A virtual power plant and vehicle network interaction resource aggregation scheduling system and method

Abstract

The application discloses a kind of virtual power plant and the resource aggregation scheduling system and method of vehicle network interaction, the above-mentioned method is: obtaining multi-source heterogeneous data and standardization processing obtains standard data set;Multi-dimensional label system is constructed, and the standard data set is labeled and processed, and finally at least one functional resource pool is constructed;The power grid operating state data in the standard data set is acquired, three-dimensional dynamic weight is calculated, and a multi-objective optimization scheduling model is constructed;Demand data in the standard data set is input into the multi-objective optimization scheduling model to solve, and the optimal solution of overall resource scheduling target is obtained;The partition scheduling target value of each functional resource pool is obtained;Based on the partition scheduling target value, the scheduling strategy model of each functional resource pool is constructed, and the scheduling instruction parameter of each resource is solved, and the control is issued and executed.This application has the advantages of realizing resource aggregation precision, scheduling dynamic adaptation, multi-objective collaborative optimization, efficient and reliable control.

Description

Technical Field

[0001] This invention relates to the field of power management, and more specifically, to a resource aggregation and scheduling system and method for virtual power plant and vehicle-grid interaction. Background Technology

[0002] With the continuous increase in the penetration rate of new energy power generation and the rapid growth in the number of electric vehicles, virtual power plants, as the core carrier for integrating distributed energy, energy storage, and flexible loads, are becoming a key path to improve grid flexibility and the capacity for new energy absorption through their integration with vehicle-to-grid (V2G) technology. Currently, grid operation faces challenges such as increased load fluctuations, highly random new energy output, and difficulties in coordinating multiple types of resources. Therefore, there is an urgent need to achieve the coordinated utilization of virtual power plant resources and V2G resources through efficient resource aggregation and scheduling methods. Existing virtual power plant (VPS) dispatching methods primarily focus on the regulation of single-type resources, lacking systematic integration of multi-source heterogeneous data from the VPS, vehicle-to-grid (V2G) interaction, and grid sides. Inconsistent data formats weaken the foundation for collaborative resource dispatching. Resource aggregation methods often rely on simple classification, failing to establish a refined labeling system that considers entities, attributes, capabilities, and scenarios, making it difficult to accurately match the differentiated dispatching needs of the grid. Optimization dispatching models often employ fixed-weight designs, unable to dynamically adjust the priorities of grid security, renewable energy consumption, and user benefits based on real-time operating conditions such as grid load factor and renewable energy output ratio, easily leading to incomplete dispatching solutions. Furthermore, the decomposition of dispatching objectives lacks targeted consideration of resource response characteristics, and the absence of effective pre-execution verification and status feedback mechanisms after the issuance of control commands affects the reliability and accuracy of dispatching execution. Therefore, existing technologies struggle to achieve efficient aggregation and dynamic optimization dispatching of VPS and V2G resources. There is an urgent need for a resource aggregation dispatching method that can integrate multi-source data, finely aggregate resources, dynamically adapt target priorities, and accurately implement control execution to improve grid operational stability, renewable energy consumption levels, and overall user benefits. Summary of the Invention

[0003] To overcome the shortcomings of existing virtual power plant and vehicle-to-grid (V2G) interaction resource aggregation scheduling, such as unsystematic data integration, insufficient refinement of resource aggregation, fixed scheduling target priority, and lack of reliability in control execution, this invention provides a virtual power plant and V2G interaction resource aggregation scheduling system and method.

[0004] A resource aggregation and scheduling method for interaction between a virtual power plant and a vehicle-to-grid system includes the following steps:

[0005] S1. Obtain multi-source heterogeneous data, which includes core data from the virtual power plant side, core data from the vehicle-to-grid interaction side, and core data from the power grid side. Standardize the multi-source heterogeneous data to obtain a standard dataset.

[0006] S2. Based on the standard dataset, construct a tag system containing four dimensions: subject, attribute, capability, and scenario. Tag the virtual power plant resources and vehicle-to-grid interaction resources in the standard dataset, and finally construct at least one functional resource pool.

[0007] S3. Obtain the power grid operation status data in the standard dataset, and based on the power grid operation status data, calculate the three-dimensional dynamic weights of power grid security, new energy consumption, and user benefits, and construct a multi-objective optimization scheduling model;

[0008] S4. Input the grid load gap demand and new energy consumption demand from the standard dataset into the multi-objective optimization scheduling model, solve the multi-objective optimization scheduling model, and obtain the optimal solution for the overall resource scheduling objective;

[0009] S5. Based on the overall resource scheduling target and the resource characteristics of the functional resource pool, the partition scheduling target value of each functional resource pool is decomposed.

[0010] S6. Based on the partitioned scheduling target value, construct the scheduling strategy model for each of the functional resource pools, solve for the scheduling instruction parameters of each resource, including the control direction, target power, and control period; finally, issue the scheduling instruction parameters and execute the control to complete the resource aggregation scheduling.

[0011] Furthermore, in one embodiment, in step S1: the core data on the virtual power plant side includes energy storage power station operation data, interruptible capacity of industrial load, distributed photovoltaic power output data, and virtual power plant contract data; the core data on the vehicle-to-grid interaction side includes real-time power of charging piles, electric vehicle battery status, and operator profile data; the core data on the grid side includes load gap data, heavy overload early warning information, and new energy consumption demand data; the standardized data format is a combination structure of resource ID, timestamp, and core parameters.

[0012] Further, in one embodiment, in step S2: the main tag includes operator type, resource type, and access area; the attribute tag includes response accuracy, adjustment precision, and failure rate; the capability tag includes maximum adjustable capacity, response latency, and continuous adjustment duration; the scenario tag includes peak shaving application scenario, valley filling application scenario, and new energy consumption application scenario; the functional resource pool includes peak shaving resource pool, valley filling resource pool, and new energy consumption resource pool, and the update frequency of the functional resource pool is every 5-15 minutes.

[0013] Further, in one embodiment, in step S3: the three-dimensional dynamic weights satisfy the following expression:

[0014] ω1+ω2+ω3=1

[0015] Where ω1 is the power grid security weight, ω2 is the renewable energy absorption weight, and ω3 is the user revenue weight; when the power grid load factor is ≥90%, ω1 ≥0.6; when the predicted renewable energy output is ≥30% of the power grid load, ω2 ≥0.5; the objectives of the multi-objective optimization scheduling model are to minimize power grid load fluctuations, maximize renewable energy absorption, and maximize comprehensive user revenue, and the constraints include adjustable resource capacity constraints, power balance constraints, response speed constraints, and cost constraints.

[0016] Further, in one embodiment, in step S5: the total adjustable capacity and response priority coefficient of each functional resource pool are extracted, the response priority coefficient being determined based on the response speed and adjustment accuracy labels obtained from the tagging process; the partition scheduling target value of each functional resource pool is calculated, expressed as:

[0017] P_i = P general tone × (Q_i × k_i) / Σ(Q_i × k_i)

[0018] Where P_total is the overall resource scheduling target, Q_i is the total adjustable capacity of the i-th functional resource pool, and k_i is the response priority coefficient of the i-th functional resource pool.

[0019] Furthermore, in one embodiment, in step S6, a corresponding scheduling strategy model is constructed by combining the response characteristics and adjustment capabilities of each functional resource pool, scheduling instructions are generated according to the resource type and the adjustment parameters are adapted, instructions are first issued to pilot resources for pre-execution, and after confirming the effectiveness of the execution, they are issued to each resource in batches, and the adjustment execution status is fed back and verified in real time to ensure that the adjustment meets the requirements of the partitioned scheduling target value.

[0020] A resource aggregation and scheduling system for virtual power plant and vehicle-to-grid interaction includes: a data perception and standardization module for acquiring multi-source heterogeneous data and performing standardization processing to output a standard dataset; a resource tagging and aggregation module, communicatively connected to the data perception and standardization module, for constructing a four-dimensional tagging system and dynamically building a functional resource pool; a scheduling model and target determination module for calculating three-dimensional dynamic weights and solving for the optimal solution of the overall resource scheduling target; a scheduling target decomposition module, communicatively connected to the resource tagging and aggregation module and the scheduling model and target determination module respectively, for decomposing the target values ​​of each partition; a scheduling strategy generation module, communicatively connected to the scheduling target decomposition module, for generating scheduling instruction parameters; and a control execution module, communicatively connected to the scheduling strategy generation module, for issuing scheduling instruction parameters and executing control.

[0021] Furthermore, in one embodiment, the resource tagging and aggregation module includes a tag engine and a resource pool construction unit; the tag engine has a built-in four-dimensional tag rule library, which supports automatic tag allocation and dynamic updates; the resource pool construction unit generates a functional resource pool based on the scene tag matching results, which supports the expansion of the resource pool and resource migration.

[0022] Furthermore, in one embodiment, the scheduling model and target determination module includes a weight calculation unit, a model construction unit, and a target solving unit; the weight calculation unit calculates three-dimensional dynamic weights in real time based on power grid operation status data; the target solving unit incorporates an improved genetic algorithm or particle swarm optimization algorithm to solve the multi-objective optimization scheduling model.

[0023] Furthermore, in one embodiment, the control execution module includes an instruction issuing unit and a backup scheduling unit; the instruction issuing unit is used to issue scheduling instruction parameters to each resource and trigger control execution; the backup scheduling unit stores backup resource pool information and backup scheduling schemes, and automatically starts switching when abnormal resource response or sudden change in power grid operating status is detected to ensure control continuity.

[0024] According to the above-mentioned solution, the beneficial effects of this invention are as follows: it can systematically integrate multi-source heterogeneous data to solidify the foundation for resource collaborative scheduling; it can achieve refined resource aggregation through a four-dimensional tagging system to accurately match the differentiated needs of the power grid; it can dynamically adapt the scheduling target priority based on three-dimensional dynamic weights to achieve synergistic optimization of power grid security, new energy consumption, and user benefits; it can scientifically decompose scheduling targets in combination with resource characteristics, and improve the accuracy and reliability of control execution with pre-execution verification and real-time feedback mechanisms, thereby effectively improving the stability of power grid operation and resource utilization efficiency. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art 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.

