Power regulation method and device based on large distributed energy storage system and medium

Abstract

The application relates to a power regulation method and device based on a large distributed energy storage system and a medium, the method comprising: maintaining a cross-mode shared data pool, the shared data pool comprising a power distribution weight vector, a reactive capacity grading parameter and a frequency disturbance identification threshold, which are periodically updated by a normal low-priority mode during normal operation; performing mutual exclusion priority arbitration on concurrent control instructions, and freezing the operating parameters of the mode before switching as a mode snapshot at the moment of mode switching; reusing corresponding parameters in the cross-mode shared data pool, and not recalculating the parameters during the effective period of the high-priority functional mode; and after the response process corresponding to the high-priority functional mode ends, continuing the normal low-priority mode that is preempted based on the mode snapshot. The application has the advantages of compressing the calculation overhead in the high-priority response window, inheriting the state of charge balance, and automatically continuing the preempted mode without relying on the upper computer to reissue instructions.

Description

Technical Field

[0001] This application relates to the field of grid energy storage, and in particular to a power regulation method based on a large-scale distributed energy storage system. Background Technology

[0002] With the increasing proportion of large-scale renewable energy integration into the grid, large-scale distributed energy storage systems, as a key regulatory resource for grid dispatch, need to respond quickly to various grid events at different time scales. Power storage monitoring systems typically need to simultaneously undertake multiple control tasks, including active power dispatch, voltage dispatch, frequency response, dynamic reactive power response, and emergency interaction between power sources, grids, and loads. These tasks are distributed across time scales of minutes, seconds, and milliseconds, and overlap with each other in terms of control objects, control objectives, and resource consumption.

[0003] Existing power storage monitoring solutions generally treat the aforementioned tasks as independent functional channels and manage them separately. Each functional channel independently completes the power allocation of each power conversion device, the occupancy of the reactive power capacity of the entire station, and the identification and judgment of disturbance events within its own response period. This independent channel architecture can guarantee basic execution accuracy when facing a single task, but it exposes multiple shortcomings due to multiple interrelationships in multi-task concurrent scenarios. First, high-priority response tasks at the millisecond level, such as primary frequency regulation and dynamic reactive power response, still require on-site calculation of the required power allocation weights and capacity occupancy parameters within the response window. The computational overhead during runtime encroaches on the effective time of the response window, resulting in action delays and further reducing the ability to support sudden disturbances. Second, each functional channel maintains its own parameter set, causing inconsistencies in their states. For example, the power allocation weights calculated by the active power dispatch channel based on the state of charge balance cannot be directly inherited when the frequency response task is triggered, causing the state of charge distribution that active power dispatch has worked hard to maintain over a long period of time to be disrupted during the frequency response process. Furthermore, reactive power tasks face similar capacity contention issues. Voltage dispatching tasks consume a significant proportion of reactive power capacity during normal operation, making it difficult for dynamic reactive power responses triggered by low-voltage ride-through events to achieve sufficient response capacity within millisecond-level time constraints. Finally, after the highest-priority source-grid-load emergency interaction event concludes, the preempted low-priority operating mode cannot autonomously resume locally based on the operating state before the event. Instead, it relies on the upper-level dispatch system to reissue instructions to recover, creating a second-level response blind spot and causing a break in grid support for the distributed energy storage system after the emergency event ends.

[0004] The aforementioned deficiencies are interconnected, stemming from the lack of a mechanism for sharing key control parameters among various functional channels, as well as the lack of the ability to solidify the state during mode switching and to locally resume operations after the event ends. Summary of the Invention

[0005] To enable multiple functional modes within a distributed energy storage system to share key control parameters, operate mutually exclusively, and automatically resume the preempted mode after an event, this application provides a power regulation method based on a large-scale distributed energy storage system.

[0006] Firstly, this application provides a power regulation method based on a large-scale distributed energy storage system, which adopts the following technical solution: A power regulation method based on a large-scale distributed energy storage system includes the following steps: S1. Maintain a cross-mode shared data pool for a distributed energy storage system, wherein the distributed energy storage system includes multiple power conversion devices connected in parallel to the power grid, and the cross-mode shared data pool includes power allocation weight vectors for each power conversion device, reactive capacity classification parameters of the distributed energy storage system, and frequency disturbance identification thresholds. The cross-mode shared data pool is periodically updated by the normal low-priority mode of the distributed energy storage system during normal operation. S2. Receive concurrent control commands for multiple functional modes within the distributed energy storage system, and perform mutual exclusion priority arbitration on the concurrent control commands based on a pre-set functional mode priority sequence to obtain a target operating mode. The mutual exclusion priority arbitration ensures that only one functional mode is active at any given time, and the functional mode priority sequence sets the priority of emergency interaction commands to the highest, the priority of rapid response commands to the second highest, and the priority of normal scheduling commands corresponding to the normal low-priority mode to the lowest. At the instant the target operating mode switches from the normal low-priority mode to a high-priority functional mode, the target power value, target voltage value, and power allocation weight vector corresponding to the normal low-priority mode before the switch are fixed as a mode snapshot. S3. During the period when the target operating mode is in effect, based on the parameters in the cross-mode shared data pool corresponding to the target operating mode, execution instructions are issued to the multiple power conversion devices, and the parameters in the cross-mode shared data pool are not recalculated during the period when the high-priority function mode is in effect; after the response process corresponding to the high-priority function mode ends, the preempted normal low-priority mode is resumed based on the mode snapshot, and the resumption does not depend on the external host computer instructions to be reissued.

[0007] By adopting the above technical solutions, the cross-mode shared data pool is periodically maintained by the normal low-priority mode during normal operation, enabling various high-priority functional modes to directly reuse existing parameters within the millisecond-level response window without the need for on-site calculation; mutual exclusion priority arbitration and mode snapshot solidification form a clear mutual exclusion sequence among multiple functional modes within the distributed energy storage system, and key operating parameters at the moment of mode switching are preserved; after the high-priority response process ends, the local succession mechanism based on the mode snapshot enables the preempted normal low-priority mode to resume execution without waiting for the host computer to reissue, thereby eliminating the response blind spot after the event ends, and ensuring that the state of charge balance maintained by active power dispatch and the reactive power capacity configuration occupied by voltage dispatch remain consistent in cross-functional scenarios.

