Virtual power plant resource real-time response control system based on edge computing

Abstract

The application discloses a virtual power plant resource real-time response control system based on edge computing, adopts a four-level architecture of "top-level scheduling-cloud coordination-edge execution-terminal response", and core contains a power grid dispatching center, an edge-cloud linkage dispatching center, a distributed edge collaborative node cluster, a dynamic resource priority scheduling module and a fault self-diagnosis and fault-tolerant response module. The distributed edge collaborative node cluster adopts a "1 master+8 slave" partition architecture, the dynamic resource priority scheduling module is integrated in the master edge node, and the fault self-diagnosis and fault-tolerant response module fully covers the master / slave edge nodes. The control method realizes real-time response through "instruction issuing-target disassembly-priority sorting-optimized execution-closed loop feedback", the response delay is less than or equal to 30 ms, the fault identification accuracy is greater than or equal to 98%, the real-time response performance, resource adaptability and operation reliability of the virtual power plant can be significantly improved, and the data transmission pressure is reduced.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plant scheduling and control technology, specifically relating to a real-time response control system and method for virtual power plant resources based on edge computing. Background Technology

[0002] With the rapid development of new energy power generation technologies, the proportion of distributed energy sources such as photovoltaics and wind power in the power system continues to increase. As the core carrier for integrating distributed energy and flexibly responding to grid dispatching needs, the real-time performance, reliability and adaptability of virtual power plants in dispatching and control have become a research hotspot in the industry.

[0003] Existing virtual power plant dispatch and control systems mostly adopt a "cloud-centralized" architecture, which involves collecting data from all energy equipment in the system, making resource dispatch decisions, and issuing control commands through a cloud platform. However, this architecture has the following significant drawbacks in practical applications: First, poor real-time response performance. The long data transmission distance and complex links between distributed energy equipment and the cloud platform result in high command transmission latency (typically >100ms), making it difficult to meet the millisecond-level response requirements for emergency peak shaving and fault handling in the power grid. Second, insufficient resource dispatch adaptability. Existing systems mostly use fixed priority dispatch strategies, failing to fully consider the dynamic changes in power grid dispatch command levels, equipment operating status, and user energy demand, easily leading to mismatches between resource dispatch and actual needs. Third, weak operational reliability. The fault diagnosis and fault tolerance mechanisms of existing systems are mostly deployed in the cloud or on a few core nodes, making it difficult to respond quickly to edge device failures, easily leading to dispatch interruptions. Fourth, high data transmission pressure. The centralized uploading of massive amounts of distributed energy equipment's operating data to the cloud consumes a large amount of network bandwidth and increases the cloud's computing load.

[0004] To address the aforementioned issues, some technical solutions have introduced edge computing technology, delegating some scheduling functions to edge nodes. However, problems such as unreasonable architecture design and poor module collaboration still exist. Existing edge computing solutions (such as the invention patent application CN119179429A, entitled "A Data Integration Method, Device and Computing Equipment Cluster Based on Cloud Computing Technology") only realize data collection from edge nodes and do not design a 'master-slave edge node hierarchical + cross-regional 5G collaboration' architecture, leading to chaotic decision-making on the edge side. This invention solves this deficiency through a 3-region '1 master 8 slave' architecture + 5G interaction of the master edge node. The dynamic scheduling module and fault tolerance module are deployed independently, lacking collaborative linkage. The edge-cloud data interaction logic is unclear, making it difficult to form an efficient closed-loop control.

[0005] Therefore, there is an urgent need to design a real-time response control system for virtual power plant resources based on edge computing that is hierarchical, responsive, adaptable, and reliable, in order to make up for the shortcomings of existing technologies. Summary of the Invention

[0006] To address the technical shortcomings of existing virtual power plant dispatch and control systems, such as slow real-time response, poor resource scheduling adaptability, low operational reliability, and high data transmission pressure, this invention provides a virtual power plant resource real-time response control system and method based on edge computing. Through a four-level hierarchical architecture design of "top-level scheduling - cloud coordination - edge execution - terminal response", combined with dynamic resource priority scheduling and a full edge node fault tolerance mechanism, it achieves rapid instruction transmission, dynamic resource adaptation, seamless fault handling, and efficient data interaction.

