carbon emission reduction benefits; low-priority loads are supplemented when control resources are sufficient, taking into account overall economic efficiency. The sequence also marks the optimal control period for each load (based on the dynamic carbon price curve, such as the optimal control period for L12 being 08:30:00-08:30:30), providing clear control targets and timing basis for subsequent collaborative optimization models.

[0079] S204, combining the flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, with the system carbon intensity not exceeding the limit as the core constraint, solve the multi-timescale collaborative optimization model and generate the source-grid-load-storage integrated regulation strategy to achieve real-time controllability of carbon emission intensity and collaborative optimization of system operation economy.

[0080] Specifically, the priority sequence of flexible load regulation, the real-time charging and discharging power boundary of energy storage system and the renewable energy output prediction curve can be integrated to construct a collaborative optimization model framework with three time scales: day-ahead, intraday, and real-time, and generate a multi-time scale optimization model parameter set. This step is the foundational framework building stage for collaborative optimization. Its core is to clarify the objectives and parameters for different control stages through multi-source data integration and time-scale division, providing structured input for subsequent optimization problems. The specific implementation method is as follows: The integration of the flexible load regulation priority sequence requires converting the ranking results into weight parameters and constraints that the model can recognize. First, the core information in the sequence is extracted: the priority ranking of the 45 flexible loads (rank 1 to 45), the comprehensive score (0 to 60 points), the maximum power reduction P_cut, the maximum transferable power E_shift, and the response time constant τ_response. The priority ranking is then converted into a regulation weight coefficient W_rank, with the formula W_rank = (46 - Rank) / 45. The higher the ranking (the smaller the Rank value), the higher the weight. For example, the W_rank of L12 (rank 1) is (46 - 1) / 45 = 1.0, the W_rank of L25 (rank 10) is (46 - 10) / 45 ≈ 0.79, and the W_rank of L78 (rank 45) is (46 - 45) / 45 ≈ 0.02. Meanwhile, the comprehensive score is correlated with the adjustment potential parameter to generate a subset of load parameters in the form of "load number - W_rank - P_cut - E_shift - τ_response", which serves as the core basis for load adjustment decisions in the model.

[0081] The determination of the real-time charge and discharge boundaries of the energy storage system is based on the real-time status data collected by the BMS and the physical characteristics of the equipment. The core boundary parameters include four dimensions: First, the charge and discharge power boundaries P_ess_max and P_ess_min. P_ess_max is the maximum charging power (taken as 100% of the rated charging power of the equipment, such as ESS-01 with a rated charging power of 5000kW, P_ess_max=5000kW), and P_ess_min is the negative value of the maximum discharge power (such as rated discharge power of 5000kW, P_ess_min=-5000kW). In actual operation, it needs to be dynamically adjusted according to the SOC. When SOC≤2 At 0% charge, P_ess_min is increased to -2000kW (limiting deep discharge); secondly, the state of charge (SOC) boundary is set, with a safe operating range of 20%≤SOC≤80% to avoid overcharging and over-discharging leading to battery life degradation. For example, ESS-01's current SOC=68%, which is within the safe range and can be freely charged and discharged; thirdly, there are charging and discharging switching constraints, requiring a minimum duration for switching between charging and discharging states in adjacent time segments (the interval between charging and discharging must be ≥30 seconds, and the interval between discharging and charging must be ≥60 seconds) to prevent frequent switching from damaging the battery; fourthly, there are constraints on the number of daily charging and discharging cycles, with a single energy storage station having ≤2 cycles per day to extend battery life. These boundary parameters are integrated into a subset of energy storage parameters: "Energy Storage Number - P_ess_max - P_ess_min - SOC_min - SOC_max - Switching Interval".

[0082] The generation of renewable energy output forecast curves employs a "multi-model fusion forecasting" method, balancing forecast accuracy and timeliness. The forecast input data includes historical output data (10Hz time-series data for the past 3 months), meteorological forecast data (irradiance, wind speed, and temperature for the next 24 hours, with a time resolution of 15 minutes), and equipment operating status data (PV panel cleanliness and wind turbine health). Day-ahead forecasts use a Long Short-Term Memory (LSTM) network model to predict the output curve for the next 24 hours, with a time resolution of 1 hour and a forecast accuracy (root mean square error) ≤12%. Intraday forecasts use a Gradient Boosting Tree (GBRT) model, continuously updating the output curve for the next 4 hours based on the latest meteorological data, with a time resolution of 15 minutes and improved accuracy to ≤8%. Real-time forecasts use a Kalman filter algorithm to correct for output fluctuations in the next 5 minutes, with a time resolution of 1 minute and an accuracy ≤5%. Taking the PV-03 photovoltaic power plant as an example, its output was predicted to be 3200kW from 08:00 to 09:00 the day before yesterday, and the intraday forecast was revised to 3350kW. The real-time forecast for the output from 08:30:00 to 08:35:00 fluctuated between 3250kW and 3400kW. The forecast results at different time scales were integrated into a subset of new energy parameters: "energy type - time scale - predicted output - accuracy index".

