Digital energy buffering algorithm and device based on dynamic reconfigurable technology

Abstract

The invention provides a digital energy buffering algorithm and device based on a dynamic reconfigurable technology. The algorithm comprises the following steps: step 1, starting and initializing a system; step 2, state monitoring and data acquisition; step 3, energy resource scheduling decision making: realizing supply and demand balance through dynamic reconstruction; step 4, resource scheduling and state switching; 5, updating the system state: returning to the step 2, and updating the system state; according to the digital energy buffering algorithm and device based on the dynamic reconfigurable technology, the structural arrangement of the device can be optimized, bilateral disturbance can be effectively isolated, and the situation that a load is impacted by sudden change of the load and the load is influenced by fluctuation of the power grid is avoided; an intermediate layer with quick response capability and an energy buffering effect is constructed to simulate a constant power load and provide an ideal voltage source; and the electric energy quality management and the energy storage scheduling function are deeply integrated, so that the system reliability is improved, and the energy use cost of a user is remarkably reduced.

Description

Technical Field

[0001] This invention relates to the fields of new energy, power systems and energy storage technologies, and more specifically, to a digital energy buffering algorithm and device based on dynamic reconfigurable technology. Background Technology

[0002] In modern electricity-dependent societies, the stability of the power supply system is fundamental to ensuring continuous industrial production, normal commercial operations, and smooth daily life. However, the power system constantly faces dual instability challenges from both the load side and the grid side. Frequent start-ups and shutdowns of loads such as large motors, electric arc furnaces, precision automated equipment, and data center server clusters can lead to severe and unpredictable power fluctuations; the grid side suffers from power quality issues such as voltage sags, harmonic pollution, and transient outages. Furthermore, with the large-scale integration of intermittent renewable energy sources such as distributed photovoltaic and wind power, and the rapid growth of fluctuating loads such as electric vehicle charging stations, the complexity and uncertainty of grid operation are further exacerbated.

[0003] The consequences of this two-sided instability far exceed the scope of traditional "power outages," easily triggering problems such as the shutdown of sensitive equipment, loss of control over production processes, and decreased product yield, resulting in huge direct and indirect economic losses and posing a serious threat to the power quality and safe operation of the regional power grid. Therefore, how to maintain the stable operation of the power supply system under complex operating conditions has become a key issue in current power system optimization and energy management.

[0004] While various power quality management devices such as voltage stabilizers, filters, and reactive power compensation exist in current technologies, most can only provide localized optimization for specific problems and lack system-level power buffering and energy dispatching capabilities. Especially when dealing with high-frequency, abrupt load fluctuations and grid disturbances, traditional devices often exhibit slow response times and limited adjustment ranges, making it difficult to achieve complete decoupling between power supply and load. Furthermore, under the context of renewable energy integration and time-of-use pricing, existing systems generally lack flexible and intelligent energy storage dispatching mechanisms, failing to effectively achieve peak-valley arbitrage and refined energy cost management. Therefore, researching how to fundamentally isolate grid and load disturbances and improve the intelligent dispatching and buffering capabilities of devices is of great significance.

[0005] Patent CN120301042B discloses a control method and system for a multi-source intelligent power manager based on 5G communication. The method includes: acquiring multi-source heterogeneous power data from the multi-source intelligent power manager; constructing a dynamic energy topology using a multi-scale pulse fusion tensor decomposition algorithm; eliminating multi-source signal interference through a phase synchronization time-frequency constraint mechanism; and generating a spatiotemporally aligned power state joint tensor. Although the system involves the setting of decoupling and filtering system buses, and although it can only improve the flexibility and response speed of power management and ensure the stability and reliability of power supply, it cannot effectively solve the problem of how to isolate the disturbances between the power grid and the load, or improve the intelligent scheduling and buffering capabilities of the device. Summary of the Invention

[0006] In view of this, the present invention aims to propose a digital energy buffering algorithm and device based on dynamic reconfigurable technology to solve a series of problems caused by instability on both the load side and the grid side in the current power system. Specifically, on the load side, frequent start-stop and sudden changes in operating conditions of loads such as large industrial motors, automated equipment, and data center server clusters can generate severe and unpredictable power surges and high-frequency interference. These disturbances are directly fed back to the grid, seriously affecting power quality and potentially causing abnormal operation or even damage to other sensitive equipment on the same power supply bus. On the grid side, voltage sags, harmonic pollution, frequency fluctuations, and intermittent and random power injections due to the large-scale integration of new energy sources degrade grid power quality, thereby threatening the safe and stable operation of load equipment, especially precision instruments and continuous production processes. The invention aims to optimize the device's structural design, effectively isolate disturbances on both sides, and prevent sudden load changes from impacting the grid and grid fluctuations from affecting the load. It constructs an intermediate layer with rapid response capabilities and energy buffering to simulate constant power loads and provide an ideal voltage source. It deeply integrates power quality management and energy storage scheduling functions, significantly reducing users' energy costs while improving system reliability.

[0007] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0008] This invention relates to a digital energy buffering algorithm and device based on dynamic reconfigurable technology. The digital energy buffering algorithm based on dynamic reconfigurable technology includes the following steps:

[0009] Step 1: System Startup and Initialization: After starting the system within the device, perform system initialization operations;

[0010] Step 2, Status Monitoring and Data Acquisition: The status of the system within the device is sensed through algorithms, and data acquisition, system status monitoring, load power calculation, and smoothing are performed on the system.

[0011] Step 3, Energy Resource Scheduling Decisions: Achieving supply and demand balance through dynamic reconfiguration;

[0012] Step 4, Resource Scheduling and State Switching: Dynamic switching algorithm for dual-port battery modules within the device;

[0013] Step 5: System Status Update: Return to Step 2 to update the system status.

[0014] Furthermore, step one includes:

[0015] Step S11: Power on the system, start the internal system of the device, and perform self-test.

[0016] Step S12: Initialize the system, preset relevant parameters and load power change threshold dP. threshold Objective function J and related constraints.

[0017] Furthermore, the relevant parameters include , , Any one or more parameters in it.

[0018] Furthermore, the objective function J can be expressed as:

[0019] ;

[0020] To provide a constant power supply to the desired power grid; It can be set by the system or dynamically adjusted according to peak and off-peak electricity prices; This represents the actual power output of the power grid. This represents the average SOC of all battery packs. ω1, ω2, and ω3 are the load demand response delays; ω1, ω2, and ω3 are all weighting coefficients.

[0021] Furthermore, in step two, data acquisition and system status monitoring refer to the algorithm periodically acquiring the state variables S(k) of different battery packs through the data bus.

[0022] Furthermore, the formulas and models used in the load power calculation and smoothing process in step two include the instantaneous load power P. load (t) Calculation formula and digital filtering formula.

