Flywheel array-based power surge mitigation method, apparatus, and electronic device

Abstract

The application provides a flywheel array-based power impact suppression method, device and electronic equipment, and particularly relates to the field of power electronics and energy storage control. In view of the problems of limited power expansion of an existing flywheel system, unbalanced array power distribution and difficulty in rapid convergence of total output under a megawatt-level burst power impact of an artificial intelligence data center, the application proposes: dividing the flywheel array into hot backup and cold backup according to the operating state, and dynamically adjusting the two groups of configurations after predicting the impact; when the impact arrives, combining the actual rotating speed state of the hot backup flywheel to distribute the transient power in layers, and continuously optimizing the output of each machine, so that the total output of the array approaches the target power within a preset time. The scheme can improve the response speed, balance the load of each flywheel, reduce the overload and circulating current loss, and improve the power supply stability.

Description

Technical Field

[0001] This application relates to the field of power electronics and energy storage control, and in particular to a power surge mitigation method, apparatus and electronic equipment based on a flywheel array. Background Technology

[0002] In high-density computing load scenarios in artificial intelligence data centers, flywheel energy storage systems are typically used to buffer and smooth out instantaneous power fluctuations.

[0003] However, when faced with sudden power surges at the megawatt level, the existing single-unit or small-scale parallel flywheel systems have limited power expansion capabilities; after array application, power distribution imbalance is prone to occur due to differences in unit speed and capacity, resulting in increased circulating current loss, insufficient dynamic response, and difficulty in balancing long-term standby energy consumption and system reliability.

[0004] Therefore, improving the rapid response capability, power distribution balance, and long-term operational economy of flywheel energy storage systems under megawatt-level sudden power surge scenarios has become a technical problem that needs to be solved. Summary of the Invention

[0005] This application provides a power surge mitigation method, apparatus, and electronic device based on a flywheel array to address the aforementioned technical problems. This method targets megawatt-level sudden power surge scenarios in artificial intelligence data centers. By combining the operating states of different flywheels in the flywheel array with the surge power demand, it dynamically organizes and coordinates the available adjustable resources within the array. This enables the flywheel array to generate output capacity adapted to the target surge power both before and after the power surge, while also considering the response coordination of each flywheel within the array and the overall operational economy, thereby improving the flywheel energy storage system's ability to mitigate sudden power fluctuations.

[0006] In a first aspect, this application provides a power surge mitigation method based on a flywheel array, the method comprising:

[0007] Based on the operating status of each flywheel, the flywheel array is divided into a hot standby group operating at rated speed and a cold standby group operating at standby speed.

[0008] When an AIDC power surge is predicted, the target number of flywheels in the hot standby group is determined based on the matching relationship between the predicted surge power and the rated power of the flywheel, and the ratio of the number of hot standby groups to the number of cold standby groups is adjusted according to the target number of flywheels.

[0009] At the moment the power surge arrives, the transient surge power borne by each flywheel is differentially allocated according to the ratio of the real-time rotational speed to the rated rotational speed of the flywheel in the hot standby group. The hot standby group includes the original hot standby flywheel that has been maintained at the rated rotational speed and the activated flywheel that has been activated from the cold standby group. The real-time rotational speed of the activated flywheel is lower than the real-time rotational speed of the original hot standby flywheel.

[0010] The power distribution of each flywheel in the hot standby group is iteratively optimized so that the total output power of the hot standby group converges to the target value corresponding to the predicted impact power within a preset time.

[0011] The iterative optimization of the power distribution of each flywheel in the hot standby group includes:

[0012] For each flywheel in the hot standby group, the actual output power of the flywheel at the instant of the impact is obtained, and the actual output power is used as the initial value for iteration.

[0013] Based on the current operating state of the flywheel, determine the impact power that the flywheel can withstand at the current moment;

[0014] The current power distribution ratio of the flywheel is determined based on the ratio of the impact power that can be borne to the sum of the impact power that can be borne by all flywheels in the hot standby group.

[0015] Adjust the power distribution of each flywheel according to the current power distribution ratio of each flywheel in the hot standby group;

[0016] Repeat the steps described above to determine the impact power that can be borne, determine the current power allocation ratio, and adjust the power allocation until the deviation between the total output power of the hot standby group and the predicted impact power is within the allowable error range.

[0017] In one possible implementation, determining the target number of flywheels in the hot standby group based on the matching relationship between the predicted impact power and the rated power of the flywheel, and adjusting the quantity ratio of the hot standby group to the cold standby group according to the target number of flywheels, includes:

[0018] The number of candidates is calculated based on the predicted impact power and the rated power of the flywheel.

[0019] The desired number is determined based on the number of candidates and the preset number of protections;

[0020] The desired number is taken as the target number of flywheels;

[0021] Adjust the ratio of the hot standby group to the cold standby group according to the desired quantity.

[0022] In one possible implementation, after iteratively optimizing the power distribution of each flywheel in the hot standby group, the method further includes:

[0023] For each flywheel in the hot standby group, upon determining the end of the power impact, the current rotational speed of the flywheel and the expected group to which the flywheel belongs after the impact are obtained;

[0024] The target charging speed of the flywheel is determined based on the expected group, and the speed difference is calculated based on the target charging speed and the current speed.

[0025] Using preset rules, the charging priority is determined based on the expected group affiliation of each flywheel in the hot standby group and the corresponding speed difference.

[0026] According to the charging priority, the flywheels that need to be charged in the hot standby group are divided into multiple charging batches, and the corresponding flywheels are recharged sequentially based on the multiple charging batches.

[0027] After the backflush is completed, the flywheel that is expected to belong to the cold standby group is adjusted to the cold standby group.

[0028] In one possible implementation, the preset rules include:

[0029] The charging priority of the flywheel that is expected to be assigned to the hot standby group is higher than that of the flywheel that is expected to be assigned to the cold standby group;

[0030] For multiple flywheels in the same expected group, the charging priority is determined in descending order of the speed difference.

[0031] In one possible implementation, determining the expected group of the flywheel after the impact includes:

[0032] Obtain the target number of the hot standby group under normal operating conditions;

[0033] The current temperature rise of each flywheel in the hot standby group at the end of the power surge is obtained, and the state of each flywheel is sorted according to the current speed and the current temperature rise.

[0034] According to the target quantity, select the corresponding number of flywheels from the state sorting results, and determine the expected group to which the selected flywheels belong as the hot standby group;

[0035] The expected group to which the flywheels that were not selected in the hot standby group belong is determined as the cold standby group.

[0036] In one possible implementation, before adjusting the ratio of the hot standby group to the cold standby group according to the target number of flywheels, the method further includes:

[0037] Acquire multidimensional feature data, including the historical power load data of the AIDC, task start instructions in the task scheduling log, and the video memory usage of the GPU cluster;

[0038] The multidimensional feature data is input into a trained multidimensional feature fusion model, which outputs the predicted impact power.

[0039] In one possible implementation, the method further includes:

[0040] Under steady-state operating conditions, obtain the rated capacity of each flywheel in the flywheel array;

[0041] Based on the rated capacity of each flywheel in the flywheel array, the power distribution ratio of each flywheel in steady state is determined, wherein the power distribution ratio of each flywheel is directly proportional to its rated capacity.

[0042] Secondly, this application provides a power surge mitigation device based on a flywheel array, comprising:

[0043] The partitioning module is used to divide the flywheel array into a hot standby group operating at rated speed and a cold standby group operating at standby speed, based on the operating status of each flywheel.

[0044] The adjustment module is used to determine the target number of flywheels in the hot standby group based on the matching relationship between the predicted impact power and the rated power of the flywheel when an AIDC power surge is predicted, and to adjust the ratio of the number of hot standby groups to the number of cold standby groups based on the predicted impact power according to the target number of flywheels.

[0045] The control module is used to differentiate the transient impact power borne by each flywheel according to the ratio of the real-time speed to the rated speed of the flywheel in the hot standby group at the moment the power impact arrives. The hot standby group includes the original hot standby flywheel that has been maintained at the rated speed and the activated flywheel that has been activated from the cold standby group. The real-time speed of the activated flywheel is lower than the real-time speed of the original hot standby flywheel.

[0046] The control module is also used to iteratively optimize the power distribution of each flywheel in the hot standby group so that the total output power of the hot standby group converges to the target value corresponding to the predicted impact power within a preset time.

[0047] The iterative optimization of the power distribution of each flywheel in the hot standby group includes:

[0048] For each flywheel in the hot standby group, the actual output power of the flywheel at the instant of the impact is obtained, and the actual output power is used as the initial value for iteration.

[0049] Based on the current operating state of the flywheel, determine the impact power that the flywheel can withstand at the current moment;

[0050] The current power distribution ratio of the flywheel is determined based on the ratio of the impact power that can be borne to the sum of the impact power that can be borne by all flywheels in the hot standby group.