[0007] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A real-time response control system for virtual power plant resources based on edge computing, employing a four-level hierarchical architecture of "top-level scheduling - cloud coordination - edge execution - terminal response," as follows: Figure 1 As shown, from top to bottom, the system includes a power grid dispatch center, an edge-cloud linkage dispatch center, a distributed edge collaborative node cluster, and terminal energy equipment. Each level forms a closed-loop control through command transmission and data feedback. The core of the system integrates four major functional modules: the edge-cloud linkage dispatch center, the distributed edge collaborative node cluster, the dynamic resource priority scheduling module, and the fault self-diagnosis and fault-tolerant response module. Each module works collaboratively according to its hierarchical affiliation. The specific structure is as follows: As the top-level command output unit of the system, the power grid dispatch center is used to generate and issue global dispatch commands based on the power grid operating status (such as load fluctuations and new energy consumption demand). The global dispatch commands include AGC commands, peak shaving and valley filling commands, and emergency response commands. At the same time, it receives the response results of virtual power plants uploaded by the edge-cloud linkage dispatch center, forming a top-level closed loop of "command issuance-result feedback".

[0008] The edge-cloud linkage dispatch center is deployed in the park's private cloud, located between the power grid dispatch center and the distributed edge collaborative node cluster. It serves as an intermediate coordination unit between top-level dispatch and edge execution. Figure 1 As shown, the core functions include command parsing, target decomposition, and status monitoring and feedback. Command parsing involves receiving global dispatch commands from the power grid dispatch center, parsing the command type (Level 1 Emergency / Level 2 Regular / Level 3 Economic), response power requirements, and response time limits, among other core parameters. Target decomposition involves breaking down the global dispatch target into sub-targets for each virtual power plant based on its partitioning, and then distributing these sub-targets to the corresponding main edge nodes of each partition. Status monitoring and feedback involves receiving operational data uploaded by the main edge nodes of each partition in real time, monitoring the overall system response progress, dynamically adjusting sub-targets for partitions that are not meeting response standards, and summarizing the response results of each partition and feeding them back to the power grid dispatch center. This module adopts a lightweight microservice architecture and does not participate in specific resource optimization calculations, thus reducing cloud computing load and network transmission pressure.

[0009] Distributed edge collaborative node clusters serve as the core execution unit of the system, such as Figure 1 , Figure 2 As shown, a "master-slave" hierarchical architecture is adopted, with the virtual power plant divided into three independent partitions. Each partition is configured with one master edge node and eight slave edge nodes, for a total of three master edge nodes and 24 slave edge nodes. The master edge node serves as the partition control core, communicating with the edge-cloud linkage dispatch center to receive partition sub-targets, coordinating data aggregation, collaborative decision-making, and command issuance from the slave edge nodes within its partition. Master edge nodes in adjacent partitions achieve cross-partition data interaction through 5G industrial modules. The slave edge nodes, as terminal execution units, connect to eight types of terminal energy devices, including photovoltaic power plants, wind turbines, energy storage power plants, industrial interruptible loads, public service loads, backup energy storage devices, sensors, and actuators. They are responsible for collecting equipment operation data and uploading it to the master edge node, while simultaneously executing control commands issued by the master edge node.

[0010] The dynamic resource priority scheduling module is integrated within each main edge node and is not deployed independently, such as... Figure 1 , Figure 2 As shown, this is used to generate a dynamic resource response priority queue based on the sub-objectives of the region issued by the edge-cloud linkage dispatch center, the level of power grid dispatch instructions, and the equipment operation data (including equipment type, operating efficiency, and SOC status) collected from edge nodes. Figure 3 As shown in the diagram. Specifically, this involves constructing a three-dimensional evaluation index system of "grid demand - equipment status - user demand," using the Analytic Hierarchy Process (AHP) to determine the weights of each index, and calculating the priority score of each resource through a fuzzy comprehensive evaluation model. Priority queues are then generated in descending order of these scores. For example, under a Level 1 emergency dispatch command, the weight of the grid demand dimension is increased to 0.6, prioritizing the dispatch of energy storage devices and industrial interruptible loads; during routine economic dispatch, the weight of user energy cost sensitivity is increased to 0.4, balancing response efficiency and economy.