[0008] Optionally, S1 includes sub-steps S11-S13: S11. Periodically update the power allocation weight vector in the normal low-priority mode; S12. Periodically update reactive capacity classification parameters under normal low-priority mode; S13. In normal low-priority mode, periodically update the frequency perturbation identification threshold.

[0009] By adopting the above technical solution, the three types of parameters in the cross-mode shared data pool are maintained independently according to their respective update mechanisms in the normal low-priority mode, so that subsequent high-priority responses can identify and reuse the corresponding parameters for power allocation, reactive capacity and frequency disturbances respectively.

[0010] Optionally, S11 includes sub-steps S111-S113: S111. Collect the state of charge, temperature, and available power of each power conversion device; S112. Perform a three-factor weighting on each power conversion device based on the normalized terms of state of charge, temperature, and available power to obtain an initial weight vector; S113. Based on the deviation between the actual station-level injected power of the distributed energy storage system in the previous cycle and the target power value corresponding to the normal low-priority mode, perform station-level closed-loop correction on the initial weight vector to obtain the power allocation weight vector. When the target operating mode switches from the normal low-priority mode to the high-priority functional mode, the high-priority functional mode directly reuses the power allocation weight vector obtained in S113 as the power response allocation basis in the high-priority functional mode, and does not re-execute S111 to S113 in the high-priority functional mode; In response to the situation where any power conversion device in the high-priority function mode is unable to fulfill the share corresponding to any power conversion device in the power response allocation base, the share will be redistributed to other available power conversion devices in the distributed energy storage system according to the pre-set nearest-neighbor transfer rules.

[0011] By adopting the above technical solutions, the three-factor weighting enables power allocation to simultaneously consider state of charge balance, temperature safety, and available power constraints. The station-level closed-loop correction adjusts the weights of the current cycle based on the measured deviation of the previous cycle, making the deviation between the total injected power at the station and the instructions from the host computer controllable. The high-priority function mode directly reuses existing weights and the nearest transfer mechanism, so that there is no need to recalculate the allocation within the millisecond-level response window, and the local anomaly of a single power conversion device does not affect the overall response depth.

[0012] Optionally, the weighting coefficients used when performing three-factor weighting on each power conversion device include a first coefficient, a second coefficient, and a third coefficient. The first coefficient corresponds to the weight of the state of charge term, the second coefficient corresponds to the weight of the temperature normalization term, and the third coefficient corresponds to the weight of the available power normalization term. The first, second, and third coefficients are periodically and adaptively adjusted based on the statistical characteristics of historical grid dispatch scenarios. The adaptive adjustment of the weighting coefficients and the self-tuning calculation thread of the frequency disturbance identification threshold share the same idle calculation time window framework and the same person-in-the-loop approval interface, but maintain their own historical sample sliding statistical windows. The sliding statistical window corresponding to the weighting coefficients stores historical dispatch scenario samples, and the sliding statistical window corresponding to the frequency disturbance identification threshold stores historical disturbance samples.

[0013] Optionally, S113 includes sub-steps S1131-S1133: S1131. Calculate the deviation between the actual station-level injected power of the distributed energy storage system in the previous cycle and the target power value corresponding to the normal low-priority mode; S1132. In response to the deviation being greater than a preset accuracy threshold, the initial weight vector of the current cycle is proportionally corrected according to the relative direction of the deviation to obtain the power allocation weight vector; S1133. In response to the deviation being less than or equal to the accuracy threshold, the initial weight vector for this cycle is directly determined as the power allocation weight vector. In response to the nearest transfer rule causing the attenuation of the total response depth of the distributed energy storage system to exceed the preset depth threshold, a response depth limitation signal is sent back to the upper-level dispatch system, and the depth threshold is synchronously injected into the proportional correction constraint condition of the next cycle S1132.

[0014] By adopting the above technical solutions, the station-level closed-loop correction is adaptively triggered according to the measured deviation of the previous cycle, so that small deviations within the accuracy threshold under normal operation do not introduce additional correction noise; the depth threshold feedback mechanism makes the power allocation weight vector of the next cycle take into account the transfer effect of the previous cycle when calculating, so as to avoid the cumulative share transfer of multiple consecutive cycles exceeding the station-level response depth boundary.

[0015] Optionally, S12 includes sub-steps S121-S123: S121. The total available reactive power capacity of the distributed energy storage system is pre-divided into normal reactive power capacity and emergency reactive power reserve. The ratio of normal reactive power capacity to emergency reactive power reserve is periodically updated based on the statistical results of historical low voltage ride-through events of the power grid. S122. In the normal low-priority mode, the emergency reactive power reserve is blocked, so that the voltage regulation mode triggered by the voltage dispatch command only performs reactive power regulation within the normal reactive power capacity; S123. When the target operating mode switches from voltage regulation mode to dynamic reactive power response mode triggered by dynamic reactive power response command, unlock emergency reactive power reserve, so that dynamic reactive power response mode performs reactive power regulation within the range of the sum of normal reactive power capacity and emergency reactive power reserve; after the low voltage ride-through event corresponding to dynamic reactive power response mode ends, as part of the normal low priority mode based on mode snapshot, re-lock emergency reactive power reserve and restore the target voltage value corresponding to voltage regulation mode from mode snapshot.

[0016] By adopting the above technical solution, reactive power capacity is divided according to the locked / unlocked state under normal operation, so that the voltage regulation mode and the dynamic reactive power response mode no longer share the same capacity pool, thereby avoiding the situation where voltage regulation occupies most of the capacity under normal operation and the reactive power response capacity is insufficient when a low voltage ride-through event is triggered; the re-locking of emergency reactive power reserve and the restoration of the target voltage value of voltage regulation mode are both completed through the mode snapshot succession mechanism, so that the capacity classification state and the state machine switching sequence are aligned.

[0017] Optionally, in the dynamic reactive power response mode, the allocation of the total available reactive power capacity among the power conversion devices is based on a weighted allocation of the current active power output of each power conversion device and the reactive power output margin corresponding to the rated capacity of the power conversion device. The reactive power output margin is calculated based on the coupling relationship between the active and reactive power capacities of the power conversion devices, and the device with the larger reactive power output margin undertakes a larger share of reactive power regulation.

[0018] By adopting the above technical solution, the dynamic reactive response is allocated according to the active and reactive coupling relationship of each power conversion device in the low voltage ride-through scenario. This prevents power conversion devices with large active output from being forced to bear a share exceeding their reactive output capacity, thus avoiding protection disconnection caused by power conversion devices exceeding the coupling capacity boundary in the low voltage ride-through scenario.