[0051] Adjust the power distribution of each flywheel according to the current power distribution ratio of each flywheel in the hot standby group;

[0052] Repeat the steps described above to determine the impact power that can be borne, determine the current power allocation ratio, and adjust the power allocation until the deviation between the total output power of the hot standby group and the predicted impact power is within the allowable error range.

[0053] Thirdly, this application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0054] The memory stores computer-executed instructions;

[0055] The processor executes computer execution instructions stored in the memory to implement the power surge mitigation method based on the flywheel array as described in the first aspect.

[0056] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a computer, are used to implement the power surge mitigation method based on a flywheel array as described in the first aspect.

[0057] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps of the scheme recommendation method as described in any of the first aspects.

[0058] This application provides a power surge mitigation method, apparatus, and electronic device based on a flywheel array. By dividing the flywheel array into a hot standby group operating at rated speed and a cold standby group operating at standby speed according to the operating state of each flywheel, and adjusting the ratio of the hot standby group to the cold standby group based on the predicted surge power when an AIDC power surge is predicted, the flywheel array can pre-form a rapid response capability that matches the surge demand. By differentially allocating the transient surge power borne by each flywheel based on the ratio of the real-time speed to the rated speed of the flywheels in the hot standby group at the moment the power surge arrives, and iteratively optimizing the power allocation of each flywheel in the hot standby group to converge the total output power of the hot standby group to the target value corresponding to the predicted surge power within a preset time, the dynamic response speed and power allocation balance under megawatt-level sudden power surges can be improved, thereby reducing the circulating current loss and insufficient response caused by differences in unit states, while also taking into account the long-term standby power consumption and operational reliability of the system. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] Figure 1 A flowchart illustrating a power surge mitigation method based on a flywheel array, provided in this application embodiment. Figure 1 ;

[0061] Figure 2 This application provides a schematic diagram of the architecture of a flywheel array system.

[0062] Figure 3 A flowchart illustrating a power surge mitigation method based on a flywheel array, provided in this application embodiment. Figure 2 ;

[0063] Figure 4 A flowchart illustrating a power surge mitigation method based on a flywheel array, provided in this application embodiment. Figure 3 ;

[0064] Figure 5 A schematic diagram illustrating the switching between steady-state flow equalization mode and impact response mode provided in an embodiment of this application;

[0065] Figure 6 A schematic diagram of a power surge mitigation device based on a flywheel array provided in this application embodiment;

[0066] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0067] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0068] Flywheel energy storage and data center power regulation technology is particularly relevant to instantaneous power buffering scenarios under high-density computing loads in artificial intelligence data centers. In this scenario, GPU clusters, power supply buses, and parallel energy storage units operate in concert. When training tasks start, phases switch, or concurrent computing occurs, the system experiences significant megawatt-level power surges.

[0069] Existing solutions typically employ a single flywheel or a small-scale parallel flywheel system, using variations in flywheel speed to release and absorb energy, thereby compensating for short-term power fluctuations. In typical peak shaving or general load fluctuation scenarios, this type of solution can provide a certain degree of rapid adjustment capability.

[0070] However, in the scenario of megawatt-level sudden power surges in artificial intelligence data centers, the power expansion capability of single machines or small-scale systems is limited. After arraying, due to the differences in the speed and capacity of each flywheel, power distribution imbalance is prone to occur, causing some flywheels to be overloaded while others are under-output. At the same time, if the power load cannot be quickly adjusted according to the actual state of each flywheel when the surge occurs, the total output power often fails to converge to the target value in time, which in turn leads to problems such as increased circulating current loss, dynamic response lag, and decreased power supply stability.

[0071] In view of this, how to balance rapid response, balanced power distribution, and output convergence capability under megawatt-level sudden power surges has become an urgent technical problem to be solved.

[0072] To address the aforementioned issues, this application provides a power surge mitigation method based on flywheel arrays. This method divides the flywheel array into hot standby and cold standby groups, adjusting the ratio between the two groups when an AIDC power surge is predicted. At the instant the power surge occurs, the transient surge power is distributed differentially based on the ratio of the real-time rotational speed to the rated rotational speed of the flywheels in the hot standby group. The power distribution among the flywheels in the hot standby group is then iteratively optimized to converge the total output power to the target value within a preset time. This method is applicable to data center power supply architectures consisting of GPU clusters, a common DC bus, and parallel flywheel arrays.

[0073] Figure 1 A flowchart illustrating a power surge mitigation method based on a flywheel array, provided in this application embodiment. Figure 1 .like Figure 1 As shown, the method includes:

[0074] S101: Based on the operating status of each flywheel, the flywheel array is divided into a hot standby group operating at rated speed and a cold standby group operating at standby speed.

[0075] In this step, the flywheel array is arranged on the common DC bus side. Each flywheel is connected to the bus in parallel via its own bidirectional power conversion unit, and the array controller uniformly receives status variables and issues power commands. The operating status includes at least the flywheel's real-time speed, temperature, state of charge, and the availability status of the power conversion unit. The real-time speed reflects the flywheel's current releaseable kinetic energy level; the rated speed represents the reference speed at which the flywheel can maintain stability and participate in rapid power regulation according to its design; and the standby speed represents the speed range maintained by the flywheel in a low-loss standby state. The array controller determines and groups the flywheels based on their operating status, assigning flywheels operating at rated speed to the hot standby group and flywheels operating at standby speed to the cold standby group.

[0076] After grouping is completed, the array controller generates a hot standby group list and a cold standby group list, and records the group and related operating status information of each flywheel as input for subsequent quantity ratio adjustment and power distribution when impact arrives.

[0077] Based on the above analysis, this step enables the flywheel array to form two sets of resources before the impact arrives: fast response resources at rated speed and backup resources at standby speed. This provides a state basis for reconstructing the number of flywheels that can be immediately deployed according to the impact prediction results, and enables the subsequent allocation to distinguish between the original hot standby flywheels and the flywheels that have been activated from cold standby to hot standby.

[0078] Figure 2 This is a schematic diagram of the architecture of a flywheel array system provided in an embodiment of this application. Figure 2 As shown, the system employs a topology of 10 flywheel energy storage units (FW1-FW10) connected in parallel via a common DC bus. Each flywheel has a rated power of 1.5MW and a rated speed of 36,000rpm, resulting in a total system capacity of 15MW, sufficient to meet power surge demands of up to 12MW while maintaining a 20% safety margin. The choice of parallel DC bus connection over AC bus connection is because it fundamentally eliminates phase synchronization issues between flywheels, reduces circulating current losses, and facilitates independent power control and fault isolation for each unit via a bidirectional DC / DC converter configured on each flywheel.

[0079] In terms of control architecture, this system adopts a hierarchical distributed design: the top layer is a centralized controller, responsible for the dynamic grouping modeling and decision-making of the hot standby group (FW1-FW4, rated speed 36000rpm) and the cold standby group (FW5-FW10, standby speed 3000rpm). The bottom layer, each flywheel, is equipped with an independent local controller and a bidirectional DC / DC converter, responsible for executing power commands, speed regulation, and fault protection. The centralized controller is connected to each local controller through the communication network shown by the dotted line, ensuring real-time transmission of control commands.

[0080] S102: When an AIDC power surge is predicted, adjust the ratio of hot standby groups to cold standby groups based on the predicted surge power.

[0081] In this step, AIDC (Artificial Intelligence Data Center) power surge refers to a megawatt-level power spike or drop event that occurs within a short period of time on the common DC bus during task initiation, training phase switching, concurrent inference, or batch parameter synchronization in an artificial intelligence data center. AIDC power surge prediction can be completed collaboratively by the upper-level energy management platform, task scheduling platform, and array controller. The array controller can perform short-term prediction of the surge based on load history, scheduling information, and bus operation information to obtain the corresponding predicted surge power.

[0082] In one possible implementation, considering that power surges in AIDC are influenced by multiple factors such as task scheduling, computing load, and historical power consumption patterns, single-dimensional load data is insufficient to accurately reflect the timing and magnitude of the surge. Multi-dimensional feature fusion prediction methods can comprehensively utilize operational information from different sources to capture precursory features of power changes, resulting in higher prediction accuracy and timeliness. Therefore, to obtain reliable target power values ​​before the surge occurs, thus providing accurate input for optimizing the ratio of hot and cold standby groups, this embodiment introduces a prediction mechanism based on a multi-dimensional feature fusion model. The corresponding specific implementation steps are as follows:

[0083] Acquire multidimensional feature data, including historical power load data of AIDC, task start instructions in task scheduling logs, and GPU cluster memory usage; input the multidimensional feature data into a trained multidimensional feature fusion model to output predicted impact power.