[0026] Figure 1 This is a schematic diagram of a resource aggregation and scheduling method for interaction between a virtual power plant and the vehicle network in this embodiment;

[0027] Figure 2 This is the core scheduling and execution flowchart of virtual power plant and vehicle-to-grid interaction resources in this embodiment;

[0028] Figure 3 This is a flowchart of the closed-loop process for exception handling and backup scheduling in this embodiment. Detailed Implementation

[0029] The present invention will now be further described with reference to the accompanying drawings and embodiments:

[0030] I. Implementation Overview

[0031] like Figure 1-3 As shown, this invention discloses a resource aggregation and scheduling method and system for virtual power plant and vehicle-to-grid interaction, aiming to solve problems such as insufficient integration of multi-source data, extensive resource aggregation, fixed scheduling targets, and low control reliability in existing technologies. The following specific application scenario (taking virtual power plant and vehicle-to-grid interaction resource scheduling within a provincial power grid area as an example) ensures that those skilled in the art can fully reproduce the technical solution of this invention based on the following implementation methods.

[0032] The application scenario of this embodiment is set as follows: within the coverage area of ​​a provincial power grid, there are 3 virtual power plants (total installed capacity of 500MW, including 3 energy storage power stations, 20 distributed photovoltaic power stations, and 10 industrial interruptible loads) and vehicle-to-grid interaction resources (500 public charging piles and 2000 private electric vehicles, managed by 3 operators). The power grid side needs to deal with the problems of large peak-valley load difference (peak-valley difference of up to 30%) and significant fluctuations in new energy output (daily fluctuation range of photovoltaic output ±25%). The scheduling cycle is set to 15 minutes / time, which is consistent with the update frequency of the functional resource pool.

[0033] II. Detailed Explanation of Implementation Steps, such as Figure 1-3 As shown;

[0034] (I) Step S1: Acquisition and Standardization of Multi-Source Heterogeneous Data

[0035] 1.1 Data Acquisition Scheme

[0036] The core of this step is to comprehensively collect multi-source heterogeneous data from the virtual power plant side, the vehicle-to-grid interaction side, and the power grid side to ensure the integrity, real-time performance, and accuracy of the data. The specific acquisition methods are as follows:

[0037] Data collection objects and content:

[0038] The core data of the virtual power plant is collected through the API interface of the virtual power plant operation platform. Specifically, it includes energy storage power station operation data, interruptible capacity of industrial load, distributed photovoltaic output data, and virtual power plant contract data. Among them, the energy storage power station operation data includes charging and discharging power, remaining power SOC, charging and discharging efficiency, and fault status; the interruptible capacity of industrial load includes current load value, maximum interruptible power, interruption duration, and recovery time; the distributed photovoltaic output data involves real-time output power, output forecast value, module temperature, and irradiance; and the virtual power plant contract data includes the adjustable capacity agreed in the contract, response time requirements, and subsidy standards.

[0039] The core data on the vehicle-to-grid (V2G) interaction side is collected through the charging pile operator management system, electric vehicle on-board terminal (OBD interface), and vehicle-to-grid platform. Specifically, it includes real-time charging pile power, electric vehicle battery status, and operator profile data. Among them, the real-time charging pile power includes charging power, discharging power, and idle status; the electric vehicle battery status includes SOC, battery capacity, cycle count, and charge / discharge rate limit; and the operator profile data involves operator qualifications, resource ownership area, and service agreement terms.

[0040] Core data on the power grid side are collected through the power grid dispatch automation system (SCADA / EMS), specifically including load gap data, heavy overload early warning information, and renewable energy consumption demand data. Among them, load gap data includes real-time load value, predicted load value, and load gap amount; heavy overload early warning information includes warning line number, warning period, and overload magnitude; and renewable energy consumption demand data involves the predicted total renewable energy output, current wind and solar curtailment, and consumption target value.

[0041] Data acquisition frequency and communication method:

[0042] The high-frequency dynamic data is collected once every minute. This data includes real-time power of charging piles, energy storage charging and discharging power, and real-time load of the power grid. It is transmitted using the MQTT lightweight communication protocol to ensure low-latency transmission.

[0043] The intermediate frequency static data is collected once per hour. This type of data includes interruptible capacity of industrial load, operator profile data, virtual power plant contract data, etc. It is transmitted using the TCP / IP protocol to ensure the stability of data transmission.

[0044] The frequency of collecting forecast data is 6 hours / time. This type of data includes photovoltaic output forecasts, load forecasts, etc. After being generated by a professional forecasting system, it is pushed to the data sensing and standardization module of this invention through the OPC UA protocol to ensure timely acquisition of forecast data.

[0045] Data acquisition reliability assurance:

[0046] The system adopts a "primary and backup dual-link" transmission mechanism. When the primary link is interrupted, the system will automatically switch to the backup link, and the switching time will not exceed 3 seconds, ensuring the continuity of data transmission.

[0047] All collected data are timestamped and have a checksum added. The checksum is generated using the CRC32 checksum algorithm. After receiving the data, the receiving end verifies the data integrity using the checksum. If the verification fails, a retransmission mechanism is immediately triggered to ensure the accuracy of the data.

[0048] 1.2 Data Standardization Processing

[0049] To address the issue of inconsistent formats in multi-source heterogeneous data, this step converts all data into a unified structure of "Resource ID - Timestamp - Core Parameters" according to a "preset standard." The specific processing flow is as follows:

[0050] Preset standard definition:

[0051] Data standardization prioritizes the adoption of national standard GB / T 38946-2020 "Technical Requirements for Smart Grid Dispatch and Control Systems" and industry standard DL / T 1860-2018 "Smart Grid Dispatch and Control Systems Part 1: Terminology". For fields without corresponding national or industry standards, custom formats are used for standardization.

[0052] The data accuracy requirements are as follows: power parameters should be retained to one decimal place, and the unit should be uniformly kW; time parameters should be accurate to the second, and the format should be uniformly YYYY-MM-DD HH:MM:SS; status parameters should be represented by enumerated values, for example, in the fault status, 0 represents normal, 1 represents alarm, and 2 represents fault.

[0053] Field mapping rules:

[0054] Construct a mapping relationship between multiple data fields, establishing a one-to-one correspondence between the original fields from different sources and standardized core parameters to ensure data consistency. The specific mapping relationships are as follows: The original field "Charging / Discharging Power Value" of an energy storage power station corresponds to the standardized core parameter "P_ess", which is a floating-point type in kW, where a positive value indicates discharging and a negative value indicates charging; the original field "Remaining Capacity Percentage" of an energy storage power station corresponds to the standardized core parameter "SOC_ess", which is a floating-point type in %; the original field "Maximum Interruptible Power" of industrial load corresponds to the standardized core parameter "P_ind_cut_max", which is a floating-point type in kW; the original field "Real-time Output" of distributed photovoltaic power corresponds to the standardized core parameter "P_pv_real", which is a floating-point type in kW; the original field "Current Charging Power" of a charging pile corresponds to the standardized core parameter "P_evse", which is a floating-point type in kW, where 0 indicates idle state; the original field "Remaining Battery Capacity" of an electric vehicle corresponds to the standardized core parameter "SOC_ev", which is a floating-point type in %; and the original field "Real-time Load" on the grid side... The corresponding standardized core parameter "P_grid_load" is a floating-point type with a unit of kW; the original field "load gap" on the grid side corresponds to the standardized core parameter "P_grid_gap" with a floating-point type with a unit of kW, where positive values ​​represent gaps and negative values ​​represent surpluses; the original field "renewable energy absorption demand" on the grid side corresponds to the standardized core parameter "P_ne_absorb_req" with a floating-point type with a unit of kW.

[0055] Missing data handling:

[0056] For missing data encountered during the data collection process, a "layered completion strategy" is used for processing:

[0057] Short-term missing data refers to data with a missing duration of no more than 5 minutes. This type of data is filled using linear interpolation. Specifically, the missing value is calculated based on the two adjacent valid data points before and after the missing period, according to the time interval ratio.

[0058] Mid-term missing data refers to data that is missing for more than 5 minutes but no more than 30 minutes. This type of data is supplemented using the average value of the same type of resources during the same period. Specifically, the historical average data of the same type of resources in the same resource pool during the same period is selected as the supplementary value.

[0059] Long-term missing data refers to data that has been missing for more than 30 minutes. This type of data is marked as "invalid data" and triggers an alarm mechanism to notify operations and maintenance personnel to investigate data acquisition link failures.

[0060] Data standardization output:

[0061] The standardized data is stored in a distributed database (HBase) in a fixed format of "Resource ID - Timestamp - Core Parameters". For example, the standardized data for an energy storage power station resource (ID: ESS-001) is "ESS-001 | 2024-06-10 14:30:00 | P_ess=120.5, SOC_ess=65.2, Eff_ess=92.3"; the standardized data for a charging pile resource (ID: EVSE-156) is "EVSE-156 | 2024-06-10 14:30:00 | P_evse=35.8, Status_evse=1", where Status_evse=1 indicates that the charging is in progress.

[0062] (II) Step S2: Tag system construction, tagging processing and functional resource pool construction

[0063] 2.1 Construction of a Four-Dimensional Tag System

[0064] This step constructs a tag system encompassing four dimensions: subject, attribute, capability, and scenario. All tags are generated based on the standard dataset output by S1. The tag rule base is built into the tag engine of the resource tag and aggregation module. The specific tag definitions, value ranges, and calculation methods are as follows:

[0065] Main tags:

[0066] The main tags are used to identify the basic attributes of resources for easy classification and management. Tag values ​​can be enumerated or fixed. Specifically, the "Operator Type" tag ranges from virtual power plant operators, charging pile operators, electric vehicle users, and power grid companies; the data comes from the "Operator Profile Data" in the standard dataset and is automatically assigned a value based on resource affiliation. The "Resource Type" tag ranges from energy storage power stations, distributed photovoltaics, industrial interruptible loads, public charging piles, and private electric vehicles; the data comes from the "Basic Resource Information" in the standard dataset and is automatically assigned a value based on equipment type. The "Access Area" tag ranges from Area A, Area B, Area C, and Area D; this area division is determined according to the provincial power grid administrative divisions; the data comes from the "Resource Location Information" in the standard dataset and is assigned a value mapped to the corresponding administrative region based on latitude and longitude.