[0019] Optionally, the update of the split ratio in S121 includes sub-steps S1211-S1213: S1211. Maintain a sliding statistical window for historical low voltage ride-through events. Each historical low voltage ride-through event includes the emergency reactive power capacity injected by the distributed energy storage system at the time the event occurred. S1212. Based on the distribution of all emergency reactive power capacity values ​​within the sliding statistical window, take the pre-set quantile value of the distribution as the target value of emergency reactive power reserve for the next period; S1213. Use the difference between the available reactive power capacity of the entire station and the target value as the normal reactive power capacity for the next cycle, and synchronize the target value and the normal reactive power capacity to the reactive power capacity classification parameters in the cross-mode shared data pool.

[0020] By adopting the above technical solution, the ratio of normal reactive power capacity to emergency reactive power reserve is dynamically adjusted based on the actual capacity demand statistics of historical low voltage ride-through events. This ensures that the target value of emergency reactive power reserve converges to the statistical representative value of historical grid disturbances during long-term operation, avoiding capacity waste or insufficient capacity caused by splitting it according to a fixed ratio.

[0021] Optionally, fast response commands include frequency response commands. Frequency disturbance identification thresholds include frequency difference dead zone value, rate of change threshold, and duration threshold. Frequency disturbance identification actions adopt a three-factor joint criterion, which includes frequency difference amplitude criterion, frequency difference rate of change criterion, and frequency difference duration criterion. The frequency difference amplitude criterion is based on the frequency difference dead zone value, the frequency difference rate of change criterion is based on the rate of change threshold, and the frequency difference duration criterion is based on the duration threshold. S2 outputs the valid identification result of the frequency response command only when all three criteria are met. S13 includes sub-steps S131-S132: S131. During the idle calculation time between the active scheduling cycle and the frequency response cycle in the normal low-priority mode, run the self-tuning calculation thread for the frequency disturbance identification threshold. S132. Based on the output periodicity of the self-tuning calculation thread, the frequency difference dead zone value, rate of change threshold, and duration threshold are synchronized to the cross-mode shared data pool.

[0022] By adopting the above technical solution, the three-factor joint criterion is identified by S2 before the priority arbitration is executed, and S13 is only responsible for the periodic maintenance of the criterion threshold in the normal low priority mode. The responsibilities of the two are clearly separated. The duration threshold, as a time dimension reinforcement for the amplitude criterion, can filter out spurious disturbance events with short-term high change rate and avoid such events triggering unnecessary battery cycles. The self-tuning calculation thread uses the idle time between the active power scheduling cycle and the frequency response cycle to periodically update the criterion threshold without additional hardware.

[0023] Optionally, S131 includes sub-steps S1311-S1313: S1311. Maintain a sliding window for historical disturbance samples. The collection of historical disturbance samples continues in all operating modes. Each historical disturbance sample includes the disturbance amplitude, disturbance duration, the identification action result of S2, and the power tracking deviation after the identification action. S1312. With the objective function of minimizing the sum of the non-response true disturbance rate and the false response cost, the optimal combination of frequency difference dead zone value, change rate threshold and duration threshold is solved based on the historical disturbance samples in the sliding window. The non-response true disturbance rate is the ratio of the real disturbance samples in the sliding window that are not identified, and the false response cost is the cumulative value of the power tracking deviation after the action is identified in the sliding window. The objective function solution for S1312 is only performed in the normal low-priority mode. S1313. Based on the optimal combination obtained in S1312, load rate adaptive correction is superimposed. When the local load rate of the distributed energy storage system is lower than the preset load rate threshold, the frequency difference dead zone value is reduced. When the local load rate is higher than the load rate threshold, the frequency difference dead zone value is amplified. The local load rate is the ratio of the current actual output power of the distributed energy storage system to the rated power of the distributed energy storage system.

[0024] By adopting the above technical solution, the collection of historical disturbance samples and the solution of the objective function are separated. Sample collection is continuously covered in all modes, while the solution of the objective function only runs in the normal low-priority mode, so that the high-priority response window is not crowded out by self-tuning calculation. The frequency disturbance identification threshold is also doubly adaptive based on the statistical learning of historical disturbance samples and the superposition of the current load rate, so that the distributed energy storage system is biased towards sensitivity under low load rate and battery life under high load rate, without the need for manual switching.

[0025] Optionally, the frequency deviation amplitude criterion and the frequency deviation duration criterion are determined by an amplitude-time linkage threshold. The amplitude-time linkage threshold makes the duration threshold inversely correlated with the frequency deviation amplitude. The first amplitude interval, the second amplitude interval, and the third amplitude interval are all located above the frequency deviation dead zone value and are arranged in ascending order of amplitude. When the frequency deviation amplitude is in the first amplitude interval, the duration threshold takes a larger value. When the frequency deviation amplitude is in the second amplitude interval, the duration threshold takes an intermediate value. When the frequency deviation amplitude is in the third amplitude interval, the duration threshold takes a smaller value. This ensures that large-amplitude short-time disturbances meet the triggering conditions, small-amplitude long-time disturbances meet the triggering conditions, and small-amplitude short-time disturbances do not meet the triggering conditions. The small amplitude refers to the frequency deviation amplitude being above the frequency deviation dead zone value and within the first amplitude interval.

[0026] By adopting the above technical solution, the frequency difference amplitude criterion and the frequency difference duration criterion are upgraded from a parallel relationship to a coupled relationship, enabling the three-factor joint criterion to make a more refined distinction between true and false disturbances based on the two-dimensional characteristics of the disturbance event, and avoiding unnecessary battery cycles triggered by small-amplitude and short-term false disturbance events.

[0027] Optionally, a pseudo-disturbance fingerprint database pre-filter layer is set before the three-factor joint criterion. The pseudo-disturbance fingerprint database stores the feature vector set of common pseudo-disturbance events of the distributed energy storage system and the upstream power grid. Pseudo-disturbance events include at least one of the following: power step events of the upper-level dispatch system, grid connection events or disconnection events of the upstream power grid, reactive power compensation equipment operation events of the system, and one-click power distribution events of the system. The pseudo-disturbance fingerprint database pre-filter layer performs similarity matching between the feature vector of the current disturbance event and each feature vector in the pseudo-disturbance fingerprint database. Only when the similarity between the feature vector of the current disturbance event and all feature vectors in the pseudo-disturbance fingerprint database is lower than a preset similarity threshold, the current disturbance event is input into the three-factor joint criterion for further identification. The pseudo-disturbance fingerprint database is periodically updated by the self-tuning calculation thread of the frequency disturbance identification threshold based on historical pseudo-disturbance samples during idle computing time.