[0084] In practice, the system collects historical power load records from the AIDC side, extracts task start instructions from the task scheduling platform, and obtains the GPU (Graphics Processing Unit) memory usage from the GPU monitoring module. Subsequently, the data from different sources undergoes timestamp alignment, outlier removal, and numerical standardization to form unified multidimensional feature data, which is then fed into a trained multidimensional feature fusion model. The model jointly analyzes the periodic changes in power load, the trigger density of task starts, and the continuous upward trend in memory usage, outputting the predicted impact power within the corresponding prediction window, thus providing input for adjusting the ratio of hot and cold standby groups.

[0085] The training of the multi-dimensional feature fusion model is completed offline. The training samples are derived from historical AIDC operation records, and the labels are the maximum power increment values ​​that actually occur within a set prediction time window (e.g., the next 5 or 10 seconds) after the corresponding time. Supervised learning is used during training to minimize the mean square error between the predicted power and the actual impact power. The model can employ a temporal fusion network based on an attention mechanism, or a multi-branch convolutional network to extract local patterns from each feature sequence before feature concatenation and fully connected regression output.

[0086] As one possible implementation, the array controller has pre-trained model parameters embedded inside, and directly calls the inference interface to complete the forward calculation during prediction. As another possible implementation, the array controller can also upload the pre-processed feature data to the edge computing server through the communication interface, and the server runs the model inference and sends the prediction results back to the array controller. The array controller then determines the predicted impact power based on the returned values.

[0087] By adopting the above method, task startup information and GPU load status can be incorporated into the impact power prediction process, so that the prediction results can correspond to the instantaneous power changes in the actual operation of AIDC. This provides a basis for the subsequent configuration of the number of hot standby groups that is closer to the actual load, and keeps the grouping adjustment of the flywheel array consistent with the power impact requirements.

[0088] Furthermore, the array controller adjusts the ratio of hot standby to cold standby groups based on the predicted impact power: if the current number of hot standby groups is lower than the required level for impact response, several flywheels from the cold standby groups are selected and activated with speed-up commands; if the current number of hot standby groups is higher than the required level for impact response, some flywheels are switched to cold standby operation. During the speed-up process, the array controller sends target speed-related commands to the activated flywheels, causing them to climb from standby speed to rated speed, and continuously monitors their speed tracking. When the predicted impact power increases, the number of hot standby groups increases accordingly; when the predicted impact power decreases, the number of hot standby groups decreases accordingly, so that the number of flywheels at or near rated speed matches the upcoming impact scale. After the above steps are completed, the system obtains an updated hot standby pool containing the original hot standby flywheels and the activated flywheels that have entered the speed-up process.

[0089] Based on the above analysis, this step converts the impact prediction results into hot standby resource scale adjustment actions, so that the number of flywheels that can respond immediately changes with the predicted power, reducing the total output gap of the array caused by insufficient hot standby resources under megawatt-level impact.

[0090] In another possible implementation, considering that power prediction inevitably has a certain deviation, and that the flywheel array may not be able to output full power during operation due to individual flywheel failures or performance degradation, if the hot standby group is configured only according to the precise calculated value of the predicted power, then when the actual impact power is higher than the predicted value or a flywheel in the hot standby group suddenly malfunctions, there will be a risk of insufficient response capacity, leading to a drop in bus voltage or a delay in power support. The preset protection quantity, as additional hot standby redundancy capacity, can withstand impact power exceeding expectations under the aforementioned non-ideal operating conditions, effectively improving the system's robustness in dealing with uncertainties. Therefore, in order to achieve a reasonable configuration of hot standby capacity while ensuring response reliability, this embodiment introduces a ratio adjustment mechanism based on the candidate quantity plus the preset protection quantity to determine the desired quantity. The corresponding specific implementation steps are as follows:

[0091] The candidate number is calculated based on the predicted impact power and the rated power of the flywheel; the expected number is determined based on the candidate number and the preset number of protections; and the ratio of the number of hot standby groups to cold standby groups is adjusted according to the expected number.

[0092] In practical implementation, the controller receives the predicted impact power output from the power prediction module and reads the rated power parameters of individual flywheels in the flywheel array. The number of candidate flywheels is obtained by dividing the predicted impact power by the rated power of each flywheel. If the calculation result contains a decimal, the smallest integer not less than this result is taken as the number of flywheels required to meet the impact response requirements. For example, if the predicted impact power is 11MW and the rated power of a single flywheel is 1.5MW, the calculated value is 7.33, so the number of candidate flywheels is 8 to ensure that the theoretical total rated power is not less than the predicted impact requirement.

[0093] The desired number of protections is then obtained by adding a preset protection number to the candidate number. This preset protection number can be pre-written into the controller's memory, for example, fixed at 1 or 2 units, suitable for scenarios where the system operating conditions are relatively stable and the flywheel failure rate is known and does not change significantly. Alternatively, it can be configured based on the current flywheel availability, historical response deviation, and target impact level. For example, the array controller periodically collects the self-test status of each flywheel (including bearing temperature, motor winding insulation resistance, speed tracking error, etc.) and calculates the ratio of the total number of flywheels currently in a healthy and usable state to the total configured number as the availability rate. Simultaneously, the array controller records the absolute values ​​of the deviations between the actual output power and the predicted value in the most recent impact responses and calculates their average as the historical response deviation. Combined with the current impact level (e.g., a larger impact level corresponds to a larger redundancy coefficient), the dynamic protection number is calculated according to a preset mapping function (e.g., a weighted sum of the reciprocal of the availability rate and the deviation coefficient, then rounded down). After determining the preset protection number, the array controller adds the candidate number to the preset protection number to obtain the desired number. For example, if the candidate number is 8 units and the preset protection number is 1 unit, the desired number is 9 units.

[0094] The controller then adjusts the ratio of hot standby groups to cold standby groups according to the desired number, transferring some flywheels that are running at standby speed to the hot standby group and correspondingly reducing the number of flywheels in the cold standby group, so that the size of the hot standby group reaches the desired number.

[0095] In this allocation adjustment, the flywheels in the hot standby group maintain their rated speed and are ready to output power immediately, while the flywheels in the cold standby group maintain their standby speed to reduce standby losses. When the desired quantity increases, the controller sends an acceleration command to the cold standby group, causing the corresponding flywheels to enter the hot standby group; when the desired quantity decreases, the excess flywheels are returned to the cold standby group to maintain a quantity balance between the hot and cold standby groups. This adjustment process is executed in conjunction with the power prediction results, enabling the hot standby group to have a response scale matching the predicted impact power before the impact arrives, and providing a basis for subsequent power distribution and convergence control.

[0096] After adopting the above-mentioned quantity allocation adjustment method, the configuration scale of the hot standby group can change synchronously with the predicted impact power, and the prediction deviation can be compensated by the preset protection quantity, so that the hot standby resources and the impact demand can be matched, thereby making it easier for the subsequent transient power allocation to converge to the target value.

[0097] S103: At the moment the power impact arrives, the transient impact power borne by each flywheel is differentiated according to the ratio of the real-time speed of the flywheel in the hot standby group to the rated speed.

[0098] The hot standby group includes the original hot standby flywheel, which has been maintained at its rated speed, and the activated flywheel, which is activated from the cold standby group to increase its speed. The real-time speed of the activated flywheel is lower than that of the original hot standby flywheel.

[0099] In this step, the moment the power surge arrives can be understood as the moment when the common DC bus detects a sudden change in load power exceeding the trigger threshold, or within a pre-set extremely short time window after the upper-level scheduling event is triggered. At this moment, the array controller no longer waits for all active flywheels to fully reach their rated speed, but directly performs transient power distribution based on the ratio of the real-time speed to the rated speed of each flywheel in the hot standby group, thereby forming the total output in the shortest possible time.

[0100] Since the energy released by a flywheel is related to its rotational speed, and the original hot-standby flywheel has maintained its rated speed while the active flywheel is still accelerating, the ratio of the real-time rotational speed of the original hot-standby flywheel to its rated speed is usually greater than that of the active flywheel. The array controller performs differentiated power allocation based on the ratio of each flywheel's real-time rotational speed to its rated speed and its current output capacity. This allows the original hot-standby flywheel to bear a higher proportion of transient impact power, while the active flywheel bears a smaller share of transient power that matches its current state.

[0101] For example, in a hot standby group comprising multiple original hot standby flywheels and active flywheels, the array controller allocates higher output to flywheels whose real-time speed is closer to their rated speed, and lower output to flywheels still in the acceleration process, allowing them to further increase their load as the speed continues to rise. Allocation commands are sent to each bidirectional power conversion unit via a high-speed communication bus, and the bidirectional power conversion unit injects or absorbs power into or from the common DC bus according to a given power reference value.

[0102] Based on the above analysis, this step directly uses the real-time speed ratio to differentiate the original hot standby flywheel and the activated flywheel, so that the array can output power consistent with the actual capacity of each unit when the impact arrives, reducing the local overload and total output mismatch caused by the low-speed flywheel being allocated too much power.