[0067] Attribute tags:

[0068] Attribute tags are used to describe the operational characteristics and reliability of resources. Tag values ​​are quantitative indicators or level classifications. The "Response Accuracy" tag is defined as the ratio of the actual number of resource responses to the number of scheduling commands issued, with values ​​ranging from 90%-100% (Excellent), 80%-89% (Good), 70%-79% (Pass), and <70% (Fail). The calculation method is Response Accuracy = (Actual Responses / Commands Issued) × 100%, with a statistical period of 7 days. The "Adjustment Accuracy" tag is defined as the deviation rate between the actual and target adjustment power of the resource, with values ​​ranging from ≤±1% (Level 1), ±1%-±3% (Level 2), ±3%-±5% (Level 3), and >±5% (Level 4). The calculation method is Adjustment Accuracy = |(Actual Adjustment Power - Target Adjustment Power) / Target Adjustment Power| × 100%, with the average of 3 tests used as the final result. "Failure Rate"... The label is defined as the ratio of failure duration to total runtime during resource operation. The value range is divided into three levels: <0.5% (low), 0.5%-1% (medium), and >1% (high). The calculation method is failure rate = (failure duration / total runtime) × 100%, and the statistical period is set to 30 days.

[0069] Ability Tags:

[0070] Capability tags are used to characterize the upper limit of a resource's technical capability to participate in scheduling, and the tag values ​​are quantified numerical values. The "Maximum Adjustable Capacity" tag is defined as the maximum adjustable power that a resource can provide under safe operating conditions, ranging from 0-500kW. The acquisition method varies depending on the resource type: for energy storage power stations, it is calculated based on battery capacity and charge / discharge rate; for charging piles, it is determined based on the equipment's rated power; and for industrial loads, it is determined based on the contractually agreed interruptible capacity. The "Response Delay Time" tag is defined as the time interval between receiving a scheduling command and starting to execute an adjustment action, ranging from 0-5s. It is obtained through field testing, and the average delay time of 10 tests is taken as the final result. The "Continuous Adjustment Duration" tag is defined as the maximum time a resource can continuously execute adjustment actions under maximum adjustable capacity, ranging from 0-24h. The acquisition method varies depending on the resource type: for energy storage power stations, it is calculated based on remaining power and charge / discharge power; and for industrial loads, it is determined based on the allowable interruption time of the production process.

[0071] Scene tags:

[0072] Scenario tags are used to match the differentiated dispatching needs of the power grid, and the tag value can be "suitable" or "unsuitable". The compatibility criteria for the "Peak Shaving Applicable Scenarios" tag are: maximum adjustable capacity of resources not less than 10kW, response delay time not exceeding 3s, adjustment accuracy not exceeding ±3%, and operating time covering the peak grid period (9:00-11:00, 19:00-21:00). The criteria are based on the "capacity label" in the standard dataset and the grid load period data. The compatibility criteria for the "Valley Filling Applicable Scenarios" tag are: maximum adjustable capacity of resources not less than 5kW, response delay time not exceeding 5s, adjustment accuracy not exceeding ±5%, and operating time covering the off-peak grid period (0:00-6:00). The criteria are based on the "capacity label" in the standard dataset and the grid load period data. The compatibility criteria for the "Renewable Energy Consumption Applicable Scenarios" tag are: absorbable power of resources not less than 20kW, continuous adjustment duration not less than 2h, and operating time covering the peak renewable energy period (10:00-16:00). The criteria are based on the "capacity label" in the standard dataset and the renewable energy output period data.

[0073] 2.2 Resource Tagging Processing

[0074] Tagging is performed automatically by the tagging engine, and the specific process is as follows:

[0075] The tag engine reads a standard dataset from a distributed database and associates various types of data one by one by "resource ID" to ensure that the relevant data for each resource is completely and completely matched.

[0076] Based on the definition and assignment rules of the above four-dimensional tags, the tag values ​​of each resource are automatically calculated or matched. Taking a certain energy storage power station (ID: ESS-001) as an example, its standard data is a maximum adjustable capacity of 200kW, a response delay time of 1.2s, a regulation accuracy of ±0.8%, and an operating period of 0:00-24:00 with stable power output during peak grid periods. The corresponding tagging results are as follows: the main tags include operator type (virtual power plant operator), resource type (energy storage power station), and access area (area A); the attribute tags include response accuracy (96%), regulation accuracy (level 1), and fault occurrence rate (0.3%); the capacity tags include maximum adjustable capacity (200kW), response delay time (1.2s), and continuous regulation duration (8h); and the scenario tags include peak shaving applicable scenarios (adaptive), valley filling applicable scenarios (adaptive), and new energy consumption applicable scenarios (adaptive).

[0077] The tagging engine binds the tagging result of each resource to the corresponding resource ID and stores it in the tag library. It also supports dynamic tag updates, with an update frequency consistent with the functional resource pool, which is 5-15 minutes / time. In this embodiment, 10 minutes / time is selected as the fixed update frequency.

[0078] For resources whose tag calculation fails or whose values ​​are abnormal, the system marks them as "pending review" and pushes them to the operation and maintenance platform, where manual verification of the relevant data is performed to supplement and improve the tag information.

[0079] 2.3 Construction of Functional Resource Pool

[0080] Based on the scene tag matching results, a functional resource pool is dynamically constructed by the resource pool construction unit. The specific process is as follows:

[0081] The resource pool construction unit reads the scene tags of all resources from the tag library and classifies and filters them according to "scene adaptability": It filters all resources with the scene tag "peak shaving applicable scene = adaptable" to construct a peak shaving resource pool; it filters all resources with the scene tag "valley filling applicable scene = adaptable" to construct a valley filling resource pool; and it filters all resources with the scene tag "new energy consumption applicable scene = adaptable" to construct a new energy consumption resource pool. It should be noted that if a resource adapts to multiple scene tags simultaneously (such as ESS-001 above), it is preferentially allocated to the resource pool with the highest matching degree according to "tag matching degree". The tag matching degree is calculated as (number of matching tags for the adapted scene / total number of tags for that scene) × 100%. In this embodiment, the peak shaving scene matching degree of ESS-001 is 100%, so it is preferentially allocated to the peak shaving resource pool.

[0082] The resource pool construction unit calculates the core parameters of each resource pool, including total adjustable capacity, average response latency, and average adjustment accuracy. The total adjustable capacity is the sum of the maximum adjustable capacities of all resources in the resource pool, the average response latency is the arithmetic mean of the response latency of all resources in the resource pool, and the average adjustment accuracy is the arithmetic mean of the adjustment accuracy of all resources in the resource pool. The above core parameters, along with the resource pool type, the number of resources contained, and the update time, are stored in the resource pool information table. Taking this embodiment as an example, the peak-shaving resource pool contains 320 resources with a total adjustable capacity of 45,000 kW, an average response delay of 1.8 s, an average adjustment accuracy of ±1.2%, and an update time of 14:30:00 on June 10, 2024; the valley-filling resource pool contains 450 resources with a total adjustable capacity of 38,000 kW, an average response delay of 2.5 s, an average adjustment accuracy of ±2.1%, and an update time of 14:30:00 on June 10, 2024; and the new energy consumption resource pool contains 280 resources with a total adjustable capacity of 52,000 kW, an average response delay of 2.2 s, an average adjustment accuracy of ±1.8%, and an update time of 14:30:00 on June 10, 2024.

[0083] The resource pool adopts a dynamic update mechanism, specifically including two methods: scheduled updates and triggered updates. It also supports resource pool expansion and cross-pool resource migration. Scheduled updates are triggered every 10 minutes. The resource pool construction unit rereads the tag library, selects resources that meet the conditions, and updates the resource pool members and core parameters. Triggered updates are triggered when there is a heavy overload warning (warning level ≥ Level 2) or a sudden change in the output of new energy sources (fluctuation amplitude ≥ 20% within 15 minutes) on the grid side, ensuring that the resource pool can quickly adapt to changes in grid status. When the total adjustable capacity of a resource pool cannot meet the grid dispatching needs, the expansion mechanism is activated to include edge resources with "scenario tag matching degree ≥ 80%" into the resource pool. When the tagging result of a resource changes (such as an increase or decrease in response accuracy) and causes a change in scenario adaptability, cross-pool resource migration is automatically triggered. For example, if the original adjustment accuracy of a charging pile is ±3.5% (Level 3), and it is improved to ±2.8% (Level 2) after operation and maintenance optimization, it will migrate from the valley filling resource pool to the peak shaving resource pool.

[0084] (III) Step S3: Three-dimensional dynamic weight calculation and construction of multi-objective optimization scheduling model

[0085] 3.1 Calculation of Three-Dimensional Dynamic Weights

[0086] This step, based on the grid operation status data output by S1 (with a focus on grid load factor and predicted renewable energy output), calculates in real time the grid security weight ω1, renewable energy absorption weight ω2, and user revenue weight ω3. The weights satisfy ω1+ω2+ω3=1. The specific calculation process and rules are as follows:

[0087] Core input parameter definitions:

[0088] The grid load factor λ is calculated as the ratio of the current real-time grid load P_grid_load to the maximum grid load P_grid_max. In this embodiment, the maximum grid load P_grid_max is set to 150,000 kW. The proportion of renewable energy output γ is calculated as the ratio of the predicted renewable energy output P_ne_pred to the real-time grid load P_grid_load.

[0089] Refinement of weight constraint rules:

[0090] When the grid load factor λ ≥ 90%, it indicates that the grid is close to overload. At this time, grid safety has the highest priority, and the grid safety weight ω1 ≥ 0.6 is forcibly constrained. The renewable energy consumption weight ω2 and the user revenue weight ω3 are dynamically allocated according to the renewable energy output ratio γ. The specific allocation formula is as follows: when γ ≥ 30%, ω2 = 0.4; when γ < 30%, ω2 = 0.4 × γ / 0.3; user revenue weight ω3 = 1 - ω1 - ω2. For example: if λ = 92% and γ = 35%, then ω1 = 0.6, ω2 = 0.4, ω3 = 0; if λ = 91% and γ = 20%, then ω1 = 0.6, ω2 = 0.4 × 20% / 30% ≈ 0.267, ω3 = 1 - 0.6 - 0.267 ≈ 0.133.

