An EV classification cooperative scheduling method and system considering carbon electric coupling and dynamic cluster

Abstract

The present application relates to an EV classification cooperative scheduling method and system considering carbon electric coupling and dynamic cluster, the method comprising: collecting environmental state observation data and packaging it as an environmental state observation vector; based on the environmental state observation vector, a flexible signing EV agent constructs a state space and an action space to set a comprehensive benefit function, iteratively optimizes the agent model according to the comprehensive benefit function, outputs the optimal charging and discharging power curve, and calculates the real-time net load shortage power; through the global static data review and cluster reconstruction of the global fixed signing vehicles in the set period by the fixed signing EV aggregator, different priority scheduling clusters are obtained, the real-time net load shortage power is used to issue and correct the macro power quota of each scheduling cluster, the final execution quota is formed, and the charging and discharging power instruction is issued until the total power accumulated in the current distribution reaches the final execution quota of each scheduling cluster.

Description

Technical Field

[0001] This invention relates to the field of collaborative scheduling technology, and in particular to an EV classification collaborative scheduling method and system that takes into account carbon-electric coupling and dynamic clustering. Background Technology

[0002] With the advancement of new power system construction and the large-scale integration of distributed photovoltaic and other renewable energy sources into microgrids, the net load fluctuation characteristics of the system are becoming increasingly significant. In microgrid scenarios without independent fixed energy storage systems (ESS), large-scale access to electric vehicles (EVs), as flexible mobile energy storage resources with bidirectional charging and discharging capabilities, is gradually becoming a key support for smoothing grid fluctuations and promoting the local consumption of new energy. To effectively manage and utilize massive EV resources, V2G collaborative optimization scheduling technology is widely used. Its core is to guide EV clusters to adjust their charging and discharging behavior through price or command mechanisms to match the grid's operational needs.

[0003] In existing vehicle-to-grid (V2G) interaction solutions, the scheduling architecture for EVs mainly includes two basic paradigms: centralized aggregation and decentralized response. In centralized aggregation schemes, the aggregator or microgrid control center typically collects the battery status, charging / discharging power boundaries, and travel constraints of all EVs within the region, establishing a global optimization model such as Mixed Integer Linear Programming (MILP) to uniformly generate and distribute charging / discharging strategies. In decentralized response schemes, the microgrid scheduling center primarily issues economic incentive signals such as time-of-use pricing, allowing each EV's built-in agent model to respond autonomously in a decentralized manner based on its own goal of minimizing electricity costs. In recent years, some solutions have begun to combine these two architectures, using master-slave game theory or bi-level programming models to achieve initial allocation of control. At the specific aggregation control and underlying scheduling power allocation levels, most existing optimal resource allocation mechanisms employ single-dimensional priority ranking or greedy algorithms. When aggregators face real-time dispatch instructions from the power grid, their conventional processing logic is as follows: A single-dimensional objective function is constructed, and the system, based on ranking scores from highest to lowest, sequentially calls upon the group of vehicles with the most abundant power or the lowest cost to participate in the response until the power deficit of the power grid is met. Furthermore, in terms of incentivizing distributed photovoltaic (PV) power consumption, existing conventional mechanisms primarily rely on a single load matching lever to achieve load-side following and consumption of PV power.

[0004] However, existing traditional vehicle-to-grid (V2G) interactive scheduling solutions, such as centralized global optimization, rely on the unified information collection of massive nodes. In operational scenarios with large-scale access and varying vehicle states, they often encounter the curse of dimensionality and communication bottlenecks, resulting in excessively long computation times and difficulty in supporting real-time and accurate responses to microgrid fluctuations. To address this, related research has proposed decentralized autonomous demand response methods, enabling vehicles to determine their charging and discharging strategies based on microgrid demand signals. However, while decentralized solutions can reduce computational overhead, in microgrid scenarios without independent fixed energy storage configurations and relying solely on electric vehicles as the only flexible adjustment resource, the lack of robust centralized control and fallback mechanisms can easily lead to inadequate execution of peak shaving and valley filling commands, resulting in the safety risk of microgrid power mismatch. On the other hand, existing scheduling and guidance mechanisms lack quantitative indicators for green energy consumption, making it difficult to effectively mobilize vehicles' proactive charging responses during periods of high solar power generation. Furthermore, most solutions rely on a single-dimensional greedy sorting based on vehicle charge state or marginal cost when performing aggregated scheduling and allocation, leading to frequent and deep calls to nodes with good battery conditions, which can easily accelerate the excessive wear and tear effect of single-vehicle battery degradation. In terms of ensuring overall available transportation capacity, the traditional absolute priority ranking lacks a macro-level fault tolerance and quota constraint mechanism, which leads to newly connected vehicles or vehicles with low initial scores losing scheduling opportunities for a long time. The system struggles to balance computational efficiency, battery life protection, and fairness of cluster participation. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention aims to provide an EV classification and collaborative scheduling method and system that considers carbon-electric coupling and dynamic clustering. It employs a "flexible pilot response + fixed aggregation fallback" architecture to improve scheduling computation efficiency and command execution determinism. It introduces a dynamic carbon emission factor coupled with a load matching coefficient to form an active absorption and cluster promotion process, enhancing low-carbon absorption capacity during periods of high photovoltaic power generation. To prevent excessive wear and tear on high-scoring vehicles and loss of new users due to single-index ranking, it introduces a multi-dimensional dynamic hierarchical and anti-starvation mechanism, and designs a tiered quota scheduling algorithm with overflow compensation. Through forced sampling quotas, it ensures scheduling opportunities for long-tail clusters, and combines an overflow difference adaptive upward transfer mechanism, enabling power allocation to dynamically match the real-time physical capacity status. This achieves the effects of improving the reliability of control commands, reducing excessive wear and tear on high-quality batteries, and eliminating hierarchical solidification in microgrid environments without independent energy storage. It realizes the synergistic optimization of computing power reduction and low-carbon response, providing a fairer and more reliable technical foundation for refined management and control of vehicle-grid interaction under new power systems.