[0083] The collaborative optimization model framework across three time scales adopts a design logic of "step-by-step progression and improved accuracy": The day-ahead scale (24 hours in advance) aims to "maximize carbon emission reduction potential," formulating the next day's control plan based on coarse forecast data, with a time resolution of 1 hour, and decision variables being the daily total adjustment amount of each load and the daily charging and discharging plan of energy storage; The intraday scale (4 hours in advance) aims to "track forecast deviations," adjusting the day-ahead plan based on updated forecast data, with a time resolution of 15 minutes, and decision variables being the periodic adjustment amount of load and the periodic charging and discharging power of energy storage; The real-time scale (5 minutes in advance) aims to "eliminate real-time fluctuations," executing precise control based on second-level real-time data, with a time resolution of 1 minute, and decision variables being the instantaneous adjustment amount of load and the instantaneous power of energy storage. The final multi-timescale optimization model parameter set contains 126 parameters across three subsets: load, energy storage, and new energy. The parameter format is standardized as "parameter name-data type-unit-timescale-value range", such as "L12_W_rank-floating-unitless-full-scale-0.02-1.0" and "ESS-01_P_ess_max-integer-kW-full-scale-0-5000", providing standardized input for model construction.

[0084] In the optimization model, the real-time carbon intensity at the interface between the regional power grid and the upper-level power grid does not exceed a preset threshold as the core constraint, and energy storage operation constraints and load regulation constraints are embedded to generate an optimization problem with carbon constraints. This step is the core definition stage of the optimization problem. By clarifying the core carbon constraints and auxiliary operational constraints, the control objectives are transformed into a mathematical optimization problem description, ensuring that the control strategy meets both carbon emission reduction requirements and grid safety operation rules. The specific implementation method is as follows: The core constraint is that the real-time carbon intensity at the interface between the regional power grid and the upper-level power grid must not exceed the limit. The setting of this threshold needs to be comprehensively determined in conjunction with the carbon emission policy of the upper-level power grid, the renewable energy absorption capacity of the regional power grid, and load characteristics. First, obtain the carbon intensity control target issued by the upper-level power grid (e.g., daily average carbon intensity ≤ 0.5 kg CO2 / kWh). Combined with the peak and valley load characteristics of the regional power grid, set time-period thresholds: the threshold for peak load periods (08:00-12:00, 17:00-21:00) is 0.55 kg CO2 / kWh (with moderate relaxation allowed), the threshold for flat load periods (12:00-17:00, 21:00-23:00) is 0.5 kg CO2 / kWh, and the threshold for low load periods (23:00-08:00 the next day) is 0.45 kg CO2 / kWh (strict control, encouraging the absorption of more renewable energy). The calculation of the carbon intensity at the threshold uses real-time carbon flow tracking results, and the formula is e_grid(t)=(∑P_import(t)×e_import(t)+∑P_re(t)×e_re(t)) / (∑P_import(t)+∑P_re(t)), where P_import(t) is the power received from the upstream grid at time t, e_import(t) is the dynamic carbon intensity of the upstream grid, P_re(t) is the output of renewable energy, and e_re(t)=0 (zero carbon). The constraint expression is e_grid(t)≤e_limit(t), where e_limit(t) is the time-period threshold at time t. For example, if 08:30:00 is during the peak period, e_limit=0.55kgCO2 / kWh. If the real-time calculated e_grid=0.56kgCO2 / kWh, then a control action will be triggered.