[0023] Furthermore, the digital filtering formula is P load_filtered (k)=α·P load (k)+(1-α)I load_filtered (k).

[0024] Furthermore, step two includes:

[0025] Step S21: Real-time monitoring of state data acquisition S(k) and calculation of P load_filtered (k);

[0026] Step S22: Health and safety judgment: Check whether all battery parameters are within the constraints. If they exceed the limits, trigger the protection logic.

[0027] Step S23: Quick Response Judgment: Real-time Calculation of Load Power Change Rate ; Determine whether If yes, then mark the system as entering "fast response mode"; proceed to step S24;

[0028] Step S24: Based on SOC, temperature, and current limits, immediately assess the output margin of the current power supply battery pack; determine if the output margin is sufficient. If not, the margin is insufficient, and proceed to step three; prioritize triggering role switching without waiting for the SOC threshold or timed triggering.

[0029] Furthermore, step three, achieving supply and demand balance through dynamic reconfiguration, is the core decision-making process of the algorithm, serving as the basis for allocating battery packs. Step three, achieving supply and demand balance through dynamic reconfiguration, includes:

[0030] Step S31: Role definition and state machine: Each battery pack is defined as a state machine, and each state machine includes three states: charging state, power supply state, and switching / standby state.

[0031] Step S32: Dynamic Reconfiguration Strategy: The system is dynamically reconfigured by executing relevant scheduling logic in each decision cycle.

[0032] Furthermore, in step S32, during the battery pack role switching process, a temporary freewheeling diode connected in parallel with at least two electronic switches in the discharge circuit is used to form a temporary freewheeling path to ensure zero interruption of load power supply; the time-of-use electricity price signal and / or the task scheduling information of the AI ​​computing cluster are obtained in real time or at regular intervals.

[0033] Based on the acquired time-of-use electricity price signal and / or the task scheduling information of the AI ​​computing cluster, the constant power charging power value of the battery pack in the charging state is dynamically adjusted.

[0034] When a power surge is predicted for the AI ​​computing cluster, the charging power of the current battery pack is reduced or suspended in advance; when a power drop is predicted for the AI ​​computing cluster, the charging power of the current battery pack is reduced or even stopped in advance to absorb excess energy. This ensures that the power supply network always sees an equivalent load with approximately constant power, while the load side obtains a highly stable voltage source, enabling electricity price arbitrage and proactive buffering against transient power changes in the AI ​​computing load.

[0035] Furthermore, the task scheduling information of the AI ​​computing cluster is directly pushed through the power consumption prediction model or job scheduler of the GPU server; the constant power charging power value is increased to the rated maximum value during low electricity prices and decreased to zero or negative value during high electricity prices.

[0036] A digital energy buffer device based on dynamic reconfigurable technology includes a digital energy buffer algorithm based on dynamic reconfigurable technology. The device is deployed as an intermediate system between the power supply network and the electrical load. The device adopts a hierarchical modular design. The device also includes a supercapacitor module connected in parallel with the DC bus. The digital energy controller uses FPGA or DSP to achieve millisecond-level control.

[0037] Compared with existing technologies, the digital energy buffering algorithm and device based on dynamic reconfigurable technology described in this invention have the following advantages:

[0038] By combining the algorithms and devices described in this application, the structural configuration of the device can be optimized to effectively isolate bilateral disturbances, avoid load surges impacting the power grid and power grid fluctuations affecting the load; a middle layer with fast response capability and energy buffering function can be constructed to simulate constant power load and provide an ideal voltage source; and power quality management and energy storage scheduling functions can be deeply integrated to improve system reliability and significantly reduce users' energy costs. Attached Figure Description

[0039] The accompanying drawings, which constitute a part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0040] Figure 1 This is a schematic diagram of the overall structure of the device;

[0041] Figure 2 This is a schematic diagram of the overall architecture of the algorithm;

[0042] Figure 3 This is a schematic diagram illustrating the overall details of the algorithm.

[0043] Figure 4 This diagram illustrates the connection relationship between the battery packs and the external interface when there are two battery packs. Detailed Implementation

[0044] The inventive concepts of this disclosure will be described below using terminology commonly used by those skilled in the art to communicate the essence of their work to others skilled in the art. However, these inventive concepts may be embodied in many different forms and should not be construed as limited to the embodiments described herein.

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0046] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] In existing technologies, common power quality management devices such as uninterruptible power supplies (UPS), dynamic voltage restorers (DVRs), or active power filters (APFs) mostly focus on local compensation and correction of voltage or current waveforms. Their response speed, energy throughput, and system-level power buffering capabilities are limited, making it difficult to fundamentally achieve complete decoupling between the power supply network and the electrical load. This means that the power grid still needs to directly withstand the dynamic impacts of the load, while the load cannot be completely immune to disturbances from the power grid. Furthermore, in scenarios with significant peak-valley electricity price differences, existing energy storage systems are often independent of power quality management systems, lacking an integrated intelligent dispatch mechanism. This makes it impossible to achieve cost-effective peak shaving and valley filling and energy arbitrage functions while ensuring power quality.

[0048] To address the series of problems caused by instability on both the load side and the grid side in current power systems, this embodiment proposes a digital energy buffering algorithm and device based on dynamic reconfigurability technology. The digital energy buffering algorithm based on dynamic reconfigurability technology includes the following steps:

[0049] Step 1: System Startup and Initialization: After starting the system within the device, perform system initialization operations;

[0050] Step 2, Status Monitoring and Data Acquisition: The status of the system within the device is sensed through algorithms, and data acquisition, system status monitoring, load power calculation, and smoothing are performed on the system.

[0051] Step 3, Energy Resource Scheduling Decisions: Achieving supply and demand balance through dynamic reconfiguration;

[0052] Step 4, Resource Scheduling and State Switching: Dynamic switching algorithm for dual-port battery modules within the device;

[0053] Step 5: System Status Update: Return to Step 2 to update the system status.

[0054] By configuring the algorithm, the digital energy buffer device can be controlled to become an ideal "energy buffer layer," achieving complete decoupling between the power supply network and the load. The basic principle of the algorithm is to calculate the load's power demand in real time and dynamically reconstruct the connection state of the internal battery modules, ensuring that the grid side always charges a constant-power load, while the load side is always powered by a near-ideal voltage source (battery pack). Furthermore, when dealing with negative power transients in the load, the method reduces or stops the charging power of the charging battery pack and utilizes it to absorb excess energy, preventing a surge in the DC bus voltage within the device.

[0055] Step one includes:

[0056] Step S11: Power on the system, start the internal system of the device, and perform self-test.

[0057] Step S12: Initialize the system, preset relevant parameters and load power change threshold dP. threshold Objective function J and related constraints. Related parameters include... , , Any one or more parameters in it.

[0058] Specifically, the objective function J can be expressed as:

[0059] ;

[0060] The constant power that the power grid is expected to provide.