[0103] S104: Iteratively optimize the power distribution of each flywheel in the hot standby group to bring the total output power of the hot standby group to the target value corresponding to the predicted impact power within a preset time.

[0104] In this step, after the initial allocation at the moment of impact, the real-time rotational speed of each flywheel is still changing. The activated flywheel continues to accelerate, while the original hot standby flywheel gradually decreases in speed after releasing power. Therefore, it is difficult to guarantee that the total output power will always match the target value in the subsequent short process based solely on the transient initial allocation. To address this, the array controller performs iterative optimization of the power allocation of each flywheel in the hot standby group within a preset time.

[0105] In each iteration, the controller acquires the current operating status, current output, and total output power of each flywheel in the hot standby group, and determines the current deviation based on the target value corresponding to the predicted impact power. When the deviation exceeds the allowable range, the power allocation is updated according to the current status of each flywheel, so that the flywheel with stronger current response capability bears more power, and the flywheel with weaker current response capability bears less power; as the speed of the activated flywheel increases, the power it bears can increase accordingly.

[0106] Through continuous iteration, the total output power of the hot standby group gradually approaches and converges to the target value corresponding to the predicted impact power within a preset time; when the total output power enters the allowable error range near the target value, the corresponding power allocation result is maintained.

[0107] Based on the above analysis, this step, by continuously correcting the power distribution between the original hot standby flywheel and the activated flywheel, gradually approaches and stabilizes the total output near the target value from the initial rapid response state. Even with inconsistent flywheel speeds, the total array power can still converge. The overall implementation principle is as follows: first, hot standby and cold standby are layered according to the flywheel operating status; then, the number of flywheels that can be immediately put into operation is adjusted based on the impact prediction results; when the impact arrives, a differentiated initial allocation is performed using the ratio of real-time speed to rated speed; and finally, iterative optimization is used to eliminate the total output deviation.

[0108] In one possible implementation, steady-state operation refers to the working stage where the system does not face power surges and the load demand is relatively stable. During this stage, the flywheel array typically undertakes the task of smoothing power fluctuations on the grid side or load side, rather than dealing with instantaneous high-power surges. If the same power command or uniform proportional allocation method is applied to flywheels with different rated capacities during steady-state operation, flywheels with smaller rated capacities may operate at higher load rates for extended periods, while flywheels with larger rated capacities may remain at lower load rates. Smaller-capacity flywheels will age faster due to frequent exposure to high power output, while the redundancy of larger-capacity flywheels will not be fully utilized. Allocating steady-state power proportionally to the rated capacity of each flywheel ensures that the power share borne by each flywheel matches its own capacity, achieving load balancing and preventing premature degradation of some flywheels due to long-term overload. Therefore, to ensure that the power output of each flywheel matches its rated capacity under steady-state operating conditions and achieves load and lifespan balancing of the flywheel array during long-term operation, this embodiment introduces a steady-state power allocation mechanism based on rated capacity proportionality. The specific implementation steps are as follows:

[0109] Under steady-state operating conditions, the rated capacity of each flywheel in the flywheel array is obtained; based on the rated capacity of each flywheel in the flywheel array, the power distribution ratio of each flywheel under steady-state conditions is determined, wherein the power distribution ratio of each flywheel is directly proportional to its rated capacity.

[0110] In actual operation, after identifying that the system has entered a steady state, the controller first obtains the rated capacity of each flywheel in the flywheel array, then sums the rated capacities and uses this sum as the total capacity benchmark. Subsequently, the ratio of each flywheel's rated capacity to the total capacity is used as the steady-state power allocation ratio for the corresponding flywheel. In this way, steady-state power can be matched and distributed according to the flywheel's capacity, ensuring that each flywheel's share of power during long-term operation remains consistent with its rated capacity. If the flywheel model, individual capacity, or number of parallel connections changes, the controller can still recalculate the allocation ratio based on the same normalization rule. In practical applications, this capacity information can also be obtained through communication reading, manual input, or configuration distribution; this application does not limit this method.

[0111] Under this scheme, the power distribution during steady-state operation is directly correlated with the rated capacity, enabling flywheels of different capacities in the array to participate in steady-state power supply according to their respective load-bearing capabilities, thereby maintaining consistency in power sharing. Since the distribution ratio is determined by the rated capacity, the controller can be configured using the same capacity benchmark before and after system state switching, thus facilitating continuous scheduling of steady-state power and balanced operation of each flywheel in the array.

[0112] It should be noted that in this embodiment and subsequent embodiments, the flywheel speed can be acquired through sensors. However, when some flywheels are not equipped with speed sensors or the speed sensors are faulty, the array controller can use a distributed extended state observer to estimate the real-time speed of the flywheels and load disturbances without speed measurement. In specific implementation, the array controller pre-establishes a simplified dynamic model of the motor system for each flywheel, using known parameters such as the flywheel's moment of inertia, electromagnetic torque coefficient, and friction coefficient as the model reference. The array controller obtains the power information of each flywheel from the common DC bus voltage and current measurements, and combines it with the current loop feedback signal uploaded by the local controller of each flywheel as the input of the observer. The distributed extended state observer treats each flywheel as an interconnected but independently observable subsystem. In each control cycle, the array controller calculates the acceleration estimate based on the difference between the input power and output power of each flywheel, and then performs numerical integration on the acceleration estimate to obtain the speed estimate. At the same time, the observer treats the parts not accurately modeled in the model (including load abrupt changes, friction torque changes, and parameter perturbations) as a total disturbance and performs real-time online estimation through extended state variables.

[0113] The observer operates without relying on physical speed sensors mounted on the flywheel shaft. It achieves a speed estimate that approximates the true speed within a preset convergence time solely through electrical quantity measurements and system model recursion. In the distributed architecture, the array controller and each flywheel's local controller exchange adjacent flywheel status information via a communication network, enabling collaborative estimation among the observers. If the flywheel is also equipped with a speed sensor, the array controller can compare the sensor's measured speed with the observer's estimated speed. If the deviation exceeds a preset threshold, it determines that the sensor or observer is malfunctioning, thus enabling self-diagnosis. This speed measurement-free estimation method reduces the system's dependence on physical sensors while ensuring the availability of speed information, improving the flywheel array's fault-tolerant operation under sensor failure conditions.

[0114] This application provides a power surge mitigation method based on a flywheel array. The method includes dividing the flywheel array into a hot standby group operating at rated speed and a cold standby group operating at standby speed, based on the operating state of each flywheel. When an AIDC power surge is predicted, the ratio of the number of hot and cold standby groups is adjusted according to the predicted surge power. At the instant the power surge arrives, the transient surge power borne by each flywheel is differentially allocated based on the ratio of the real-time speed to the rated speed of the flywheels in the hot standby group. The hot standby group includes original hot standby flywheels maintained at rated speed and activated flywheels activated from the cold standby group, with the real-time speed of the activated flywheels being lower than that of the original hot standby flywheels. The power allocation of each flywheel in the hot standby group is iteratively optimized to converge the total output power of the hot standby group to the target value corresponding to the predicted surge power within a preset time. In this application, by combining grouping, predictive reconstruction, transient allocation based on speed ratio, and iterative optimization, the flywheel array is able to bear the output power according to the actual available state of each flywheel under an AIDC megawatt-level power surge, and the total output converges to the target value within a preset time. Furthermore, a prediction mechanism based on a multi-dimensional feature fusion model is introduced, incorporating task startup information and GPU load status into the impact power prediction process. This ensures that the prediction results correspond to instantaneous power changes during actual AIDC operation, providing a more realistic basis for configuring the number of hot standby groups and aligning the flywheel array grouping adjustments with power impact demands. Simultaneously, a ratio adjustment mechanism is introduced to determine the desired number based on a combination of candidate numbers and preset protection numbers. The configuration scale of the hot standby groups can change synchronously with the predicted impact power, and the preset protection numbers compensate for prediction deviations, maintaining a match between hot standby resources and impact demands. This makes subsequent transient power allocation more likely to converge to the target value. It should be understood that the above examples are for illustrative purposes only and are not limiting.

[0115] Figure 3 A flowchart illustrating a power surge mitigation method based on a flywheel array, provided in this application embodiment. Figure 2 This embodiment provides a detailed explanation of the steps involved in iteratively optimizing the power distribution of each flywheel in the hot standby group. For example... Figure 3 As shown, the method includes:

[0116] S301. For each flywheel in the hot standby group, obtain the actual output power of the flywheel at the moment of impact, and use the actual output power as the initial value for iteration.