[0028] By adopting the above technical solution, the input end of the three-factor joint criterion pre-eliminates known spurious disturbance events through fingerprint database matching, so that the distributed energy storage system does not enter the identification action when a known spurious disturbance event is triggered, avoiding the frequency feedback caused by common control actions of the local station and the upstream power grid being misidentified as real disturbances; the fingerprint database and the three-factor joint criterion form a two-layer identification architecture of "pre-matching + back-end criterion", and the two share the same self-tuning infrastructure, so that the evolution direction of the fingerprint database and the criterion threshold converges in a coordinated manner.

[0029] Optionally, after the self-tuning calculation thread of the frequency disturbance identification threshold produces a new threshold, the new threshold does not directly overwrite the currently effective value in the cross-mode shared data pool. Instead, it first enters the shadow identification channel. The shadow identification channel uses the new threshold to identify real-time frequency disturbance samples, but the identification result does not trigger any action. After the shadow identification channel runs for a preset grayscale duration, it compares the difference between the shadow identification result and the formal identification result. In response to the difference rate falling into the preset tolerance range, the new threshold is submitted to human-in-the-loop approval. In response to the difference rate exceeding the tolerance range, the new threshold is discarded, and the difference is fed back as a counterexample to the objective function of the next cycle of the self-tuning calculation thread. Since the self-tuning output of the weighted coefficient cannot be verified by the shadow identification channel, it adopts a simplified path of direct submission to human-in-the-loop approval, without going through the shadow identification channel.

[0030] By adopting the above technical solution, the self-tuning output of the frequency disturbance identification threshold is first verified by the shadow identification channel before taking effect, so that the actual identification behavior of the new threshold can be verified without affecting the formal response channel; the new threshold whose difference rate exceeds the tolerance range is discarded and fed back to the objective function, so that the self-tuning iteration process has the ability to self-correct; the weighting coefficient, because its target is the allocation process rather than the identification process, adopts a simplified path of direct approval to avoid introducing gray-scale verification overhead that is incompatible with its nature.

[0031] Optionally, each type of parameter in the cross-mode shared data pool maintains two sets of values, including the currently effective value and the self-tuning recommended value. The self-tuning recommended value is available for operators to view through the monitoring interface and switch to the currently effective value through the approval interface. The switching action is recorded in the version log of the cross-mode shared data pool. In response to the approval interface being triggered during the high-priority functional mode, the parameter version switch is delayed until the state machine returns to the normal low-priority mode before execution, to avoid parameter mutations during the high-priority functional mode response process.

[0032] By adopting the above technical solution, parameter changes in the cross-mode shared data pool are explicitly controlled by operators through a dual-track approval interface, ensuring that the self-tuning results must be manually approved before they are put into effect. Furthermore, the timing of parameter version switching is coordinated with the state machine, preventing parameter mutations from occurring during high-priority functional mode response processes.

[0033] Optionally, during the high-priority functional mode, the power allocation weight vector, reactive capacity classification parameters, and frequency disturbance identification threshold in the cross-mode shared data pool are not recalculated on-site. However, the sample acquisition thread of the normal low-priority mode continues to collect the latest state of charge, latest temperature, latest available power, and historical disturbance samples of each power conversion device in all operating modes. The latest data collected is stored in the offline observation area of ​​the cross-mode shared data pool. The latest data in the offline observation area is only read by the weight recalculation of the next cycle S111-S113 during the normal low-priority mode transition phase after the high-priority functional mode ends.

[0034] Optionally, emergency interaction commands include emergency interaction response commands, and high-priority function modes include emergency interaction modes triggered by emergency interaction response commands. When the target operating mode is emergency interaction mode, S2 also includes sub-steps S21-S23: S21. After the emergency interaction mode takes effect, start the main timeout timer and initialize the extension count counter to zero. During the pre-set secondary confirmation window period before the main timeout timer reaches the pre-set main timeout duration, collect the local frequency and voltage samples of the distributed energy storage system and perform matching determination between the local frequency and voltage samples and the normal range of the power grid. If all local frequency and voltage samples fall into the normal range of the power grid within the secondary confirmation window period, exit the emergency interaction mode when the main timeout timer reaches the main timeout duration. If any local frequency or voltage sample does not fall into the normal range of the power grid within the secondary confirmation window period, reset the main timeout timer and increment the extension count counter by 1. If the extension count counter reaches the pre-set maximum extension count, report a continuous abnormal alarm and return the decision to exit the emergency interaction mode to the operators. S22. After exiting the emergency interaction mode, the system enters a transition state. In the transition state, the output power in the emergency interaction mode is reduced back to the target power value in the mode snapshot using a slope-limited ramp function. The slope of the ramp function is adaptively selected based on the deviation of the local station frequency and voltage from the normal range of the power grid. The closer the deviation is to the normal range of the power grid, the larger the slope. In response to receiving an emergency interaction response command again during the transition state, the slope function reduction process is interrupted, and the output power of the power conversion equipment is increased back to the full power response level corresponding to the emergency interaction mode. S23. After the transition state ends, mark the current operating mode of the distributed energy storage system as the normal low-priority mode, and restore the target power value, target voltage value and power allocation weight vector corresponding to the normal low-priority mode based on the mode snapshot.

[0035] Optionally, concurrent control instructions also include group control instructions. When S3 issues execution instructions to the power conversion device, it is also subject to the local veto power of the group control instructions. The application of the local veto power includes sub-steps S31-S33: S31. Receive the rejection signal of a group control command for a single power conversion device. The rejection signal is triggered when the state of charge or temperature of a single power conversion device exceeds the corresponding safe range. S32. Set the share of a single power conversion device in the power response allocation base to zero without affecting the overall effectiveness of the target operating mode, and redistribute the share to other available power conversion devices in the distributed energy storage system according to the nearest transfer rule; S33. In response to the target operating mode being emergency interaction mode, the local veto power is applied to a single power conversion device with its share set to zero. When the total response depth of the distributed energy storage system decreases beyond the preset emergency depth threshold due to the limited available power of other available power conversion devices caused by the proximity transfer rule, a response depth limitation signal is sent back to the upper-level dispatch system and the application of subsequent veto signals is suspended.