[0061] It can be set by the system or dynamically adjusted according to peak and off-peak electricity prices.

[0062] This represents the actual power output of the power grid.

[0063] This represents the average SOC of all battery packs.

[0064] Load demand response delay is a metric that measures the time required from a sudden change in load power until the digital energy buffer fully accommodates the demand through internal adjustments (such as changes in battery pack output power), thereby restoring grid power stability. Ideally, all power surges in the load are entirely compensated for by changes in the battery pack's state of charge and discharge within the buffer, requiring no grid-side response. This latency is affected by the controller's processing speed, switching time, and algorithm strategy.

[0065] ω1, ω2, and ω3 are all weighting coefficients used to balance the priorities of the three objectives: "grid power stability," "battery pack balance," and "load dynamic response performance," respectively. In AI load scenarios, the weight of response speed ω3 should be significantly increased because a slow response means that the load voltage may fluctuate, directly affecting the stable operation of the computing task and equipment safety.

[0066] By setting the objective function, the core objective of the algorithm is to maintain stable power on the grid side and preserve energy balance among battery packs. The algorithm's decision-making process pursues the optimal system objective while satisfying a series of hardware and safety constraints. Therefore, setting the objective function effectively improves the algorithm's reliability.

[0067] The relevant constraints include any one or more of the following: battery pack operation constraints, power path constraints, dynamic performance constraints, and power quality constraints.

[0068] Specifically, battery pack operating constraints include: SOC operating range, charging current limit, discharging current limit, and temperature operating range constraints. The SOC operating range is as follows: The charging current is limited to: , This is the minimum charging current. This represents the maximum charging current; the discharge current is limited to: ; This is the minimum discharge current. This represents the maximum discharge current; the operating temperature range is: .

[0069] Power path constraints include: at any given time, exactly one battery pack is connected to the grid for constant-power charging. At any given time, at least one battery pack is supplying power to the load, and the power supply loop is open. Any combination of switching states that could lead to circulating currents between battery packs is prohibited.

[0070] Dynamic performance constraints are also known as maximum response time constraints. The maximum response time constraint is... This is the maximum allowable response latency for the system. For AI workloads, this value should be set in the millisecond range, for example... ≤10ms, to ensure effective shielding against GPU transients.

[0071] Power quality constraints are load-side constraints. Power quality constraints include load voltage ripple. Load voltage ripple is... During sudden changes in load power and internal system switching, voltage fluctuations at the load end must be limited to a very small range; for example... ±2% to ensure the normal operation of sensitive AI devices.

[0072] Algorithm-based decision-making must strictly adhere to the above constraints, which can effectively improve the accuracy and reliability of the algorithm, as well as its computational efficiency.

[0073] Step two, data acquisition and system status monitoring, refers to the algorithm periodically collecting the state variables S(k) of different battery packs via the data bus. Specifically, SOC... i (k) represents the real-time state of charge of the i-th battery pack. V i (k), I i (k), T i (k) represents the voltage, current, and temperature of the i-th battery pack, respectively. SW i,j (k) represents the current state of switch j in the switch array corresponding to the i-th battery pack. The state of switch j is usually closed or open.

[0074] The formulas and models used in step two for load power calculation and smoothing include the instantaneous load power P. load (t) Calculation formula and digital filtering formula. Instantaneous load power P load The formula for calculating (t) is: P load (t)=V load (t)×I load (t), where V load (t) and I load (t) is measured in real time by a sensor at the load end of the device. The digital filtering formula is P. load_filtered (k)=α·P load (k)+(1-α)I load_filtered (k). P serves as the basis for algorithmic decision-making. load_filtered (k) represents the load power value after filtering in the k-th sampling period. α is the filtering coefficient, and 0 < α < 1. The value of α determines the trade-off between the system's response speed and smoothness. The larger α is, the faster the response, but the less smooth it is. In this embodiment, the digital filtering is a first-order low-pass filter.

[0075] By setting up load power calculation and smoothing processing as the basis for the algorithm to perceive the system state and make decisions, the instantaneous power demand of the load can be obtained in real time and filtered to obtain a reference power value for scheduling decisions, thus avoiding frequent switching due to instantaneous noise.

[0076] Step two includes:

[0077] Step S21: Real-time monitoring of state data acquisition S(k) and calculation of P load_filtered (k);

[0078] Step S22: Health and safety judgment: Check whether all battery parameters are within the constraints. If they exceed the limits, trigger the protection logic.

[0079] Step S23: Quick Response Judgment: Real-time Calculation of Load Power Change Rate ; Determine whether If yes, then mark the system as entering "fast response mode"; proceed to step S24;

[0080] Step S24: Based on SOC, temperature, and current limits, immediately assess the output margin of the current power supply battery pack; determine if the output margin is sufficient. If not, the margin is insufficient, and proceed to step three; prioritize triggering role switching without waiting for the SOC threshold or timed triggering.

[0081] By coordinating the steps in Step Two, which is part of the main loop, the security and stability of the system can be effectively improved. By setting health and safety judgments, the accuracy of the system operation can be enhanced.

[0082] Step three, achieving supply and demand balance through dynamic reconfiguration, is the core decision-making process of the algorithm, which determines how to allocate battery packs. Specifically, step three, dynamic reconfiguration to achieve supply and demand balance, includes:

[0083] Step S31: Role definition and state machine: Each battery pack is defined as a state machine, and each state machine includes three states: charging, supplying, and switching / standby.

[0084] Step S32: Dynamic Reconfiguration Strategy: The system is dynamically reconfigured by executing relevant scheduling logic in each decision cycle.

[0085] In step S31, the charging state refers to being connected to the power grid via a constant power module, with a power P charge Charging. Power supply status refers to connecting to the load via a switch array and supplying power to the load. Switching / standby status refers to a brief transitional state during the switching process.

[0086] The scheduling logic in step S32 includes:

[0087] Step S321: Demand Judgment: Compare the filtered load power P load_filtered With the current charging power P charge Determine whether dynamic reconstruction is needed. If yes, proceed to step S322.

[0088] Step S322: Role swap trigger: When the SOC of the i-th battery pack currently supplying power is... iThe SOC drops to the set low threshold. low Reconfiguration is triggered when the system needs to switch based on time or electricity pricing policies.

[0089] Step S323: Optimal group selection: From all non-charging battery groups, select the battery group j with the highest SOC and in good condition as the next power supply unit.

[0090] Step S324: Seamless Switching Planning: The algorithm generates a preset switching sequence to ensure that the load side is always powered by the battery pack during the switching process, achieving "zero power interruption". The load side and the battery pack are connected via switching diodes or directly.