[0117] In this step, the flywheels in the hot standby group include the original hot standby flywheels that maintain their rated speed over a long period and the activated flywheels that are activated from the cold standby group and accelerate before the impact. The actual speeds of these two types of flywheels differ at the moment of impact. The activated flywheels typically haven't fully reached their rated speed, and their instantaneous output capacity is lower than that of the original hot standby flywheels. If all hot standby flywheels are evenly distributed or only proportionally to their rated power when the impact arrives, the activated flywheels with lower speeds may be instructed to output power exceeding their current carrying capacity, causing their speed to drop sharply or even triggering a protective shutdown. Meanwhile, the original hot standby flywheels with higher speeds fail to fully utilize their output potential, thus prolonging the time it takes for the total output power to converge to the target value. Determining the impact power each flywheel can currently bear based on its real-time operating status and iteratively adjusting the distribution ratio accordingly ensures that the distribution result matches the actual physical capabilities of each flywheel, avoiding the imbalance of overload and underload. Therefore, in order to enable the total output power of the hot standby group to converge quickly and stably to the target value corresponding to the predicted impact power within a limited time after the impact occurs, this embodiment introduces an iterative optimization allocation mechanism driven by the proportion of real-time bearable impact power.

[0118] Specifically, after an impact occurs, the controller first reads the speed feedback, motor current feedback, and power command feedback of each flywheel in the hot standby group to determine the actual output power of each flywheel and writes it into the initial allocation vector. For example, for a hot standby group containing 9 flywheels, the initial allocation vector stores the actual output power values ​​of the 1st to 9th flywheels at the moment of impact in sequence.

[0119] S302. Based on the current operating state of the flywheel, determine the impact power that the flywheel can withstand at the current moment.

[0120] In this step, after initialization, the array controller enters an iterative optimization loop. In each iteration, the array controller first determines the current impact power that each flywheel in the hot standby group can withstand based on its current operating status. This determination process specifically includes: the array controller acquiring the current speed of the flywheel, the current output power value, and whether the flywheel is in a normal power output state (e.g., by judging through the fault flag and enable flag bits in the status word). Subsequently, the array controller calculates the current impact power that each flywheel can withstand based on the preset safe lower speed limit and power upper limit model. The safe lower speed limit refers to the speed threshold that the flywheel is not allowed to fall below during power output. If the real-time speed is lower than this threshold, it is determined that the flywheel currently does not have the additional power handling capacity. The power upper limit model constructs the maximum power increment that the flywheel can safely output at the current speed based on parameters such as the ratio of the current speed of the flywheel to the rated speed, the rated torque of the motor, and the maximum allowable current of the converter.

[0121] The array controller subtracts the current output power from the maximum power increment and takes the difference as the impact power that the flywheel can withstand at the current moment. If the difference is positive, the flywheel has positive impact power to withstand; if the difference is less than or equal to zero, the flywheel no longer has the ability to withstand additional power and the impact power to withstand is recorded as zero.

[0122] As one optional implementation, the calculation of the withstandable impact power adopts a linear simplification model. That is, the maximum output power is linearly calculated based on the ratio of the current speed to the rated speed, and then the current actual output power is subtracted to obtain the withstandable impact power. As another optional implementation, the array controller has a pre-stored power output capability curve table of each flywheel at different speed ranges. When determining the withstandable impact power, the corresponding maximum allowable power is directly obtained by looking up the table based on the current speed, and then the current actual output power is subtracted to obtain the withstandable impact power value that more closely matches the actual external characteristic curve of the flywheel.

[0123] S303. Determine the current power distribution ratio of the flywheel based on the ratio of the impact power it can withstand to the sum of the impact power it can withstand of all flywheels in the hot standby group.

[0124] In this step, after determining the impact power that each flywheel in the hot standby group can withstand, the array controller calculates the sum of the impact power that all flywheels in the hot standby group can withstand. Then, the impact power that a single flywheel can withstand is divided by this sum to obtain the power distribution ratio of that flywheel at the current moment. This ratio reflects the proportion of the additional output capacity that the flywheel has in the current operating state to the total output capacity of the hot standby group.

[0125] S304. Adjust the power distribution of each flywheel according to the current power distribution ratio of each flywheel in the hot standby group.

[0126] In this step, the array controller multiplies the predicted impact power by the current power allocation ratio of each flywheel to obtain the power setpoint that each flywheel should bear in the current iteration. This setpoint is then sent via the bus to the local controller of the corresponding flywheel. The local controller adjusts the drive signal of the power conversion unit, causing the flywheel to output power again according to the new allocation ratio. It should be noted that if the impact power that a flywheel can bear is zero, its allocation ratio is zero, and that flywheel does not increase its additional power output in the current iteration.

[0127] S305. Repeat the above steps of determining the impact power that can be borne, determining the current power allocation ratio, and adjusting the power allocation until the deviation between the total output power of the hot standby group and the predicted impact power is within the allowable error range.

[0128] In this step, after each round of power allocation adjustment, the array controller re-collects and accumulates the real-time output power of each flywheel in the hot standby group via the bus to obtain the current total output power of the hot standby group. The controller compares the deviation between the total output power of the hot standby group and the predicted impact power, and compares the deviation with the allowable error range. When the deviation does not meet the threshold requirement, the controller continues to recalculate the impact power that each flywheel can bear based on the new speed and output state, and repeatedly corrects the allocation ratio until the total output power converges to the target range. This method allows the initial allocation result to be quickly incorporated into the iterative correction process after the impact occurs, and allows each flywheel to redistribute power according to its real-time load-bearing capacity.

[0129] With this implementation, the output of each flywheel in the hot standby group can be continuously corrected according to the real-time status, the matching accuracy between the power distribution result and the predicted impact power is improved, and the total output power can converge within the limited error, thereby enabling the hot standby group to maintain stable power support capability under megawatt-level sudden load.

[0130] It should be noted that the above iterative optimization process mainly targets the scenario where the hot standby group contains both the original hot standby flywheel and the activated flywheel that has been activated and accelerated from the cold standby group. In this case, the real-time speeds of each flywheel differ, requiring differentiated allocation based on the proportion of impact power each flywheel can withstand. When the hot standby group only contains the original hot standby flywheel that has been maintained at its rated speed, since all flywheels have the same speed and are in their rated state, their respective impact power is the same. The array controller can directly use an average allocation method to distribute the predicted impact power equally among the flywheels in the hot standby group, thus achieving rapid matching of the total output power after the impact arrives. In addition, to maintain the uniformity of the control strategy and code reusability, the array controller can also choose to continue executing the above iterative optimization process. In this case, since the initial operating state of each flywheel is consistent, the impact power that can be withheld calculated in the first round is equal to each other. The iterative process usually converges to the target value within a few rounds, achieving the same effect of accurately matching the predicted impact power. Both methods can operate reliably when the hot standby group consists entirely of flywheels at rated speed.

[0131] This application provides a power surge mitigation method based on flywheel arrays. Through the aforementioned iterative optimization techniques, the array controller enables the power distribution of each flywheel in the hot standby group to be continuously corrected after the surge occurs based on the real-time speed maintenance capability and power margin of each flywheel. The imbalance deviation in the initial distribution result is gradually eliminated during the iteration process. Each flywheel shares the power reasonably according to its actual load-bearing capacity, avoiding the overall convergence delay caused by overload or underload of individual flywheels. Ultimately, the total output power of the hot standby group accurately matches the predicted surge power within a preset time, providing stable and reliable megawatt-level power support for AIDC loads.

[0132] Figure 4A flowchart illustrating a power surge mitigation method based on a flywheel array, provided in this application embodiment. Figure 3 .like Figure 4 As shown, the method includes:

[0133] S401. For each flywheel in the hot standby group, when the power impact is determined to be over, obtain the current speed of the flywheel and determine the expected group to which the flywheel belongs after the impact.

[0134] During this step, the flywheels in the hot standby group release their stored kinetic energy to support the load, causing their speeds to drop to varying degrees. If all flywheels are simultaneously recharged in a disorderly manner after the impact, the total charging power drawn by the array controller from the grid side will instantaneously reach a high peak, potentially causing a secondary impact on the common DC bus. Simultaneous competition for charging resources among the flywheels can also easily lead to reduced charging efficiency and bus voltage fluctuations. Determining charging priority based on the expected group affiliation and speed recovery requirements of each flywheel, and then recharging in batches, can stagger the charging load in time, smoothing the bus power demand, while prioritizing the rapid recovery of critical flywheels to their target speeds. Therefore, to safely, orderly, and efficiently complete the speed recovery and group reconfiguration of the hot standby group flywheels after the impact, this embodiment introduces a charging priority division and batch recharge mechanism based on the expected group affiliation and speed difference.

[0135] Specifically, the array controller first determines whether the power surge has ended. This determination can be based on the detection that the load-side power demand has decreased to below a preset surge end threshold over multiple consecutive sampling periods, and that the deviation between the current total output power of the hot standby group and the predicted surge power has remained within the allowable error range for a preset duration (e.g., 1 second), indicating that the power surge support phase has been completed. After determining that the power surge has ended, the array controller performs status acquisition and attribution planning operations for each flywheel within the hot standby group.