[0036] Optionally, the execution entities of the power regulation method based on large-scale distributed energy storage systems include an energy storage monitoring system and a power conversion coordination control host. The energy storage monitoring system is responsible for receiving and arbitrating routine dispatch and group control commands on minute and second time scales. The power conversion coordination control host is responsible for identifying and issuing frequency response and dynamic reactive power response commands on millisecond time scales. The energy storage monitoring system and the power conversion coordination control host jointly undertake the reception and execution of emergency interactive response commands. The primary copy of the cross-mode shared data pool is maintained by the energy storage monitoring system, and the power conversion coordination control host maintains a local copy of the cross-mode shared data pool. The local copy is periodically updated from the primary copy through the routine parameter synchronization channel in the hierarchical heartbeat synchronization mechanism.

[0037] By adopting the above technical solutions, the execution entities are physically divided according to the time scale, so that the second-level and millisecond-level responses are isolated from each other at the hardware level and do not compete for computing resources; the ownership of the primary replica and the synchronization mechanism of the local replica in the cross-mode shared data pool are clear, so that the data flow of the data pool in the dual-host architecture is clear and controllable.

[0038] Optionally, the energy storage monitoring system maintains a second-level state machine, which covers the switching between functional modes corresponding to normal scheduling commands. The power conversion coordination control host maintains a millisecond-level state machine, which covers the switching between functional modes corresponding to fast response commands and emergency interaction commands. The second-level and millisecond-level state machines make state switching decisions independently. When the millisecond-level state machine preempts a normal scheduling functional mode, it does not wait for a response from the second-level state machine but directly executes the response based on the currently effective parameters in the cross-mode shared data pool. The second-level and millisecond-level state machines are asynchronously synchronized through the state version number in the cross-mode shared data pool. The mode snapshot at the moment of mode switching is written to the cross-mode shared data pool by the preemption initiator and then asynchronously read by the other host. In response to any host detecting that the state version number maintained by this host is inconsistent with the state version number of the other host, this host's decision is based on the latest version number of the millisecond-level state machine, and the current decision of the second-level state machine is postponed until the state version number synchronization is completed.

[0039] By adopting the above technical solutions, the state machines of the two hosts operate independently according to time scales, enabling millisecond-level preemption to trigger responses without traversing the communication link between the two hosts, thus avoiding communication delays between the two hosts crowding out the millisecond-level response window. The asynchronous synchronization mechanism of the state version number ensures that the second-level state machine and the millisecond-level state machine remain eventually consistent after state switching. When the state version numbers are inconsistent, the millisecond-level state machine takes priority, ensuring that the distributed energy storage system's decisions during the inconsistent window period between the two hosts always favor a time-sensitive high-priority mode.

[0040] Optionally, the energy storage monitoring system and the power conversion coordination control host maintain the consistency of the cross-mode shared data pool, mode snapshots, and state machine states through a hierarchical heartbeat synchronization mechanism. The hierarchical heartbeat synchronization mechanism is divided into normal parameter synchronization channels, transient snapshot synchronization channels, and state version synchronization channels according to the data lifecycle. The normal parameter synchronization channel synchronizes the cross-mode shared data pool at a second-level cycle, the transient snapshot synchronization channel synchronizes the mode snapshots according to an event-triggered method, and the state version synchronization channel synchronizes the state version number at a millisecond-level cycle. In response to any synchronization channel's consecutive timeouts exceeding a pre-set synchronization timeout threshold, the corresponding channel's fault degradation strategy is triggered. The fault degradation strategy includes at least one of the following: second-level cycle extension, transient snapshot local caching pending recovery and retransmission, and state version synchronization degradation to single-host local decision.

[0041] Optional, normal dispatch instructions include voltage dispatch instructions for buses of different voltage levels. Each voltage dispatch instruction for a bus of different voltage levels corresponds to a pre-set control step size and control dead zone. The control step size and control dead zone are increased sequentially from low to high voltage level.

[0042] Optionally, the execution parameters of the frequency response command include response delay, response arrival time, inequality rate, and stabilization duration. The response delay shall not exceed the preset delay limit. The response arrival time is the time it takes for the output power to rise from the trigger of the recognition action to the power value corresponding to the preset arrival ratio. The inequality rate is the proportional coefficient between the frequency deviation and the power response. The stabilization duration is the time during which the frequency response process is continuously supported.

[0043] Optionally, the performance constraints of the dynamic reactive response mode include that the time elapsed from voltage drop triggering to the reactive output reaching the preset proportion of the target reactive regulation amount does not exceed the preset upper limit of transient response time.

[0044] Optionally, the performance constraints of the emergency interaction mode include that the time elapsed from the triggering of the emergency interaction response command to the output power reaching the full power response level does not exceed a preset upper limit of the emergency response duration.

[0045] Optionally, group control commands also include panoramic monitoring commands and one-click power delivery commands. Panoramic monitoring commands are used to obtain the operating status parameters and battery data of each power conversion device in real time, while one-click power delivery commands are used to execute batch power command delivery to all or some power conversion devices in normal low-priority mode.

[0046] Optionally, emergency interactive commands include load shedding level commands. When S2 receives a load shedding level command, it first performs load shedding error prevention judgment before executing mode switching. Load shedding error prevention judgment includes: determining the target frequency range corresponding to the load shedding level command based on a pre-stored level-frequency range mapping table; performing a judgment on the local station frequency using a three-factor joint criterion to obtain the local station frequency judgment result; executing mode switching corresponding to the load shedding level command only when the local station frequency falls into the target frequency range and the local station frequency judgment result is valid; responding to the local station frequency judgment result being valid but the local station frequency not falling into the target frequency range, sending back a criterion inconsistency alarm to the control center station and temporarily suspending mode switching; responding to the local station frequency falling into the target frequency range but the local station frequency judgment result being invalid, executing mode switching with half the action amount corresponding to the load shedding level command and continuing to monitor the local station frequency.

[0047] Secondly, the computer device provided in this application adopts the following technical solution: A computer device comprising: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: The above-described power regulation method based on a large-scale distributed energy storage system is implemented.

[0048] Thirdly, this application provides a computer-readable storage medium that adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above.

