A Computing Power Allocation Method and System Based on Dynamic Reorganization of Resource Forms

Abstract

This invention discloses a computing power configuration method and system based on dynamic resource reconfiguration, belonging to the field of computing power resource scheduling technology. The method includes: abstracting fixed computing units and portable computing power carriers into dynamically divisible heterogeneous computing power pools; establishing a two-way interlocking comprehensive protection mechanism, the mechanism definition including at least a set of interlocking constraint conditions such as thermal-computing interlock, electrical-computing interlock, and pressure-computing interlock. By establishing a two-way interlocking comprehensive protection mechanism, this invention hard-couples computing power reconfiguration with energy storage systems and shared cooling circuits (heat load margin). When any interlocking condition is triggered, the system generates a joint adjustment vector and synchronously sends it to the computing power scheduler and energy storage controller, realizing "single instruction dual-domain collaboration". This fundamentally eliminates the safety risks of energy storage over-discharge, overheating, or lithium plating caused by sudden changes in computing power load, significantly improving the operational safety and task continuity of composite nodes under dynamic operating conditions.

Description

Technical Field

[0001] This invention relates to the field of computing power resource scheduling technology, and in particular to a computing power configuration method and system based on dynamic reorganization of resource forms. Background Technology

[0002] With the rapid development of artificial intelligence and edge computing, the demand for computing power has exploded. At the same time, energy storage systems (including battery clusters, charging piles, and battery swapping stations) have been widely deployed. These systems integrate a large number of computing units, which are in a low-load or completely idle state most of the time, forming a massive amount of "dormant computing power" resources.

[0003] Existing computing power scheduling technologies are primarily geared towards data center or cloud computing environments, focusing on the virtualization and load balancing of resources such as CPUs and GPUs. However, these technologies have the following shortcomings when applied to energy storage-edge computing composite nodes: The disconnect between computing power and energy: Current scheduling decisions only consider computing load and latency requirements, ignoring energy constraints such as the battery state of health (SOH), state of charge (SOC), charging and discharging power limits, and heat dissipation resources of the energy storage system. This may lead to safety issues such as battery over-discharge, thermal overload, and lithium plating during computing power reorganization operations.

[0004] Lack of inter-module coordination: There is a lack of closed-loop feedback mechanism between the adjustment of computing power forms (sharding, aggregation, hibernation, wake-up) and the power control, thermal management, and grid command response of the energy storage system, making it impossible to achieve coordinated regulation of "computing power-power-thermal". In existing technologies, computing power scheduling and energy storage control are often executed by two independent systems, without interlocking or joint optimization between them.

[0005] Insufficient adaptation to edge computing scenarios: The computing resources of distributed nodes such as charging piles and battery swapping stations are characterized by strong time-varying nature, limited power consumption, and strong coupling with charging services. Furthermore, there exists "portable computing power" that can move with energy carriers (such as electric vehicle batteries and backup battery packs). Existing technologies have failed to incorporate this portable computing power into a unified resource pool, nor can they achieve computing power drift with energy flow. Based on this, a computing power configuration method and system based on dynamic reorganization of resource forms is proposed. Summary of the Invention

[0006] The purpose of this invention is to solve the problems existing in the prior art by proposing a computing power allocation method and system based on dynamic reorganization of resource forms.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A computing power allocation method based on dynamic reorganization of resource forms, the method comprising: The fixed computing unit and the portable computing power carrier are uniformly abstracted into a dynamically divisible heterogeneous computing power pool. The portable computing power carrier includes a mobile energy storage unit that can serve as an energy carrier. A two-way interlocking integrated protection mechanism is established. The mechanism defines a set of interlocking constraints including at least thermal-computation interlocking, electrical-computation interlocking, and pressure-computation interlocking. When any interlocking constraint is triggered, a joint adjustment vector is generated and sent synchronously to the computing power scheduler and the energy storage controller. Based on the aforementioned bidirectional interlocking integrated protection mechanism, at least one linkage decision is executed, including: Computing power drift and lossless migration: When it is predicted that the migrateable computing power carrier will leave the current composite node, a state snapshot is taken of the computing power slice running on it, and the task image is migrated to the target node using the energy flow channel and hot-swapping is completed. Thermal coupling interlock: Based on the thermal load model of the shared cooling circuit, the total power consumption of the computing power slice and the charging and discharging power of the energy storage system are jointly adjusted; Virtual discharge inversion: Based on the grid dispatch instructions, a virtual discharge computing task is generated. By waking up or increasing the frequency of the computing power slice, the corresponding electrical energy is consumed so that the actual discharge power of the energy storage system meets the grid instructions. Based on the execution result of the linkage decision, the computing units in the heterogeneous computing pool are subjected to operations such as fragmentation, fragment aggregation, drift migration, or computational precision adjustment. The actual state parameters after execution are read back, and the threshold of the interlock constraint and the parameters of the linkage decision model are corrected according to the deviation value.