[0085] Embedding energy storage operation constraints requires transforming physical boundaries into mathematical constraints to ensure that the model output conforms to the operating rules of energy storage devices. First, there is a power constraint: P_ess_min(t) ≤ P_ess(t) ≤ P_ess_max(t), where P_ess_min(t) and P_ess_max(t) change dynamically with SOC. For example, when SOC(t) ≤ 20%, P_ess_min(t) = -2000kW and P_ess_max(t) = 5000kW (charging is allowed but discharging is restricted). Second, there is a SOC constraint: SOC_min ≤ SOC(t) ≤ SOC_max, and SOC(t) = SOC(t-1) + (P_ess(t) × Δt) / E_ess_rated, where Δt is the time step (real-time scale Δt = 60 seconds), and E_ess_rated is the rated capacity of the energy storage (e.g., ESS-01 is 10000kWh). For example, if SOC(t-1) = 20%, P_ess(t) = 5000kW, and Δt = 60 seconds, then... SOC(t) = 20% + (5000 × 60 / 3600) / 10000 × 100% = 20% + 0.083% ≈ 20.083%, which satisfies the constraint; the third constraint is the charging and discharging switching constraint: if P_ess(t) × P_ess(t-1) < 0 (state switching), then Δt_switch ≥ T_switch (T_switch is the minimum switching interval, charging to discharging T_switch = 30 seconds). For example, at time t-1, P_ess = -3000kW (discharging), and if it is to switch to P_ess = 2000kW (charging) at time t, it is necessary to ensure that the time interval between t and t-1 is ≥ 60 seconds, otherwise the model does not allow the switching; the fourth constraint is the number of cycles: ∑|sign(P_ess(t)) - sign(P_ess(t-1))| / 2 ≤ N_cycle_max (N_cycle_max = 2), to avoid excessive cycling.

[0086] The embedding of load adjustment constraints needs to consider both the adjustment effect and user energy demand. The core constraints include four aspects: First, adjustment amount constraints: the instantaneous power reduction of a single load ≤ P_cut. For example, for L12, P_cut = 500kW, and the adjustment amount P_adjust(t) calculated by the model must satisfy 0 ≤ P_adjust(t) ≤ 500kW; the total adjustment amount of transferable electricity ≤ E_shift. For example, for L12, E_shift = 12000kWh, and the cumulative transferred electricity on a given day must not exceed this value. Second, response time constraints: the rate of change of load adjustment amount ≤ P_cut / τ_response, ensuring a smooth adjustment process. For example, for L12, τ_response = 120 seconds, P_cut = 500kW, and the rate of change of adjustment ≤ 500 / 120 ≈ 4.17kW / second, avoiding... The first constraint is the impact of sudden power changes on user equipment. The second constraint is the continuity constraint: the duration of load adjustment (reduction / non-reduction) must be ≥ τ_cont (τ_cont = 300 seconds for industrial load, 120 seconds for commercial load, and 60 seconds for residential load). For example, after L35 (commercial load) is called for reduction, it must remain for at least 120 seconds to recover. The third constraint is the user comfort constraint: the adjustment of commercial and residential loads must be controlled within the acceptable range for users. For example, the adjustment of commercial air conditioning load must not cause the indoor temperature to deviate from the set value by ±2℃, and the adjustment of residential electric heating load must not cause the room temperature to be lower than 18℃. This constraint is achieved by associating load characteristic parameters (temperature sensitivity coefficient). For example, the temperature sensitivity coefficient of L35 is 1℃ / 100kW. An adjustment of 300kW corresponds to a temperature change of 3℃, which needs to be divided into 3 time steps to avoid exceeding the limit in a single adjustment.

[0087] The carbon-constrained optimization problem takes "minimizing the total system operating cost" as its objective function. The total cost consists of three parts: first, the electricity purchase cost C_purchase, which is the cost of purchasing electricity from the upstream grid, C_purchase=∑P_import(t)×λ(t)×Δt, where λ(t) is the time-of-use electricity price (1.0 yuan / kWh during peak hours, 0.6 yuan / kWh during flat hours, and 0.3 yuan / kWh during off-peak hours); second, the load adjustment cost C_adjust, which is the adjustment incentive fee paid to flexible load users, C_adjust=∑P_adjust(t)×P_cpool(t)×Δt, where P_cpool(t) is the dynamic carbon price of the virtual carbon pool; and third, the energy storage operation cost C_ess, which includes battery depreciation cost and energy loss cost, C_ess=∑|P_ess(t)|×(0.02 yuan / kWh)×Δt (depreciation cost 0.01 yuan / kWh, energy loss cost 0.01 yuan / kWh). The objective function is expressed as MinimizeC_total = C_purchase + C_adjust + C_ess. The optimal control strategy is to be solved under the constraints of core carbon constraints and various operational constraints.