[0006] To achieve the above objectives, the present invention provides the following solution: An EV classification and cooperative scheduling method considering carbon-electric coupling and dynamic clustering includes: Historical base load data and historical output data of distributed power sources such as photovoltaics within the microgrid are collected and encapsulated into environmental status observation vectors. Based on the flexible contract EV agent, the state space and action space are constructed using the environmental state observation vector to set the comprehensive benefit function. The agent model is iteratively optimized according to the comprehensive benefit function, the optimal charging and discharging power curve is output, and the real-time net load deficit power is calculated. By having fixed-contract EV aggregators perform global static data review and cluster reconstruction on all fixed-contract vehicles across the network at set intervals, scheduling clusters with different priorities are obtained. The real-time net load deficit power is used to issue and correct macro power quotas to each scheduling cluster to form the final execution quota. At the same time, charging and discharging power instructions are issued until the total power currently allocated reaches the final execution quota of each scheduling cluster.

[0007] Optionally, the environmental state observation vector includes: Based on the historical base load data and the historical output data, the power of future scheduling cycles is predicted, and the photovoltaic power prediction for the target future period is output. At the same time, time-series modeling is performed on the non-EV base load, and the base load prediction value for the target future period is output. The initial net load of the microgrid in the target future period is calculated using the photovoltaic predicted power and the base load predicted value, and a dynamic load matching coefficient is generated by combining the baseline operating status of the upper-level power grid. Based on the network topology and power flow distribution of the microgrid, the node carbon potential of the nodes in the microgrid in the target future time period is calculated, and the node carbon potential is weighted and averaged to obtain the dynamic carbon emission factor of the entire microgrid area in the target future time period. The dynamic load matching coefficient, the dynamic carbon emission factor, and the global low-carbon guidance factor are encapsulated into the environmental state observation vector.

[0008] Optionally, setting the comprehensive return function includes: ; ; ; ; ; in, For the comprehensive benefit function, For grid load matching gain, Gain control for low-carbon consumption. This is a penalty for physical battery wear and tear. As a penalty for range anxiety, For the agent in time period The charging and discharging power of the decision. This is the dynamic load matching factor. For charge / discharge time, The dynamic carbon emission factor issued by the microgrid dispatch center. As the benchmark carbon emission factor for the power grid, As a global low-carbon guiding factor, For vehicles Current state of charge, For the expected off-line time, For the desired off-grid target power, , These are the weighting coefficients.

[0009] Optionally, calculating the real-time net load deficit power includes: The optimal charge / discharge power curve is superimposed with the initial net load to calculate the final real-time net load deficit power for this period: ; in, This represents the final real-time net load deficit power for this period. This indicates that the autonomous charging of flexible EVs has not yet fully absorbed photovoltaic power, and there is still a surplus within the microgrid. Aggregators are needed to organize fixed EVs for aggregated charging. This indicates that even after the flexible EV response, there is still a power supply gap in the grid, requiring aggregators to organize fixed EVs for aggregated discharge. To achieve the optimal charge / discharge power curve, This represents the initial net load of the microgrid.

[0010] Optionally, fixed-contract EV aggregators can perform a global static data review and cluster reconstruction of all fixed-contract vehicles across the network at set intervals, including: The fixed-contract EV aggregator extracts the operation log data of each vehicle in the fixed-contract vehicle pool within a set period, and extracts the physical quantities of the core evaluation dimensions in the operation log data; the physical quantities include: historical carbon emission reduction, response participation and battery health. The physical quantities are converted into normalized indices, and a dynamic reward and punishment factor based on conditional branch triggering is introduced to calculate the final comprehensive score of each fixed contracted EV. The final comprehensive scores of all fixed-contract EVs across the network are sorted in descending order, and the vehicle list is divided into scheduling clusters of different priorities using a set score threshold or proportion threshold; the scheduling clusters include: core scheduling cluster, regular scheduling cluster and backup scheduling cluster.

[0011] Optionally, the final overall score for each fixed-contract EV includes: ; in, The final overall score for each contracted EV. As a positive reward factor, if a vehicle is detected If a system remains online for three consecutive days but is not selected for scheduling, the wear-and-tear compensation mechanism will be triggered, granting it a positive reward factor. As a negative penalty factor, if a vehicle is detected If a system rejects or interrupts a scheduled instruction three times consecutively, a breach of trust penalty mechanism will be triggered, imposing a negative penalty factor. , , These are normalized indicators corresponding to historical carbon emission reductions, response participation, and battery health, respectively. The preset objective weighting coefficients satisfy... This is used to balance the importance of low-carbon contribution, scheduling reliability, and battery physical safety.

[0012] Optionally, forming the final execution quota includes: By polling the current physical status of all fixed-contract EVs in the network through the vehicle network gateway, and removing vehicles that are not currently plugged in and online, have malfunctioned, or whose current state of charge is not greater than the first target threshold and are in the process of discharging, or whose current state of charge is not less than the second target threshold and are in the process of charging, a filtered pool of available vehicles in real time is obtained. The maximum physical available response power of the core scheduling cluster, the regular scheduling cluster, and the backup scheduling cluster is calculated by summing the filtered real-time available vehicle pools during this period. A tiered basic scheduling ratio matrix is ​​then introduced to proportionally decompose the total shortage power during this period to obtain the initial target quota. The initial target quota includes: the initial quota of the core cluster, the initial quota of the regular cluster, and the initial quota of the backup cluster. The final execution quota is obtained by verifying and transferring the overflow compensation mechanism using the maximum available physical response power and the initial target quota.