[0011] The fault self-diagnosis and fault-tolerant response module is deployed on all master and slave edge nodes to achieve full edge node fault protection. Its core functions include fault diagnosis and fault-tolerant response. Fault diagnosis employs a CNN-LSTM hybrid deep learning model, collecting multi-dimensional data such as device voltage, current, temperature, and vibration to identify 10 common fault types, including sensor faults, inverter faults, and communication interruptions, with a fault identification accuracy of ≥98%. The fault-tolerant response specifically works as follows: if a slave edge node fault is detected, the master edge node activates a backup slave edge node, synchronizing historical data and current instructions via the consortium blockchain, completing a hot switch within 50ms; if a terminal device fault is detected, a backup resource of the same priority from the priority queue is selected for replacement, while simultaneously adjusting control command parameters to ensure uninterrupted scheduling.

[0012] Terminal energy equipment serves as the final response unit of the system, including a 20MW photovoltaic power station, a 500kW wind turbine, a 100MWh energy storage power station, industrial interruptible loads, public service loads, backup energy storage equipment, and various types of sensors and actuators (inverters, contactors). Each type of terminal energy equipment is connected to a corresponding edge node, receiving instructions from the edge node and adjusting its operating status (such as energy storage charging and discharging power, load switch status).

[0013] Furthermore, this invention also discloses a real-time response control method for virtual power plant resources based on edge computing, applied to the aforementioned control system, such as... Figure 4 As shown, it includes the following steps: S1: Top-level command issuance: The power grid dispatch center generates global dispatch commands and issues them to the edge-cloud linkage dispatch center; S2: Cloud target decomposition: The edge-cloud linkage scheduling center parses the global scheduling instructions, decomposes the global scheduling target into sub-targets of each partition, and sends them to the main edge nodes of the corresponding partitions respectively; S3: Edge data acquisition and priority sorting: Each partition collects terminal energy equipment operation data from the edge node and uploads it to the main edge node. The main edge node generates a resource response priority queue for this partition through the dynamic resource priority scheduling module. S4: Edge optimization decision-making and command execution: The master edge node solves the optimal control strategy based on the partition sub-objectives, priority queues and cross-partition collaborative data through a distributed model predictive control algorithm, generates control commands and sends them to the slave edge nodes in the same partition, and the slave edge nodes drive the terminal energy equipment to adjust the operating status. S5: Fault tolerance and closed-loop feedback: The fault self-diagnosis and fault-tolerant response module monitors the entire process in real time and triggers a fault-tolerant response when a fault is detected; the main edge nodes of each partition upload the actual response results to the edge-cloud linkage dispatch center, which summarizes and feeds them back to the power grid dispatch center to complete the closed-loop control.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Excellent real-time response performance: It adopts a four-level architecture of "top-level scheduling - cloud coordination - edge execution - terminal response", with core optimized computing pushed down to edge nodes. Only instructions and summary data are transmitted between the edge and the cloud, and the response latency is ≤30ms, which can meet the millisecond-level requirements of emergency power grid dispatch. 2. Strong adaptability of resource scheduling: The dynamic resource priority scheduling module is based on a three-dimensional evaluation index system and dynamic weight allocation. It can adjust the resource order in real time according to the power grid command level, equipment status and user needs, and balance response timeliness and economy. 3. High operational reliability: The fault self-diagnosis and fault-tolerant response module achieves full coverage of master-slave edge nodes, with a fault identification accuracy of ≥98% and a hot switching time of ≤50ms for backup nodes, ensuring uninterrupted scheduling. 4. High-efficiency data transmission: The edge-cloud linkage scheduling center adopts a lightweight design, only undertaking coordination functions. Core data processing is completed on the edge side, reducing network transmission volume by more than 70% and significantly reducing bandwidth pressure and cloud computing load. Attached Figure Description

[0015] Figure 1 This is a block diagram of the overall system architecture of the present invention.

[0016] Figure 2 This is a schematic diagram of a distributed edge collaborative node cluster "master-slave" architecture.

[0017] Figure 3 This is a schematic diagram of the workflow of the dynamic resource priority scheduling module.

[0018] Figure 4 This is a flowchart of the control method steps of the present invention.