[0091] By setting up dynamic reconfiguration in step three, each battery pack can support dynamic reconfiguration in the parallel direction, enabling dynamic reconfiguration of battery cells within the battery pack, constructing pulse charging and discharging conditions, and extending battery life.

[0092] In step S32, during the battery pack role switching process, a temporary freewheeling diode connected in parallel with at least two electronic switches in the discharge circuit is used to form a temporary freewheeling path to ensure zero interruption of load power supply; the time-of-use electricity price signal and / or task scheduling information of the AI ​​computing cluster are obtained in real time or at regular intervals.

[0093] Based on the acquired time-of-use electricity price signal and / or the task scheduling information of the AI ​​computing cluster, the constant power charging power value of the battery pack in the charging state is dynamically adjusted.

[0094] When a power surge in the AI ​​computing cluster is predicted, the charging power of the current battery pack is reduced or paused in advance; when a power drop in the AI ​​computing cluster is predicted, the charging power of the current battery pack is reduced or even stopped in advance to absorb excess energy. This ensures that the power supply network always sees an equivalent load of approximately constant power, while providing the load side with a highly stable voltage source, and enabling electricity price arbitrage and proactive buffering against transient power fluctuations in the AI ​​computing load. The task scheduling information of the AI ​​computing cluster is directly pushed through the power consumption prediction model of the GPU server or the job scheduler; the constant power charging power value is increased to the rated maximum value during periods of low electricity prices and decreased to zero or a negative value during periods of high electricity prices.

[0095] Advanced system optimization is performed during the decision gap in step three. Optimization includes any one or more of predictive scheduling, state preservation, peak-valley arbitrage, and load balancing.

[0096] Specifically, predictive scheduling refers to: by integrating with the cluster management system, obtaining advance notice of task start / stop, and pre-setting a Δt time to initiate the switchover, so that the system is already in an optimal state before sudden load changes.

[0097] Status maintenance: In fast response mode, if the current power supply battery pack has sufficient capacity, the originally scheduled routine switching can be appropriately postponed to maintain the stability of the power supply link and avoid unnecessary switching actions.

[0098] Peak-valley arbitrage: dynamically adjusting P based on electricity price signals grid_ref During valley hours, set P. grid_ref For larger values, actively store electricity; during peak hours, if feedback is allowed, set P. grid_ref If the value is 0 or negative, the battery pack discharges, reducing the purchase of electricity from the grid.

[0099] Balanced maintenance: During long-term operation, if the SOC difference between battery packs is found to be increasing continuously, additional switching cycles can be actively triggered to allow battery packs with high SOC to discharge more and battery packs with low SOC to charge more.

[0100] Through the algorithms described in steps two and three, the digital energy buffer device can intelligently act as a "decoupling" between the power grid and the load, which not only ensures extremely high power supply quality on the load side, but also greatly improves the stability and economy of the power grid side.

[0101] Step four, resource scheduling and state switching, includes: executing a preset switching sequence according to the switching command issued in step three; and enabling different battery packs to switch accordingly according to the required sequence.

[0102] A digital energy buffer device based on dynamic reconfigurable technology is disclosed. The device employs a digital energy buffering algorithm based on this technology and is deployed as an intermediate system between the power supply network and the electrical load. The device adopts a hierarchical modular design and includes a supercapacitor module connected in parallel with the DC bus. The digital energy controller uses an FPGA or DSP to achieve millisecond-level control. The number and capacity of the battery packs can be expanded according to the application scenario.

[0103] A central digital energy controller coordinates and manages multiple parallel, identically structured battery packs and switch array modules, thereby enabling intelligent energy scheduling and dynamic reconfiguration of power paths.

[0104] By deploying an intermediate system called a "digital energy buffer device" between the power supply network and the electrical load, and relying on a dedicated energy scheduling algorithm and power electronic switch array within the device, a virtual power buffer layer is constructed. This layer acts as a "decoupling" in the energy transmission path, its core function being to dynamically and intelligently manage the storage and release of energy. This ensures that the equivalent load viewed from the grid side appears as a nearly constant power load, while the power source viewed from the load side appears as a highly stable voltage source with near-ideal characteristics. Furthermore, as a digital energy buffer system, this device achieves efficient decoupling between power supply and load through a dedicated scheduling algorithm and modular circuit design. Its core lies in constructing a virtual power buffer layer, enabling the power supply network to handle a nearly constant power equivalent load while simultaneously providing the load with a highly stable ideal voltage source. This technology is suitable for scenarios with high power quality requirements, such as industrial equipment, data centers, and communication base stations, and supports integrated peak shaving and valley filling backup and storage applications, possessing good scalability and engineering practical value.

[0105] The digital energy buffer device includes a power routing module, an energy unit module, a digital energy controller, and a human-machine interface and communication module. The power routing module has bidirectional communication connections with the energy unit module, the digital energy controller, and the human-machine interface and communication module. The power supply network in the external system is connected to the power routing module, the energy unit module, and the electrical loads via external power interfaces. The energy unit module is connected to the electrical loads via an external power interface. The power routing module is programmable. Additionally, the device integrates a supercapacitor module connected in parallel with the DC bus to absorb extremely high-frequency power pulsations ranging from microseconds to milliseconds.

[0106] By strategically configuring its components, the device serves as an integrated hardware and software platform combining power electronics, energy storage, and digital control. This significantly enhances the stability of the device structure, ensures its functionality, and fully embodies its modularity, scalability, and intelligence. Through the precise control of the hardware platform combined with intelligent scheduling via software algorithms, efficient decoupling and high-quality energy management between the power supply network and the electrical load are achieved. The human-machine interface and communication module enable parameter setting, status display, and communication with the higher-level energy management system. The communication module receives task scheduling information from the AI ​​cluster management system and uses this information to predict load power changes, thereby optimizing the battery pack's scheduling strategy.

[0107] The power routing module is the physical execution layer of energy dispatch, responsible for constructing and switching energy flow paths between the power supply network and the load. Its core consists of multiple power switch nodes directly defined by software, responsible for constructing arbitrary energy flow paths. The power routing module includes an electronic switch array, an internal chip, and diodes. Some switches within the electronic switch array are connected in parallel with diodes. The electronic switch array is connected to the internal chip, the power grid, the load, and the corresponding battery pack. At least one electronic switch array and at least two diodes are required. The diodes are reverse-biased.

[0108] The electronic switch array includes three high-power electronic switches, which together form a programmable three-port power router. The three ports are connected to the power grid, the load, and the corresponding battery pack, respectively. The switches connected to the load and battery pack are equipped with reverse-biased diodes in parallel. The high-power electronic switches are power devices based on wide-bandgap semiconductor materials. The high-power electronic switches can be either MOSFETs or IGBTs. In this embodiment, the switching devices must have high frequency and low loss characteristics to support millisecond-level state switching. The selection of the freewheeling diode must meet the maximum instantaneous current requirement of the load. Preferably, each electronic switch array includes a constant power module and three electronic switches. At least two of the three electronic switches are equipped with reverse-biased diodes in parallel to provide a freewheeling path when the switch is open and to prevent circulating current between battery packs. The constant power module is activated when the battery pack is charging to ensure a constant power absorption from the power supply network.