[0013] Optionally, obtaining the final execution quota includes: Backup scheduling cluster verification: The maximum physical available response power of the backup scheduling cluster during this period is compared with the initial quota of the core cluster. If the maximum physical available response power of the backup scheduling cluster during this period is not less than the initial quota of the core cluster, it proves that the actual execution quota of the backup scheduling set has not overflowed. If the maximum physical available response power of the backup scheduling cluster during this period is less than the initial quota of the core cluster, it proves that the backup scheduling set cannot meet the demand, and the resulting first difference in capacity will be extracted and added to the tasks of the regular scheduling cluster. Regular scheduling cluster verification: If the first difference capacity exists, the initial quota of the regular cluster is updated using the first difference capacity, and the maximum physical available response power of the regular scheduling cluster during this period is compared with the updated initial quota of the regular cluster. If the maximum physical available response power of the regular scheduling cluster during this period is not less than the updated initial quota of the regular cluster, it proves that the actual execution quota of the regular scheduling cluster has not overflowed. If the maximum physical available response power of the regular scheduling cluster during this period is less than the updated initial quota of the regular cluster, it proves that the actual execution quota of the regular scheduling cluster cannot meet the demand, and the resulting second difference capacity will be extracted and added to the tasks of the core scheduling cluster. Core scheduling cluster verification: If the second difference in capacity exists, the initial quota of the core cluster is updated using the second difference in capacity, and the maximum physical available response power of the core scheduling cluster during this period is compared with the updated initial quota of the core cluster. If the maximum physical available response power of the core scheduling cluster during this period is not less than the updated initial quota of the core cluster, it proves that the actual execution quota of the core scheduling cluster has not overflowed. If the maximum physical available response power of the core scheduling cluster during this period is less than the updated initial quota of the core cluster, it proves that the actual execution quota of the core scheduling cluster cannot meet the demand, and the resulting third difference in capacity will be extracted, triggering a secondary global backfill mechanism. When the third difference capacity is greater than the target threshold, the uncompleted indicators are looped back to the cluster with remaining capacity. If the remaining capacity of the regular scheduling cluster is available, it is filled first. If the regular scheduling cluster is full or there is still a gap after filling, the remaining gap is added down to the remaining capacity of the backup scheduling cluster to obtain the final execution quota. If the limit difference capacity after looping backfilling is still greater than the sum of all remaining capacities at the bottom layer, it indicates that the scheduling instruction issued by the microgrid has exceeded the physical response limit of the current fixed contract EV cluster. Based on the fixed contract EV aggregator, a full capacity load instruction is issued, and the final gap that cannot be smoothed out is reported to the microgrid scheduling center in real time.

[0014] Optionally, issuing the charging / discharging power command until the currently allocated total power accumulates to the final execution quota of each scheduling cluster includes: Within each scheduling cluster, the final comprehensive score of each vehicle node is retrieved and sorted in descending order of the final comprehensive score to generate an absolute scheduling execution sequence within the cluster. According to the absolute scheduling execution sequence, charging and discharging power instructions that reach their physical limit boundaries are issued to each vehicle node from top to bottom until the total power currently allocated reaches the final execution quota of each scheduling cluster. After the charging and discharging scheduling action is completed, the battery physical aging parameters caused by discharge, the tiered excess loss subsidy for a single scheduling, and the dynamic collaborative incentive points finally obtained by each fixed contracted EV are calculated to update the comprehensive evaluation accumulation value of the fixed contracted vehicle node.

[0015] To achieve the above objectives, the present invention also provides an EV classification and collaborative scheduling system that considers carbon-electric coupling and dynamic clustering, comprising: The observation vector encapsulation module is used to collect historical base load data and historical output data of distributed power sources such as photovoltaics within the microgrid, and encapsulate them into environmental status observation vectors. The revenue function design module is used to construct a state space and action space based on the environmental state observation vector of the flexible contracted EV agent, so as to set the comprehensive revenue function; The power deficit acquisition module is used to iteratively optimize the agent model according to the comprehensive benefit function, output the optimal charging and discharging power curve, and calculate the real-time net load deficit power. The collaborative scheduling module is used to perform global static data review and cluster reconstruction of all fixed-contract EV vehicles in the network at a set period through fixed-contract EV aggregators to obtain scheduling clusters with different priorities. It also uses the real-time net load deficit power to issue and correct macro power quotas to each scheduling cluster to form the final execution quota. At the same time, it issues charging and discharging power instructions until the current total allocated power accumulates to reach the final execution quota of each scheduling cluster.

[0016] The beneficial effects of this invention are as follows: This invention effectively mitigates the disaster of computing power dimensionality and improves the reliability of scheduling execution: the dual-track decoupling architecture avoids communication delays and computing power explosions caused by directly incorporating massive unordered nodes into the global optimization solution matrix; at the same time, the fixed aggregation fallback mechanism compensates for the uncertainty brought about by pure decentralized response and ensures the absolute suppression of real-time power shortage in the microgrid.

[0017] This invention prevents excessive wear and breaks down the rigidity of cluster hierarchy: by combining SOH and anti-exploitation bonus mechanisms, it prevents high-frequency and deep utilization of a single high-quality vehicle with good battery condition from the source of physical regulation; it ensures the wear balance and life extension of the overall battery assets of the system, and maintains the long-term stickiness of microgrid users.

[0018] This invention addresses the "Matthew effect" in scheduling while balancing physical constraints with system fairness: the innovative overflow compensation quota algorithm not only forces low-priority / newly registered vehicles to accept orders, but also ensures that when some vehicles have insufficient power, the overall peak shaving instructions of the microgrid can still be fully satisfied by the high-priority clusters through the differential adaptive upward transfer mechanism, thereby enhancing the system's robustness.

[0019] This invention achieves a closed-loop carbon-electric incentive system, enhancing users' willingness to engage in low-carbon interactions: the introduction of dynamic carbon factors transforms the potential for on-site photovoltaic consumption into visible benefits; while the underlying nonlinear weighted state integral update and the step-by-step physical loss algorithm exemption replace the extremely complex physical fatigue hard constraint equations of large-scale battery arrays with a unified mathematical mapping function, and utilizes the integral feedback mechanism at the algorithm level to guide edge hardware nodes to actively optimize their charging and discharging behavior, providing a highly robust automated control paradigm for the large-scale collaboration of massive distributed vehicle-to-grid interactions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a framework diagram of an EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to an embodiment of the present invention; Figure 2 This is a flowchart of an EV classification and collaborative scheduling method that takes into account carbon-electric coupling and dynamic clustering, according to an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1-2As shown, this embodiment discloses an EV classification and collaborative scheduling method that considers carbon-electric coupling and dynamic clustering. The method includes: collecting historical base load data and historical output data of distributed power sources such as photovoltaics within the microgrid, and encapsulating them into an environmental state observation vector; constructing a state space and action space based on the environmental state observation vector using a flexible contracted EV agent to set a comprehensive benefit function; iteratively optimizing the agent model according to the comprehensive benefit function to output the optimal charging and discharging power curve and calculate the real-time net load deficit power; and obtaining scheduling clusters of different priorities by having a fixed-contracted EV aggregator perform global static data review and cluster reconstruction on all fixed-contracted vehicles in the network at a set period. Then, using the real-time net load deficit power, macro-level power quotas are issued and corrected for each scheduling cluster to form the final execution quota. Simultaneously, charging and discharging power instructions are issued until the currently allocated total power accumulates to reach the final execution quota of each scheduling cluster.