[0008] As a preferred embodiment, the computing power drift and lossless migration further include: Based on the predicted movement path of the portable computing power carrier, computing power capacity is pre-registered at at least one composite node along the route. The task image is transmitted synchronously while energy flows bidirectionally, utilizing the V2G charging channel, battery cluster equalization charging and discharging channel, or power line communication channel between the portable computing power carrier and the target node. Without interrupting service, the task image is restored and the hot-swap mounting of the computing power slice is completed at the target node. Once the migration is complete, release the computing power slice resources on the source carrier and set the source carrier to a hibernation or reallocation state.

[0009] As a preferred embodiment, the bidirectional interlocking integrated protection mechanism further includes: The interlock priority order is defined as follows: safety-calculation interlock > electrical-calculation interlock > thermal-calculation interlock > pressure-calculation interlock > aging-calculation interlock; When multiple interlocking constraints are triggered simultaneously, a joint adjustment vector is generated according to the priority arbitration, and the joint adjustment vector is represented as follows: ,in This is the total power consumption adjustment amount for computing power. For the adjustment of energy storage charging and discharging power, This refers to the frequency adjustment of the computing chip. This is an adjustment factor for the number of computing power slices; The joint adjustment vector is sent to both the computing power scheduler and the energy storage controller simultaneously, enabling a single instruction to perform dual-domain collaborative execution of the computing power domain and the energy domain. Record the actual response effect of each interlock adjustment, dynamically adjust the trigger threshold of each interlock type and the gain coefficient of the joint adjustment vector to form an adaptive interlock closed loop.

[0010] As a preferred embodiment, the thermal coupling interlock includes: real-time acquisition of the total heat load margin of the shared cooling circuit and the real-time temperature and power consumption of each computing unit; and establishment of constraint relationships. ,in This represents the total power consumption of the current computing chip. The charging and discharging heat power of the energy storage battery cluster, The maximum heat dissipation power of the shared cooling circuit; when the total heat power exceeds the total heat load margin, a joint adjustment scheme is generated, which simultaneously reduces the number of computing power slices or the operating frequency of at least one computing unit, and reduces the charging and discharging power of the energy storage system; The virtual discharge inversion includes: monitoring grid dispatch commands and calculating the active power deviation that the energy storage system needs to increase or decrease. ;when When the energy storage system is required to increase its discharge power and the external load is insufficient, the target power consumption is automatically generated. The virtual discharge computing task dynamically wakes up or upscales at least one idle computing unit's computing power slice to execute an interruptible computing task; the virtual discharge computing task is marked as an interruptible priority at any time, and the corresponding computing power slice is automatically released after the power grid command is revoked or the risk of battery lithium plating is eliminated.

[0011] As a preferred solution, bus voltage coupling logic and lithium plating risk mitigation and recombination logic are also included: The bus voltage coupling logic is as follows: real-time monitoring of DC bus voltage ripple; when the voltage drops below the first threshold, freezing the wake-up operation of non-critical computing power slices and reducing the frequency of activated computing power slices; when the voltage drops below the second threshold, triggering the energy storage converter to increase the discharge current while reducing the frequency. The lithium plating risk avoidance and reorganization logic is as follows: Real-time estimation of the lithium plating edge state of the battery negative electrode. When the negative electrode potential is lower than the safety threshold and the current charging current cannot be reduced quickly, a risk avoidance computing task is automatically generated, which wakes up or increases the frequency of several interruptible computing slices. By consuming part of the charging current, the actual current flowing into the battery is reduced to below the safety value. After the lithium plating risk is eliminated, the computing slice occupied by the risk avoidance computing task is released.

[0012] As a preferred embodiment, the composite node includes a battery swapping station, and the portable computing power carrier includes a spare power battery pack within the battery swapping station, wherein the BMS processor of the spare power battery pack is incorporated into the heterogeneous computing power pool; the method further includes: Based on the charging pile occupancy prediction model, the probability of each charging pile being occupied by charging tasks within a future time window T is estimated. When it is necessary to perform computing power slice migration, the built-in computing unit of the charging pile or the on-site edge server with a predicted probability lower than the set threshold shall be selected as the target node. For the backup power battery that is about to be replaced, during the remaining time it stays in the station, the idle computing power of its BMS processor is divided into temporary computing power slices and non-real-time computing tasks are attached. Before being replaced, a state snapshot and computing power drift and lossless migration are automatically performed.

[0013] As a preferred option, the information-energy joint routing step is also included: Maintain a dynamic joint routing table, which records the current available energy transmission direction, transmission power, and signal-to-noise ratio of power line communication between any two nodes; When high-bandwidth or low-latency computing power slice data needs to be transmitted, the scheduler queries the dynamic joint routing table and prioritizes the communication path with an energy flow direction at the current moment and a signal-to-noise ratio higher than the threshold for data transmission. When the energy flow direction changes or the power line communication channel quality fluctuates, the joint routing table is updated in real time and dynamic reselection of the computing power slice transmission path is triggered.