[0019] Figure 5 This is a logical diagram of the fault self-diagnosis and fault-tolerant response mechanism. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: System Hardware Deployment like Figure 1 , Figure 2 and Figure 3 As shown, the system in this embodiment is deployed in a virtual power plant within an industrial park. This virtual power plant covers three production zones and one residential support zone within the park, with a total installed capacity of 200MW, including 60MW of distributed photovoltaic power, 40MW of wind power, and 100MWh of energy storage. The maximum interruptible load reduction capacity for industrial applications is 30MW, and the minimum guaranteed capacity for residential loads is 20MW. The system strictly follows a four-level architecture of "top-level scheduling - cloud coordination - edge execution - terminal response." The specific hardware configuration, software deployment, and network topology are as follows: The power grid dispatch center adopts the State Grid standard version of the dispatch automation system (SCADA / EMS), configured with two redundant servers (model: Huawei RH2288H V5), each with an Intel Xeon Gold 6226R CPU, 32GB of memory, a 1TB SSD + 4TB SATA hard drive, and Windows Server 2019 Standard operating system. Secure data interaction is established with the edge-cloud linkage dispatch center through two independent fiber optic communication links (1000Mbps bandwidth per link), using the IEC 61850 protocol for command and data transmission, with link latency ≤10ms and data transmission reliability ≥99.99%.

[0022] Edge-Cloud Linked Dispatch Center: Deployed on a private cloud node within the campus, using the Huawei FusionSphere OpenStack virtualization platform, and configured with a cluster of 3 Huawei FusionServer Pro 2288H V5 servers. Each server is equipped with an Intel Xeon Silver 4214 processor (12 cores, 24 threads), 8GB x 4 DDR4 memory, a 1TB SSD system disk + a 4TB SATA data disk, and runs Ubuntu 20.04 LTS Server 64-bit operating system. The software modules are developed using a microservice architecture, implemented based on the Spring Cloud Alibaba framework. Specifically, they include an instruction parsing service (port 8081, concurrent processing capacity ≥1000 TPS), a target decomposition service (port 8082, decomposition error ≤2%), a status monitoring service (port 8083, data sampling frequency 10Hz), and a data aggregation and feedback service (port 8084). Each microservice uses Nacos for service registration and discovery, and RocketMQ for message queue communication, ensuring high availability of inter-module interactions.

[0023] Distributed edge collaboration node cluster: Each of the three production partitions is configured with one Advantech UNO-2484G master edge node and eight Advantech EPC-R4670 slave edge nodes. The main edge node is configured with an Intel Core i7-12700E processor (14 cores, 20 threads), 16GB×2 DDR4 3200MHz memory, a 512GB NVMe SSD + 2TB SATA III hard drive, and integrates an Intel I225-LM Gigabit Ethernet controller and a 5G industrial module (model: Huawei ME909s-821, supporting SA / NSA dual-mode, downlink speed ≥1Gbps, uplink speed ≥100Mbps); the slave edge node is configured with an Intel Celeron N5105 processor (4 cores, 4 threads), 8GB DDR4 memory, a 256GB NVMe SSD, and integrates two Gigabit Ethernet ports and a LoRa-WAN communication module (model: Semtech SX1302, operating frequency band 470-510MHz, communication distance ≥3km, speed selectable 100bps-37.5kbps). The main edge nodes form an SA network using 5G industrial modules, employing IPSec VPN for encrypted data transmission to achieve cross-regional collaborative data interaction. The main and slave edge nodes communicate via the LoRa-WAN protocol, operating in Class A mode. The uplink uses an SF12 spreading factor, and the downlink uses an SF8 spreading factor to ensure low power consumption and interference resistance. On the software side, the main edge nodes run Ubuntu 20.04 LTS Server, deploying a dynamic resource priority scheduling module, a distributed model predictive control algorithm module, and a fault self-diagnosis main module. The slave edge nodes run Debian 11, deploying a data acquisition module (supporting Modbus RTU / TCP, MQTT, OPC UA, etc.), an instruction execution module, and a fault self-diagnosis slave module. Data interaction between the main and slave modules occurs via the MQTT protocol.