[0025] S1. Dynamic generation of carbon-electric dual-track guidance signal: The microgrid dispatch center is responsible for executing this step. Its core task is to construct a dynamic guidance signal reflecting the system's carbon footprint and power shortage based on the real-time matching of renewable energy output and load within the microgrid. This signal is then broadcast as a carbon-electricity dual-track guidance signal to all flexible and fixed-contract EV aggregators across the network. The specific process is as follows: S1.1 Microgrid Basic Data Prediction and Net Load Calculation: (1) Before the start of the day-ahead or intraday scheduling cycle, the microgrid dispatch center collects historical basic load data and historical output data of distributed power sources such as photovoltaics within the microgrid.

[0026] (2) The dispatch center predicts the power of future dispatch cycles and outputs the power for future time periods. Photovoltaic power forecast Similarly, time-series modeling is performed on the non-EV base load to output future time periods. Basic load forecast .

[0027] (3) Based on the obtained basic load forecast, calculate the microgrid in the time period. Initial net load : ; when When this occurs, it indicates that there is a surplus of green electricity to be consumed during the current period due to high photovoltaic power generation; when... When the power output of photovoltaic power is insufficient to support the base load, it indicates that there is a power shortage that requires external power supply or power discharge from electric vehicles.

[0028] S1.2 Calculation of dynamic carbon emission factor: Traditional static carbon emission factors cannot reflect real-time changes in renewable energy output within microgrids, making it difficult to accurately measure the true historical carbon emission reductions on the electricity consumption side. This invention calculates dynamic carbon emission factors based on carbon emission flow theory.

[0029] (1) Calculate the number of nodes in the microgrid based on the network topology and power flow distribution. During the period node carbon potential : ; in, For time period By node To the node Injected active power; For nodes carbon potential; and They are nodes The power generation capacity and direct carbon emission intensity of the connected generators, photovoltaic power. =0; To the node The set of adjacent nodes into which the current is injected. For nodes A collection of directly connected power supplies.

[0030] Based on the obtained carbon potential at each node By performing load spatial weighted averaging, the entire microgrid area is obtained in the time period. Dynamic carbon emission factors : ; in, It is the set of all load nodes within the microgrid; For time period node The actual load. When the proportion of photovoltaic output within the microgrid increases... This will significantly reduce emissions, thereby transforming the high proportion of green electricity in the microgrid into a quantified low-carbon emission signal.

[0031] S1.3, Dynamic load matching coefficient distribution: The microgrid dispatch center calculates the initial net load based on S1.1. The curve, combined with the baseline operating status of the upper-level power grid, generates a dynamic load matching coefficient. When net load is below a certain threshold When the load exceeds a certain threshold, a valley period coefficient is set to encourage charging. At that time, the peak segment coefficient is set to encourage discharge.

[0032] S1.4, Carbon electrical signal transmission: The microgrid dispatch center will use the dynamic carbon emission factor calculated in S1.2. The dynamic load matching coefficient generated in S1.3 and the system's global low-carbon guiding factors Encapsulated as an environmental state observation vector It is also broadcast synchronously to all flexible contracted EV agents and fixed contracted EV aggregators within the micronet via the communication network.

[0033] S2. Edge-side autonomous demand response and net shortfall calculation for flexible contract EVs: The execution entity for this step is the local intelligent agent built into the flexible subscription EV and the microgrid scheduling center. Considering the high randomness of travel and strong privacy protection requirements of flexible subscription users, this invention abandons traditional centralized information collection and adopts edge-side autonomous decision-making. The specific implementation process is as follows: S2.1 Construction of Local State Space and Action Space: The flexible contract EV intelligent agent runs on the local terminal. During the specified time period... The intelligent agent perceives the environment and constructs a state space. and action space : state space It includes global signals received from the microgrid and local vehicle privacy data: ; in, The dynamic carbon emission factor issued by the S1 microgrid dispatch center. This refers to the dynamic load matching coefficient. For vehicles Current state of charge, For the expected off-line time, The desired off-grid target power.

[0034] Action space For continuous variables, that is, the agent's time period Decision charging and discharging power . Indicates charging. This indicates discharge, and is subject to the vehicle's maximum charge and discharge physical constraints.

[0035] S2.2 Design of a multi-objective comprehensive reward function considering carbon-electric coupling: To achieve carbon-electricity synergy, a comprehensive benefit function was designed. : ; For grid load matching gain: ; in, This refers to the charging and discharging time.

[0036] To control carbon emissions: When vehicles are charging during periods of high solar power generation, the system assigns a carbon reduction bonus weight. ; in, The baseline carbon emission factor for the power grid is the carbon emission equivalent per unit of electricity generated by the upstream regional power grid to which the microgrid is connected. This parameter is derived from real-time dynamic carbon emission intensity data of the power grid issued by the upstream dispatch center. It serves as a global low-carbon guiding factor.

[0037] Battery physical wear and tear penalty: To ensure the health of the underlying hardware, a secondary penalty function for battery degradation is introduced here: ; Penalty for range anxiety: when disconnected from the network If the actual battery consumption does not reach the expected value, an exponential penalty is applied: .