[0014] As a preferred option, it also includes: Fault-adaptive reconfiguration: Real-time assessment of the health score of each computing unit, which integrates the unit's temperature, operating voltage, error rate, and communication latency; when the score falls below the health threshold, active withdrawal is triggered, gradually migrating the computing power slice on that unit to other healthy nodes. After the migration is complete, the unit is placed in an unschedulable state, and the energy storage control system is notified to lower the maximum charging and discharging power limit; when the faulty unit recovers, the computing power slice expansion deployment is reversed and the power control margin of the energy storage system is restored. Joint scheduling of carbon credits and computing power revenue: establishing a comprehensive revenue model ,in The business revenue generated from executing computing tasks on computing power slices The carbon credits earned by the energy storage system in the current period for participating in carbon trading. To mitigate battery aging costs; to acquire real-time electricity price signals, carbon credit prices, and the carbon reduction rate of energy storage systems, in order to maximize... With the goal of dynamically determining whether to wake up, reduce frequency, migrate, or hibernate computing power slices.

[0015] A computing power allocation system based on dynamic reorganization of resource forms includes a resource pooling module, a multi-source sensing module, a two-way interlock arbitration module, a computing power drift module, a linkage decision-making module, a reorganization execution module, and a feedback correction module. The resource pooling module is configured to abstract the fixed computing units and the portable computing power carriers within the energy storage-edge computing composite node into a dynamically divisible heterogeneous computing power pool. The multi-source sensing module is configured to collect energy parameters, thermal parameters, power grid parameters, and computing power parameters in real time. The bidirectional interlocked arbitration module is configured to execute a bidirectional interlocked integrated protection mechanism and generate a joint adjustment vector; The computing power drift module is configured to execute the computing power drift and lossless migration process; The linkage decision module is configured to execute at least one linkage logic among thermal coupling interlock, virtual discharge inversion, bus voltage coupling, and lithium plating risk avoidance recombination. The reorganization execution module is configured to perform sharding, sharding aggregation, drift migration, or calculation accuracy adjustment operations on the computing units in the heterogeneous computing pool based on the output of the linkage decision module or the computing power drift module. The feedback correction module is configured to read back the actual state parameters after execution and correct the interlock threshold of the bidirectional interlock arbitration module and the decision model parameters of the linkage decision module.

[0016] As a preferred embodiment, the system is deployed in at least one of the following physical entities: an integrated photovoltaic-storage-charging power station, a battery swapping station, a microgrid energy storage station, a V2G charging pile group for electric vehicles, or a mobile energy storage vehicle.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention establishes a bidirectional interlocking comprehensive protection mechanism, which rigidly couples the reconfiguration of computing power with the energy storage system and the shared cooling circuit (heat load margin). When any interlocking condition is triggered, the system does not adjust the computing power or power separately, but generates a joint adjustment vector and synchronously sends it to the computing power scheduler and the energy storage controller, realizing "single instruction dual-domain collaboration". Compared with the existing technology where computing power scheduling and energy control are independent and operate independently, this invention fundamentally eliminates the safety risks of over-discharge, overheating or lithium plating of energy storage caused by sudden changes in computing power load. At the same time, it avoids interruption of edge computing services due to energy storage power adjustment, and significantly improves the operational safety and task continuity of composite nodes under dynamic conditions.

[0018] 2. This invention introduces a portable computing power carrier and incorporates it into a unified heterogeneous computing power pool. Through computing power drift and lossless migration technology, it utilizes the existing energy flow channel between the carrier and the target node to simultaneously migrate the computing power task image while transmitting energy, and completes the hot switching and mounting of the target node without interrupting the service of the source slice. This mechanism enables the "walking computing power" that was originally idle when the carrier moved to be dynamically captured and reused, and computing power resources are no longer limited to fixed physical locations. Compared with the traditional edge computing solution that can only schedule static node computing power, it greatly improves the spatiotemporal utilization of edge computing power, and is especially suitable for mobile energy storage scenarios such as battery swapping stations and V2G charging stations.

[0019] 3. In the technical decision-making process, this invention establishes a comprehensive benefit function through a joint scheduling model of carbon credit and computing power revenue. The scheduler obtains electricity price signals, carbon credit prices, computing power task business value, and the current aging cost coefficient of energy storage in real time, and dynamically decides on the wake-up, frequency reduction, migration, or hibernation of computing power slices with the goal of maximizing comprehensive benefits. At the same time, the virtual discharge inversion mechanism enables the energy storage system to direct excess discharge power to the computing power chip to execute interruptible computing tasks when responding to grid AGC commands without external load, transforming "forced discharge" into "valuable computing power output". Existing technologies often only focus on the single-objective optimization of computing power utilization or energy storage peak-shaving benefits. This invention is the first to quantitatively play a game between computing power revenue, carbon credit revenue, and battery life loss at the system bottom layer, upgrading the composite node from a single energy facility to a "dual service provider of computing power and energy", thereby significantly improving overall economic benefits and market competitiveness. Attached Figure Description

[0020] Figure 1 This is a framework diagram of the computing power allocation system based on dynamic reorganization of resource forms proposed in this invention; Figure 2 This is a flowchart of the computing power allocation method based on dynamic reorganization of resource forms proposed in this invention; Figure 3 This is a signal flow and arbitration logic diagram of the bidirectional interlocking integrated protection mechanism in this invention. Detailed Implementation

[0021] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0023] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0024] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0025] Example, refer to Figures 1 to 3 A computing power configuration method based on dynamic reorganization of resource forms is proposed. This method is applied to energy storage-edge computing composite nodes. The composite node includes fixed computing units and portable computing power carriers. Fixed computing units include GPUs / NPUs built into charging piles, edge servers of battery swapping stations, and BMS controllers of battery clusters. Portable computing power carriers include mobile energy storage units that can serve as energy carriers, such as backup power battery packs of battery swapping stations, power batteries of electric vehicles, and mobile energy storage cabinets. While providing energy storage and transmission functions, the computing units (such as BMS processors and battery monitoring chips) inside these carriers can be scheduled as computing power resources.