[0049] The storage medium stores at least one instruction, at least one program, a code set, or an instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement the following: The power regulation method based on large-scale distributed energy storage systems, as described above. Attached Figure Description

[0050] Figure 1 is a schematic diagram of the overall architecture and dual-host division of labor of the distributed energy storage system used in the power regulation method based on a large-scale distributed energy storage system in an embodiment of the present invention.

[0051] Figure 2 is a main flowchart of a power regulation method based on a large-scale distributed energy storage system in one embodiment of the present invention.

[0052] Figure 3 is a schematic diagram of the cross-mode shared data pool and the maintenance process of three types of parameters in one embodiment of the present invention.

[0053] Figure 4 is a timing diagram of the emergency interaction mode activation and deactivation process in one embodiment of the present invention. Detailed Implementation

[0054] The present application will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the application and are not intended to limit the scope of the application.

[0055] This application provides a power regulation method based on a large-scale distributed energy storage system. This method is applied to large-scale distributed energy storage systems on the grid side. As a key regulation resource for grid dispatch, the distributed energy storage system needs to simultaneously undertake multiple control tasks, including active power dispatch, voltage dispatch, frequency response, dynamic reactive power response, and emergency interaction, at three time scales: minute, second, and millisecond. This method establishes a cross-mode shared data pool within the distributed energy storage system. During normal operation, the low-priority mode periodically maintains the power allocation weight vector, reactive power capacity classification parameters, and frequency disturbance identification threshold in the data pool, allowing various high-priority functional modes to directly reuse existing parameters within the millisecond-level response window. A mutual exclusion priority arbitration state machine coordinates concurrent command conflicts among multiple functional modes, and at the moment of mode switching, the operating parameters of the preempted mode are fixed as a mode snapshot. This allows the preempted mode to resume operation based on the local snapshot without relying on a higher-level dispatch system to reissue commands after an emergency event. (Refer to...) Figure 2 The following is a detailed explanation of each step S1-S3.

[0056] For ease of explanation, let's first define the core entities involved. A distributed energy storage system is a battery energy storage power station composed of multiple power conversion devices connected in parallel to a designated bus of the power grid and corresponding battery clusters. A power conversion device is a bidirectional converter connecting the battery clusters and the power grid. A battery cluster is an energy storage unit consisting of battery modules connected in series and parallel, corresponding to a single power conversion device. A cross-mode shared data pool is a set of key control parameters maintained by the energy storage monitoring system and shared by multiple functional modes. The normal low-priority mode is the operating mode of the distributed energy storage system when no high-priority functional modes are active, including active power dispatch mode, voltage regulation mode, and group control batch distribution mode. High-priority functional modes are functional modes that have preemptive rights over the normal low-priority mode, including frequency response mode, dynamic reactive power response mode, and emergency interaction mode.

[0057] To facilitate the numerical calculations in subsequent steps, the following baseline scenario is established. The total rated active power of the distributed energy storage system is 50MW, consisting of 10 power conversion devices, each with a rated power of 5MW, connected in parallel. These 10 power conversion devices are designated as power conversion devices 1 to 10. The distributed energy storage system is connected to a 35kV bus via a step-up transformer and is connected to the upper-level dispatch system via a fiber optic communication link. The total available reactive power capacity is 30Mvar, the grid rated frequency is 50Hz, and the normal grid range is a frequency of 49.95Hz to 50.05Hz and a bus voltage of 0.95 to 1.05 times the per-unit value. The main timeout duration for the emergency interaction mode is 5 minutes, with a maximum of 3 extensions. The numerical examples in subsequent steps will all be based on this baseline scenario, with differences in operating conditions explicitly marked as Operating Condition 1, Operating Condition 2, etc. The overall maintenance mechanism of S1 will be explained below.

[0058] S1. Maintain a cross-mode shared data pool for a distributed energy storage system, wherein the distributed energy storage system includes multiple power conversion devices connected in parallel to the power grid, and the cross-mode shared data pool includes the power allocation weight vector of the multiple power conversion devices, the reactive capacity classification parameters of the distributed energy storage system, and the frequency disturbance identification threshold. The cross-mode shared data pool is periodically updated by the normal low-priority mode of the distributed energy storage system during normal operation.

[0059] The cross-mode shared data pool is physically implemented as an in-memory data structure of the energy storage monitoring system, organized as a set of key-value pairs with version numbers. Each type of parameter in the key-value pair set corresponds to a fixed key, and the parameter value occupies a contiguous memory area. In the baseline scenario of 50MW / 10 power conversion devices, the power allocation weight vector in the cross-mode shared data pool is a floating-point array of length 10, corresponding to the power allocation weight of each of the 10 power conversion devices; the reactive power capacity classification parameter is a combination of two floating-point values: normal reactive power capacity and emergency reactive power reserve; the frequency disturbance identification threshold is a combination of three floating-point values: frequency difference dead zone value, rate of change threshold, and duration threshold. In addition to the currently effective values ​​mentioned above, the cross-mode shared data pool also maintains a self-tuning recommended value with the same structure for each type of parameter, as well as an offline observation area for continuously receiving the latest observation data during long-term high-priority modes.

[0060] Write operations for the cross-mode shared data pool are handled by the normal low-priority mode. This mode recalculates various parameters according to a pre-set update cycle and writes the new values ​​back to the data pool. The update cycle is differentiated based on the parameter type: power allocation weight vectors are updated on a second-by-second basis, reactive capacity classification parameters on a minute-by-minute basis, and frequency disturbance identification thresholds on an hourly basis. Read operations for the cross-mode shared data pool are handled by various high-priority function modes. Upon activation, a high-priority function mode directly reads the parameters corresponding to the target operating mode from the data pool and uses these parameters as the basis for its response process. Separating read and write time periods ensures that no weight calculation operations occur within the response window.

[0061] Specifically, refer to Figure 3 S1 includes sub-steps S11-S13.

[0062] S11. In the normal low-priority mode, periodically update the power allocation weight vector.

[0063] S12. Periodically update reactive capacity classification parameters under normal low-priority mode.

[0064] S13. In normal low-priority mode, periodically update the frequency perturbation identification threshold.

[0065] The power allocation weight vector maintained by S11 serves the active power domain. Each element in the vector represents the share of the corresponding power conversion device in the total active power response at the station level. The vector's values ​​are limited to non-negative floating-point numbers for all elements, and the sum of all elements is 1. In the baseline scenario with 10 power conversion devices, if the operating states of each power conversion device are completely uniform, the power allocation weight vector is a uniform vector with all elements equal to 0.1.