[0024] Terminal energy equipment: Each zone is equipped with one 20MW photovoltaic power station (consisting of 1000 200W monocrystalline silicon photovoltaic modules and 50 400kW inverters, inverter model: Sungrow SG125HV), one 500kW wind turbine (model: Goldwind GW130-5.0MW, suitable for wind speeds of 3-25m / s), and one 100MWh energy storage power station (using lithium iron phosphate batteries, single cell voltage 3.2V, capacity 200Ah, configured with 20 5MW units). PCS energy storage converter, model: Kelun Electronics CL-ESS-5000, industrial interruptible loads (including motor loads of 3 production lines, total power 20MW, equipped with intelligent contactor model: Schneider LC1D65M7C), public service loads (including air conditioning and lighting systems for park dormitories, total power 10MW, equipped with smart meters model: Huali DDSY188), backup energy storage equipment (2 sets of 10MWh mobile energy storage containers, model: CATL CETC-10MWh), various types of sensors (voltage sensor model: ABB CVM300, measurement range 0-35kV; current sensor model: Siemens 7SJ68, measurement range 0-2000A; temperature sensor model: PT100, measurement range -50-200℃; vibration sensor model: Bentley 330180) and actuators (inverters, contactors, etc., matched with the above equipment). Various devices are connected to the system through corresponding edge nodes. The communication protocol between the devices and the edge nodes is adapted according to the device type. For example, photovoltaic inverters use the Modbus TCP protocol, and sensors use 4-20mA analog signals + Modbus RTU protocol.

[0025] Software module deployment: The dynamic resource priority scheduling module is developed using Python 3.9, implementing the weight calculation of the Analytic Hierarchy Process (AHP) based on the Scikit-learn library, and implementing the fuzzy comprehensive evaluation model based on the FuzzyLogic library. The module's memory usage is ≤2GB, and the priority calculation time is ≤50ms. The fault self-diagnosis and fault-tolerant response module is built on the TensorFlow 2.8 framework to construct a CNN-LSTM hybrid deep learning model. The CNN part uses 4 convolutional layers + 2 pooling layers, and the LSTM part uses 2 hidden layers (128 neurons per layer). The model training dataset contains 100,000 normal operation data and 50,000 fault data (covering 10 common faults). After deployment, the model's memory usage is ≤4GB, the fault identification time is ≤30ms, and the identification accuracy is ≥98%. The distributed model predictive control algorithm module is developed using C++17, implementing matrix operations based on the Eigen library, using IPOPT as the optimization solver, setting the optimization time domain to 10 minutes, the rolling step size to 1 minute, and the single optimization solution time to ≤100ms. The algorithm sets multi-dimensional constraints to ensure operational safety and compliance, including: energy storage charging and discharging power ≤ 5MW (rated power limit of a single PCS device), maximum single reduction of industrial load ≤ 8MW (to avoid large-scale production line shutdowns), and grid connection point voltage deviation ≤ ±7% (the allowable fluctuation range for a 380V low-voltage three-phase system is 26.6V, which complies with Article 4.2 of GB / T12325-2008 "Power Quality Supply Voltage Deviation").

[0026] Example 2: System Working Process like Figure 4 , Figure 5 As shown, this embodiment takes the issuance of a Level 1 emergency peak-shaving command by the power grid (command number: AGC-20240615-001, command type: Level 1 emergency, response requirements: increase response power by 50MW within 1 minute, response duration: 30 minutes, allowable error ±2MW) as an example, combined with the attached... Figure 4 The real-time response control process of this system is described in detail below, including specific interaction steps and data flow: S1: Top-level command issuance phase (0ms): Based on real-time load monitoring data of the regional power grid (at this time, the regional power grid load gap is 50MW), the SCADA / EMS system of the power grid dispatch center automatically generates a Level 1 emergency peak-shaving command and sends the command data concurrently to the edge-cloud linkage dispatch center through two redundant fiber optic links. The command data is encapsulated using the IEC 61850-8-1 protocol. The data frame format includes a command header (16 bytes, including command number and command type), a data body (32 bytes, including response power, response time limit, and duration), and a checksum (4 bytes, using CRC32 check). The command parsing service of the edge-cloud linkage dispatch center receives the command data from the two links at 10ms. After verifying the data integrity through the checksum, it selects the first valid command to arrive for parsing, discards redundant commands, and obtains the core parameters: Level 1 emergency peak-shaving, 50MW response power, 60s response time limit, and 30min duration.