[0038] S2.3, Edge-side autonomous decision-making and execution: The flexibly contracted EV iteratively optimizes the agent model locally based on the improved reward function described above. During the decision-making process, the vehicle's privacy data is processed entirely locally and is not uploaded to the microgrid dispatch center. The agent only outputs the optimal charge / discharge power curve. Uploaded to the microgrid dispatch center. While achieving a fully market-driven response, it also ensures the user's travel privacy and security through physical isolation.

[0039] S2.4 Calculation of real-time net load deficit power of microgrid: Microgrid dispatch center collects data from the entire network. Autonomous response power curve uploaded by flexibly contracted EVs Compare this with the initial net load of the microgrid obtained in S1. The values ​​are superimposed to calculate the final real-time net load deficit power for this period. : ; when This indicates that the autonomous charging of flexible EVs has not yet fully absorbed photovoltaic power, and there is still a surplus within the microgrid, requiring aggregators to organize fixed EVs for aggregated charging. This indicates that even after the flexible EV response, a power supply gap still exists in the grid, requiring aggregators to organize fixed EVs for aggregated discharge. As an absolute physical hard constraint indicator, it will be directly issued to the aggregator as the overall target basis for the fixed contracted EV tiered quota scheduling in the subsequent S4 steps.

[0040] S3. Multi-dimensional comprehensive evaluation and anti-exploitation dynamic clustering of fixed-contract EVs: The EV aggregator is responsible for executing this step. This invention performs a global static data review and cluster reconstruction of all fixed-contract vehicles across the network every day at 00:00. The specific implementation process is as follows: S3.1 Multidimensional extraction and normalization of historical operating data: The aggregator system extracts each vehicle from the fixed contract vehicle pool. Extract physical quantities from the operational log data over a defined period of time, focusing on three core evaluation dimensions: Historical carbon emission reductions: i.e., vehicle emissions reductions The cumulative amount of green electricity generated through past participation in photovoltaic power consumption.

[0041] Response engagement: i.e., vehicle participation The ratio of the number of times scheduling was actually executed to the number of times the system issued scheduling instructions.

[0042] Battery health: The percentage of the current battery's maximum available capacity that has been degraded, as read by the battery management system.

[0043] The system uses the Z-score standardization method and then the sigmoid function to transform the three data points with different physical dimensions into normalized indices: , , This maps all values ​​to the interval [0,1].

[0044] S3.2 Calculation of triggering rules for wear leveling and anti-exploitation reward / penalty factors: To avoid hierarchical solidification, the system introduces a dynamic reward and punishment factor based on conditional branching: Anti-starvation reward determination: Scan log system; if a vehicle is detected... If a user is online for three consecutive days but not selected by the system for scheduling, the wear-and-tear compensation mechanism will be triggered, granting them a positive reward factor. .

[0045] Determination of refusal to carry penalty: Scan log system; if the vehicle is detected... If a system rejects or interrupts a scheduled instruction three times consecutively, a breach of trust penalty mechanism will be triggered, imposing a negative penalty factor. , .

[0046] S3.3, Calculation and Ranking of Overall Score: Combining the above normalized indicators and reward / penalty factors, the aggregator calculates the final comprehensive score for each fixed-contract EV. : ; in, The preset objective weighting coefficients satisfy... This is used to balance the importance of low-carbon contribution, scheduling reliability, and battery physical safety.

[0047] S3.4 Dynamic Hierarchy and Cluster Tier Reconstruction: The system will calculate the overall score of all fixed-contract EVs across the entire network. The vehicles are sorted in descending order, and the list is divided into three scheduling clusters with different priorities using a set score threshold or ratio threshold.

[0048] (1) Core scheduling cluster: Vehicles ranked in the top 30% of the score. They will be given the highest scheduling priority and the highest battery physical loss compensation coefficient.

[0049] (2) Regular scheduling cluster: Vehicles with scores ranking between 31% and 70% are used as basic scheduling capacity.

[0050] (3) Backup scheduling cluster: Vehicles ranked in the bottom 30% (including newly registered vehicles with no historical data) are assigned the lowest scheduling priority and baseline loss compensation weight. The cluster partitioning results will be written to the aggregator's real-time database as the sole macro-level priority basis for power allocation in response to sudden instructions in step S4 the following day.

[0051] S4. Tiered quota allocation based on real-time physical capacity and overflow compensation: The EV aggregator is responsible for executing this step. This is done upon receiving the real-time net load deficit power from the microgrid dispatch center. Subsequently, the aggregator needs to allocate and adjust macro-level power quotas based on the static cluster partitioned by S3 and the real-time physical status of the vehicles. The specific implementation process is as follows: S4.1 Real-time available physical capacity scanning and assessment: When a real-time dispatch command is triggered, the aggregator system polls the current physical status of all fixed-contract EVs across the entire network via the vehicle network gateway: (1) Status filtering: Remove items that are not currently plugged in online, have a fault, or are currently in a charged state. And it is in the process of discharging, or And vehicles that are in need of charging.

[0052] (2) Cluster capacity aggregation: For the filtered real-time available vehicle pool, the core scheduling cluster, regular scheduling cluster and standby scheduling cluster are calculated respectively during this period. The maximum physically available response power is denoted as . , and .

[0053] S4.2, Proportional decomposition of initial fair quota: To overcome the Matthew effect caused by traditional greedy algorithms, the system enforces a tiered basic scheduling ratio matrix. .

[0054] The system will calculate the total power deficit for this period. Decompose and allocate initial target quotas proportionally: Initial quota for core clusters: ; Initial quota for a standard cluster: ; Initial quota for standby clusters: ; This step ensures that even newly registered vehicles or low-scoring vehicles in the backup cluster have a basic opportunity to participate and earn rewards in each round of scheduling.

[0055] S4.3 Verification and Indicator Transfer of Overflow Compensation Mechanism: Due to the extreme randomness of physical conditions, the actual physical capacity of low-priority clusters may not meet their initial quota. The system employs bottom-up conditional branching logic for overflow compensation: (1) Backup cluster verification: compare its available capacity With initial quota ; like The standby cluster will actually execute the quota. No overflow. If the backup cluster cannot meet the demand, its actual execution quota will be reduced. The resulting difference in capacity It will be extracted and superimposed onto the tasks in the regular cluster.