[0088] A distributed robust optimization algorithm is used to solve the carbon-constrained optimization problem. The optimal adjustment amount of each flexible load, the optimal charging and discharging plan of the energy storage system, and the renewable energy consumption scheme are calculated under multiple time sections in the future, and a preliminary coordinated control strategy is generated. This step is the core of solving the optimization problem. It uses a distributed robust optimization algorithm to handle the uncertainty of renewable energy output, and outputs specific action instructions for each regulated object while ensuring carbon constraints are met, thus forming a preliminary regulation strategy. The specific implementation is as follows: The core advantage of the distributed robust optimization algorithm lies in its balance between uncertainty handling and computational efficiency. Compared to the conservatism of traditional robust optimization and the computational complexity of stochastic optimization, this algorithm constructs an "uncertainty set" to describe the fluctuation range of renewable energy output, solves for the optimal solution in the worst-case scenario within the set, and employs a distributed computing architecture to decompose the problem into sub-nodes, thereby improving the solution speed. The algorithm's applicable scenarios are highly compatible with the requirements of this step—fluctuations in renewable energy output are the main source of uncertainty, requiring the optimization to reserve adjustment margins to ensure that carbon intensity constraints do not exceed limits under any fluctuation scenario.

[0089] The construction of the uncertainty set is the foundation of the algorithm. Centered on the predicted output value of renewable energy, the fluctuation range is determined by combining the prediction accuracy. The output fluctuation range for photovoltaic and wind power is set as [P_re_pred(t)×(1-α), P_re_pred(t)×(1+α)], where α is the fluctuation coefficient, dynamically adjusted according to the prediction time scale: day-ahead scale α=0.2 (lower prediction accuracy, reserving a 20% fluctuation margin), intraday scale α=0.1 (improved accuracy, margin reduced to 10%), real-time scale α=0.05 (highest accuracy, 5% margin). For example, if the predicted output of PV-03 at the real-time scale is P_re_pred=3300kW and α=0.05, then the uncertainty set is [3135kW, 3465kW]. The algorithm must ensure that the threshold carbon intensity does not exceed the threshold for any output value within this range. Meanwhile, the uncertainty set satisfies the constraints of "energy conservation" and "time correlation", with the output fluctuation amplitude of adjacent time sections ≤5%, avoiding extreme jump scenarios.

[0090] The algorithm's distributed computing architecture decomposes the optimization problem into a two-layer structure of "central node - sub-nodes": the central node is the regional power grid control center, responsible for handling global constraints (critical carbon intensity constraints, system power balance constraints) and coordinating with sub-nodes; the sub-nodes are divided into three categories, corresponding to flexible load clusters, energy storage power station clusters, and renewable energy power station clusters, respectively, responsible for handling their respective local constraints (such as load regulation constraints, energy storage SOC constraints). The problem decomposition adopts the "Alternating Direction Multiplier Method (ADMM)," which introduces Lagrange multipliers to coordinate global and local objectives, iteratively solving until convergence (convergence condition is that the difference between the objective functions of two adjacent iterations is ≤10^-4). Taking real-time scale optimization as an example, the decomposed subproblems include: load subproblem (solving for the optimal regulation of each load), energy storage subproblem (solving for the charging and discharging power of each energy storage unit), and new energy subproblem (determining the maximum absorption of renewable energy). The central node ensures that the global carbon constraint is met by coordinating the solutions to the subproblems.

[0091] The solution process for the objective function requires dynamic adjustment of weights based on time scale characteristics. At the day-ahead scale, minimizing electricity purchase cost is the primary objective (weight 0.6), while also considering carbon emission reduction (weight 0.4). At the intraday scale, electricity purchase cost and carbon emission reduction each have a weight of 0.5. At the real-time scale, carbon intensity control is the primary objective (weight 0.7), while also considering adjustment costs (weight 0.3). Taking the solution at a certain time segment at the real-time scale as an example, given the upstream grid carbon intensity e_import = 0.52 kgCO2 / kWh, the PV-03 output uncertainty set [3135 kW, 3465 kW], the receiving power at the gate P_import = 6000 kW, and the current carbon intensity e_grid = 0.56 kgCO2 / kWh (exceeding the threshold of 0.55), the receiving power needs to be reduced through load shedding and energy storage discharge. After 5 iterations, the algorithm converges and outputs the optimal solution: High-priority load L12 is reduced by 500kW (full capacity regulation), L9 is reduced by 450kW, for a total load regulation of 950kW; ESS-01 discharges 1000kW, ESS-02 discharges 50kW, for a total energy storage discharge of 1050kW; renewable energy is fully utilized at 3465kW (taking the upper limit of the uncertainty set to maximize zero-carbon energy utilization). After adjustment, P_import decreases to 6000-950-1050=4000kW, e_grid=(4000×0.52+3465×0) / 7465≈2080 / 7465≈0.279kgCO2 / kWh, which is below the threshold. Simultaneously, the total cost increases by 210 yuan (regulation incentive + energy storage cost), achieving a balance between carbon constraints and economic efficiency.