[0026] The method includes the following steps: Step 1: Resource pooling and perception; Fixed computing units and portable computing power carriers are uniformly abstracted into a dynamically shardable heterogeneous computing power pool. Each computing unit can be divided into multiple virtual computing power instances (called "computing power slices").

[0027] The following status parameters are collected in real time: Energy parameters: State of charge (SOC), State of health (SOH), charge / discharge current, negative electrode potential, and temperature for each battery cluster; Thermal parameters: total heat load margin of the shared cooling loop, real-time temperature and power consumption of each computing unit; Grid parameters: real-time electricity price, grid dispatch instructions (AGC signal), DC bus voltage ripple; Computing power parameters: available computing power capacity of each computing unit, current task load, power consumption-precision adjustable curve (e.g., power consumption and performance data of FP16 / INT8 / INT4 at different precisions).

[0028] Step 2: Establish a two-way interlocking integrated protection mechanism; Define a set of interlocking constraints that includes at least thermal-computation interlocking, electrical-computation interlocking, and pressure-computation interlocking. The core idea of ​​this mechanism is to impose hard constraints on the adjustment of computing power configuration and the safety boundary of the energy storage system. When any interlocking condition is triggered, the system is not allowed to adjust computing power or power independently. Instead, it must generate a joint adjustment vector and send it synchronously to the computing power scheduler and the energy storage controller.

[0029] Specifically, the interlock priority order is defined as: safety-calculation interlock > electrical-calculation interlock > thermal-calculation interlock > pressure-calculation interlock > aging-calculation interlock. When multiple interlock conditions are triggered simultaneously, a joint adjustment vector is generated according to priority arbitration. This vector is represented as follows: ,in This is the total power consumption adjustment amount for computing power. For the adjustment of energy storage charging and discharging power, This refers to the frequency adjustment of the computing chip. This refers to the adjustment of the number of computing power slices; the joint adjustment vector is simultaneously sent to the computing power scheduler and the energy storage controller to achieve dual-domain collaborative execution of a single instruction across the computing power domain and the energy domain. After each interlock adjustment, the actual response effect is recorded, and the trigger threshold and gain coefficient of each interlock type are dynamically adjusted to form an adaptive interlock closed loop. Step 3: Execute coordinated decisions based on the interlock protection mechanism; At least one or more of the following linked decisions should be executed: Computing power drift and lossless migration: When it is predicted that the migrateable computing power carrier will leave the current composite node (e.g., the backup battery is about to be replaced, or the electric vehicle is about to leave), a state snapshot (including model parameters, intermediate calculation results, and execution context) is taken of the computing power slice running on it. The task image is migrated to the target node using energy flow channels (such as V2G charging channels, battery cluster equalization charging and discharging channels, and power line communication PLC channels). Hot switching and mounting are completed on the target node without interrupting the service of the source carrier computing power slice. After the migration is completed, the computing power slice resources on the source carrier are released.

[0030] Thermal coupling interlock: Real-time acquisition of the total heat load margin of the shared cooling loop and the real-time temperature and power consumption of each computing unit. Establishment of constraint relationships. ,in This represents the total power consumption of the current computing chip. The charging and discharging heat power of the energy storage battery cluster, The maximum heat dissipation power of the shared cooling loop. When the total heat power exceeds... At the same time, a joint adjustment scheme is generated, which simultaneously reduces the number of computing power slices or the operating frequency of at least one computing unit, and reduces the charging and discharging power of the energy storage system.

[0031] Virtual discharge inversion: Monitoring grid dispatch commands and calculating the required increase or decrease in active power deviation for the energy storage system. .when When the energy storage system is required to increase its discharge power and the external load is insufficient, the target power consumption is automatically generated. The virtual discharge computing task dynamically wakes up or upscales a slice of computing power from at least one idle computing unit to perform interruptible computing tasks (such as model retraining or data preprocessing). This virtual discharge task is marked as interruptible at any time, and the computing power slice is automatically released after the grid command is revoked or the risk of lithium plating in the battery is eliminated.

[0032] Bus voltage coupling logic: Real-time monitoring of DC bus voltage ripple. When the voltage drops below the first threshold, the wake-up operation of non-critical computing slices is frozen and frequency reduction is performed on the activated computing slices. When the voltage drops below the second threshold, the energy storage converter (PCS) is triggered to increase the discharge current while reducing the frequency.