[0056] (2) Verification of Regular Clusters: The quota to be assessed for regular clusters is updated to... ; like The actual quota executed by the regular cluster is then... No further overflow. If The actual quota executed by the regular cluster is then... The resulting secondary differential capacity It will continue to be stacked up to the core cluster's tasks.

[0057] (3) Core Cluster Verification: The core cluster, as the strongest and highest-scoring backup force, is responsible for the final aggregated quota. Its actual quota Same as above. Its actual execution quota is maximized to This results in three limiting difference capacities. The system will trigger a secondary global backfill mechanism.

[0058] (4) Secondary global backfill: When Upon completion of the initial upward migration, the system initiates a top-down redundancy optimization process, backfilling unfinished metrics to clusters with remaining capacity. The remaining capacity of a typical cluster... If there is any remaining capacity, it will be filled first; if the regular cluster is full or there is still a gap after filling, the remaining gap will be added to the remaining capacity of the standby cluster. The standby and regular clusters, based on their original actual execution quotas, are supplemented with the received backfilling metrics to form the final execution quota. .

[0059] If, after closed-loop backfilling, the ultimate capacity shortfall is still greater than the sum of all remaining capacities at the underlying level, it indicates that the scheduling command issued by the microgrid has absolutely exceeded the physical response limit of the current fixed-subscription EV cluster. The aggregator issues a full-capacity load command and reports the final unmanageable shortfall to the microgrid scheduling center in real time.

[0060] S5. Fine-grained power scheduling and tiered physical loss compensation integral update for fixed clusters: This step is executed by the EV aggregator. After obtaining the cluster-specific adjusted quota indicators output from step S4... Afterwards, the aggregator needs to perform hard execution of micro-power allocation within each cluster and update the response integral and weight status of each vehicle node after scheduling is complete. The specific implementation process is as follows: S5.1 Fine-grained sorting and precise power allocation within the cluster: (1) Internal grouping and sorting: Within each cluster, retrieve the comprehensive score calculated by each vehicle node in stage S3. Sort the results in descending order of scores to generate an absolute scheduling execution sequence within the cluster.

[0061] (2) Greedy Power Allocation: The system sequentially issues charging and discharging power commands to each vehicle node from top to bottom along the execution sequence, up to the physical limit boundary, until the total allocated power accumulates to reach the corrected quota index of the cluster. Taking the core cluster as an example, its internal allocation must satisfy the following equality and inequality constraints: ; ; in, This refers to the set of vehicle nodes selected for scheduling within the core cluster. Power is actually allocated to a single node; and This represents the current physical charging and discharging limit of the vehicle node. A strategy of first performing coarse dimensionality reduction across the cluster, followed by optimal selection of high-resolution internal components, compresses the computational load.

[0062] S5.2 Quantification of battery physical loss parameters in a single scheduling: After completing the charge / discharge scheduling, the aggregator system reads the actual charge / discharge data of each participating vehicle through the vehicle-to-everything (V2G) gateway. Especially for vehicles participating in nighttime V2G discharge, the system needs to calculate the battery physical aging parameters caused by the discharge. : ; in, For vehicles Battery reset baseline weight parameters; This represents the change in depth of discharge caused by this discharge. This represents the total number of nominal cycle lives of the battery at the corresponding depth of discharge.

[0063] S5.3, Fixed-subscribed EV physical loss exemption and comprehensive state integral update based on cluster ladder: To incentivize users to provide high-quality scheduling resources in the long term and alleviate their battery anxiety, aggregators will combine the overall contribution to global load smoothing at the macro level with the overall low-carbon operation index by setting tiered loss exemption weights. Mapped to each fixed contracted vehicle node. A tiered excess loss subsidy for each dispatch. The calculation is as follows: ; Core Cluster: Configuration Core cluster users not only completely offset the physical aging penalty at the algorithm level, but also receive a 20% positive weight overflow, transforming high-frequency physical responses into high-priority state maintenance within the system.

[0064] Standard Cluster: Configuration Regular cluster users receive 100% compensation for physical losses, achieving lossless parity iteration in internal system evaluation. Backup cluster: The backup cluster nodes need to retain a small amount of aging penalty in the algorithm, which serves as a control lever for system feedback, forcing low-scoring nodes to actively improve their daily green energy consumption and response coordination at the edge, so as to advance to a higher-dimensional cluster.

[0065] To further allocate global coordination metrics, aggregators utilize the globally available adjustment pool accumulated through centralized responses. Used to generate dynamic collaborative incentive integrals for each node. .

[0066] in, The calculation is as follows: ; in, and These represent the total contribution of aggregators to grid load stabilization and the total low-carbon operation indicators achieved during the period, respectively. This is to account for the inherent penalty dissipation in maintaining the system's basic communication and control computations; and This is the base amount of physical capacity already allocated to each node during this period, plus the total amount of dynamic loss compensation. The preset integral extraction ratio ensures a balance between the system's overall adjustment margin and the incentive distribution to lower-level nodes.

[0067] To further widen the revenue gap between the core cluster and the backup cluster, and to reflect the principle of "higher scores, higher benefits," the aggregator system does not use a simple linear ratio. Instead, it introduces an exponential weighting function based on temperature parameters to calculate vehicle... State update and weight assignment : ; in, For vehicles The final overall score; This refers to the set of all fixed-contract vehicles participating in the custody during the settlement period; The preset excitation temperature parameters for the system ( ).when When it is larger, the score A slight increase in its dividend payout ratio This will also amplify exponentially, which can greatly incentivize regular cluster users to increase their green energy consumption and responsiveness in pursuit of higher scores; when When it approaches 0, it degenerates into an average distribution.

[0068] The system allocates weights based on the calculated weights. With the points pool Calculate the dynamic collaborative incentive points ultimately obtained by each fixed-contract EV. : ; The collaborative incentive integral will be combined with the capacity maintenance baseline and physical loss compensation amount, and written as the latest state parameter into the local state registers of the aggregator and the vehicle.