[0092] The preliminary coordinated control strategy is a structured presentation of the algorithm's solution results, divided into three categories according to the time scale: the day-ahead control strategy is a 24-hour time-period instruction, in the format of "time period - total load adjustment - energy storage charging and discharging plan - power purchase plan", such as "08:00-09:00 - load reduction of 2000kW - ESS charging of 1000kW - power purchase of 5000kW"; the intraday control strategy is a 4-hour 15-minute level instruction, supplementing and refining the day-ahead plan, such as "08:00-08:15 - L12 reduction of 300kW - ESS-01 discharge of 500kW"; the real-time control strategy is a 5-minute level instruction, accurately tracking fluctuations, such as "08:30:00-08:31:00 - L12 maintains reduction of 500kW - ESS-01 discharge power adjusted to 1000kW". The strategy also includes contingency plans for dealing with uncertainties. For example, if the PV output is lower than the predicted lower limit of 3135kW, the secondary priority load L89 will be automatically activated to reduce the output by 150kW to ensure that the carbon intensity does not exceed the limit.

[0093] The initial coordinated control strategy is verified and dynamically adjusted to form the final executable integrated source-grid-load-storage control instruction set, achieving the coordinated goal of full controllable carbon emission intensity and optimal total system operating cost.

[0094] This step is a crucial guarantee for the implementation of the control strategy. By eliminating potential risks through safety verification and dynamically adjusting the strategy based on real-time data, it ensures the executability of instructions and the safety of grid operation, ultimately achieving the synergistic goal of carbon emission reduction and economic efficiency. The specific implementation method is as follows: The core content of the safety verification includes three main categories: power flow verification, voltage level verification, and frequency stability verification, which are simulated and calculated using professional power system analysis software. Power flow verification uses the Newton-Raphson method to calculate the real-time power flow distribution of each line, ensuring that the power transmitted by each line does not exceed 80% of its rated capacity (a safety margin of 20%). For example, line L1-05, connecting junction G1 and load L12, has a rated capacity of 8000kW. The verification found that its power flow in the initial strategy was 7200kW (90% of rated capacity), exceeding the safety margin and requiring adjustment. Voltage level verification ensures that the voltage deviation of all nodes is within ±5% of the rated voltage. For example, the voltage of a 10kV load node needs to be maintained between 9.5kV and 10.5kV. If the voltage of a node drops to 9.3kV, the voltage needs to be increased through reactive power regulation using energy storage. Frequency stability verification ensures that the system frequency is maintained at 50Hz ±0.2Hz. When renewable energy output fluctuates or the load changes abruptly, the frequency deviation must not exceed this range; otherwise, emergency regulation will be triggered.

[0095] The dynamic adjustments after verification adopt the principle of "priority-based adjustment," prioritizing adjustments to the lowest-cost loads to avoid significantly increasing the overall cost. For power flow exceeding limits, the adjustment order is as follows: first, increase local renewable energy consumption (zero cost); second, adjust energy storage charging and discharging power (low cost); and finally, adjust the regulation of low-priority loads (higher cost). Taking the power flow exceeding limit of line L1-05 as an example, the initial strategy indicates a power flow of 7200kW, a rated capacity of 8000kW, and a safety limit of 6400kW, requiring a reduction of 800kW in power flow. First, we attempted to increase the output of the wind power WT-02 within the power supply range of this line from the predicted 2800kW to 3000kW (to absorb local wind power and reduce the power received), thus reducing the power flow to 7000kW. Second, we adjusted the energy storage ESS-03 in this area from charging at 500kW to discharging at 300kW, further reducing the power flow to 6700kW. Finally, we called upon the low-priority load L78 (ranked 45th, W_rank=0.02) supplied by this line to reduce its power flow by 300kW, thus reducing the power flow to 6400kW, meeting safety constraints. The total cost of the adjustment only increased by 300 × 0.02 yuan / kWh × 1 hour = 6 yuan, which is a very small increase in cost.