[0109] The use of three high-power switches provides a freewheeling path when the switches are open, ensuring continuous power supply to the load during state switching and preventing circulating current between battery packs. The reverse-biased diodes connected in parallel with the required switches provide a preset, passive freewheeling path for instantaneous energy during reconfiguration, ensuring uninterrupted power supply to the load during topology switching transients and achieving dynamic freewheeling and protection for the device.

[0110] Specifically, the three electronic switches include: a first switch connected between the positive terminal of the battery pack and the constant power module, used to control the charging circuit; a second switch connected between the positive terminal of the battery pack and the common terminal; and a third switch connected between the negative terminal of the battery pack and the load terminal. Both the second and third switches are connected in parallel with reverse-biased diodes. The coordinated switching of the second and third switches controls whether the battery pack is connected to the load terminal to supply power to the load.

[0111] The built-in chip has a digital instruction dynamic configuration and is softly connected to the electronic switch array. Through digital instruction dynamic configuration, it controls each high-power electronic switch in the electronic switch array to achieve "on" and "off".

[0112] Through a software-defined connection between the built-in chip and the electronic array switch, the connection relationship between the battery pack and the system's external ports can be changed instantaneously. The system's external interfaces are the power supply terminal, the common terminal, and the load terminal. For example, a battery pack can be reconfigured in real time from a "connected to the grid and load" configuration to a "connected to the grid only" or "connected to the load only" configuration.

[0113] The energy unit module features heterogeneity and pooling capabilities. Each module includes at least one battery pack. Each battery pack is configured with a switch array. Each battery pack utilizes a modularly designed lithium-ion battery or other types of energy storage battery packs. When two or more battery packs are used, they are arranged in parallel. The capacity and number of battery packs can be flexibly configured according to the application scenario. Each battery pack is abstracted as a standardized energy block with a specific state, which refers to any one of the following: State of Charge (SOC), State of Health (SOH), and temperature. Each battery pack monitors its battery status in real time through its built-in sensors or its built-in battery management functions and reports this information to the digital energy controller via a data bus. Battery status includes any one or more parameters such as voltage, current, and temperature. In this embodiment, the battery pack needs to have high power density and fast charging / discharging capabilities to cope with frequent role switching. The modular design facilitates system expansion and maintenance.

[0114] Working principle: The digital energy controller dynamically selects one or more of the most suitable energy blocks from the energy pool (energy unit module) based on the real-time needs of the system (such as load power and grid dispatch instructions), and puts them into the "power supply" role through the power routing module; at the same time, it selects other energy blocks to put them into the "charging" or "standby" role. This allocation is not fixed, but is continuously reconfigured as conditions and optimization objectives change.

[0115] By treating all battery packs in the system as a unified, pooled energy resource, rather than as independent units fixed to a particular load or function, it is possible to provide a near-ideal voltage source for the load.

[0116] The digital energy controller internally includes a data bus and a control bus. The data bus is responsible for collecting status information from all battery packs and feedback signals from each switch. The control bus is responsible for sending precise on / off commands to each electronic switch array. Additionally, the digital energy controller has functions for state awareness and modeling, reconfiguration decision engine, and timing and safety logic. Specifically, the digital energy controller includes a scheduling algorithm core and a state management and timing logic module. The scheduling algorithm core makes decisions based on the system state, and the state management and timing logic module generates conflict-free switch state switching sequences. Preferably, the digital energy controller is implemented using a field-programmable gate array (FPGA) or a digital signal processor (DSP) to meet millisecond-level control timing requirements.

[0117] By configuring the digital energy controller, it can connect to each battery pack and electronic switch array via a data bus to collect status information, and connect to each electronic switch array via a control bus to control their on / off states. Furthermore, the digital energy controller is configured to schedule the electronic switch arrays, causing the multiple battery packs to alternately be in charging and discharging states.

[0118] The state awareness and modeling function refers to the continuous acquisition of the state of each switch in all energy unit modules and power routing modules through the internal data bus, and the real-time maintenance of a dynamic digital twin model of the system in the digital space.

[0119] The reconfigured decision engine function refers to the scheduling algorithm core acting as the decision engine, performing calculations and decisions based on preset objectives and real-time constraints. Its output is not a simple PID control signal, but a sequence of instructions for system topology reconfiguration. The preset objectives are any one or more of the following: constant power, optimal efficiency, and lowest cost. Real-time constraints are based on battery SOC.

[0120] The timing and safety logic function refers to the digital energy controller's transformation of reconfiguration commands into a series of precisely synchronized and logically conflict-free switching actions, ensuring a smooth, fast, and safe transition from one stable topology to another. In this embodiment, the controller requires a high-performance processor to ensure the real-time performance of the complex calculations of the scheduling algorithm and the millisecond-level control commands.

[0121] By setting up a digital energy controller as the core control module, it can act as the brain of the system, responsible for sensing the global state, executing energy scheduling algorithms, coordinating the collaborative work of all modules, and issuing reconfiguration commands to the physical layer.

[0122] The human-machine interface and communication module includes a local interface and an uplink communication interface. The local interface can be configured with a local display and buttons for displaying system status and setting parameters (such as peak / off-peak electricity pricing periods). The uplink communication interface provides a standard communication interface for connecting to the upper-level energy management system. It receives charging and discharging strategy commands and uploads detailed operating data and alarm information. System status includes one or more of the following: current operating mode, battery SOC, and alarm information. The communication interface includes one or more of the following: Ethernet, RS485, and CAN.

[0123] The human-machine interface and communication module enable interaction with external systems and operators.

[0124] The external power interface includes a power supply terminal, a common terminal, and a load terminal. These three terminals are all located on the device. Specifically, the power supply terminal is connected to the negative output of the power supply system to provide electrical energy input to the device. The common terminal is connected to the positive output of the power supply system and the positive input of the electrical load, forming the main power circuit. The load terminal is connected to the negative input of the electrical load to provide a stable voltage to the load.

[0125] The external communication and control interface includes an upper-level system communication interface and an internal bus interface. Both are located on the device. The upper-level system communication interface is used to connect with the power grid dispatching system, cloud platform, or park energy management system to receive macro-level dispatching commands. The control bus interface within the internal bus interface serves as the internal channel for the digital energy controller to send drive signals to each switch array. The data bus interface within the internal bus interface serves as the internal channel for each battery pack status sensor / BMS to report data to the digital energy controller. Additionally, the external communication and control interface also includes a relay control interface. The relay control interface is located on the device. The relay control interface is used to control the grid connection and off-grid connection of battery packs at the system level.