[0033] Lithium plating risk mitigation and reorganization logic: Real-time estimation of the lithium plating edge state of the battery's negative electrode (e.g., through a negative electrode potential estimation model). When the negative electrode potential falls below a safety threshold and the current charging current cannot be reduced quickly, a risk mitigation computing task is automatically generated, waking up or increasing the frequency of several interruptible computing slices. By consuming part of the charging current, the actual current flowing into the battery is reduced to below a safe value. After the lithium plating risk is eliminated, the computing slices occupied by the risk mitigation computing task are released. Further; Step 4: Execution of computing power restructuring; Based on the results of the collaborative decision generation, perform one or more of the following operations on the computing units in the heterogeneous computing power pool: Slicing: Physically dividing a single GPU / NPU into multiple independent computing power slices (e.g., using MIG technology or virtualization drivers). Sharding aggregation: Combining the idle computing power slices of multiple computing units into a logical computing cluster; Drift migration: Migrate computing power slices from the source carrier to the target node according to the computing power drift process; Calculation precision adjustment: Switch the calculation precision of the computing slice between FP16 / INT8 / INT4 to dynamically adjust power consumption and performance.

[0034] Step 5: Closed-loop feedback correction; The actual state parameters after execution (including actual power consumption, temperature change, bus voltage recovery time, battery negative electrode potential change, etc.) are read back, and the deviation from the expected values ​​is calculated. Based on the deviation value, the threshold of the interlock constraint condition and the parameters of the linkage decision model (such as thermal coupling coefficient, voltage response gain, lithium plating safety margin) are adjusted to achieve adaptive optimization of the system.

[0035] A computing power allocation system based on dynamic reorganization of resource forms; The system includes the following modules: Resource pooling module: Configured to abstract the fixed computing units and portable computing power carriers within the energy storage-edge computing composite node into a dynamically divisible heterogeneous computing power pool.

[0036] Multi-source sensing module: configured to collect energy parameters, thermal parameters, power grid parameters and computing power parameters in real time.

[0037] Two-way interlock arbitration module: configured to execute the above-mentioned two-way interlock integrated protection mechanism and generate a joint adjustment vector.

[0038] Computing power drift module: Configured to perform computing power drift and lossless migration processes, including sub-functions such as path prediction, capacity pre-registration, energy-data collaborative transmission, and hot-swapping mounting.

[0039] Linked decision module: configured to execute at least one of the following linked logics: thermal coupling interlock, virtual discharge inversion, bus voltage coupling, and lithium plating risk avoidance reorganization.

[0040] Reorganization Execution Module: Configured to perform sharding, sharding aggregation, drift migration, or computational precision adjustment operations on computing units in the heterogeneous computing pool based on the output of the linkage decision module or computing power drift module.

[0041] Feedback Correction Module: Configured to read back the actual state parameters after execution and correct the interlock threshold of the two-way interlock arbitration module and the decision model parameters of the linkage decision module.

[0042] The above system can be deployed in various physical entities such as integrated photovoltaic-storage-charging power stations, battery swapping stations, microgrid energy storage stations, V2G charging pile groups for electric vehicles, or mobile energy storage vehicles.

[0043] Specific Implementation Example 1: Thermal Coupling Interlock and Virtual Discharge Linkage in an Integrated Photovoltaic-Storage-Charging Power Station; In this embodiment, the composite node is an integrated photovoltaic, energy storage, and charging power station, comprising: a group of lithium iron phosphate battery clusters (total capacity 1MWh, nominal heat dissipation design power consumption). ), 8 DC fast charging piles (each pile has a built-in Jetson Orin GPU with a nominal maximum power consumption of 45W), a liquid cooling system, and a PCS energy storage converter.

[0044] Initial state: The power station is operating at night, electricity prices are low, battery SOC is 60%, and the liquid cooling system is currently at its maximum heat dissipation capacity. (Ambient temperature 25℃, coolant flow rate normal). 3 charging stations are idle and in compute sleep mode, while 5 charging stations are charging vehicles (GPU used for vehicle-to-everything communication and charging management; total computing power consumption...). The battery clusters are charged at a rate of 0.3C, with a charging heat output of [missing information]. At this time, the total heat power far below System security.

[0045] The cloud platform issues a batch of AI inference tasks (image recognition, which need to be completed within 3 minutes), which requires waking up 3 idle GPUs.

[0046] Thermally Coupled Interlock Trigger: The scheduler pre-calculates the new total thermal power if 3 GPUs are woken up to run at full speed (each with a peak power consumption of 45W, an increase of 135W). , The total thermal power is still 15kW, approximately 15.255kW, which is still below 50kW. Therefore, the thermal coupling interlock is not triggered. The scheduler wakes up the GPU normally and allocates inference tasks.