[0027] S2: Cloud-based target decomposition stage (15ms): The target decomposition service of the edge-cloud linkage scheduling center calls the energy equipment capacity data of each partition in the historical operation database (Partition 1: PV 20MW, Wind Power 15MW, Energy Storage 30MW, Industrial Load 10MW; Partition 2: PV 20MW, Wind Power 15MW, Energy Storage 35MW, Industrial Load 6MW; Partition 3: PV 20MW, Wind Power 10MW, Energy Storage 35MW, Industrial Load 4MW), and allocates targets based on the decomposition algorithm of "capacity ratio + response capability weight". Among them, the response capability weight is calculated according to the average response rate of each partition's equipment in the recent period (within 1 hour), with a response capability weight of 0.3 for Partition 1, 0.35 for Partition 2, and 0.35 for Partition 3. Finally, the sub-targets of each partition are obtained: Partition 1: 15MW (50MW×0.3), Partition 2: 17.5MW (50MW×0.35), Partition 3: 17.5MW (50MW×0.35), with a decomposition error of 0, meeting the design requirements. The target decomposition service uses RocketMQ message queues to send sub-target instructions for each partition to the corresponding partition master edge node. The sub-target instructions are encapsulated in JSON format and include information such as partition number, sub-target power, and response time limit. The message sending timeout is set to 50ms to ensure reliable delivery of instructions.

[0028] S3: Edge Data Acquisition and Prioritization Phase (30ms): After receiving the sub-target instruction, each partition's master edge node immediately sends data acquisition instructions to its eight slave edge nodes via the LoRa-WAN network. The acquisition frequency is set to 500Hz, and the acquisition duration is 100ms. The slave edge nodes synchronously acquire real-time operating data of the corresponding terminal devices according to the instructions: for example, the real-time output of the photovoltaic power station in partition 1 is 18MW, the SOC of the energy storage power station is 75%, the current power of industrial load is 8MW, the real-time output of wind turbines is 12MW, and the current power of residential load is 5MW. Simultaneously, they acquire equipment status data (such as inverter operating temperature 42℃, contactor switch status closed) and environmental data (wind speed 5m / s, illuminance 800W / m²). After acquisition, the slave edge nodes preprocess the data (removing outliers and standardizing data), and upload the preprocessed data to the master edge node via the MQTT protocol. The data upload time is ≤20ms (the master edge node completes the aggregation of all slave edge node data at 60ms). At 65ms, the main edge node initiates the dynamic resource priority scheduling module, inputting the partition sub-target (15MW), instruction type (Level 1 Emergency), and summarized equipment operation data to construct a three-dimensional evaluation index system of "grid demand - equipment status - user demand": the grid demand dimension has a weight of 0.6 (Level 1 Emergency Instruction), including two secondary indicators: response rate and response capacity; the equipment status dimension has a weight of 0.3, including two secondary indicators: equipment health and current output; and the user demand dimension has a weight of 0.1, including one secondary indicator: priority of people's livelihood. The weights of each secondary indicator are calculated using the analytic hierarchy process (AHP), and the priority scores of each resource are calculated using a fuzzy comprehensive evaluation model: energy storage (0.92) > industrial interruptible load (0.85) > photovoltaic (0.78) > wind turbine (0.75) > people's livelihood load (0.60). At 75ms, a priority queue is generated and stored in the local cache.

[0029] S4: Edge Optimization Decision and Command Execution Phase (80ms): The main edge node calls the distributed model predictive control algorithm module, inputting the partition sub-objective (15MW), priority queue, equipment operation data, and cross-partition collaborative data obtained through the 5G network (preliminary response schemes for partitions 2 and 3). The optimization objective is set as "response power meets target + lowest operating cost," and the constraints are the equipment operating limits (specifically: single energy storage PCS charging / discharging power ≤ 5MW, maximum single reduction in industrial load ≤ 8MW, grid connection point voltage deviation ≤ ±7%). The algorithm module performs optimization calculations using the IPOPT solver, obtaining the optimal control strategy at 95ms: Partition 1: Energy storage power station discharges 10MW (PCS operating mode switches to discharge, output voltage 380V, frequency 50Hz), industrial load is reduced by 5MW (disconnecting the contactor of production line 2, reducing power by 5MW), photovoltaic power station maximizes output to 18MW (no adjustment required), wind turbine maximizes output to 12MW (no adjustment required), total response power 10 + 5 = 15MW, meeting the sub-objective requirements. At 100ms, the master edge node breaks down the control command into specific execution commands for each slave edge node. For example, it sends a "discharge power 10MW" command to the energy storage slave edge node and a "disconnect contactor of production line 2" command to the industrial load slave edge node. The commands are encapsulated using the Modbus RTU protocol and sent to the slave edge nodes via the LoRa-WAN network. At 110ms, the slave edge nodes receive the commands and drive the terminal devices to execute them: the energy storage PCS completes the working mode switch and begins discharging at 10MW; after receiving the command, the industrial load contactor completes the disconnection action at 115ms, reducing the industrial load power to 3MW. At 120ms, the terminal devices complete the execution, and the slave edge nodes collect the status data after the device execution (such as energy storage discharge current 26.3kA, contactor switch status open), and upload it to the master edge node. The master edge node confirms that the response power meets the standard (15MW).