[0091] When the proportion of new energy output γ ≥ 30% and the grid load factor λ < 90%, it indicates that the pressure of new energy consumption is high. In this case, consumption has the highest priority, and the new energy consumption weight ω2 ≥ 0.5 is forcibly constrained. The grid security weight ω1 and the user benefit weight ω3 are dynamically allocated according to the grid load factor λ. The specific allocation formula is as follows: when λ ≥ 50%, ω1 = 0.5 × λ / 0.9; when λ < 50%, ω1 = 0.278; user benefit weight ω3 = 1 - ω1 - ω2. For example: if γ = 32% and λ = 85%, then ω2 = 0.5, ω1 = 0.5 × 85% / 90% ≈ 0.472, ω3 = 1 - 0.472 - 0.5 ≈ 0.028; if γ = 40% and λ = 45%, then ω2 = 0.5, ω1 = 0.278, ω3 = 1 - 0.278 - 0.5 ≈ 0.222.

[0092] When the grid load factor λ < 90% and the proportion of renewable energy output γ < 30%, it indicates that the grid security and absorption pressure are relatively low. In this case, user benefits are taken into account, and the weights are allocated to a base value according to a fixed ratio, and then fine-tuned according to λ and γ. The base values ​​are set as ω1=0.4, ω2=0.3, and ω3=0.3; the fine-tuning rule is that for every 1% increase in λ, ω1 increases by 0.005 and ω3 decreases by 0.005; for every 1% increase in γ, ω2 increases by 0.005 and ω3 decreases by 0.005; after fine-tuning, the weights must satisfy ω1≥0.3, ω2≥0.2, and ω3≥0.1. For example: if λ=75% and γ=25%, then ω1=0.4 + (75%-70%)×0.005=0.425, ω2=0.3 + (25%-20%)×0.005=0.325, and ω3=1-0.425-0.325=0.25.

[0093] Weight calculation results output:

[0094] The weight calculation unit stores the real-time calculated power grid security weight ω1, renewable energy consumption weight ω2, and user revenue weight ω3 in a cache, with an update frequency of 1 minute / time, to ensure that the multi-objective optimization scheduling model can obtain the latest objective priorities in real time.

[0095] 3.2 Construction of Multi-Objective Optimization Scheduling Model

[0096] The multi-objective optimization scheduling model in this embodiment takes "minimizing grid load fluctuations, maximizing renewable energy absorption, and maximizing overall user benefits" as its core objectives. It combines three-dimensional dynamic weights and constraints to construct a complete mathematical model, as detailed below:

[0097] Objective function construction:

[0098] To achieve multi-objective collaborative optimization, each individual objective is first normalized (to eliminate dimensional differences), and then fused into a total objective function through three-dimensional dynamic weighting. The total objective function adopts a "minimization" form (to facilitate algorithm solution).

[0099] Single-objective function definition:

[0100] Objective 1 (Minimize grid load fluctuations): The objective function expression is as follows: =∑t=1^T |P_grid_load (t)- P_grid_avg|, where T is the total number of time periods in the scheduling cycle. In this embodiment, T=96 (i.e., 15 minutes / time period, 24 hours a day), P_grid_load (t) is the real-time load of the power grid in time period t, and P_grid_avg is the average load of the power grid within the scheduling cycle. The calculation method is P_grid_avg=∑t=1^T P_grid_load (t) / T.

[0101] Objective 2 (Maximize the amount of new energy consumed): The objective function expression is as follows: =∑t=1^T P_ne_used (t), where P_ne_used (t) is the actual renewable energy power consumed in time period t, and P_ne_used (t)≤P_ne_pred(t) (P_ne_pred (t) is the predicted renewable energy output in time period t).

[0102] Objective 3 (Maximizing overall user benefit): The objective function expression is as follows: =∑t=1^T ∑ᵢ=1^N [R_i (t)- C_i (t)], where N is the total number of resources participating in the scheduling, R_i (t) is the revenue of resource i in time period t, and C_i (t) is the control cost of resource i in time period t.

[0103] The revenue R_i(t) consists of two parts: first, the peak-valley electricity price difference revenue R1_i(t), calculated as R1_i(t) = |P_i(t)| × (P_peak - P_valley), where P_peak is the peak electricity price and P_valley is the valley electricity price. In this example, P_peak = 1.0 yuan / kWh and P_valley = 0.3 yuan / kWh; second, the demand response subsidy R2_i(t), implemented according to the grid company's standards, with a subsidy of 0.5 yuan / kWh for peak shaving scenarios, 0.3 yuan / kWh for valley filling scenarios, and 0.4 yuan / kWh for renewable energy consumption scenarios, i.e., R_i(t) = R1_i(t) + R2_i(t).

[0104] The cost C_i(t) consists of two parts: first, the energy consumption and loss cost C1_i(t), which is calculated according to resource type. For energy storage power stations, the calculation method is C1_i(t) = |P_i(t)| × t × (1 - η_ess) × P_e (η_ess is the energy storage charging and discharging efficiency, and P_e is the electricity price). For electric vehicles, the calculation method is C1_i(t) = |P_i(t)| × t × C_battery (C_battery is the battery degradation cost coefficient, and in this embodiment, C_battery = 0.05 yuan / kWh); second, the equipment operation and maintenance cost C2_i(t), which is a fixed value according to resource type: 0.02 yuan / kWh for energy storage power stations, 0.01 yuan / kWh for charging piles, and 0.03 yuan / kWh for industrial loads, i.e., C_i(t) = C1_i(t) + C2_i(t).

[0105] Normalization process:

[0106] Linear normalization is performed on each single objective function to map its value range to [0,1], thus eliminating dimensional differences. For the minimization objective (f1), the normalization formula is f1_norm(t)=(f1(t) - f1_min) / (f1_max - f1_min), where f1_min is the theoretical minimum value of f1 (0), and f1_max is the historical maximum value of f1. In this embodiment, the maximum load fluctuation value in the past 30 days is taken, and f1_max=50000kW·h. For the maximization objectives (f2, f3), the normalization formulas are f2_norm(t)=(f2(t) - f2_min) / (f2_max - f2_min) and f3_norm(t)=(f3(t) - f3_min) / (f3_max - f3_min), respectively, where f2_min=0, f2_max is the maximum renewable energy absorption capacity (∑t=1^T P_ne_pred(t)), and f3_min is the minimum user benefit (-∑t=1^T ∑ᵢ=1^N). C_i(t)), f3_max is the user's maximum profit (theoretical maximum value, f3_max = 200000 yuan in this embodiment).

[0107] Overall objective function:

[0108] The overall objective function is expressed as F = ω1 × f1_norm + ω2 × (1 - f2_norm) + ω3 × (1 - f3_norm), where (1 - f2_norm) and (1 - f3_norm) transform the maximization objective into the minimization objective. The smaller the overall objective F, the better the multi-objective collaborative optimization effect.

[0109] Constraint definition:

[0110] The model constraints ensure the feasibility and safety of the scheduling scheme, and all constraints are quantified into mathematical expressions:

[0111] Adjustable capacity constraint for resources: The expression is P_i_min(t) ≤ P_i(t) ≤ P_i_max(t), where P_i(t) is the adjustable power of resource i in time period t (positive value is discharging / reducing load, negative value is charging / increasing load), P_i_max(t) is the maximum adjustable power of resource i (taken from the capacity label "maximum adjustable capacity"), and P_i_min(t) is the minimum adjustable power of resource i. The calculation is differentiated according to resource type: P_i_min(t) for energy storage power station = -SOC_i(t)×C_i×η_ess / t (C_i is the battery capacity), P_i_min(t) for charging pile = -P_evse_rated (P_evse_rated is the rated power of charging pile), and P_i_min(t) for industrial load = -P_ind_cut_max (P_ind_cut_max is the maximum interruptible power of industrial load).

[0112] Power balance constraint: The expression is ∑ᵢ=1^N P_i (t) + P_ne_used (t) + P_conv (t) =P_grid_load (t), where P_conv (t) is the output of the traditional power source in time period t, which is given by the power grid dispatch center. In this embodiment, P_conv (t) ≥ 30000kW.

[0113] Response speed constraint: The expression is |P_i(t) - P_i(t-1)| ≤ ΔP_i_max, where ΔP_i_max is the maximum adjustment rate of resource i, which is set according to resource type: energy storage power station ΔP_i_max=50kW / s, charging pile ΔP_i_max=10kW / s, industrial load ΔP_i_max=20kW / s.

[0114] Cost constraint: The expression is ∑t=1^T ∑ᵢ=1^N C_i (t) ≤ C_total_max, where C_total_max is the total scheduling budget. In this embodiment, C_total_max = 50,000 yuan / day.

[0115] New energy consumption constraint: The expression is P_ne_used (t) ≤ P_ne_pred (t) and P_ne_used(t) ≥ P_ne_absorb_min (t), where P_ne_absorb_min (t) is the minimum amount of new energy consumption in time period t (given by the grid side, in this embodiment P_ne_absorb_min (t) ≥ 0.8 × P_ne_pred (t)).

[0116] Model storage and retrieval:

[0117] The model building unit stores the above objective function, constraints and parameters in the model library, supports dynamic loading according to scheduling needs, and automatically triggers model parameter updates when the power grid operating state changes suddenly (such as λ≥95% or γ≥40%), ensuring that the model is adapted to the actual state of the power grid.

[0118] II. Detailed Explanation of Implementation Steps

[0119] (iv) Step S4: Demand input preprocessing, multi-objective optimization scheduling model solution and optimal solution output

[0120] 4.1 Demand Preprocessing Process

[0121] This step first preprocesses the grid load gap demand and renewable energy consumption demand in the standard dataset output by S1 to ensure the accuracy and suitability of the input data, providing a reliable foundation for model solving. The specific process is as follows:

[0122] Spatiotemporal correlation matching:

[0123] The core of spatiotemporal correlation matching is to precisely bind the grid load gap demand with the renewable energy consumption demand according to the scheduling period, ensuring consistency between the two in both time and spatial dimensions. In the time dimension, the entire day is divided into 96 time periods, each lasting 15 minutes. The grid load gap demand (including real-time and predicted gap amounts) for each time period is correlated one-to-one with the renewable energy consumption demand (including consumption targets and curtailment warnings) for the corresponding time period, forming a three-dimensional data set of "time period - load gap - consumption demand". In the spatial dimension, the demand data is partitioned according to resource access areas (areas A, B, C, and D). The load gap demand of a specific area is preferentially correlated with the renewable energy consumption demand within that area, avoiding transmission losses and delays caused by cross-regional scheduling. For example, if the load gap demand in region A is 8000kW during the period from 19:00 to 19:15, and the demand for renewable energy consumption in the same region is 6000kW, then the associated data group is formed as "19:00-19:15 | Region A | Load gap 8000kW | Consumption demand 6000kW".