[0069] At the end of the scheduling cycle, the aggregator is the fixed contracted vehicle node. Calculate and update its cumulative comprehensive evaluation value. : ; in, To maintain a baseline for physical capacity, as long as vehicles are connected online and managed by the system within the agreed time period, they can obtain a fixed basic status score, which provides a lower limit guarantee for the node's activity rating in the system.

[0070] S5.4, Scheduling Data Archiving and Periodic Refresh: After settlement, the aggregator system timestamps the actual charging and discharging volume, green electricity traceability volume, and performance status of each vehicle in this dispatch and stores them in the historical operation database. This data will serve as the direct input parameter for triggering the multi-dimensional comprehensive evaluation and anti-exploitation dynamic cluster partitioning in step S3 at the start of the next 24-hour dispatch cycle, thereby completing the microgrid's dynamic optimization across multiple time scales within the day.

[0071] This embodiment also provides an EV classification and collaborative scheduling system that takes into account carbon-electric coupling and dynamic clustering, including: The observation vector encapsulation module is used to collect historical base load data and historical output data of distributed power sources such as photovoltaics within the microgrid, and encapsulate them into environmental status observation vectors. The revenue function design module is used to construct the state space and action space based on the environmental state observation vector of the flexible contracted EV agent, so as to set the comprehensive revenue function; The power deficit acquisition module is used to iteratively optimize the agent model based on the comprehensive benefit function, output the optimal charging and discharging power curve, and calculate the real-time net load deficit power. The collaborative scheduling module is used to perform global static data review and cluster reconstruction of all fixed-contract EV vehicles in the network at a set period through fixed-contract EV aggregators, to obtain scheduling clusters with different priorities, and to issue and correct macro power quotas for each scheduling cluster using real-time net load deficit power, forming the final execution quota. At the same time, charging and discharging power instructions are issued until the current total allocated power accumulates to reach the final execution quota of each scheduling cluster.

[0072] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for classifying and coordinating EV scheduling that considers carbon-electric coupling and dynamic clustering, characterized in that, include: Historical base load data and historical output data of distributed power sources such as photovoltaics within the microgrid are collected and encapsulated into environmental status observation vectors. Based on the flexible contract EV agent, the state space and action space are constructed using the environmental state observation vector to set the comprehensive benefit function. The agent model is iteratively optimized according to the comprehensive benefit function, the optimal charging and discharging power curve is output, and the real-time net load deficit power is calculated. By having fixed-contract EV aggregators perform global static data review and cluster reconstruction on all fixed-contract vehicles across the network at set intervals, scheduling clusters with different priorities are obtained. The real-time net load deficit power is used to issue and correct macro power quotas to each scheduling cluster to form the final execution quota. At the same time, charging and discharging power instructions are issued until the total power currently allocated reaches the final execution quota of each scheduling cluster.

2. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 1, characterized in that, The environmental state observation vector includes: Based on the historical base load data and the historical output data, the power of future scheduling cycles is predicted, and the photovoltaic power prediction for the target future period is output. At the same time, time-series modeling is performed on the non-EV base load, and the base load prediction value for the target future period is output. The initial net load of the microgrid in the target future period is calculated using the photovoltaic predicted power and the base load predicted value, and a dynamic load matching coefficient is generated by combining the baseline operating status of the upper-level power grid. Based on the network topology and power flow distribution of the microgrid, the node carbon potential of the nodes in the microgrid in the target future time period is calculated, and the node carbon potential is weighted and averaged to obtain the dynamic carbon emission factor of the entire microgrid area in the target future time period. The dynamic load matching coefficient, the dynamic carbon emission factor, and the global low-carbon guidance factor are encapsulated into the environmental state observation vector.

3. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 1, characterized in that, Setting the comprehensive return function includes: ； ； ； ； ； in, For the comprehensive benefit function, For grid load matching gain, Gain control for low-carbon consumption. This is a penalty for physical battery wear and tear. As a penalty for range anxiety, For the agent in time period The charging and discharging power of the decision. This is the dynamic load matching factor. For charge / discharge time, The dynamic carbon emission factor issued by the microgrid dispatch center. As the benchmark carbon emission factor for the power grid, As a global low-carbon guiding factor, For vehicles Current state of charge, For the expected off-line time, For the desired off-grid target power, , These are the weighting coefficients.

4. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 1, characterized in that, Calculating the real-time net load deficit power includes: The optimal charge / discharge power curve is superimposed with the initial net load to calculate the final real-time net load deficit power for this period: ； in, This represents the final real-time net load deficit power for this period. This indicates that the autonomous charging of flexible EVs has not yet fully absorbed photovoltaic power, and there is still a surplus within the microgrid. Aggregators are needed to organize fixed EVs for aggregated charging. This indicates that even after the flexible EV response, there is still a power supply gap in the grid, requiring aggregators to organize fixed EVs for aggregated discharge. To achieve the optimal charge / discharge power curve, This represents the initial net load of the microgrid.

5. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 1, characterized in that, By having fixed-contract EV aggregators conduct a global static data review and cluster reconstruction of all fixed-contract vehicles across the network at set intervals, including: The fixed-contract EV aggregator extracts the operation log data of each vehicle in the fixed-contract vehicle pool within a set period, and extracts the physical quantities of the core evaluation dimensions in the operation log data; the physical quantities include: historical carbon emission reduction, response participation and battery health. The physical quantities are converted into normalized indices, and a dynamic reward and punishment factor based on conditional branch triggering is introduced to calculate the final comprehensive score of each fixed contracted EV. The final comprehensive scores of all fixed-contract EVs across the network are sorted in descending order, and the vehicle list is divided into scheduling clusters of different priorities using a set score threshold or proportion threshold; the scheduling clusters include: core scheduling cluster, regular scheduling cluster and backup scheduling cluster.

6. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 5, characterized in that, The final overall score for each fixed-contract EV is calculated as follows: ； in, The final overall score for each contracted EV. As a positive reward factor, if a vehicle is detected If a system remains online for three consecutive days but is not selected for scheduling, the wear-and-tear compensation mechanism will be triggered, granting it a positive reward factor. As a negative penalty factor, if a vehicle is detected... If a system rejects or interrupts a scheduled instruction three times consecutively, a breach of trust penalty mechanism will be triggered, imposing a negative penalty factor. , , These are normalized indicators corresponding to historical carbon emission reductions, response participation, and battery health, respectively. The preset objective weighting coefficients satisfy... This is used to balance the importance of low-carbon contribution, scheduling reliability, and battery physical safety.

7. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 1, characterized in that, The final execution quota is formed by: By polling the current physical status of all fixed-contract EVs in the network through the vehicle network gateway, and removing vehicles that are not currently plugged in and online, have malfunctioned, or whose current state of charge is not greater than the first target threshold and are in the process of discharging, or whose current state of charge is not less than the second target threshold and are in the process of charging, a filtered pool of available vehicles in real time is obtained. The maximum physical available response power of the core scheduling cluster, the regular scheduling cluster, and the backup scheduling cluster is calculated by summing the filtered real-time available vehicle pools during this period. A tiered basic scheduling ratio matrix is ​​then introduced to proportionally decompose the total shortage power during this period to obtain the initial target quota. The initial target quota includes: the initial quota of the core cluster, the initial quota of the regular cluster, and the initial quota of the backup cluster. The final execution quota is obtained by verifying and transferring the overflow compensation mechanism using the maximum available physical response power and the initial target quota.

8. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 7, characterized in that, Obtaining the final execution quota includes: Backup scheduling cluster verification: The maximum physical available response power of the backup scheduling cluster during this period is compared with the initial quota of the core cluster. If the maximum physical available response power of the backup scheduling cluster during this period is not less than the initial quota of the core cluster, it proves that the actual execution quota of the backup scheduling set has not overflowed. If the maximum physical available response power of the backup scheduling cluster during this period is less than the initial quota of the core cluster, it proves that the backup scheduling set cannot meet the demand, and the resulting first difference in capacity will be extracted and added to the tasks of the regular scheduling cluster. Regular scheduling cluster verification: If the first difference capacity exists, the initial quota of the regular cluster is updated using the first difference capacity, and the maximum physical available response power of the regular scheduling cluster during this period is compared with the updated initial quota of the regular cluster. If the maximum physical available response power of the regular scheduling cluster during this period is not less than the updated initial quota of the regular cluster, it proves that the actual execution quota of the regular scheduling cluster has not overflowed. If the maximum physical available response power of the regular scheduling cluster during this period is less than the updated initial quota of the regular cluster, it proves that the actual execution quota of the regular scheduling cluster cannot meet the demand, and the resulting second difference capacity will be extracted and added to the tasks of the core scheduling cluster. Core scheduling cluster verification: If the second difference in capacity exists, the initial quota of the core cluster is updated using the second difference in capacity, and the maximum physical available response power of the core scheduling cluster during this period is compared with the updated initial quota of the core cluster. If the maximum physical available response power of the core scheduling cluster during this period is not less than the updated initial quota of the core cluster, it proves that the actual execution quota of the core scheduling cluster has not overflowed. If the maximum physical available response power of the core scheduling cluster during this period is less than the updated initial quota of the core cluster, it proves that the actual execution quota of the core scheduling cluster cannot meet the demand, and the resulting third difference in capacity will be extracted, triggering a secondary global backfill mechanism. When the third difference capacity is greater than the target threshold, the uncompleted indicators are looped back to the cluster with remaining capacity. If the remaining capacity of the regular scheduling cluster is available, it is filled first. If the regular scheduling cluster is full or there is still a gap after filling, the remaining gap is added down to the remaining capacity of the backup scheduling cluster to obtain the final execution quota. If the limit difference capacity after looping backfilling is still greater than the sum of all remaining capacities at the bottom layer, it indicates that the scheduling instruction issued by the microgrid has exceeded the physical response limit of the current fixed contract EV cluster. Based on the fixed contract EV aggregator, a full capacity load instruction is issued, and the final gap that cannot be smoothed out is reported to the microgrid scheduling center in real time.

9. The EV classification and collaborative scheduling method considering carbon-electric coupling and dynamic clustering according to claim 1, characterized in that, Issuing the charging and discharging power command until the currently allocated total power accumulates to the final execution quota of each scheduling cluster includes: Within each scheduling cluster, the final comprehensive score of each vehicle node is retrieved and sorted in descending order of the final comprehensive score to generate an absolute scheduling execution sequence within the cluster. According to the absolute scheduling execution sequence, charging and discharging power instructions that reach their physical limit boundaries are issued to each vehicle node from top to bottom until the total power currently allocated reaches the final execution quota of each scheduling cluster. After the charging and discharging scheduling action is completed, the battery physical aging parameters caused by discharge, the tiered excess loss subsidy for a single scheduling, and the dynamic collaborative incentive points finally obtained by each fixed contracted EV are calculated to update the comprehensive evaluation accumulation value of the fixed contracted vehicle node.

10. An EV classification and collaborative scheduling system considering carbon-electric coupling and dynamic clustering, implemented by the method according to any one of claims 1-9, characterized in that, include: The observation vector encapsulation module is used to collect historical base load data and historical output data of distributed power sources such as photovoltaics within the microgrid, and encapsulate them into environmental status observation vectors. The revenue function design module is used to construct a state space and action space based on the environmental state observation vector of the flexible contracted EV agent, so as to set the comprehensive revenue function; The power deficit acquisition module is used to iteratively optimize the agent model according to the comprehensive benefit function, output the optimal charging and discharging power curve, and calculate the real-time net load deficit power. The collaborative scheduling module is used to perform global static data review and cluster reconstruction of all fixed-contract EV vehicles in the network at a set period through fixed-contract EV aggregators to obtain scheduling clusters with different priorities. It also uses the real-time net load deficit power to issue and correct macro power quotas to each scheduling cluster to form the final execution quota. At the same time, it issues charging and discharging power instructions until the current total allocated power accumulates to reach the final execution quota of each scheduling cluster.

Images