[0047] Virtual discharge inversion triggered: During the inference task, the grid AGC issued a "increase discharge by 200kW" command, but there was no new load added to the station at this time (vehicle charging power had stabilized). The dispatcher detected this. Given insufficient external load, a virtual discharge computing task with a target power consumption of 200kW is automatically generated. It checks the current idle computing power: the three awakened GPUs have spare computing power (inference tasks only utilize 60%), and the GPUs in the other five charging piles can be briefly upclocked. The scheduler upclocks the three GPUs to 90% utilization, while temporarily removing some computing power slices from the five charging pile GPUs in the charging management task (charging management has low computing power requirements, allowing for a 70% relinquishment), resulting in a total additional power consumption of 200kW. A data cleaning task is then launched as the virtual discharge load. The actual discharge power of the energy storage PCS increases by 200kW, satisfying the AGC instruction, while the inference task still completes within 2.9 seconds. One minute later, the grid instruction is revoked, the scheduler releases the virtual discharge task, and the GPUs return to their original frequencies.

[0048] Results: The inference task was completed on time, the power grid command was satisfied and the charging service was not interrupted, and the battery did not over-discharge or overheat.

[0049] Example 2: Computing power drift and lossless migration in battery swapping stations; In this embodiment, the composite node is a battery swapping station with 30 spare power batteries. Each battery's BMS has a built-in ARM Cortex-A78 processor (quad-core, each core can be used as an independent computing power slice). There are also 2 edge servers (GPUs) in the station. The battery swapping station is equipped with charging racks and a power line communication (PLC) network.

[0050] Scenario: A spare battery (number BAT-12) has been reserved and will be replaced in 10 minutes for an electric vehicle. The battery's BMS processor is currently completely idle. A federated learning training task is issued by the cloud platform, which requires a large amount of distributed computing power (approximately 30 ARM cores). The fixed computing units on site only have two GPUs on the edge server and some BMS processors, which are already occupied and cannot meet the demand.

[0051] Computing power migration process: Prediction and pre-registration: The scheduler predicts that BAT-12 will leave in 10 minutes based on the reservation information; it queries the expected movement path of the battery (it will travel with the vehicle after being swapped out, and will pass through three other battery swapping stations along the way), and coordinates through the cloud to pre-register the computing power capacity at the first battery swapping station (target node) along the way - reserving 4 ARM core resources.

[0052] Task Deployment: The scheduler divides the federated learning training task into 30 subtasks, deploys 4 of these subtasks to the BMS processor of the BAT-12 (each subtask occupies one large core, and a small core is reserved for BMS monitoring), and marks the computing power slice as "driftable".

[0053] Energy-data collaborative transmission: Before the BAT-12 is swapped out, the battery swapping station charges it briefly at 10kW using the charging rack (energy flows to the battery); while transmitting energy, the scheduler uses the PLC channel to pre-synchronize the task image (including model parameters and current gradient state) to the pre-registered resources of the target node, and the source slice continues to run.

[0054] Hot-swap mounting: When BAT-12 is removed from the charging rack, a migration command is triggered, the source slice performs a state snapshot (incremental changes only), and sends it to the target node via short-range wireless communication (such as UWB). The target node resumes its task within 50ms and continues execution from the snapshot point; the source slice is released, and BAT-12's BMS returns to normal monitoring mode.

[0055] Results: The federated learning task was completed within the expected time. BAT-12 contributed computing power during its 8-minute stay. The migration process was transparent to the task (the task did not perceive the interruption). This embodiment realizes that computing power drifts with the energy carrier, making full use of the originally idle "walking computing power".

[0056] Example 3: Multi-condition arbitration using a two-way interlocking mechanism; In this embodiment, the composite node is under extreme operating conditions: a summer afternoon with an ambient temperature of 40°C, and the liquid cooling system's maximum heat dissipation power is reduced due to partial fan failure. Reduced to 35kW. Battery clusters can be fast charged at a 1C rate. Total power consumption of computing chips The total thermal power is 40kW, which exceeds 35kW, triggering a thermal-computation interlock and requiring a joint reduction in computing power and charging power.

[0057] During the load reduction process executed by the dispatcher, the power grid suddenly issued a command to "increase discharge by 50kW". The energy storage PCS was preparing to discharge. At this time, the estimated value of the negative electrode potential of the battery was close to the lithium plating threshold (due to high temperature and fast charging). The power-computer interlock was triggered, requiring the charging current to be diverted.

[0058] Arbitration Process: The bidirectional interlock arbitration module detects that two interlock conditions are triggered simultaneously. Based on priority "electrical-computational interlock > thermal-computational interlock," the electrical-computational interlock is executed first: a risk-avoidance computing task is generated, waking up two idle GPU shards, consuming 20kW of current (reducing the current flowing into the battery by 20kW), while simultaneously reducing the charging power from 30kW to 10kW, and the battery negative electrode potential returns to the safe zone; the joint adjustment vector of the thermal-computational interlock is merged and executed: the already executed computing power wake-up action also helps reduce thermal power (because GPU power consumption is converted into heat, but the total thermal power needs to be recalculated). From 10kW to (Efficiency conversion) (The power output was reduced from 30kW to 10kW, resulting in a total thermal power of 38kW, still higher than 35kW). The arbitrator then issued an additional instruction to reduce the GPU frequency by 10%. The power was reduced to 25kW, with a total thermal power of 35kW, which met the constraints.