[0124] Outlier removal:

[0125] Outlier removal employs a dual mechanism of "rule-based judgment + trend verification" to ensure the accuracy of the removal results. First, basic judgment rules are established: the reasonable range for grid load shortfall demand is no more than 20% of the grid's maximum load, and the reasonable range for renewable energy consumption demand is no less than 0 (i.e., no negative consumption demand). Data exceeding these ranges is initially identified as outliers. Subsequently, trend verification is performed by comparing the demand data trends of three adjacent time periods. If the outlier deviates from the trend by more than 30%, it is ultimately determined to be an outlier requiring removal. If the deviation is within 30%, it is considered a reasonable fluctuation, and the data is retained and marked as "trend fluctuation data." For example, if the demand shortfall in the power grid during a certain period is 35,000 kW (the maximum power grid load is 150,000 kW, and 20% is 30,000 kW), which exceeds the basic range, and the deviations from the load shortfalls in the preceding and following periods (28,000 kW and 29,000 kW) are 25% and 20.7% respectively, neither exceeding 30%, then it is judged as a reasonable fluctuation and retained. If the demand shortfall in the power grid during a certain period is 40,000 kW, and the deviations from the preceding and following periods both exceed 35%, then it is judged as an outlier, removed, and supplemented using the historical average demand data of the same period in the region.

[0126] 4.2 Improved Implementation of Intelligent Optimization Algorithm (Taking the Improved Particle Swarm Optimization Algorithm as an Example)

[0127] This step employs an improved particle swarm optimization (PSO) algorithm that integrates global search and local convergence optimization to solve the multi-objective optimization scheduling model. This algorithm addresses the problem of traditional PSO algorithms easily getting trapped in local optima by optimizing inertia weights and introducing a local search strategy. Specific implementation details are as follows:

[0128] Core principle of the algorithm:

[0129] The Particle Swarm Optimization (PSO) algorithm treats each possible scheduling scheme as a "particle," and all particles form a "particle swarm." Within the solution space, particles update their positions by following the best individual particle and the globally best particle, gradually approaching the optimal solution. The improvements of this invention are: first, dynamically adjusting the inertia weight to balance global search capability and local convergence speed; second, introducing a local search strategy from the simulated annealing algorithm to prevent the particle swarm from converging prematurely to a local optimum; and third, designing a fitness function in conjunction with model constraints to ensure that the search direction meets scheduling requirements.

[0130] Algorithm parameter settings:

[0131] The particle swarm size is set to 100 (i.e., generating 100 candidate scheduling schemes simultaneously). The solution space dimension is consistent with the total number of resources N participating in the scheduling (in this embodiment, N = 320 + 450 + 280 = 1050). The position vector of each particle corresponds to the control power P_i(t) of each resource in each scheduling period. The maximum number of algorithm iterations is set to 150. The individual learning factor c1 and the global learning factor c2 are both set to 2.0 (to balance the influence of individual experience and group experience). The initial inertia weight w0 is set to 0.9, and the minimum inertia weight w_min is set to 0.4.

[0132] Dynamic inertia weight adjustment:

[0133] The inertia weight employs a combination of linear decreasing and adaptive adjustment. The linear decreasing formula is w(k) = w0 - (w0 - w_min) × k / K_max, where k is the current iteration number and K_max is the maximum iteration number. Based on this, if the global optimal solution remains unchanged for 5 consecutive iterations, the inertia weight is temporarily increased to 0.7 to trigger a global search and avoid getting trapped in local optima. For example, when the iteration reaches the 50th iteration, if the global optimal solution has not been updated for 5 consecutive iterations, the inertia weight is increased from the current 0.67 to 0.7. The particle swarm then performs a broader search within the solution space. If a better solution is found, the iteration continues; otherwise, the linear decreasing trend resumes.

[0134] Fitness function design:

[0135] The fitness function directly adopts the overall objective function F of the multi-objective optimization scheduling model, i.e., the fitness value f = F = ω1 × f1_norm + ω2 × (1 - f2_norm) + ω3 × (1 - f3_norm). The smaller the fitness value, the better the multi-objective collaborative optimization effect of the candidate scheduling scheme. Simultaneously, a penalty mechanism is introduced in the fitness calculation process. For particles that violate constraints, the fitness value is increased according to the degree of violation: for violations of adjustable resource capacity constraints, the fitness value increases by 0.001 for every 1kW exceeding the limit; for violations of power balance constraints, the fitness value increases by 0.01 for every 1% deviation; and for violations of renewable energy consumption constraints, the fitness value increases by 0.02 for every 1% below the guaranteed minimum. This penalty mechanism guides particles to search for feasible solutions.

[0136] Iterative update process:

[0137] Initialize the particle swarm: Randomly generate initial position vectors for 100 particles. The position component of each particle (i.e., the control power of each resource) is randomly selected within the adjustable capacity range of the corresponding resource, while ensuring that the initial position satisfies the approximate condition of power balance constraint (deviation not exceeding 5%).

[0138] Calculate fitness value: For each particle, substitute it into the multi-objective optimization scheduling model, calculate its fitness value, and record the individual optimal position of each particle (the position corresponding to its own historical optimal fitness) and the global optimal position of the particle swarm (the position corresponding to the historical optimal fitness of all particles).

[0139] Particle position and velocity update: The particle velocity update formula is v_i (k+1)=w (k)×v_i (k) + c1×r1×(p_best_i - x_i (k)) + c2×r2×(g_best - x_i (k)), where v_i (k) is the velocity of particle i in the k-th iteration, x_i (k) is the position of particle i in the k-th iteration, p_best_i is the individual optimal position of particle i, g_best is the global optimal position, and r1 and r2 are random numbers between 0 and 1; The particle position update formula is x_i (k+1)=x_i (k) + v_i (k+1). The updated position needs to be checked again to see if it meets all constraints. If it does not meet the constraints, it needs to be corrected (if it exceeds the adjustable capacity, the boundary value is taken).

[0140] Local search strategy: When the number of iterations reaches 70% of the total number of iterations (105 in this embodiment), a local search is performed on the region where the global optimal particle is located. The Metropolis criterion of the simulated annealing algorithm is adopted to accept worse solutions with a certain probability to avoid premature convergence. Specifically, 5 candidate positions are randomly generated near the position of the global optimal particle, and their fitness values ​​are calculated. If the fitness value of a candidate position is better than that of the current global optimal position, it is directly replaced; if it is worse than that of the current global optimal position, the acceptance probability P=exp (-Δf / T) is calculated, where Δf is the difference in fitness values ​​between the candidate position and the current global optimal position, and T is the annealing temperature (decreasing with the number of iterations, with an initial temperature T0=10 and a cooling coefficient of 0.95). If the random number r≤P, the candidate position is accepted; otherwise, it is discarded.

[0141] Convergence criteria:

[0142] The algorithm convergence is determined by two conditions: first, the rate of change of the global optimal fitness value over 10 consecutive iterations does not exceed 0.1%; second, the current iteration number reaches the maximum number of iterations (150). The algorithm stops iterating when either condition is met. The rate of change of fitness value is calculated as |f(k) - f(k-10)| / f(k-10)×100%, where f(k) is the global optimal fitness value in the k-th iteration, and f(k-10) is the global optimal fitness value in the (k-10)-th iteration. For example, the global optimal fitness value in the 120th iteration is 0.256, and the global optimal fitness value in the 110th iteration is 0.2563, with a change rate of |0.256 - 0.2563| / 0.2563×100%≈0.117%, which does not meet the condition; the global optimal fitness value in the 125th iteration is 0.2562, and the global optimal fitness value in the 115th iteration is 0.2565, with a change rate of ≈0.117%, which still does not meet the condition; the global optimal fitness value in the 130th iteration is 0.2561, and the global optimal fitness value in the 120th iteration is 0.2563, with a change rate of ≈0.078%, which meets the condition of change rate ≤0.1% for 10 consecutive iterations. The algorithm stops iterating and outputs the current global optimal position as the optimal scheduling scheme.

[0143] 4.3 Output of the optimal solution for overall resource scheduling objectives

[0144] After the algorithm converges, the optimal solution for the overall resource scheduling objective output contains four types of core parameters. The specific content and generation logic of each type of parameter are as follows:

[0145] Total control power:

[0146] The total control power is the sum of the control power of all participating resources in each scheduling period. It is calculated separately for each scheduling period and reflects the total control capacity that needs to be provided by the virtual power plant and vehicle-grid interaction resources in each period. For example, the total control power for the period 19:00-19:15 is 8200kW, of which the peak shaving resource pool provides 5500kW, the valley filling resource pool provides 1200kW, and the renewable energy consumption resource pool provides 1500kW. The total control power must be no less than 1.05 times the grid load gap demand for this period (in this example, the load gap is 8000kW, 1.05 times is 8400kW, 8200kW does not meet the standard, so the algorithm will be re-optimized until the redundancy requirement is met, and the final output total control power is 8450kW).

[0147] Time-segmented scheduling objectives:

[0148] The time-shaving scheduling targets are set separately for each scheduling period and resource pool type, specifying the specific control tasks for each resource pool in each time period, including the control power range, regulation rate requirements, and execution start time. For example, the scheduling target for the peak-shaving resource pool during the 19:00-19:15 period is "control power of 5600kW-5800kW, regulation rate ≥30kW / s, execution to start precisely at 19:00:00"; the scheduling target for the renewable energy consumption resource pool is "control power of 1600kW-1700kW, continuous absorption of renewable energy output, regulation accuracy ≤±1.5%".