[0096] To address the low voltage issue, a combined strategy of "energy storage reactive power regulation + load power factor adjustment" is adopted. The energy storage system has four-quadrant operation capability, which can increase voltage by adjusting reactive power. For example, if the voltage at a certain node is 9.3kV, the ESS-04 is instructed to output 100kvar of capacitive reactive power, increasing the voltage to 9.6kV. If this is still insufficient, the power factor of the industrial load at that node is adjusted from 0.85 to 0.95 to reduce reactive power consumption, further increasing the voltage to 9.8kV and meeting the constraint. To address the frequency deviation issue, a frequency response threshold is set. When the frequency is ≤49.8Hz, the energy storage system automatically switches to discharge mode (increasing active power output), and the load automatically reduces high-priority regulation by 10%. When the frequency is ≥50.2Hz, the energy storage switches to charging mode, the load reduction pauses, and some regulation is restored, quickly smoothing out frequency fluctuations.

[0097] The final integrated source-grid-load-storage control command set adopts a "hierarchical" command format, divided into three levels: control center level, site level, and equipment level. Control center level commands are global coordination commands, issued to each site, with the format "timestamp-command type-global target," such as "2025-10-10 08:30:00-real-time control-critical carbon intensity ≤ 0.55 kgCO2 / kWh, frequency 50±0.2Hz." Site level commands are subsystem commands, issued to load aggregators, energy storage power stations, and new energy power stations, with the format "timestamp-site number-regulation task," such as "2025-10-10 08:30:00-load aggregator A-coordinate L12 reduction of 500kWh." "W, L9 reduce 450kW"; "2025-10-10 08:30:00-ESS-01-Discharge 1000kW, SOC maintained ≥65%"; Device-level instructions are specific execution instructions issued to terminal devices, in the format of "timestamp-device number-action parameter", such as "2025-10-10 08:30:00-L12-Cooling system power reduced from 2000kW to 1500kW" and "2025-10-10 08:30:00-ESS-01-Charging and discharging power set to -1000kW". The instruction set uses an encrypted communication protocol for transmission, with a transmission delay of ≤100ms, ensuring real-time performance and security.

[0098] The post-implementation effect evaluation and closed-loop optimization mechanism assesses the control effect by real-time monitoring of three dimensions: carbon intensity at the control point, system cost, and grid operation status. Carbon intensity assessment uses a "moving average" method to calculate the 15-minute average carbon intensity, ensuring that it does not exceed the time-period threshold. For example, the average carbon intensity from 08:00 to 08:15 is 0.54 kg CO2 / kWh, meeting the peak-period threshold of 0.55. Economic assessment calculates the deviation between the actual total cost and the optimization target cost; a deviation ≤5% is considered acceptable. Safety assessment counts the number of voltage, frequency, and power flow exceedances; exceedances ≤0.1 times / hour are considered compliant. If a certain indicator fails to meet the standard, the cause is analyzed, and the optimization model parameters for the next cycle are adjusted. For example, if carbon intensity remains consistently high, the weight of carbon constraints in the objective function is increased, forming a closed-loop control mechanism of "optimization-execution-evaluation-adjustment." Ultimately, this achieves the synergistic optimization goal of real-time controllable carbon emission intensity (carbon intensity compliance rate ≥99%) and system operation economy (total cost reduction ≥8%).

[0099] As can be seen, by collecting real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals from the upper-level power grid, and using a source-load-storage full-process carbon flow tracking algorithm based on the dynamic carbon emission intensity signals, a load-side carbon emission intensity spectrum with spatiotemporal resolution is generated. Based on the load-side carbon emission intensity spectrum, a virtual carbon pool with carbon cost as a weighting factor is established, and a flexible load regulation priority sequence is generated. Combining the flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, and taking the system carbon intensity not exceeding the limit as the core constraint, a multi-timescale collaborative optimization model is solved, and an integrated source-grid-load-storage regulation strategy is generated. This enables refined measurement and proactive regulation of load-side carbon emissions, improving the system's response capability to dynamic carbon constraints.

[0100] Another embodiment of the present invention provides a demand-side resource and energy storage coordinated regulation system based on carbon index constraints, see [link to relevant documentation]. Figure 3 The system may include: The data acquisition module 301 is used to acquire real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals issued by the upper-level power grid. The generation module 302 is used to generate a load-side carbon emission intensity spectrum with spatiotemporal resolution based on the dynamic carbon emission intensity signal and through a source-load-storage full-process carbon flow tracing algorithm. Module 303 is used to establish a virtual carbon pool with carbon cost as a weighting factor and generate a flexible load regulation priority sequence based on the load-side carbon emission intensity spectrum and the demand-side resource adjustability potential model. The regulation module 304 is used to combine the flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, with the system carbon intensity not exceeding the limit as the core constraint, to solve the multi-time-scale collaborative optimization model and generate an integrated source-grid-load-storage regulation strategy, so as to achieve real-time controllability of carbon emission intensity and collaborative optimization of system operation economy.