[0059] Results: Under multiple conflict constraints, the system arbitrates according to priority and generates a joint adjustment vector, which ensures both battery safety and thermal safety, and partially satisfies the grid command.

[0060] In summary, the following conclusions can be drawn: the computing power configuration method and system can be widely applied to scenarios such as energy storage power stations, charging stations, battery swapping stations, microgrids, and virtual power plants. It can effectively improve the utilization rate of edge computing power, enhance the flexibility and economy of energy storage systems participating in grid interaction, and at the same time ensure battery safety and equipment lifespan. This invention can be deployed using existing hardware conditions (such as BMS processors, charging pile GPUs, and PLC communication) without large-scale modifications, and has good prospects for industrial application.

[0061] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A computing power configuration method based on dynamic resource reorganization, applied to an energy storage-edge computing composite node, wherein the composite node comprises fixed computing units and portable computing power carriers, characterized in that, The method includes: The fixed computing unit and the portable computing power carrier are uniformly abstracted into a dynamically divisible heterogeneous computing power pool. The portable computing power carrier includes a mobile energy storage unit that can serve as an energy carrier. A two-way interlocking integrated protection mechanism is established. The mechanism defines a set of interlocking constraints including at least thermal-computation interlocking, electrical-computation interlocking, and pressure-computation interlocking. When any interlocking constraint is triggered, a joint adjustment vector is generated and sent synchronously to the computing power scheduler and the energy storage controller. Based on the aforementioned bidirectional interlocking integrated protection mechanism, at least one linkage decision is executed, including: Computing power drift and lossless migration: When it is predicted that the migrateable computing power carrier will leave the current composite node, a state snapshot is taken of the computing power slice running on it, and the task image is migrated to the target node using the energy flow channel and hot-swapping is completed. Thermal coupling interlock: Based on the thermal load model of the shared cooling circuit, the total power consumption of the computing power slice and the charging and discharging power of the energy storage system are jointly adjusted; Virtual discharge inversion: Based on the grid dispatch instructions, a virtual discharge computing task is generated. By waking up or increasing the frequency of the computing power slice, the corresponding electrical energy is consumed so that the actual discharge power of the energy storage system meets the grid instructions. Based on the execution result of the linkage decision, the computing units in the heterogeneous computing pool are subjected to operations such as fragmentation, fragment aggregation, drift migration, or computational precision adjustment. The actual state parameters after execution are read back, and the threshold of the interlock constraint and the parameters of the linkage decision model are corrected according to the deviation value.

2. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, The aforementioned computing power drift and lossless migration further include: Based on the predicted movement path of the portable computing power carrier, computing power capacity is pre-registered at at least one composite node along the route. The task image is transmitted synchronously while energy flows bidirectionally, utilizing the V2G charging channel, battery cluster equalization charging and discharging channel, or power line communication channel between the portable computing power carrier and the target node. Without interrupting service, the task image is restored and the hot-swap mounting of the computing power slice is completed at the target node. Once the migration is complete, release the computing power slice resources on the source carrier and set the source carrier to a hibernation or reallocation state.

3. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, The bidirectional interlocking integrated protection mechanism further includes: The interlock priority order is defined as follows: safety-calculation interlock > electrical-calculation interlock > thermal-calculation interlock > pressure-calculation interlock > aging-calculation interlock; When multiple interlocking constraints are triggered simultaneously, a joint adjustment vector is generated according to the priority arbitration, and the joint adjustment vector is represented as follows: in This is the total power consumption adjustment amount for computing power. For the adjustment of energy storage charging and discharging power, This refers to the frequency adjustment of the computing chip. This is an adjustment factor for the number of computing power slices; The joint adjustment vector is sent to both the computing power scheduler and the energy storage controller simultaneously, enabling a single instruction to perform dual-domain collaborative execution of the computing power domain and the energy domain. Record the actual response effect of each interlock adjustment, dynamically adjust the trigger threshold of each interlock type and the gain coefficient of the joint adjustment vector to form an adaptive interlock closed loop.

4. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, The thermal coupling interlock includes: real-time acquisition of the total heat load margin of the shared cooling loop and the real-time temperature and power consumption of each computing unit; and establishment of constraint relationships. ,in This represents the total power consumption of the current computing chip. The charging and discharging heat power of the energy storage battery cluster, The maximum heat dissipation power of the shared cooling circuit; when the total heat power exceeds the total heat load margin, a joint adjustment scheme is generated, which simultaneously reduces the number of computing power slices or the operating frequency of at least one computing unit, and reduces the charging and discharging power of the energy storage system; The virtual discharge inversion includes: monitoring grid dispatch commands and calculating the active power deviation that the energy storage system needs to increase or decrease. ;when When the energy storage system is required to increase its discharge power and the external load is insufficient, the target power consumption is automatically generated. The virtual discharge computing task dynamically wakes up or upscales at least one idle computing unit's computing power slice to execute an interruptible computing task; the virtual discharge computing task is marked as an interruptible priority at any time, and the corresponding computing power slice is automatically released after the power grid command is revoked or the risk of battery lithium plating is eliminated.

5. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, It also includes bus voltage coupling logic and lithium plating risk mitigation and reorganization logic: The bus voltage coupling logic is as follows: real-time monitoring of DC bus voltage ripple; when the voltage drops below the first threshold, freezing the wake-up operation of non-critical computing power slices and reducing the frequency of activated computing power slices; when the voltage drops below the second threshold, triggering the energy storage converter to increase the discharge current while reducing the frequency. The lithium plating risk avoidance and reorganization logic is as follows: Real-time estimation of the lithium plating edge state of the battery negative electrode. When the negative electrode potential is lower than the safety threshold and the current charging current cannot be reduced quickly, a risk avoidance computing task is automatically generated, which wakes up or increases the frequency of several interruptible computing slices. By consuming part of the charging current, the actual current flowing into the battery is reduced to below the safety value. After the lithium plating risk is eliminated, the computing slice occupied by the risk avoidance computing task is released.

6. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, The composite node includes a battery swapping station, and the portable computing power carrier includes a spare power battery pack within the battery swapping station. The BMS processor of the spare power battery pack is incorporated into the heterogeneous computing power pool. The method further includes: Based on the charging pile occupancy prediction model, the probability of each charging pile being occupied by charging tasks within a future time window T is estimated. When it is necessary to perform computing power slice migration, the built-in computing unit of the charging pile or the on-site edge server with a predicted probability lower than the set threshold shall be selected as the target node. For the backup power battery that is about to be replaced, during the remaining time it stays in the station, the idle computing power of its BMS processor is divided into temporary computing power slices and non-real-time computing tasks are attached. Before being replaced, a state snapshot and computing power drift and lossless migration are automatically performed.

7. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, It also includes the information-energy joint routing step: Maintain a dynamic joint routing table, which records the current available energy transmission direction, transmission power, and signal-to-noise ratio of power line communication between any two nodes; When high-bandwidth or low-latency computing power slice data needs to be transmitted, the scheduler queries the dynamic joint routing table and prioritizes the communication path with an energy flow direction at the current moment and a signal-to-noise ratio higher than the threshold for data transmission. When the energy flow direction changes or the power line communication channel quality fluctuates, the joint routing table is updated in real time and dynamic reselection of the computing power slice transmission path is triggered.

8. The computing power allocation method based on dynamic resource reorganization according to claim 1, characterized in that, Also includes: Fault-adaptive reconfiguration: Real-time evaluation of the health score of each computing unit, which integrates the temperature, operating voltage, error rate and communication latency of the computing unit; When the score is lower than the health threshold, an active withdrawal is triggered, and the computing power slice on the unit is gradually migrated to other healthy nodes. After the migration is completed, the unit is placed in an unschedulable state, and the energy storage control system is notified to lower the maximum charging and discharging power limit. When the faulty unit recovers, reverse the deployment of computing power slices and restore the power control margin of the energy storage system. Joint scheduling of carbon credits and computing power revenue: establishing a comprehensive revenue model ,in The business revenue generated from executing computing tasks on computing power slices The carbon credits earned by the energy storage system in the current period for participating in carbon trading. To mitigate battery aging costs; to acquire real-time electricity price signals, carbon credit prices, and the carbon reduction rate of energy storage systems, in order to maximize... With the goal of dynamically determining whether to wake up, reduce frequency, migrate, or hibernate computing power slices.

9. A computing power allocation system based on dynamic resource reorganization, used to implement the method described in any one of claims 1-8, characterized in that, It includes a resource pooling module, a multi-source sensing module, a two-way interlock arbitration module, a computing power drift module, a linkage decision-making module, a reorganization execution module, and a feedback correction module; The resource pooling module is configured to abstract the fixed computing units and the portable computing power carriers within the energy storage-edge computing composite node into a dynamically divisible heterogeneous computing power pool. The multi-source sensing module is configured to collect energy parameters, thermal parameters, power grid parameters, and computing power parameters in real time. The bidirectional interlocking arbitration module is configured to execute the bidirectional interlocking integrated protection mechanism as described in claim 1 or 3, and generate a joint adjustment vector. The computing power drift module is configured to execute the computing power drift and lossless migration process as described in claim 2; The linkage decision module is configured to execute at least one linkage logic as described in claim 4 or 5, namely thermal coupling interlock, virtual discharge inversion, bus voltage coupling, and lithium plating risk avoidance recombination. The reorganization execution module is configured to perform sharding, sharding aggregation, drift migration, or calculation accuracy adjustment operations on the computing units in the heterogeneous computing pool based on the output of the linkage decision module or the computing power drift module. The feedback correction module is configured to read back the actual state parameters after execution and correct the interlock threshold of the bidirectional interlock arbitration module and the decision model parameters of the linkage decision module.

10. The computing power allocation system based on dynamic resource reorganization according to claim 9, characterized in that, The system is deployed in at least one of the following physical entities: integrated photovoltaic, energy storage and charging power station, battery swapping station, microgrid energy storage station, electric vehicle V2G charging pile group, and mobile energy storage vehicle.

Images