[0126] Through the configuration of the components within the aforementioned device, it is able to fundamentally isolate the power grid from load disturbances and possess intelligent scheduling and buffering capabilities. Furthermore, this digital energy buffer device integrates a dedicated scheduling algorithm and multi-battery pack collaborative control. By constructing a virtual power buffer layer in the energy transmission path, it achieves bidirectional isolation and dynamic support for the power supply network and load, thereby significantly improving the system's stability, reliability, and economy. Based on the concept of dynamic reconfiguration, this device is no longer a traditional, fixed-function "black box" power supply device, but a software-defined, topology-programmable open energy platform. Its core modules work together to achieve on-demand, real-time, and dynamic intelligent scheduling and routing of energy between the "energy pool" and "demand points," thus fundamentally solving the mismatch problem between the power supply network and dynamic loads.

[0127] Working Principle: The device employs a modular, scalable hardware architecture combined with intelligent scheduling algorithms. At the hardware level, the device mainly consists of multiple independently controllable battery packs, a matching electronic switch array, and a core digital energy controller. The device has three external ports (common port, power supply port, and load port) to achieve a clear energy interface. Multiple battery packs, through the coordinated operation of the electronic switches, rapidly and seamlessly switch between "charging" and "discharging to power the load."

[0128] The core logic of the operation is "time-sharing" and "state switching". At any given time, at least one battery pack is providing a stable voltage source to the load, while another one or more battery packs can be charged by the grid at a constant power. The digital energy controller collects information such as voltage, current, temperature, and switch status of each battery pack in real time through the internal data bus and control bus, and controls the electronic switch array to perform a series of preset, rapid (millisecond-level) state switches according to a dedicated scheduling algorithm.

[0129] By adopting the overall approach of "building a buffer platform with hardware architecture" and "achieving intelligent decoupling through algorithm scheduling", efficient isolation between power supply and load is achieved. On this basis, peak-valley electricity pricing strategies can be further integrated to store energy during off-peak hours and discharge during peak hours, achieving comprehensive economic benefits of integrated backup and storage.

[0130] Example 1: Suppose there are two battery packs, and the initial states of the two battery packs are: Bat1 is charging and Bat2 is supplying power.

[0131] A digital energy buffering algorithm based on dynamic reconfigurable technology includes the following steps:

[0132] Step 1: Initialization: Power on the system and perform a self-test. Set P grid_ref SOC high SOC low Any one or more parameters in the table. Set the load power change rate threshold dP. threshold .

[0133] Step 2, Status Monitoring: Real-time acquisition of S(k) and calculation of P load_filtered (k).

[0134] Health and safety assessment: Check whether all battery parameters are within the constraints. If they exceed the limits, trigger protection logic (such as disconnecting the relay).

[0135] Quick response judgment:

[0136] Real-time calculation of load power change rate .

[0137] if Then: the system is marked to enter "fast response mode".

[0138] Immediately assess the output margin of the current power supply battery pack (based on SOC, temperature, and current limits). If the margin is insufficient, proceed directly to step three, prioritizing role switching without waiting for the SOC threshold or timed triggering.

[0139] Step 3, Scheduling Decision:

[0140] If SOC supplying ≤SOC low If the preset switching time is reached or the system is in "fast response mode": a switching command is issued; and a preset switching sequence is executed. Among these, SOC... supplying That is, when the SOC of the i-th battery pack that is currently supplying power... i .

[0141] Taking Bat2 switching to charging and Bat1 switching to power supply as an example, the state switching method in step four includes:

[0142] State 1 (Current): Bat1[SW1:ON, SW2:OFF, SW3:OFF], Bat2[SW1:OFF, SW2:ON,SW3:ON].

[0143] State 2 (Switch 1): Bat1 [All OFF], Bat2 [SW1:OFF, SW2:ON, SW3:OFF]. (Bat2 is powered by diode freewheeling).

[0144] State 3 (Switch 2): Bat1[SW1:OFF, SW2:ON, SW3:OFF], Bat2[SW1:OFF, SW2:ON,SW3:OFF]. (Dual diode freewheeling to prevent circulating current).

[0145] State 4 (Switch 3): Bat1 [SW1:OFF, SW2:ON, SW3:OFF], Bat2 [All OFF]. (Bat1 is powered by diode freewheeling).

[0146] State 5 (Switch 4): Bat1 [SW1:OFF, SW2:ON, SW3:ON], Bat2 [All OFF]. (Bat1 is directly powered).

[0147] State 6 (Completed): Bat1 [SW1:OFF, SW2:ON, SW3:ON], Bat2 [SW1:ON, SW2:OFF, SW3:OFF]. (Bat1 is powered, Bat2 is charged).

[0148] Step 5: Update system status.

[0149] The above details the specific execution steps of the state switching, which is the physical realization of "dynamic reconfiguration." The initial state is that battery pack 1 is charging and battery pack 2 is supplying power. Through state switching (during which reverse bias diodes and other devices are used to ensure the continuity of power supply to the load), the system eventually smoothly transitions to a state where battery pack 1 is supplying power and battery pack 2 is charging. Through this periodic and rapid switching of charging and discharging roles, from a macroscopic and continuous perspective, the power grid is charging an equivalent constant power load (i.e., the charging battery pack), thereby shielding the actual load from power surges. At the same time, the load is also supplied by the discharging battery pack, obtaining a stable voltage and thus becoming immune to disturbances on the grid side.

[0150] Example 2: An AI computing center of a large Internet company, a GPU server cluster equipped with the "digital energy buffer device" described in this invention.

[0151] Monitoring objective: To address the sudden surge in instantaneous power generated when GPU servers start training large models with hundreds of billions of parameters, and to achieve stable power supply on the grid side and uninterrupted power supply on the load side.

[0152] Implementation process:

[0153] Step 1: System Initialization and Steady-State Operation

[0154] The digital energy buffer device is activated; it contains two battery packs (Bat 1 and Bat 2). After system initialization, it enters a steady state.

[0155] Bat 1: Connected to the power grid by closing switch S1 in its switch array, and supplied with power by the power grid through a constant power module at a preset power P. charge Charging is performed at 200kW. At this time, Bat 1 is disconnected from the load.

[0156] Bat 2: Connected to the load by closing switches S2 and S3 in its switch array, Bat 2 provides a near-ideal voltage source to the entire GPU cluster. At this time, Bat 2 is disconnected from the power grid.

[0157] Step 2: The digital energy controller continuously monitors the total load power P. load And state parameters such as SOC, voltage, and temperature of each battery pack.

[0158] Event triggered (training task started):

[0159] At 09:00 one day, the cluster scheduling system issued an instruction to start a large-scale distributed training task.