[0149] Guaranteed minimum amount of new energy consumption:

[0150] The minimum renewable energy consumption guarantee is set according to the scheduling period and represents the minimum renewable energy consumption that must be achieved in each period to ensure that renewable energy consumption demand is guaranteed. The logic for determining the minimum consumption guarantee is that it should not be less than 95% of the input renewable energy consumption demand. For example, if the input renewable energy consumption demand in a certain period is 6000kW, then the minimum consumption guarantee for that period is 5700kW, and the actual consumption in the optimal solution for that period must be ≥5700kW.

[0151] User revenue floor:

[0152] The minimum user revenue threshold is the lowest total revenue for all participating users during the scheduling period, ensuring user participation. The threshold is calculated as total revenue not less than 1.2 times the total control cost. In this example, the total scheduling budget is 50,000 yuan / day, and the total control cost is estimated at 40,000 yuan. Therefore, the minimum user revenue threshold is set at 48,000 yuan. In the optimal solution, the total user revenue must be ≥ 48,000 yuan.

[0153] (V) Step S5: Decomposition of target value for partitioned scheduling

[0154] This step, based on the overall resource scheduling objective and the resource characteristics of each functional resource pool, decomposes the overall objective into partitioned scheduling objective values ​​for each resource pool, ensuring that the decomposition results are scientific, reasonable, and executable. The specific decomposition process is as follows:

[0155] 5.1 Calculation of Response Priority Coefficient k_i

[0156] The response priority coefficient k_i is determined based on the response speed label and adjustment accuracy label obtained through tagging, comprehensively reflecting the response capability and control reliability of the resource pool. The specific calculation rules are as follows:

[0157] Individual tag rating:

[0158] The response speed label is divided into four scoring levels based on latency: 10 points for latency ≤ 1s, 8 points for 1s < latency ≤ 2s, 6 points for 2s < latency ≤ 3s, and 4 points for latency > 3s. The adjustment accuracy label is also divided into four scoring levels: Level 1 (≤ ±1%), 10 points for Level 2 (±1% - ±3%), 8 points for Level 3 (±3% - ±5%), and 4 points for Level 4 (> ±5%). The score for each resource pool's individual label is the arithmetic mean of the individual label scores for all resources within that pool. For example, the average response speed label score for the 320 resources in the peak-shaving resource pool is 8.5 points, and the average adjustment accuracy label score is 9.2 points.

[0159] k_i fusion computing:

[0160] The response priority coefficient k_i is calculated using a weighted summation method, where the weight of the response speed label score is 0.4 and the weight of the adjustment accuracy label score is 0.6. The calculation formula is k_i = 0.4 × S_speed + 0.6 × S_precision, where S_speed is the average response speed score of the resource pool and S_precision is the average adjustment accuracy score of the resource pool. In this embodiment, the peak-shaving resource pool has k1 = 0.4 × 8.5 + 0.6 × 9.2 = 8.92; the valley-filling resource pool has an average response speed score of 7.3 and an average adjustment accuracy score of 7.8, k2 = 0.4 × 7.3 + 0.6 × 7.8 = 7.6; and the renewable energy consumption resource pool has an average response speed score of 7.9 and an average adjustment accuracy score of 8.3, k3 = 0.4 × 7.9 + 0.6 × 8.3 = 8.14.

[0161] 5.2 Decomposition and Calculation of Target Values ​​for Partition Scheduling

[0162] The partitioned scheduling target value is decomposed according to the proportion of the product of the total adjustable capacity of each resource pool and the response priority coefficient k_i. The decomposition formula is P_i = P_total_schedule × (Q_i × k_i) / Σ(Q_i × k_i), where P_total_schedule is the total control power of the overall resource scheduling target, Q_i is the total adjustable capacity of the i-th resource pool, and Σ(Q_i × k_i) is the sum of the products of the total adjustable capacity of all resource pools and k_i.

[0163] Example of decomposition calculation:

[0164] In this embodiment, the total control power P_total_control = 8450kW, the total adjustable capacity of the peak shaving resource pool Q1 = 45000kW, the valley filling resource pool Q2 = 38000kW, and the new energy consumption resource pool Q3 = 52000kW; Σ(Q_i × k_i) = 45000 × 8.92 + 38000 × 7.6 + 52000 × 8.14 = 401400 + 288800 + 423280 = 1113480.

[0165] The target value for partitioned scheduling of the peak-shaving resource pool is P1 = 8450 × (45000 × 8.92) / 1113480 = 8450 × 401400 / 1113480 ≈ 8450 × 0.3605 ≈ 3046 kW;

[0166] The target value for zoned scheduling of the valley-filling resource pool is P2 = 8450 × (38000 × 7.6) / 1113480 = 8450 × 288800 / 1113480 ≈ 8450 × 0.2594 ≈ 2192 kW;

[0167] The target value for zoned scheduling of the new energy consumption resource pool is P3 = 8450 × (52000 × 8.14) / 1113480 = 8450 × 423280 / 1113480 ≈ 8450 × 0.3799 ≈ 3212 kW;

[0168] Decomposition result verification: P1+P2+P3≈3046+2192+3212=8450kW, consistent with the overall P, indicating a balanced decomposition; at the same time, the partition scheduling target value of each resource pool does not exceed its total adjustable capacity (3046kW≤45000kW, 2192kW≤38000kW, 3212kW≤52000kW), satisfying the constraint requirements.

[0169] 5.3 Adjustment and Confirmation of Decomposition Results

[0170] After the decomposition calculation is completed, adjustments need to be made based on the actual operating status of each resource pool to ensure the executability of the decomposition results:

[0171] If the partition scheduling target value of a resource pool accounts for more than 30% of its total adjustable capacity, the decomposition value of that resource pool will be appropriately reduced, and the excess portion will be allocated to other resource pools with lower load rates to avoid overloading a single resource pool. For example, if the partition scheduling target value of a resource pool is 15,000 kW, and its total adjustable capacity is 40,000 kW, accounting for 37.5%, which exceeds 30%, then it will be reduced to 12,000 kW (accounting for 30%), and the excess 3,000 kW will be allocated to other resource pools.

[0172] If a resource pool has a large number of faulty or pending resources (accounting for more than 10%), the decomposition value of that resource pool will be reduced proportionally to the total adjustable capacity of the faulty resources to ensure the reliability of scheduling execution. For example, if a valley-filling resource pool has 45 faulty resources (accounting for 10%), and the total adjustable capacity of the faulty resources is 3800kW (accounting for 10% of the total adjustable capacity of the resource pool), then its partition scheduling target value will be adjusted from 2192kW to 2192×(1-10%)≈1973kW, and the excess 219kW will be allocated to the peak-shaving resource pool.

[0173] After the adjustment is completed, the final partitioned scheduling target value table is generated, which clarifies the total decomposition target and the time-period decomposition target for each resource pool (allocated on an average basis according to 96 scheduling periods, or allocated according to the difference in load gaps between time periods), and is pushed to the scheduling strategy generation module.

[0174] (vi) Step S6: Construction of scheduling strategy model, generation of scheduling instructions and control execution

[0175] 6.1 Construction of Scheduling Strategy Models for Each Functional Resource Pool

[0176] Based on the partitioned scheduling target values ​​and resource characteristics of each resource pool, a differentiated scheduling strategy model is constructed to ensure the relevance and effectiveness of the scheduling scheme. The specific model construction logic is as follows:

[0177] Peak-shaving resource pool scheduling strategy model:

[0178] The core requirement of a peak-shaving resource pool is rapid response to peak grid load. The control priority is response speed > adjustment accuracy > duration; therefore, a PID (Proportional-Integral-Derivative) control model is adopted. The model inputs are the zone scheduling target value and the real-time grid load deviation (actual load - predicted load), and the output is the real-time control power command for each resource. The proportional coefficient Kp of the PID controller is set to 0.8, the integral coefficient Ki to 0.2, and the derivative coefficient Kd to 0.1. The proportional element rapidly responds to load deviations, the integral element eliminates steady-state errors, and the derivative element predicts load change trends and adjusts the control power in advance. For example, when the real-time grid load exceeds the predicted load by 5%, the proportional element rapidly increases the control power, and the derivative element further optimizes the control amplitude based on the load increase rate to ensure rapid peak shaving.

[0179] Valley-filling resource pool scheduling strategy model:

[0180] The core requirement of a valley-filling resource pool is to stably absorb off-peak loads from the power grid. The priority of regulation is duration > regulation accuracy > response speed; therefore, a fuzzy control model is adopted. The model uses the grid load factor (λ) and remaining resource capacity (such as energy storage SOC and electric vehicle SOC) as input fuzzy variables, and the regulation power as the output fuzzy variable. The fuzzy subsets of the input variables are defined as "low," "medium," and "high," and the fuzzy subsets of the output variables are defined as "small," "medium," and "large." Inference and decision-making are performed using a fuzzy rule base (containing 27 fuzzy rules). For example, when the grid load factor is "low" and the remaining resource capacity is "high," the output regulation power is "large"; when the grid load factor is "medium" and the remaining resource capacity is "medium," the output regulation power is "medium," ensuring that the valley-filling demand is met while avoiding excessive charging and discharging of resources.

[0181] New energy consumption-type resource pool scheduling strategy model:

[0182] The core requirement of a renewable energy consumption-oriented resource pool is to accurately match the fluctuations in renewable energy output. The priority of regulation is: adjustment accuracy > duration > response speed. Therefore, a Model Predictive Control (MPC) model is adopted. Based on renewable energy output forecast data (output curves for the next hour) and resource adjustable capacity constraints, a predictive model is constructed to predict the renewable energy output change trend over six scheduling periods (90 minutes). The optimal regulation power command for each period is generated through rolling optimization. The predictive model uses an LSTM (Long Short-Term Memory) network. Inputs include historical renewable energy output data, irradiance, temperature, and other environmental parameters. The output is the predicted output value for the future period, with the prediction error controlled within ±5%. The rolling optimization cycle is 15 minutes, updating the latest actual output data and correcting the predictive model during each optimization to ensure regulation accuracy.