[0066] The reactive capacity classification parameters maintained by S12 serve the reactive power domain. The parameters include two values: normal reactive power capacity and emergency reactive power reserve. The sum of the two is the total available reactive power capacity of the entire station. Under the baseline scenario of a total station capacity of 30 Mvar, the separation of the two values ​​is periodically adjusted according to the statistical driving mechanism defined by S12.

[0067] The frequency disturbance identification threshold maintained by S13 serves the event identification domain. The three threshold values ​​together determine whether the frequency disturbance of this station is identified as a valid frequency response command.

[0068] The update cycles for the three branches, S11, S12, and S13, are set independently. The power allocation weight vector is directly related to dynamic quantities such as battery state of charge and temperature, with an update cycle on the order of seconds. The reactive capacity classification parameter depends on the statistical results of historical low-voltage ride-through events in the power grid, with an update cycle on the order of minutes. The frequency disturbance identification threshold depends on the sliding statistical learning of historical disturbance samples and human-in-the-loop approval, with an update cycle on the order of hours. The independence of the three types of cycles allows each type of parameter to be updated according to its own engineering time scale.

[0069] S11, S12, and S13 maintain their respective parameters independently within their respective update cycles, but share the same type of infrastructure. All three branches use the same idle computing time window framework, which refers to the idle period between the active power scheduling cycle and the frequency response cycle of the power conversion coordination control host. All three branches use the same human-in-the-loop approval interface, uniformly exposing the recommended self-tuning values ​​of all three branches to operators for review. All three branches use the same version log mechanism, recording each parameter change in a unified format in the version log for backtracking. The specific update mechanisms for the three branches will be detailed in the subsequent expansions of S111-S113, S121-S123, and S131-S132.

[0070] Specifically, S11 includes sub-steps S111-S113.

[0071] S111. Collect the state of charge, temperature, and available power of each power conversion device. The state of charge is estimated by the battery management system corresponding to the battery cluster at millisecond intervals and uploaded to the energy storage monitoring system via a dual-host communication link; the temperature is the current measured temperature at the temperature measurement point inside the battery cluster; the available power is the maximum active power that the power conversion device can currently output under the constraints of the upper limit of the battery-side current and the upper limit of the grid-side voltage.

[0072] S112. Based on the normalized terms of state of charge, temperature, and available power, a three-factor weighting is performed on each power conversion device to obtain an initial weight vector. The three-factor weighting adopts either a linear weighted sum or a product weighting form. In the linear weighted sum form, the... Initial weights of power conversion devices satisfy: in, It is the first Normalized state of charge terms of the power conversion equipment It is the first Temperature normalization term of the power conversion equipment It is the first The normalized term of available power for a power conversion device. , , These are the corresponding first, second, and third coefficients, respectively. In the product-weighted form, the first... Initial weights of power conversion devices satisfy: The linear weighted sum form is suitable for operating conditions where the three factors are relatively independent, while the product weighted form is suitable for operating conditions where the overall weight should be zero when any factor is zero. Normalized state of charge term. Take in charging scenarios This assigns a larger weight to power conversion devices with low state of charge; and in discharge scenarios, it selects... This assigns a larger weight to power conversion devices with high states of charge, among which This is the upper limit of the normalized state of charge. Temperature normalization term. The value is close to 1 when the temperature is within the safe range, and approaches 0 when it approaches the boundary. The power normalization term can be used. It equals the ratio of the current available power to the rated power of the power conversion device.

[0073] Construct a set of specific numerical calculations for discharge scenarios. Let... , , The normalized values ​​of the 10 power conversion devices are as follows: the normalized state of charge (SBC) values ​​for devices 1 to 10 are 0.50, 0.55, 0.60, 0.65, 0.70, 0.75, 0.80, 0.85, 0.90, and 0.95, respectively; the temperature normalization value is 0.95 for all devices; and the available power normalization value is 1.00 for all devices. Substituting these values ​​into a linear weighted sum, the initial weight for device 1 is... The initial weights of the 10 devices were 0.740, 0.765, 0.790, 0.815, 0.840, 0.865, 0.890, 0.915, 0.940, and 0.965, respectively. After normalization (i.e., dividing each value by the sum of 8.525), the resulting power allocation weights were 0.087, 0.090, 0.093, 0.096, 0.099, 0.101, 0.104, 0.107, 0.110, and 0.113, respectively. Devices with higher states of charge received larger weights, consistent with the energy balance guidelines in discharge scenarios.

[0074] The first, second, and third coefficients support engineered settings. The value range for each coefficient is preset by operators via an approval interface; a typical value range is [insert range here]. , , Once the coefficients are determined, the weights are calculated using the weighting formula in S112.

[0075] S113. Based on the deviation between the actual station-level injected power of the distributed energy storage system in the previous cycle and the target power value corresponding to the normal low-priority mode, perform station-level closed-loop correction on the initial weight vector to obtain the power allocation weight vector. The sub-steps of S113 are expanded in subsequent blocks.

[0076] When the target operating mode switches from the normal low-priority mode to the high-priority functional mode, the high-priority functional mode directly reuses the power allocation weight vector obtained in S113 as the power response allocation basis in the high-priority functional mode, and does not re-execute S111 to S113 in the high-priority functional mode. When any power conversion device in the high-priority functional mode cannot assume the corresponding share in the power response allocation basis, the share is reallocated to other available power conversion devices in the distributed energy storage system according to the pre-set proximity transfer rules. The proximity transfer rules sort the transfer targets according to the physical adjacency relationship in the distributed energy storage system layout, and adjacent devices take priority in assuming the transfer share. For example, in the baseline scenario, if the 5th power conversion device cannot assume its weight share of 0.099 due to the SOC dropping to the critical value, the share is transferred to the physically adjacent 4th and 6th devices according to the rules, and each of them assumes 0.0495.

[0077] The first, second, and third coefficients also support periodic adaptive adjustments based on statistical characteristics of historical power grid dispatch scenarios. The adaptive adjustment of the weighting coefficients and the self-tuning calculation thread for the frequency disturbance identification threshold share the same idle computation time window framework and the same person-in-the-loop approval interface, but maintain their own historical sample sliding statistical windows: the sliding statistical window corresponding to the weighting coefficients stores historical dispatch scenario samples, with each sample recording the dispatch period, target power value, actual station-level injected power value, and corresponding SOC distribution deviation; the sliding statistical window corresponding to the frequency disturbance identification threshold stores historical disturbance samples. The two have different sample objects and are maintained independently. The self-tuning output of the weighting coefficients takes effect through the approval interface, allowing the weighting coefficients to automatically follow the long-term statistical characteristics of power grid dispatch.