[0160] Thousands of GPU cores are activated within hundreds of milliseconds, with a load power of P.load It began to rise sharply, increasing rapidly from 150kW in standby mode to 550kW.

[0161] Step 3: The core scheduling algorithm inside the digital energy controller detects this positive power transient. Simultaneously, the algorithm monitors that the State of Charge (SOC) of the currently powering Bat 2 has decreased from an initial 90% to 65%, approaching the set switching threshold SOC. low =60%.

[0162] Step 4: Scheduling Decisions and State Switching

[0163] The algorithm core immediately makes a decision: initiate the battery pack role switching process, switch Bat 1 to power supply mode, and switch Bat 2 out and charge it.

[0164] The digital energy controller sends commands to the electronic switch arrays of the two battery packs via the control bus, according to a preset timing logic, to execute seamless switching at the millisecond level.

[0165] Switching to State 1: Disconnect switch S1 of Bat 1 to stop charging. Switch S3 of Bat 2 is open, but its switch S2 remains closed. The load current freewheels through the reverse-biased diode of S2, ensuring uninterrupted power supply.

[0166] Switching to State 2: Close switch S2 of Bat 1, and its current supplies power to the load through the reverse-biased diode. At this time, Bat 1 and Bat 2 supply power to the load through their respective reverse-biased diodes connected in parallel. The presence of the diodes prevents circulating current between the battery packs.

[0167] Switching to state 3: Disconnect switch S2 of Bat 2, completely disconnecting it from the load. At this time, Bat 1 continues to supply power to the load through its reverse-biased diode.

[0168] Switching to state 4: Close switch S3 of Bat 1 to switch it from diode freewheeling mode to low impedance direct power supply mode.

[0169] Completion status: Switch S1 of Bat 2 is closed, its constant power module is activated, and the power grid begins to operate at P... charge =200kW to charge Bat 2. At this time, Bat 1 has become the main power supply unit, and Bat 2 has become the charging unit.

[0170] Effect verification:

[0171] For the load side: Throughout the entire switching process (lasting approximately 10-20 milliseconds), the load (GPU cluster) is always powered by the battery pack (Bat 2 or Bat 1), with highly stable input voltage fluctuations of less than ±1.5%. The GPU training task was not affected by any power supply disturbances and successfully initialized and entered a stable computing state.

[0172] On the grid side: From a macroscopic perspective, the grid is constantly providing constant power (P) to a single battery pack (first Bat 1, then Bat 2). charge =200kW) charging. The power surge of hundreds of kilowatts generated when the GPU cluster starts up is completely absorbed inside the digital energy buffer device. The grid side only perceives a smooth and constant power load, effectively avoiding the malfunction of the upstream power distribution protection or voltage drop that may be caused by a sudden power surge.

[0173] Further optimization (peak-valley arbitrage):

[0174] During off-peak electricity prices at night, the digital energy controller can intelligently adjust P. charge Up to higher values ​​to quickly store energy for the battery pack; during peak daytime electricity price periods, P can be appropriately reduced. charge Alternatively, the battery pack can be used to discharge and support part of the load, thereby achieving peak shaving and valley filling and reducing overall operating costs.

[0175] This embodiment demonstrates that the algorithm and device of this application successfully transform the drastic and unpredictable power dynamics of AI loads into a stable and predictable constant power input on the grid side, perfectly resolving the core power supply contradiction in high-performance computing scenarios.

[0176] In summary, the device and method described in this application fundamentally solve the dynamic mismatch problem between the power supply network and the electrical load. The device achieves efficient decoupling and dynamic scheduling of energy by constructing an "energy buffer layer" based on digital control and multi-battery pack collaboration between the two. Specifically, on the load side, the device provides a highly stable, near-ideal voltage source, capable of responding in real-time to millisecond-level power surges generated by AI accelerators such as GPUs, effectively shielding grid voltage dips, harmonics, and other disturbances, ensuring the reliable operation of precision equipment. On the grid side, the device makes its equivalent load exhibit constant power, completely eliminating current surges and frequency flicker caused by drastic load fluctuations, significantly improving the power quality and reliability of the grid. Simultaneously, its modular and scalable architecture allows for flexible adaptation to diverse scenarios, from industrial equipment to large data centers. In terms of economic benefits, the device supports intelligent peak-valley scheduling strategies, enabling energy storage during off-peak hours and discharge during peak hours, achieving integrated backup and storage applications and significantly reducing users' energy operating costs. This invention provides a systematic solution that combines stability, economy, and intelligence to address the power supply challenges of next-generation high-density, dynamic computing facilities.

[0177] Furthermore, the device also includes a digital energy scheduling method, the scheduling method comprising:

[0178] Deploy a digital energy buffer between the power supply network and the electrical load;

[0179] The digital energy buffer device uses multiple battery packs inside to perform charging and discharging operations in turn.

[0180] The battery packs are controlled such that at any given time, at least one battery pack is in a discharging state that provides a stable voltage source to the electrical load, while at least another battery pack is in a charging state that is being charged by the power supply network.

[0181] By periodically and seamlessly switching the charging and discharging roles of the battery pack, the equivalent load viewed from the power supply network side appears as an approximately constant power load, while the power source viewed from the electrical load side appears as a stable voltage source.

[0182] The switching of the charging and discharging roles of the battery packs is based on a preset switching sequence that includes multiple intermediate states. This switching sequence ensures that during the switching process, the electrical load is always powered by at least one battery pack through a direct connection or a freewheeling circuit. The triggering conditions for the switching sequence include: the state of charge (SOC) of the currently discharging battery pack falling below a first threshold, or reaching a preset switching time point.

[0183] Charging via the power supply network is constant power charging, and the power value of the constant power charging can be set and adjusted according to system configuration or external commands.

[0184] The power value of constant power charging is dynamically adjusted according to the time-of-use electricity price signal to realize the integrated application of peak shaving and valley filling for backup and storage.

[0185] The scheduling method further includes: real-time acquisition of voltage, current, temperature and switching status of each battery pack; based on the acquired data, calculating and issuing control commands through a scheduling algorithm to execute the role switching of the battery pack.

[0186] The external system includes a power supply network, an electrical load, and the aforementioned digital energy buffer device, which is connected in series between the power supply network and the electrical load. The electrical load can be an artificial intelligence (AI) computing cluster, a data center server, or precision industrial equipment.

[0187] The device includes one or more processors;

[0188] Storage device for storing one or more programs;

[0189] When the one or more programs are executed by the one or more processors, the one or more processors implement the digital energy buffer algorithm and / or scheduling method.

[0190] The device includes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the digital energy buffering algorithm and / or scheduling method.