[0030] S5: Fault Tolerance and Closed-Loop Feedback Phase (125ms): The fault self-diagnosis and fault-tolerant response module monitors the entire process data in real time. At 130ms, it detects an interruption in the communication link of a certain energy storage device from the edge node in partition 1 (determined by heartbeat timeout), and immediately triggers the fault-tolerant response (corresponding to the attached...). Figure 5The fault-tolerant logic is as follows: The primary edge node starts the backup edge node, synchronizing historical data (energy storage operation data and control commands from the previous 130ms) with the current command (discharging 10MW) via the Hyperledger Fabric consortium blockchain. The backup edge node quickly loads the configuration and establishes communication, completing the hot switch at 170ms without interruption of the energy storage discharge process. At 180ms, the primary edge nodes of each zone upload the actual response results (zone 1: 15MW, zone 2: 17.5MW, zone 3: 17.5MW) and equipment operating status to the edge-cloud linkage dispatch center. At 190ms, the cloud summarizes the response results (total response power 50MW, meeting the global target), generates a feedback report (including response completion time of 120ms, response details of each zone, and fault handling records), and feeds it back to the power grid dispatch center via fiber optic link. At 200ms, the power grid dispatch center receives the feedback report, confirms that the response meets the standard, and initiates the continuous monitoring process. The entire emergency peak-shaving response process is completed within 200ms, far less than the 60s response time limit requirement. During the subsequent 30-minute continuous response, the system re-optimizes the control strategy every minute to ensure that the response power remains stable at around 50MW, while monitoring the equipment status in real time to ensure operational safety.

[0031] Those skilled in the art should understand that the present invention is not limited to the specific embodiments described above. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention shall still fall within the protection scope of the present invention. For example, the number of partitions in a virtual power plant can be adjusted according to the actual scenario, and the number of edge nodes can be flexibly configured according to the type of terminal device. These adjustments do not deviate from the core technical solution of the present invention.

Claims

1. A real-time response control system for virtual power plant resources based on edge computing, characterized in that: The system adopts a four-tiered architecture of "top-level scheduling - cloud coordination - edge execution - terminal response," consisting of a power grid dispatch center, an edge-cloud linkage dispatch center, a distributed edge collaborative node cluster, and terminal energy equipment, arranged from top to bottom. Each level forms a closed-loop control system through command transmission and data feedback. The core of the system integrates four major functional modules: the edge-cloud linkage dispatch center, the distributed edge collaborative node cluster, the dynamic resource priority scheduling module, and the fault self-diagnosis and fault-tolerant response module. Each module works collaboratively according to its hierarchical affiliation, specifically including: The power grid dispatch center, as the top-level instruction output unit, is used to generate and issue global dispatch instructions, and at the same time receive feedback from the virtual power plant response results. The edge-cloud linkage dispatch center is deployed in the park's private cloud, located between the power grid dispatch center and the distributed edge collaborative node cluster. It is used to parse global dispatch instructions, break down global dispatch targets into sub-targets of each partition, monitor response progress, and summarize and feedback results. The distributed edge collaborative node cluster adopts a "master-slave" hierarchical architecture. The virtual power plant is divided into 3 independent partitions, and each partition is configured with 1 master edge node and 8 slave edge nodes. The master edge node is used to receive the sub-objectives of the partition, coordinate the data aggregation and collaborative decision-making of the slave edge nodes, and the slave edge nodes are used to connect to the terminal energy equipment and execute control commands. The dynamic resource priority scheduling module is integrated into each main edge node and is used to generate a dynamic resource response priority queue based on partition sub-objectives, scheduling instruction levels and device operation data. The fault self-diagnosis and fault-tolerant response module is deployed on all master edge nodes and slave edge nodes to monitor the operating status in real time, identify fault types and locations, and trigger targeted fault-tolerant responses. The terminal energy device, as the final response unit, is used to receive instructions from the edge node and adjust its operating status.