[0183] 6.2 Scheduling Instruction Generation

[0184] After the scheduling strategy model outputs the power control instructions for each resource, it combines information such as control direction and control period to generate a complete scheduling instruction. The specific generation rules are as follows:

[0185] The direction of regulation has been determined:

[0186] The control direction is determined based on resource type and scheduling scenario: In peak shaving scenarios, the control direction for energy storage power stations and industrial interruptible loads is "discharging / reducing load" (outputting positive power), while the control direction for charging piles and electric vehicles is "stopping charging / reverse discharging" (outputting positive power); In valley filling scenarios, the control direction for energy storage power stations and electric vehicles is "charging / increasing load" (outputting negative power), while the control direction for charging piles is "priority charging" (outputting negative power); In renewable energy consumption scenarios, the control direction for all resources is "charging / absorbing power" (outputting negative power), ensuring full consumption of renewable energy output.

[0187] Target power determined:

[0188] The target power is determined based on the control power output by the scheduling strategy model and the maximum adjustable capacity of the resource. It must not exceed the maximum adjustable capacity of the resource and must also meet the response speed constraint. For example, if the maximum adjustable capacity of a certain energy storage power station is 200kW, the model output control power is 180kW, and the response speed constraint is ≤50kW / s, then the target power is set to 180kW, and the adjustment time is set to 4s (180kW / 50kW / s=3.6s, take 4s).

[0189] The period for regulation has been determined:

[0190] The control period is consistent with the scheduling cycle, which is 15 minutes per instance. The instruction clearly indicates the start and end times, accurate to the second. For example, if a scheduling instruction has a control period of "2024-06-10 19:00:00 - 2024-06-10 19:15:00", the resource must start execution on time at the start time and complete the control before the end time.

[0191] Command format specifications:

[0192] The dispatch instructions are generated using a standardized format, including fields such as instruction ID, resource ID, control direction, target power, control period, execution priority, and verification threshold. For example: "Instruction ID: CMD-202406101900-001 | Resource ID: ESS-001 | Control direction: Discharge | Target power: 180kW | Control period: 2024-06-10 19:00:00 - 2024-06-10 19:15:00 | Execution priority: High | Verification threshold: ±2%", where the execution priority is divided into three levels: high, medium, and low. The instruction priority is high for peak shaving scenarios, medium for valley filling scenarios, and medium-high for renewable energy consumption scenarios.

[0193] 6.3 Pre-execution and verification of scheduling instructions

[0194] To ensure the effectiveness of scheduling instructions, a "pilot pre-execution + batch issuance" model is adopted, with the specific process as follows:

[0195] Pilot resource selection:

[0196] Each resource pool randomly selects 3% of its resources as pilot resources based on resource type. Pilot resources must meet conditions such as response accuracy ≥85%, failure rate <0.5%, and normal tag status to ensure the representativeness of the pilot results. In this embodiment, the peak-shaving resource pool selects 10 pilot resources (320×3%≈10), the valley-filling resource pool selects 14 pilot resources (450×3%≈14), and the new energy consumption resource pool selects 8 pilot resources (280×3%≈8).

[0197] Pre-execution and data acquisition:

[0198] A scheduling instruction is issued to the pilot resource, with a pre-execution duration of one scheduling period (15 minutes). Execution data of the pilot resource is collected in real time, including actual start-up time, actual controlled power, regulation accuracy, and fault information. The data collection frequency is 1 second per instance. For example, the pre-execution data for pilot resource ESS-001 is: start-up time 2024-06-10 19:00:00.2 (delay 0.2s), average actual controlled power 179.5kW, regulation accuracy ±0.3%, and no fault information.

[0199] Pre-execution effect verification:

[0200] The verification indicators include three items: startup response rate, adjustment accuracy compliance rate, and failure rate. Startup response rate = (Number of pilot resources that started on time / Total number of pilot resources) × 100%, must be ≥95%; Adjustment accuracy compliance rate = (Number of pilot resources whose adjustment accuracy meets the threshold requirement / Total number of pilot resources) × 100%, must be ≥90%; Failure rate = (Number of pilot resources that failed during pre-execution / Total number of pilot resources) × 100%, must be ≤5%. In this embodiment, the peak-shaving resource pool pilot startup response rate is 100% (10 / 10), the adjustment accuracy compliance rate is 100% (10 / 10), and the failure rate is 0%, meeting the verification requirements. If the adjustment accuracy compliance rate of a resource pool is 85% (not reaching 90%), the scheduling strategy model is returned to re-optimize parameters (such as adjusting the PID controller coefficients), instructions are generated again, and pre-execution is performed until the verification requirements are met.

[0201] 6.4 Batch Issuance and Real-time Feedback of Scheduling Instructions

[0202] After the pre-execution verification passes, scheduling instructions are issued in batches to all resources in each resource pool. The specific issuance strategy and feedback mechanism are as follows:

[0203] Batch distribution order:

[0204] Resource requests are distributed in batches according to the resource access area, with a 3-second interval between each region to avoid communication congestion caused by simultaneous distribution. For example, resource instructions for region A are distributed first, followed by instructions for region B 3 seconds later, and so on. Within the same region, resources are distributed according to their type, prioritizing resources with faster response times (such as energy storage power stations) and then resources with slower response times (such as industrial loads).

[0205] Real-time feedback mechanism:

[0206] After receiving the command, the resource provides real-time feedback on its execution status, including status codes such as "Command received successfully," "Executing," "Execution completed," and "Execution failure." The feedback frequency is once every 5 seconds. Simultaneously, actual controlled power data is fed back every 30 seconds to ensure the dispatch center has real-time control of the execution. For example, the feedback information for resource EVSE-156 is: 2024-06-10 19:00:00.5 (Command received successfully), 2024-06-10 19:00:01 (Executing, actual power 35.6kW), 2024-06-10 19:15:00 (Execution completed, average power 35.7kW).

[0207] Verification of the regulation effect:

[0208] After each scheduling period ends, the regulation performance indicators for each resource pool are calculated, including the regulation power completion rate, regulation accuracy, renewable energy consumption completion rate, and user revenue achievement rate.

[0209] Controlled power completion rate = Actual total controlled power / Zoned scheduling target value × 100%, which must be ≥95%;

[0210] Adjustment accuracy = |Actual average control power - Target power| / Target power × 100%, must be ≤ ±3%;

[0211] New energy consumption completion rate = (Actual new energy power consumed / Guaranteed consumption amount) × 100%, which must be ≥ 100%;

[0212] User revenue achievement rate = (Actual total user revenue / Minimum user revenue) × 100%, must be ≥ 100%.

[0213] If a certain indicator fails to meet the target, analyze the reasons and take remedial measures: if the control power completion rate is insufficient, activate the backup resource pool (as described in claim 10) to supplement the control; if the control accuracy exceeds the target, adjust the scheduling strategy model parameters; if the new energy consumption completion rate is insufficient, increase the control power of the new energy consumption resource pool.

[0214] 6.5 Exception Handling and Backup Scheduling Switchover

[0215] When an abnormal resource response or a sudden change in the power grid's operating state is detected, the backup dispatch scheme is automatically activated to ensure continuous control. The specific processing procedure is as follows:

[0216] Anomaly detection criteria:

[0217] Abnormal resource response includes: failure to receive "successful" feedback within 3 seconds after the instruction is issued, feedback of "fault" during execution, and deviation of actual controlled power from target power exceeding 10% for more than 5 seconds; sudden changes in grid operation status include: grid load rate suddenly rising to over 95%, fluctuation of new energy output exceeding 30% within 15 minutes, and heavy overload warning level rising to level one.

[0218] Backup resource pool activated:

[0219] The backup scheduling unit stores backup resource pool information (composed of edge resources with a scene tag matching degree of 70%-80%) and backup scheduling schemes (scheduling strategies preset based on extreme scenarios). When an anomaly is detected, the backup resource pool is immediately activated, and the backup scheduling scheme is distributed to the backup resources to supplement or replace the control tasks of the abnormal resources. For example, if an energy storage power station reports a "fault" during execution, its 180kW control task is replaced by another energy storage power station in the backup resource pool. The backup resource starts execution within 3 seconds to ensure that the total control power does not decrease.

[0220] Anomaly recovery and switchover rollback:

[0221] Once the abnormal resource is repaired or the grid operation stabilizes, the control tasks are rolled back from the backup resource pool to the original resource pool according to the "gradual switchover" principle. During the rollback process, the total control power is ensured to transition smoothly to avoid load fluctuations. For example, after a faulty energy storage power station is repaired, the control power of the backup resource is first reduced from 180kW to 90kW, while the control power of the original resource is increased from 0 to 90kW, keeping the total power unchanged at 180kW. After the original resource operation stabilizes, the backup resource power is then reduced to 0, and the original resource power is increased to 180kW, completing the rollback.

[0222] III. Complete Dispatch Case Demonstration (Typical Weekday Evening Peak Hour)

[0223] (I) Case Scenario Setting

[0224] A typical scenario is selected based on the period from 19:00 to 19:15 on a certain weekday (the evening peak period of the power grid). During this period, the grid load factor λ=92% (close to overload), the proportion of renewable energy output γ=35% (significant pressure on absorption), and the overall resource dispatch target is a total dispatch power of 8450kW, a guaranteed renewable energy absorption capacity of 5700kW, and a minimum user revenue of 48,000 yuan / day. The resources involved in the dispatch include 320 peak-shaving resource pools, 450 valley-filling resource pools, 280 renewable energy absorption resource pools, and a reserve resource pool containing 50 edge resources.

[0225] (II) Execution process of the entire scheduling process

[0226] Data collection and standardization (S1):

[0227] From 18:55 to 18:59, multi-source data was collected through various communication links. High-frequency dynamic data (such as charging pile power and grid load) was collected every minute, medium-frequency static data every hour, and predictive data every 6 hours. After collection, the data was standardized according to preset standards, mapping the "charging / discharging power value of 120.5kW" of the energy storage power station to "P_ess=120.5", and the "current charging power of 35.8kW" of the charging pile to "P_evse=35.8". The processed data was stored in the HBase database, with no missing or abnormal data.

[0228] Tagging and Resource Pool Construction (S2):

[0229] The tagging engine tags resources according to a four-dimensional tagging system. All 320 resources in the peak-shaving resource pool meet the requirement of "peak-shaving applicable scenario = suitable". The average response latency is 1.8s and the average adjustment accuracy is ±1.2%. The resource pool is updated every 10 minutes. The current total adjustable capacity is 45,000kW and there are no faulty resources.