[0136] The controller synchronously collects the current rotational speed of each flywheel in the hot standby group and determines the expected group to which each flywheel belongs based on the configuration requirements after the impact. The expected group can be determined by the number of targets in the group, the current rotational speed of the flywheel, and the state ranking result after the impact. The state ranking can be based on a comprehensive ranking of the current rotational speed and the current temperature rise. The flywheel with the higher ranking result is assigned to the hot standby group, and the rest are assigned to the cold standby group.

[0137] In one possible implementation, when the array controller determines the expected group assignment of each flywheel after the power surge, it needs to comprehensively consider the speed changes and temperature rise accumulation of each flywheel during the surge. If group assignment is based solely on the current speed, flywheels with low speeds but near-maximum temperature rises may be assigned to the hot standby group and required to continue operating at their rated speeds. This would exacerbate heat accumulation in these flywheels, shorten their lifespan, or even cause overheating failures. By comprehensively evaluating both current speed and current temperature rise when determining the expected group assignment, flywheels with severely reduced speeds and high temperature rises that need to exit hot standby for cooling can be identified. Simultaneously, flywheels with good speeds and low temperature rises that can continue to perform hot standby tasks can be selected, thus achieving a more reasonable group configuration. Therefore, to retain the most suitable flywheels in the hot standby group based on their overall health status after the surge, while allowing flywheels in poor condition to promptly return to the cold standby group for recovery, this embodiment introduces a mechanism for determining the expected group assignment based on a comprehensive ranking of current speed and current temperature rise. The corresponding specific implementation steps are as follows:

[0138] Obtain the current temperature rise of each flywheel in the hot standby group at the end of the power surge, and sort the flywheels according to the current speed and current temperature rise; select the corresponding number of flywheels from the state sorting results according to the target number, and determine the expected group of the selected flywheels as the hot standby group; determine the expected group of the flywheels that were not selected in the hot standby group as the cold standby group.

[0139] After a power surge has ended, the array controller first obtains the target number of flywheels in the hot standby group under normal operating conditions. This target number is determined by the array controller based on configuration instructions issued by the upper-level energy management system, reflecting the number of flywheels that the system needs to retain in the hot standby group to maintain basic hot standby response capability during non-surge periods. For example, under typical operating conditions, this target number can be set to 4 flywheels to ensure that the hot standby group always has a certain immediate response capacity. The array controller reads the current value of this target number from internal registers or non-volatile memory.

[0140] Subsequently, the array controller collects the current temperature rise and current speed of each flywheel in the hot standby group at the end of the power surge. The current temperature rise refers to the difference between the temperature of the flywheel winding or bearing and the ambient temperature, which is measured by thermocouples or thermistors embedded in the flywheel motor windings and bearing housings and reported after conversion by the local controller. The array controller also acquires the current speed of each flywheel, which is a state variable at the same moment as the current speed collected during the aforementioned recharge phase.

[0141] After acquiring the current speed and temperature rise of each flywheel, the array controller sorts the flywheels according to these two parameters. As an optional implementation, the array controller has a pre-set state evaluation model that normalizes the current speed and temperature rise. Specifically, the state evaluation model first obtains the ratio of the current speed to the rated speed as the speed normalization value; the closer this value is to 1, the better the flywheel speed is maintained. Next, it obtains the ratio of the current temperature rise to a preset temperature rise alarm threshold and inverts it to obtain the temperature rise normalization value; the closer this value is to 0 (or the larger the positive value), the lower the flywheel temperature rise and the safer the thermal state. Subsequently, the state evaluation model weights and sums the speed normalization value and the temperature rise normalization value according to preset weighting coefficients (e.g., speed weight 0.6, temperature rise weight 0.4) to generate a comprehensive state value for each flywheel. The flywheel with the higher the comprehensive state value, the closer its current speed is to the rated speed and the lower its temperature rise, indicating a better overall state, and thus ranking higher in the sorting results.

[0142] As an alternative implementation, the array controller can also employ a temperature rise constraint-based sorting method: first, flywheels whose temperature rise exceeds a preset safety threshold are directly marked as unsuitable to remain in the hot standby group and placed at the end of the sorting; then, the remaining flywheels are sorted from highest to lowest current speed, with those having higher speeds ranked first. This method is suitable for scenarios with stricter thermal safety requirements, ensuring that any flywheel with overheating risk will not be preferentially selected for the hot standby group simply because of its higher speed.

[0143] After completing the state sorting, the array controller selects a corresponding number of flywheels from the sorting results according to the target number of hot standby groups. Specifically, the array controller selects the flywheels with the highest overall state value in the sorting results, whose number equals the target number, and determines their expected group assignment as the hot standby group. Simultaneously, the array controller determines the expected group assignment of the remaining flywheels in the current hot standby group as the cold standby group. This determination reflects the allocation plan of each flywheel in subsequent operating cycles after the impact: flywheels ranked higher have smaller speed drops and lower temperature accumulation during the impact, and are suitable to remain in the hot standby group; flywheels ranked lower have larger speed drops or higher temperature rises, and are suitable to revert to the cold standby group for low-speed standby to accelerate heat dissipation and recovery. After determination, the array controller writes the expected group assignment of each flywheel into the flywheel array's group management table for subsequent recharge scheduling and group switching operations.

[0144] By employing the aforementioned method of comprehensively ranking flywheels based on current speed and temperature rise, and selecting the optimal number of flywheels for the hot standby group, the array controller ensures that the composition of the hot standby group after a power surge is no longer solely determined by the fixed configuration before the surge. Instead, it dynamically adjusts the composition based on the actual performance of each flywheel during the surge. Flywheels with the best overall condition are retained in the hot standby group to maintain immediate response capability, while flywheels with high temperature rise or significant speed drop are promptly moved to the cold standby group to obtain opportunities for low-speed operation and natural cooling. This method effectively avoids the risk of overheating accumulation caused by forcing high-temperature flywheels to remain in the hot standby group to continue undertaking standby tasks, extending the overall service life of the flywheel array. Simultaneously, it ensures that the hot standby group always maintains a configuration consisting of flywheels in optimal condition, providing a more reliable hot standby resource foundation for responding to the next power surge.

[0145] S402. Determine the target charging speed of the flywheel according to the expected group, and calculate the speed difference between the target charging speed and the current speed.

[0146] In this step, the controller then calls the corresponding target speed table according to the expected group to obtain the charging target speed of each flywheel, and obtains the speed difference by subtracting the current speed from the target speed.

[0147] S403. Using preset rules, the charging priority is determined based on the expected group affiliation of each flywheel in the hot standby group and the corresponding speed difference.

[0148] In this step, the array controller uses preset rules to determine the charging priority based on the expected group affiliation of each flywheel in the hot standby group and the corresponding speed difference.

[0149] As an optional implementation, the default rule is to first divide the priority levels according to the expected belonging group: the flywheels expected to belong to the hot standby group have a higher charging priority than the flywheels expected to belong to the cold standby group, because the flywheels in the hot standby group need to be restored to the rated speed first to maintain the instantaneous response capability of the hot standby group; within the same expected belonging group, the priority is then determined in descending order of the speed difference, that is, the flywheel with the larger the speed difference and the farther the current speed is from the target speed has a higher charging priority in that group.

[0150] By adopting the aforementioned preset rules, the backflush control can prioritize restoring the hot standby flywheel to a state where it can once again bear the impact power, and allow the flywheels in the same group that deviate more from the target state to obtain charging resources first. This ensures that the charging sequence is consistent with the flywheel state changes, thereby improving the determinism of backflush scheduling and the consistency of output recovery.

[0151] As an alternative implementation, when the system has a more urgent requirement for the hot standby group recovery time, the preset rules can also disregard grouping and directly sort all flywheels in descending order of speed difference, with the flywheel with the largest speed difference receiving the highest charging priority, so that the flywheel with the lowest speed can get out of the deep discharge state as soon as possible.

[0152] S404. According to the charging priority, the flywheels that need to be charged in the hot standby group are divided into multiple charging batches, and the corresponding flywheels are recharged in sequence based on the multiple charging batches.

[0153] In this step, after determining the charging priority, the array controller divides the flywheels in the hot standby group that need charging into multiple charging batches according to the charging priority from high to low. The number of flywheels included in each charging batch is set according to the array controller's charging power management strategy. For example, it can be set to ensure that the number of flywheels charging simultaneously in each batch does not exceed a preset concurrent charging limit (such as 3 or 4), so as to ensure that the total charging power drawn from the DC bus when each flywheel returns to charge does not exceed the power limit allowed by the grid interface or the common DC bus.

[0154] The array controller assigns the flywheels with the highest priority (equal to the concurrent charging limit) to the first batch, then assigns the same number of flywheels from the next priority level to the second batch, and so on, until all flywheels requiring charging are assigned to a batch. Subsequently, the array controller sends recharge commands to the local controller of each flywheel in the order of the first batch, second batch, and finally the last batch. Each flywheel then accelerates from its current speed towards the target charging speed according to the command.