2. The edge computing-based virtual power plant resource real-time response control system according to claim 1, characterized in that: The edge-cloud linkage scheduling center adopts a lightweight microservice architecture. Its core functions include instruction parsing, target decomposition, status monitoring and feedback, and it does not participate in specific resource optimization calculations.

3. The edge computing-based virtual power plant resource real-time response control system according to claim 1, characterized in that: In the distributed edge collaborative node cluster, the master edge nodes of adjacent partitions realize cross-partition data interaction through 5G industrial modules, and the master and slave edge nodes communicate through the LoRa-WAN protocol.

4. The edge computing-based virtual power plant resource real-time response control system according to claim 1, characterized in that: The dynamic resource priority scheduling module constructs a three-dimensional evaluation index system of "grid demand - equipment status - user demand", uses the analytic hierarchy process to determine the index weights, and calculates the priority score of each resource through a fuzzy comprehensive evaluation model. Under the first-level emergency dispatch command, the weight of the grid demand dimension is not less than 0.6, and under the regular economic dispatch command, the weight of the user energy cost sensitivity is not less than 0.

4.

5. The edge computing-based virtual power plant resource real-time response control system according to claim 1, characterized in that: The fault self-diagnosis and fault-tolerant response module uses a CNN-LSTM hybrid deep learning model for fault diagnosis. By collecting multi-dimensional data on equipment voltage, current, temperature, and vibration, it identifies 10 common faults with a fault identification accuracy of ≥98%.

6. The edge computing-based virtual power plant resource real-time response control system according to claim 1, characterized in that: The fault-tolerant response strategy of the fault self-diagnosis and fault-tolerant response module includes: when the edge node fails, a backup edge node is started and data is synchronized through the consortium blockchain, with a hot switching time of ≤50ms; when the terminal device fails, a backup resource of the same level is selected from the priority queue to replace it and the control command parameters are adjusted.

7. The virtual power plant resource real-time response control system based on edge computing according to claim 1, characterized in that: The terminal energy equipment includes a 20MW photovoltaic power station, a 500kW wind turbine, a 100MWh energy storage power station, industrial interruptible loads, livelihood security loads, backup energy storage equipment, and various types of sensors and actuators. Each type of terminal energy equipment is connected to a corresponding edge node.

8. A real-time response control method for virtual power plant resources based on edge computing, characterized in that: Applied to the control system according to any one of claims 1-7, the method comprises the following steps: S1: Top-level command issuance: The power grid dispatch center generates global dispatch commands and issues them to the edge-cloud linkage dispatch center; S2: Cloud target decomposition: The edge-cloud linkage scheduling center parses the global scheduling instructions, decomposes the global scheduling target into sub-targets of each partition, and sends them to the main edge nodes of the corresponding partitions respectively; S3: Edge data acquisition and priority sorting: Each partition collects terminal energy equipment operation data from the edge node and uploads it to the main edge node. The main edge node generates a resource response priority queue for this partition through the dynamic resource priority scheduling module. S4: Edge optimization decision-making and command execution: The master edge node solves the optimal control strategy based on the partition sub-objectives, priority queues and cross-partition collaborative data through a distributed model predictive control algorithm, generates control commands and sends them to the slave edge nodes in the same partition, and the slave edge nodes drive the terminal energy equipment to adjust the operating status. S5: Fault tolerance and closed-loop feedback: The fault self-diagnosis and fault-tolerant response module monitors the entire process in real time and triggers a fault-tolerant response when a fault is detected; the main edge nodes of each partition upload the actual response results to the edge-cloud linkage dispatch center, which summarizes and feeds them back to the power grid dispatch center to complete the closed-loop control.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the steps of the control method of claim 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the control method of claim 8.

Images