[0101] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0102] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0103] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0104] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. A method for coordinated regulation of demand-side resources and energy storage based on carbon index constraints, characterized in that, The method includes: Collect real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals issued by the upper-level power grid in the region; Based on the dynamic carbon emission intensity signal, a load-side carbon emission intensity spectrum with spatiotemporal resolution is generated by a source-load-storage full-process carbon flow tracking algorithm. Based on the load-side carbon emission intensity spectrum and combined with the demand-side resource adjustability potential model, a virtual carbon pool with carbon cost as the weighting factor is established and a flexible load regulation priority sequence is generated. By combining the aforementioned flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, and taking the system carbon intensity not exceeding the limit as the core constraint, a multi-timescale collaborative optimization model is solved and an integrated source-grid-load-storage regulation strategy is generated to achieve real-time controllability of carbon emission intensity and collaborative optimization of system operation economy.

2. The method according to claim 1, characterized in that, The data collected includes real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals from the upper-level power grid. The system collects power data of load nodes in real time by deploying IoT sensors at various nodes of the regional power grid, obtains real-time output data of renewable energy power plants through the energy management system, and collects state of charge and charging / discharging power data of energy storage devices through the battery management system, generating a raw heterogeneous dataset. The original heterogeneous dataset is timestamped and cleaned to unify data with different sampling frequencies to the same time base, and bad data caused by abnormal jump points and communication interruptions is removed to generate a cleaned and aligned time series dataset. The system receives dynamic carbon emission intensity signals from the upper-level power grid through the power dispatch data network. These signals are updated at a minute-level frequency and are associated with the regional power grid gateway metering points to generate dynamic carbon emission intensity time-series signals. The cleaned and aligned time-series dataset is fused with the dynamic carbon emission intensity time-series signal and encapsulated according to a standardized data model to generate a standardized multi-source dataset for subsequent carbon flow tracing analysis.

3. The method according to claim 2, characterized in that, The process of generating a load-side carbon emission intensity spectrum with spatiotemporal resolution based on the dynamic carbon emission intensity signal using a source-load-storage full-process carbon flow tracing algorithm includes: The dynamic carbon emission intensity time series signal in the standardized multi-source dataset is analyzed, and a carbon flow network topology is established with the regional power grid gateway as the starting point and the load node as the ending point, generating a carbon flow network model. Based on the carbon flow network model and real-time power flow data, and using the proportional sharing principle, the distribution of electrical carbon flow from the generation side to the load side is calculated, and a node carbon flow intensity matrix is ​​generated. Establish a carbon state transition model for energy storage devices, track the carbon emissions corresponding to the electrical energy absorbed during the charging period of energy storage, and perform carbon flow tracing and allocation during the discharging period to generate an energy storage carbon flow distribution matrix; By combining the carbon flow intensity matrix of the integrated nodes and the carbon flow distribution matrix of the energy storage, a mixed-integer linear programming algorithm is used to accurately allocate carbon emissions to each load node and time segment, ultimately generating a load-side carbon emission intensity spectrum with spatiotemporal resolution.

4. The method according to claim 3, characterized in that, The step of establishing a virtual carbon pool with carbon cost as a weighting factor and generating a flexible load regulation priority sequence based on the load-side carbon emission intensity spectrum and the demand-side resource adjustability potential model includes: Analyze the carbon emission intensity spectrum on the load side, calculate the marginal carbon emission intensity of each flexible load node at different time sections, and convert it into carbon cost per unit of electricity to generate a carbon cost vector. The demand-side resource adjustability potential model is invoked. Based on historical operating data and load characteristics, this model quantifies the power reduction, transferable electricity, and response time constant of various flexible loads, and generates an adjustability potential mapping table. Using the carbon cost vector as the core weighting factor and combining it with the adjustable potential mapping table, a pricing and settlement mechanism for the virtual carbon pool is designed, and a dynamic carbon price curve that correlates carbon cost with adjustment value is established. Based on the dynamic carbon price curve and the adjustment costs of each flexible load, a priority sequence for flexible load regulation is generated, which is optimally ranked in terms of both carbon emission reduction benefits and economic efficiency.