[0155] After each batch of recharge is completed, the array controller waits for the rotational speed of each flywheel in that batch to reach or approach the corresponding target charging speed (e.g., the speed deviation is less than the preset speed tolerance) before starting the recharge operation for the next batch.

[0156] S405. After completing the backflush, adjust the flywheel that was originally intended to belong to the cold standby group to the cold standby group.

[0157] In this step, after all batches have been recharged, the array controller adjusts the flywheels that were originally intended for the cold standby group to the cold standby group. This adjustment specifically includes: the array controller updating its internally maintained group status table, changing the group label of the corresponding flywheel from the hot standby group to the cold standby group, and sending a mode switching command to the local controllers of these flywheels. This switches their operating control strategy from the rated speed maintenance mode of the hot standby group to the standby speed maintenance mode of the cold standby group, subsequently operating at a lower standby speed to reduce standby losses.

[0158] The above method first confirms the status after the power surge, then organizes the recharge sequence according to the group and speed difference, and completes the recovery process through batch recharge. Finally, it completes the switchover of the cold standby group, thus ensuring that the recovery process of the hot standby group is consistent with subsequent group adjustments. By associating the recharge sequence with the expected group and speed difference, recovery resources can be allocated as needed, and the status update of the flywheel after the surge can be more consistent.

[0159] This application provides a power surge mitigation method based on a flywheel array. By employing the aforementioned methods of determining charging priority based on the expected group and speed difference, batch recharging, and group switching upon completion, the array controller achieves on-demand resource allocation during the flywheel speed recovery process after the power surge. Flywheels expected to belong to the hot standby group are prioritized for recharging to their rated speed to quickly restore hot standby response capability, while flywheels expected to belong to the cold standby group are recharged to their standby speed in subsequent batches. Batch recharging effectively mitigates the instantaneous impact of charging power on the DC bus, avoiding secondary power disturbances caused by concentrated charging. Simultaneously, the group switching after recharging ensures that the actual operating state of each flywheel remains consistent with the system-maintained configuration information, providing an accurate state basis for predicting and adjusting the ratio of the next power surge. Meanwhile, by employing the aforementioned method of comprehensively ranking flywheels based on current speed and temperature rise, and selecting the optimal number of flywheels for the hot standby group, the array controller ensures that the composition of the hot standby group after the impact is no longer solely determined by the fixed configuration before the impact. Instead, it dynamically adjusts the composition based on the actual performance of each flywheel during the impact. Flywheels with the best overall condition are retained in the hot standby group to maintain subsequent immediate response capabilities, while flywheels with high temperature rises or significant speed drops are promptly moved to the cold standby group to obtain opportunities for low-speed operation and natural cooling. This method effectively avoids the risk of overheating accumulation caused by forcing high-temperature flywheels to remain in the hot standby group to continue undertaking standby tasks, extending the overall service life of the flywheel array. Simultaneously, it ensures that the hot standby group always maintains a configuration consisting of flywheels in optimal condition, providing a more reliable hot standby resource foundation for responding to the next power surge.

[0160] Figure 5 This diagram illustrates the switching between a steady-state current sharing mode and an impulse response mode, as provided in an embodiment of this application. Figure 5 As shown, the entire working process of the system is divided into three stages: steady-state current sharing mode, impulse response mode, and automatic recharge mode, and the power distribution behavior in each stage is shown.

[0161] The steady-state current sharing mode occurs from t=0-2s. During this stage, the four flywheels FW1 to FW4 in the hot standby group maintain their rated speed (represented by solid lines), while the six flywheels FW5 to FW10 in the cold standby group are in a low-speed standby state (represented by dashed lines). During steady-state operation, the power distribution of each flywheel follows the principle of proportionality to rated capacity, that is, the power share undertaken by each flywheel is consistent with the ratio of its rated capacity to the total capacity, thereby keeping the energy state of each flywheel synchronized and effectively eliminating the circulating current problem within the group that may be caused by uneven distribution.

[0162] The period t = 2-4 seconds represents the response process after the power surge. During this phase, the total power demand rapidly increases from 1.5MW to 10MW, and the system instantly switches to surge response mode. In this mode, the power allocation principle shifts to differentiated allocation based on the real-time speed and inertia ratio of each flywheel. FW1 bears 2.5MW, FW2 bears 2.0MW, FW3 bears 1.8MW, FW4 bears 1.5MW, while FW5, as a flywheel newly activated from the cold standby group, bears 1.0MW due to its slightly lower speed. This allocation difference reflects the basic physical law of flywheel energy storage: flywheels with higher speeds possess greater releaseable energy and thus bear a higher share of the surge power. As shown in the figure, the total surge power is dynamically redistributed within 50ms, and the response time is controlled within 5ms.

[0163] The t=4-6s interval is the automatic recharge mode. During this stage, the array controller recharges the flywheels that have released energy in batches according to the degree of speed drop of each flywheel after the impact and their expected group assignment, restoring their speed to their respective target speeds. After recharging is completed, the flywheels expected to be assigned to the cold standby group switch to low-speed standby state, and the system as a whole returns to steady-state operation configuration.

[0164] Figure 6 This is a schematic diagram of a power surge mitigation device based on a flywheel array, provided as an embodiment of this application. Figure 6 As shown, the flywheel array-based power surge mitigation device 60 includes:

[0165] The partitioning module 601 is used to divide the flywheel array into a hot standby group operating at rated speed and a cold standby group operating at standby speed according to the operating status of each flywheel.

[0166] The adjustment module 602 is used to adjust the ratio of hot standby groups to cold standby groups according to the predicted impact power when an AIDC power surge is predicted.

[0167] The control module 603 is used to differentiate the transient impact power borne by each flywheel according to the ratio of the real-time speed to the rated speed of the flywheel in the hot standby group at the moment the power impact arrives. The hot standby group includes the original hot standby flywheel that has been maintained at the rated speed and the activated flywheel that has been activated from the cold standby group. The real-time speed of the activated flywheel is lower than the real-time speed of the original hot standby flywheel.

[0168] The control module 603 is also used to iteratively optimize the power distribution of each flywheel in the hot standby group so that the total output power of the hot standby group converges to the target value corresponding to the predicted impact power within a preset time.

[0169] In one possible implementation, the adjustment module 602 is further configured to:

[0170] The number of candidates is calculated based on the predicted impact power and the rated power of the flywheel.

[0171] The desired number is determined based on the number of candidates and the preset number of protections;

[0172] Adjust the ratio of hot standby groups to cold standby groups according to the expected quantity.

[0173] In one possible implementation, the control module 603 is further configured to:

[0174] For each flywheel in the hot standby group, obtain the actual output power of the flywheel at the instant of the impact, and use the actual output power as the initial value for iteration;

[0175] Based on the current operating state of the flywheel, determine the impact power that the flywheel can withstand at the current moment;

[0176] The current power distribution ratio of the flywheel is determined based on the ratio of the impact power that can be borne to the sum of the impact power that can be borne by all flywheels in the hot standby group.

[0177] Adjust the power distribution of each flywheel according to the current power distribution ratio of each flywheel in the hot standby group;

[0178] Repeat the steps described above to determine the impact power that can be handled, determine the current power allocation ratio, and adjust the power allocation until the deviation between the total output power of the hot standby group and the predicted impact power is within the allowable error range.

[0179] In one possible implementation, after iteratively optimizing the power distribution of each flywheel in the hot standby group, the control module 603 is further configured to:

[0180] For each flywheel in the hot standby group, at the end of the power impact, the current speed of the flywheel is obtained and the expected group to which the flywheel belongs after the impact is determined;

[0181] The target charging speed of the flywheel is determined based on the expected group, and the speed difference is calculated based on the target charging speed and the current speed.

[0182] Using preset rules, charging priority is determined based on the expected group affiliation of each flywheel in the hot standby group and the corresponding speed difference;

[0183] According to the charging priority, the flywheels that need to be charged in the hot standby group are divided into multiple charging batches, and the corresponding flywheels are recharged in sequence based on the multiple charging batches.

[0184] After the backflush is completed, the flywheel that was originally intended to belong to the cold standby group will be moved to the cold standby group.

[0185] In one possible implementation, the preset rules within the control module 603 include:

[0186] The flywheels expected to be assigned to the hot standby group have a higher charging priority than those expected to be assigned to the cold standby group.

[0187] For multiple flywheels that are expected to belong to the same group, the charging priority is determined in descending order of the speed difference.

[0188] In one possible implementation, the control module 603 is further configured to:

[0189] Obtain the target number of hot standby groups under normal operating conditions;

[0190] Obtain the current temperature rise of each flywheel in the hot standby group at the end of the power surge, and sort the flywheels according to the current speed and the current temperature rise.