[0230] 3D Dynamic Weight Calculation and Model Construction (S3):

[0231] With a grid load factor λ=92% and a renewable energy output ratio γ=35%, and weighted by rules, ω1=0.6, ω2=0.4, and ω3=0; the overall objective function of the multi-objective optimization scheduling model is F=0.6×f1_norm + 0.4×(1 - f2_norm), which focuses on optimizing grid security and renewable energy consumption.

[0232] Demand preprocessing and model solving (S4):

[0233] During this period, the grid load gap demand is 8000kW, and the renewable energy consumption demand is 6000kW. A three-dimensional data set was formed through spatiotemporal correlation matching, with no outliers. An improved particle swarm optimization algorithm was used to solve the model, which converged after 130 iterations (with a change rate of 0.078% for 10 consecutive iterations), outputting the optimal solution: total control power of 8450kW, time-segmented scheduling targets of 3046kW for peak shaving resource pool, 2192kW for valley filling resource pool, and 3212kW for renewable energy consumption resource pool during the 19:00-19:15 period, with a guaranteed renewable energy consumption capacity of 5700kW and a minimum user revenue of 48,000 yuan.

[0234] Partition scheduling target value decomposition (S5):

[0235] Calculate the response priority coefficient k_i for each resource pool: peak shaving type k1=8.92, valley filling type k2=7.6, and new energy consumption type k3=8.14. Calculate the target values ​​P1=3046kW, P2=2192kW, and P3=3212kW for each zone according to the decomposition formula. The decomposition results are balanced and meet the constraints, so no adjustment is needed.

[0236] Scheduling execution and feedback (S6):

[0237] Construct scheduling strategy models: PID control is used for peak shaving, fuzzy control is used for valley filling, and MPC model is used for renewable energy consumption.

[0238] Generate scheduling instructions: For example, the instruction for ESS-001 is "Discharge, 180kW, 19:00:00-19:15:00, High Priority";

[0239] Pre-execution verification: Select 10 peak shaving pilot resources, with a pre-execution start-up response rate of 100%, an adjustment accuracy compliance rate of 100%, and a failure rate of 0%. Verification passed.

[0240] Batch distribution: Instructions are distributed in batches according to regions, and execution will start precisely at 19:00:00;

[0241] Real-time feedback: The status of resources is reported every 5 seconds, and the power data is reported every 30 seconds. At 19:05, a charging pile reported a "fault", and the backup resource was immediately activated to replace it. The control was restored within 3 seconds.

[0242] Performance verification: After the 19:15 period ended, the power control completion rate was 98% (8281kW / 8450kW), the control accuracy was ±1.1%, the renewable energy consumption completion rate was 105% (5985kW / 5700kW), and the user income was 52,000 yuan / day, all of which met the target requirements.

[0243] (III) Case Effect Analysis

[0244] The scheduling results during this typical period demonstrate that the method of this invention can effectively integrate multi-source resources, rapidly providing sufficient control power during peak grid periods, reducing the grid load factor from 92% to 88%, and preventing grid overload. The renewable energy absorption rate is increased by 12% compared to traditional methods, and the curtailment rate is reduced from 5% to 1.2%. User benefits are increased by 8% compared to traditional scheduling, fully mobilizing user participation. Simultaneously, the backup scheduling mechanism successfully addresses resource failures, ensuring continuous control without load fluctuations or interruptions, verifying the feasibility and superiority of the technical solution of this invention.

[0245] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

[0246] The present invention has been described above with reference to the accompanying drawings. Obviously, the implementation of the present invention is not limited to the above-described manner. Any improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A resource aggregation and scheduling method for interaction between a virtual power plant and a vehicle-to-grid system, characterized in that, Includes the following steps: S1. Obtain multi-source heterogeneous data, which includes core data from the virtual power plant side, core data from the vehicle-to-grid interaction side, and core data from the power grid side. Standardize the multi-source heterogeneous data to obtain a standard dataset. S2. Based on the standard dataset, construct a tag system containing four dimensions: subject, attribute, capability, and scenario. Tag the virtual power plant resources and vehicle-to-grid interaction resources in the standard dataset, and finally construct at least one functional resource pool. S3. Obtain the power grid operation status data in the standard dataset, and based on the power grid operation status data, calculate the three-dimensional dynamic weights of power grid security, new energy consumption, and user benefits, and construct a multi-objective optimization scheduling model; S4. Input the grid load gap demand and new energy consumption demand from the standard dataset into the multi-objective optimization scheduling model, solve the multi-objective optimization scheduling model, and obtain the optimal solution for the overall resource scheduling objective; S5. Based on the overall resource scheduling target and the resource characteristics of the functional resource pool, the partition scheduling target value of each functional resource pool is decomposed. S6. Based on the partitioned scheduling target value, construct the scheduling strategy model for each of the functional resource pools, solve for the scheduling instruction parameters of each resource, including the control direction, target power, and control period; finally, issue the scheduling instruction parameters and execute the control to complete the resource aggregation scheduling.

2. The resource aggregation and scheduling method for virtual power plant and vehicle-to-grid interaction according to claim 1, characterized in that, In step S1: the core data on the virtual power plant side includes energy storage power station operation data, interruptible capacity of industrial load, distributed photovoltaic power output data, and virtual power plant contract data; the core data on the vehicle-to-grid interaction side includes real-time power of charging piles, electric vehicle battery status, and operator profile data; the core data on the grid side includes load gap data, heavy overload early warning information, and new energy consumption demand data; the standardized data format is a combination structure of resource ID, timestamp, and core parameters.

3. The resource aggregation and scheduling method for virtual power plant and vehicle-to-grid interaction according to claim 1, characterized in that, In step S2: the main tags include operator type, resource type, and access area; the attribute tags include response accuracy, adjustment precision, and failure rate; the capability tags include maximum adjustable capacity, response latency, and continuous adjustment duration; the scenario tags include peak shaving applicable scenarios, valley filling applicable scenarios, and renewable energy consumption applicable scenarios; the functional resource pools include peak shaving resource pools, valley filling resource pools, and renewable energy consumption resource pools, and the update frequency of the functional resource pools is every 5-15 minutes.

4. The resource aggregation and scheduling method for virtual power plant and vehicle-to-grid interaction according to claim 1, characterized in that, In step S3, the three-dimensional dynamic weights satisfy the following expression: ω1+ω2+ω3=1 Where ω1 is the power grid security weight, ω2 is the renewable energy absorption weight, and ω3 is the user revenue weight; when the power grid load factor is ≥90%, ω1 ≥0.6; when the predicted renewable energy output is ≥30% of the power grid load, ω2 ≥0.5; the objectives of the multi-objective optimization scheduling model are to minimize power grid load fluctuations, maximize renewable energy absorption, and maximize comprehensive user revenue, and the constraints include adjustable resource capacity constraints, power balance constraints, response speed constraints, and cost constraints.

5. The resource aggregation and scheduling method for virtual power plant and vehicle-to-grid interaction according to claim 1, characterized in that, In step S5: the total adjustable capacity and response priority coefficient of each functional resource pool are extracted, the response priority coefficient being determined based on the response speed and adjustment accuracy labels obtained from the tagging process; the partition scheduling target value of each functional resource pool is calculated, expressed as: P_i = P general tone × (Q_i × k_i) / Σ(Q_i × k_i) Where P_total is the overall resource scheduling target, Q_i is the total adjustable capacity of the i-th functional resource pool, and k_i is the response priority coefficient of the i-th functional resource pool.

6. The resource aggregation and scheduling method for virtual power plant and vehicle-to-grid interaction as described in claim 1, characterized in that, In step S6, a corresponding scheduling strategy model is constructed by combining the response characteristics and adjustment capabilities of each functional resource pool. Scheduling instructions are generated according to the resource type and the adjustment parameters are adapted. Instructions are first issued to pilot resources for pre-execution. After confirming the effectiveness of the execution, they are issued to each resource in batches. The adjustment execution status is fed back and verified in real time to ensure that the adjustment meets the requirements of the partitioned scheduling target value.

7. A resource aggregation and scheduling system for virtual power plant and vehicle-to-grid interaction, characterized in that, include: The data perception and standardization module is used to acquire multi-source heterogeneous data and perform standardization processing to output a standard dataset. The resource tagging and aggregation module, which communicates with the data perception and standardization module, is used to construct a four-dimensional tagging system and dynamically build a functional resource pool; the scheduling model and target determination module is used to calculate the three-dimensional dynamic weights and solve for the optimal solution of the overall resource scheduling target. The scheduling target decomposition module is communicatively connected to the resource tag and aggregation module and the scheduling model and target determination module, respectively, and is used to decompose the partition scheduling target value; The scheduling strategy generation module is communicatively connected to the scheduling target decomposition module and is used to generate scheduling instruction parameters; The control execution module is communicatively connected to the scheduling strategy generation module and is used to issue scheduling instruction parameters and perform control.

8. The resource aggregation and scheduling system for virtual power plant and vehicle-to-grid interaction according to claim 7, characterized in that, The resource tagging and aggregation module includes a tag engine and a resource pool construction unit; the tag engine has a built-in four-dimensional tag rule library, which supports automatic tag allocation and dynamic updates; the resource pool construction unit generates a functional resource pool based on the scene tag matching results, which supports the expansion of the resource pool and resource migration.

9. The resource aggregation and scheduling system for virtual power plant and vehicle-to-grid interaction according to claim 7, characterized in that, The scheduling model and objective determination module includes a weight calculation unit, a model construction unit, and an objective solution unit; the weight calculation unit calculates three-dimensional dynamic weights in real time based on power grid operation status data; the objective solution unit incorporates an improved genetic algorithm or particle swarm optimization algorithm to solve the multi-objective optimization scheduling model.

10. The resource aggregation and scheduling system for virtual power plant and vehicle-to-grid interaction according to claim 7, characterized in that, The control execution module includes an instruction issuing unit and a backup scheduling unit. The instruction issuing unit is used to issue scheduling instruction parameters to each resource and trigger control execution. The backup scheduling unit stores backup resource pool information and backup scheduling schemes. When an abnormal resource response or a sudden change in the power grid operating status is detected, it automatically starts switching to ensure the continuity of control.

Images