5. The method according to claim 4, characterized in that, The method combines the flexible load regulation priority sequence with the real-time charge and discharge boundary of the energy storage system, taking the system carbon intensity not exceeding the limit as the core constraint, to solve a multi-timescale collaborative optimization model and generate an integrated source-grid-load-storage regulation strategy. This achieves real-time controllability of carbon emission intensity and collaborative optimization of system operation economy, including: By integrating the priority sequence of flexible load regulation, the real-time charging and discharging power boundary of the energy storage system, and the renewable energy output prediction curve, a collaborative optimization model framework with three time scales—day-ahead, intraday, and real-time—is constructed, generating a multi-time-scale optimization model parameter set. In the optimization model, the real-time carbon intensity at the interface between the regional power grid and the upper-level power grid does not exceed a preset threshold as the core constraint, and energy storage operation constraints and load regulation constraints are embedded to generate an optimization problem with carbon constraints. A distributed robust optimization algorithm is used to solve the carbon-constrained optimization problem. The optimal adjustment amount of each flexible load, the optimal charging and discharging plan of the energy storage system, and the renewable energy consumption scheme are calculated under multiple time sections in the future, and a preliminary coordinated control strategy is generated. The initial coordinated control strategy is verified and dynamically adjusted to form the final executable integrated source-grid-load-storage control instruction set, achieving the coordinated goal of full controllable carbon emission intensity and optimal total system operating cost.

6. A demand-side resource and energy storage coordinated regulation system based on carbon index constraints, characterized in that, The system includes: The data acquisition module is used to collect real-time load data, renewable energy output data, energy storage device status data, and dynamic carbon emission intensity signals issued by the upper-level power grid. The generation module is used to generate a load-side carbon emission intensity spectrum with spatiotemporal resolution based on the dynamic carbon emission intensity signal and through a source-load-storage full-process carbon flow tracing algorithm. A module is established to create a virtual carbon pool with carbon cost as a weighting factor and generate a flexible load regulation priority sequence based on the load-side carbon emission intensity spectrum and the demand-side resource adjustability potential model. The control module is used to combine the flexible load control priority sequence with the real-time charge and discharge boundary of the energy storage system, with the system carbon intensity not exceeding the limit as the core constraint, to solve the multi-timescale collaborative optimization model and generate an integrated source-grid-load-storage control strategy, so as to achieve real-time controllability of carbon emission intensity and collaborative optimization of system operation economy.

7. The system according to claim 6, characterized in that, The acquisition module is specifically used for: The system collects power data of load nodes in real time by deploying IoT sensors at various nodes of the regional power grid, obtains real-time output data of renewable energy power plants through the energy management system, and collects state of charge and charging / discharging power data of energy storage devices through the battery management system, generating a raw heterogeneous dataset. The original heterogeneous dataset is timestamped and cleaned to unify data with different sampling frequencies to the same time base, and bad data caused by abnormal jump points and communication interruptions is removed to generate a cleaned and aligned time series dataset. The system receives dynamic carbon emission intensity signals from the upper-level power grid through the power dispatch data network. These signals are updated at a minute-level frequency and are associated with the regional power grid gateway metering points to generate dynamic carbon emission intensity time-series signals. The cleaned and aligned time-series dataset is fused with the dynamic carbon emission intensity time-series signal and encapsulated according to a standardized data model to generate a standardized multi-source dataset for subsequent carbon flow tracing analysis.

8. The system according to claim 7, characterized in that, The generation module is specifically used for: The dynamic carbon emission intensity time series signal in the standardized multi-source dataset is analyzed, and a carbon flow network topology is established with the regional power grid gateway as the starting point and the load node as the ending point, generating a carbon flow network model. Based on the carbon flow network model and real-time power flow data, and using the proportional sharing principle, the distribution of electrical carbon flow from the generation side to the load side is calculated, and a node carbon flow intensity matrix is ​​generated. Establish a carbon state transition model for energy storage devices, track the carbon emissions corresponding to the electrical energy absorbed during the charging period of energy storage, and perform carbon flow tracing and allocation during the discharging period to generate an energy storage carbon flow distribution matrix; By combining the carbon flow intensity matrix of the integrated nodes and the carbon flow distribution matrix of the energy storage, a mixed-integer linear programming algorithm is used to accurately allocate carbon emissions to each load node and time segment, ultimately generating a load-side carbon emission intensity spectrum with spatiotemporal resolution.

9. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-5 when it is run.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1-5.

Images