[0191] The method for switching sequences in the algorithm includes:

[0192] Initial state: The first battery pack is charging, and the second battery pack is supplying power to the load;

[0193] First switching state: Disconnect all switches of the first battery pack, disconnect the switch in the second battery pack that is connected in parallel with the freewheeling diode, and allow the load to freewheel through the freewheeling diode;

[0194] Second switching state: Close the switch in the first battery pack that is connected in parallel with the freewheeling diode, so that it supplies power to the load through the diode, while the second battery pack continues to supply power to the load through the diode;

[0195] Third switching state: Disconnect the switch in the second battery pack that is connected in parallel with the freewheeling diode, so that it is completely disconnected from the load, and the first battery pack supplies power to the load through the diode;

[0196] Fourth switching state: Close the remaining switch of the first battery pack to directly supply power to the load;

[0197] Completed status: Close the charging switch of the second battery pack to put it into constant power charging state.

[0198] The device is applied to the scheduling system, the system comprising:

[0199] The status monitoring module is used to collect real-time status data of the power supply network, electrical load, and internal battery packs and switches of the device;

[0200] The scheduling decision module is used to determine whether and when to trigger battery pack role switching based on the status data and a preset scheduling algorithm.

[0201] The timing control module is used to generate and execute the preset switching sequence when a switching is triggered, and to send control commands to the electronic switch array.

[0202] The scheduling decision module employs a hybrid triggering strategy based on battery pack SOC threshold and timed triggering. The scheduling decision module also employs an active triggering strategy based on load power change rate prediction.

[0203] The scheduling method or apparatus includes a method for improving the power supply reliability of AI computing clusters, wherein the method is used to shield the upstream power supply network from the power surges caused by GPU server startup / shutdown and task scheduling.

[0204] The scheduling method or apparatus further includes a method for realizing electricity price arbitrage, wherein the method is used to increase charging power for energy storage during periods of low electricity prices and reduce charging power or utilize battery pack discharge during periods of high electricity prices, so as to reduce the overall cost of electricity.

[0205] The scheduling method includes a battery pack health state balancing method, which dynamically adjusts the residence time of each battery pack in the charging and discharging states during the operation of the scheduling method, so that the cumulative charge and discharge amounts of each battery pack tend to be consistent, thereby achieving passive balancing.

[0206] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A digital energy buffering algorithm based on dynamic reconfigurable technology, characterized in that, Includes the following steps: Step 1: System Startup and Initialization: After starting the system within the device, perform system initialization operations; Step 2, Status Monitoring and Data Acquisition: The status of the system within the device is sensed through algorithms, and data acquisition, system status monitoring, load power calculation, and smoothing are performed on the system. Step 3, Energy Resource Scheduling Decisions: Achieving supply and demand balance through dynamic reconfiguration; Step 4, Resource Scheduling and State Switching: Dynamic switching algorithm for dual-port battery modules within the device; Step 5: System Status Update: Return to Step 2 to update the system status.

2. The digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 1, characterized in that, Step one includes: Step S11: Power on the system, start the internal system of the device, and perform self-test. Step S12: Initialize the system, preset relevant parameters and load power change threshold dP. threshold Objective function J and related constraints; related parameters include , , Any one or more parameters in it.

3. The digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 2, characterized in that, The objective function J can be expressed as: ； To provide a constant power supply to the desired power grid; It can be set by the system or dynamically adjusted according to peak and off-peak electricity prices; This represents the actual power output of the power grid. This represents the average SOC of all battery packs. ω1, ω2, and ω3 are the load demand response delays; ω1, ω2, and ω3 are all weighting coefficients.

4. The digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 1, characterized in that, In step two, data acquisition and system status monitoring refer to the algorithm periodically acquiring the state variables S(k) of different battery packs via the data bus; the formulas and models used for load power calculation and smoothing include instantaneous load power P. load (t) Calculation formula and digital filtering formula.

5. A digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 4, characterized in that, The digital filtering formula is P load_filtered (k)=α·P load (k)+(1-α)I load_filtered (k).

6. The digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 5, characterized in that, Step two includes: Step S21: Real-time monitoring of state data acquisition S(k) and calculation of P load_filtered (k); Step S22: Health and safety judgment: Check whether all battery parameters are within the constraints. If they exceed the limits, trigger the protection logic. Step S23: Quick Response Judgment: Real-time Calculation of Load Power Change Rate Determine whether If yes, then mark the system as entering "fast response mode"; proceed to step S24; Step S24: Based on SOC, temperature, and current limits, immediately assess the output margin of the current power supply battery pack; determine if the output margin is sufficient. If not, the margin is insufficient, and proceed to step three; prioritize triggering role switching without waiting for the SOC threshold or timed triggering.

7. A digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 6, characterized in that, In step three, achieving supply and demand balance through dynamic reconstruction is the core decision-making process of the algorithm, which is used to determine how to allocate battery packs. Step three, dynamic restructuring to achieve supply and demand balance, includes: Step S31: Role definition and state machine: Each battery pack is defined as a state machine, and each state machine includes three states: charging state, power supply state, and switching / standby state. Step S32: Dynamic Reconfiguration Strategy: The system is dynamically reconfigured by executing relevant scheduling logic in each decision cycle.

8. A digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 7, characterized in that, In step S32, during the battery pack role switching process, a temporary freewheeling diode connected in parallel with at least two electronic switches in the discharge circuit is used to form a temporary freewheeling path to ensure zero interruption of load power supply; the time-of-use electricity price signal and / or the task scheduling information of the AI ​​computing cluster are acquired in real time or at regular intervals. Based on the acquired time-of-use electricity price signal and / or the task scheduling information of the AI ​​computing cluster, the constant power charging power value of the battery pack in the charging state is dynamically adjusted. When a power surge is predicted for the AI ​​computing cluster, the charging power of the current battery pack is reduced or suspended in advance; when a power drop is predicted for the AI ​​computing cluster, the charging power of the current battery pack is reduced or even stopped in advance to absorb excess energy. This ensures that the power supply network always sees an equivalent load with approximately constant power, while the load side obtains a highly stable voltage source, enabling electricity price arbitrage and proactive buffering against transient power changes in the AI ​​computing load.

9. A digital energy buffering algorithm based on dynamic reconfigurable technology according to claim 8, characterized in that, The task scheduling information of the AI ​​computing cluster is directly pushed through the power consumption prediction model or job scheduler of the GPU server; the constant power charging power value is increased to the rated maximum value when the electricity price is low, and decreased to zero or negative value when the electricity price is high.

10. A digital energy buffer device based on dynamically reconfigurable technology, characterized in that, The device comprises a digital energy buffering algorithm based on dynamic reconfigurable technology according to any one of claims 1-9, wherein the device is deployed as an intermediate system between the power supply network and the electrical load; the device adopts a hierarchical modular design; the device also includes a supercapacitor module connected in parallel with the DC bus; Digital energy controllers use FPGAs or DSPs to achieve millisecond-level control.

Images