[0191] Based on the target quantity, select the corresponding number of flywheels from the results of the state sorting, and determine the expected group to which the selected flywheels belong as the hot standby group;

[0192] The expected group for flywheels that were not selected in the hot standby group will be determined as the cold standby group.

[0193] In one possible implementation, before adjusting the ratio of hot standby units to cold standby units based on the predicted impact power, the control module 603 is further configured to:

[0194] Acquire multidimensional feature data, including historical power load data of AIDC, task start instructions in task scheduling logs, and GPU cluster memory usage.

[0195] Multidimensional feature data is input into a trained multidimensional feature fusion model, which outputs the predicted impact power.

[0196] In one possible implementation, the control module 603 is further configured to:

[0197] Under steady-state operating conditions, obtain the rated capacity of each flywheel in the flywheel array;

[0198] Based on the rated capacity of each flywheel in the flywheel array, the power distribution ratio of each flywheel in steady state is determined, wherein the power distribution ratio of each flywheel is directly proportional to its rated capacity.

[0199] This embodiment provides a power surge suppression device based on a flywheel array, which can perform the power surge suppression method based on a flywheel array provided in the above-described method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0200] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device 70 includes a processor 701 and a memory 702 communicatively connected to the processor 701. Optionally, the electronic device 70 also includes a communication component 703. The processor 701, memory 702, and communication component 703 are connected via a bus 704.

[0201] Memory 702 stores instructions executed by the computer;

[0202] The processor 701 executes computer execution instructions stored in the memory 702 to implement the power surge mitigation method based on the flywheel array as described above.

[0203] At least one processor 701 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0204] Optionally, in specific implementations, the processor 701 and memory 702 are implemented independently. In this case, the processor 701 and memory 702 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0205] Optionally, in a specific implementation, if the processor 701 and the memory 702 are integrated on a single chip, the processor 701 and the memory 702 can communicate through an internal interface.

[0206] This application also provides a computer storage medium storing computer execution instructions, which, when executed by a processor, implement the aforementioned power surge mitigation method based on a flywheel array.

[0207] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The computer-readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0208] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Alternatively, the readable storage medium can be an integral part of the processor. Both the processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the control device of a garment handling apparatus.

[0209] This unit division is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0210] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0211] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0212] If this function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods indicated in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0213] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0214] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it is readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A power surge mitigation method based on a flywheel array, characterized in that, The method includes: Based on the operating status of each flywheel, the flywheel array is divided into a hot standby group operating at rated speed and a cold standby group operating at standby speed. When an AIDC power surge is predicted, the target number of flywheels in the hot standby group is determined based on the matching relationship between the predicted surge power and the rated power of the flywheel, and the ratio of the number of hot standby groups to the number of cold standby groups is adjusted according to the target number of flywheels. At the moment the power surge arrives, the transient surge power borne by each flywheel is differentially allocated according to the ratio of the real-time rotational speed to the rated rotational speed of the flywheel in the hot standby group. The hot standby group includes the original hot standby flywheel that has been maintained at the rated rotational speed and the activated flywheel that has been activated from the cold standby group. The real-time rotational speed of the activated flywheel is lower than the real-time rotational speed of the original hot standby flywheel. The power distribution of each flywheel in the hot standby group is iteratively optimized so that the total output power of the hot standby group converges to the target value corresponding to the predicted impact power within a preset time. The iterative optimization of the power distribution of each flywheel in the hot standby group includes: For each flywheel in the hot standby group, the actual output power of the flywheel at the instant of the impact is obtained, and the actual output power is used as the initial value for iteration. Based on the current operating state of the flywheel, determine the impact power that the flywheel can withstand at the current moment; The current power distribution ratio of the flywheel is determined based on the ratio of the impact power that can be borne to the sum of the impact power that can be borne by all flywheels in the hot standby group. Adjust the power distribution of each flywheel according to the current power distribution ratio of each flywheel in the hot standby group; Repeat the steps of determining the impact power that can be borne, determining the current power allocation ratio, and adjusting the power allocation until the deviation between the total output power of the hot standby group and the predicted impact power is within the allowable error range.

2. The method according to claim 1, characterized in that, The step of determining the target number of flywheels in the hot standby group based on the matching relationship between the predicted impact power and the rated power of the flywheel, and adjusting the quantity ratio of the hot standby group and the cold standby group according to the target number of flywheels, includes: The number of candidates is calculated based on the predicted impact power and the rated power of the flywheel. The desired number is determined based on the number of candidates and the preset number of protections; The desired number is taken as the target number of flywheels; Adjust the ratio of the hot standby group to the cold standby group according to the desired quantity.

3. The method according to claim 1, characterized in that, After iteratively optimizing the power distribution of each flywheel in the hot standby group, the method further includes: For each flywheel in the hot standby group, upon determining the end of the power impact, the current rotational speed of the flywheel and the expected group to which the flywheel belongs after the impact are obtained; The target charging speed of the flywheel is determined based on the expected group, and the speed difference is calculated based on the target charging speed and the current speed. Using preset rules, the charging priority is determined based on the expected group affiliation of each flywheel in the hot standby group and the corresponding speed difference. According to the charging priority, the flywheels that need to be charged in the hot standby group are divided into multiple charging batches, and the corresponding flywheels are recharged sequentially based on the multiple charging batches. After the backflush is completed, the flywheel that is expected to belong to the cold standby group is adjusted to the cold standby group.

4. The method according to claim 3, characterized in that, The preset rules include: The charging priority of the flywheel that is expected to be assigned to the hot standby group is higher than that of the flywheel that is expected to be assigned to the cold standby group; For multiple flywheels in the same expected group, the charging priority is determined in descending order of the speed difference.

5. The method according to claim 3, characterized in that, Determining the expected group of the flywheel after the impact includes: Obtain the target number of the hot standby group under normal operating conditions; The current temperature rise of each flywheel in the hot standby group at the end of the power surge is obtained, and the state of each flywheel is sorted according to the current speed and the current temperature rise. According to the target quantity, select the corresponding number of flywheels from the state sorting results, and determine the expected group to which the selected flywheels belong as the hot standby group; The expected group to which the flywheels that were not selected in the hot standby group belong is determined as the cold standby group.

6. The method according to claim 1, characterized in that, Before adjusting the ratio of the hot standby group to the cold standby group according to the target number of flywheels, the method further includes: Acquire multidimensional feature data, including the historical power load data of the AIDC, task start instructions in the task scheduling log, and the video memory usage of the GPU cluster; The multidimensional feature data is input into a trained multidimensional feature fusion model, which outputs the predicted impact power.

7. The method according to claim 1, characterized in that, The method further includes: Under steady-state operating conditions, obtain the rated capacity of each flywheel in the flywheel array; Based on the rated capacity of each flywheel in the flywheel array, the power distribution ratio of each flywheel in steady state is determined, wherein the power distribution ratio of each flywheel is directly proportional to its rated capacity.

8. A power surge mitigation device based on a flywheel array, characterized in that, include: The partitioning module is used to divide the flywheel array into a hot standby group operating at rated speed and a cold standby group operating at standby speed, based on the operating status of each flywheel. The adjustment module is used to determine the target number of flywheels in the hot standby group based on the matching relationship between the predicted impact power and the rated power of the flywheel when an AIDC power surge is predicted, and to adjust the ratio of the number of hot standby groups to the number of cold standby groups based on the target number of flywheels. The control module is used to differentiate the transient impact power borne by each flywheel according to the ratio of the real-time speed to the rated speed of the flywheel in the hot standby group at the moment the power impact arrives. The hot standby group includes the original hot standby flywheel that has been maintained at the rated speed and the activated flywheel that has been activated from the cold standby group. The real-time speed of the activated flywheel is lower than the real-time speed of the original hot standby flywheel. The control module is also used to iteratively optimize the power distribution of each flywheel in the hot standby group so that the total output power of the hot standby group converges to the target value corresponding to the predicted impact power within a preset time. The iterative optimization of the power distribution of each flywheel in the hot standby group includes: For each flywheel in the hot standby group, the actual output power of the flywheel at the instant of the impact is obtained, and the actual output power is used as the initial value for iteration. Based on the current operating state of the flywheel, determine the impact power that the flywheel can withstand at the current moment; The current power distribution ratio of the flywheel is determined based on the ratio of the impact power that can be borne to the sum of the impact power that can be borne by all flywheels in the hot standby group. Adjust the power distribution of each flywheel according to the current power distribution ratio of each flywheel in the hot standby group; Repeat the steps of determining the impact power that can be borne, determining the current power allocation ratio, and adjusting the power allocation until the deviation between the total output power of the hot standby group and the predicted impact power is within the allowable error range.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the power surge mitigation method based on a flywheel array as described in any one of claims 1 to 7.

Images