[0078] Furthermore, S113 includes sub-steps S1131-S1133.

[0079] S1131. Calculate the deviation between the actual station-level injected power of the distributed energy storage system in the previous cycle and the target power value corresponding to the normal low-priority mode. The actual station-level injected power is the measured active power at the connection bus of the distributed energy storage system, which is read from the metering device and stored in the offline observation area of ​​the cross-mode shared data pool by the energy storage monitoring system at the end of each second-level cycle. The deviation is the difference between the actual station-level injected power and the target power value, which can be positive or negative. A positive value indicates injection overshoot, and a negative value indicates injection undershoot.

[0080] S1132. When the deviation exceeds a preset accuracy threshold, a proportional correction is performed on the initial weight vector for this period according to the relative direction of the deviation, resulting in a power allocation weight vector. The accuracy threshold is a preset percentage of the target power value, typically 0.5% of the target power value. The proportional correction scales the initial weight vector according to the relative ratio of the deviation to the target power value, and the scaled power allocation weight satisfies: in, This is the actual station-level injection power from the previous cycle. It is the target power value. It is the first The initial weights of the power conversion equipment. It is the revised version of the first Power allocation weights for power conversion devices. This is the proportional correction factor. The actual injected power after execution converges towards the target value.

[0081] S1133. When the deviation is less than or equal to the accuracy threshold, the initial weight vector of this cycle is directly determined as the power allocation weight vector.

[0082] Scenario 1: Assume the target power value of the previous cycle is 30MW, the actual station-level injected power is 29.7MW, and the deviation is -0.3MW. 0.5% of 30MW is 0.15MW. The absolute value of the deviation of 0.3MW is greater than the accuracy threshold. Proceed to S1132, and perform a proportional correction on the initial weight vector according to the relative direction of the deviation, so that the weight vector of this cycle is adjusted upward to compensate for the under-adjustment gap. Scenario 2: Assume the target power value of the previous cycle is 30MW, the actual station-level injected power is 30.05MW, and the deviation is 0.05MW, which is less than the accuracy threshold of 0.15MW. Proceed to S1133, and directly determine the initial weight vector of this cycle as the power allocation weight vector without proportional correction.

[0083] When the nearest-neighbor transfer rule causes the total response depth of the distributed energy storage system to decrease beyond a pre-set depth threshold, a response depth-limited signal is transmitted back to the upper-level dispatch system, and the depth threshold is synchronously injected into the proportional correction constraint of the next cycle S1132. The depth threshold is a pre-set percentage of the station-level target response power, typically 5% of the target response power. In the baseline scenario, if the full-power response level required by the emergency interaction mode is 50MW, the depth threshold is 2.5MW. When the actual achievable response depth of the distributed energy storage system falls below 47.5MW due to share transfers over multiple consecutive cycles, a response depth-limited signal is triggered. In the next cycle S1132, when performing proportional correction, this depth threshold is used as a constraint to ensure that the corrected weight vector prioritizes the share of devices with sufficient available power, preventing further reduction in response depth attenuation during the correction process.

[0084] Specifically, S12 includes sub-steps S121-S123.

[0085] S121. The total available reactive power capacity of the distributed energy storage system is pre-divided into normal reactive power capacity and emergency reactive power reserve. The division ratio between normal reactive power capacity and emergency reactive power reserve is periodically updated based on the statistical results of historical low voltage ride-through events of the power grid. The division ratio update process of S121 is described in subsequent blocks.

[0086] S122. In the normal low-priority mode, the emergency reactive power reserve is blocked, so that the voltage regulation mode triggered by the voltage dispatch command only performs reactive power regulation within the normal reactive power capacity. The blocking is achieved by marking the emergency reactive power reserve field as blocked in the cross-mode shared data pool. When reading the reactive power capacity classification parameters during the voltage regulation mode, only the normal reactive power capacity field can be accessed, and the emergency reactive power reserve field is not visible to the voltage regulation mode.

[0087] S123. When the target operating mode switches from voltage regulation mode to dynamic reactive power response mode triggered by dynamic reactive power response command, the emergency reactive power reserve is unlocked, allowing the dynamic reactive power response mode to perform reactive power regulation within the sum of normal reactive power capacity and emergency reactive power reserve. After the low-voltage ride-through event corresponding to the dynamic reactive power response mode ends, as part of the normal low-priority mode continuation based on mode snapshot, the emergency reactive power reserve is re-locked, and the target voltage value corresponding to the voltage regulation mode is restored from the mode snapshot. The determination of the end of the low-voltage ride-through event is based on the local bus voltage recovering to above a preset recovery threshold and maintaining a preset stable duration.

[0088] In the baseline scenario of a total station capacity of 30 Mvar, the normal reactive power capacity is taken as 18 Mvar, and the emergency reactive power reserve is taken as 12 Mvar. During the voltage regulation mode, the reactive power regulation is limited to the range of -18 Mvar to +18 Mvar. After a low-voltage ride-through event is triggered, the state machine switches to dynamic reactive power response mode, and the reactive power regulation range is expanded to -30 Mvar to +30 Mvar, enabling the distributed energy storage system to have the transient reactive power support capability of full station capacity. After the event ends, the target voltage value corresponding to the voltage regulation mode is restored from the fixed value in the mode snapshot, the emergency reactive power reserve is re-locked, and the voltage regulation mode returns to steady-state regulation within the range of 18 Mvar.

[0089] In dynamic reactive power response mode, the allocation of available reactive power capacity across all power conversion devices is based on a weighted average of the current active power output and the reactive power output margin corresponding to the rated capacity of each device. The reactive power output margin is calculated based on the coupling relationship between the active and reactive power capacities of the power conversion devices, with those having larger reactive power output margins undertaking a larger share of reactive power regulation. Based on the coupling relationship of the PQ circle diagram, the [missing information - likely a specific configuration or scale]... Reactive power output margin of power conversion equipment satisfy: in, It is the first The rated capacity of the power conversion equipment It is the first The current active power output of the power conversion equipment. The reactive power regulation share of each piece of equipment is calculated as follows: The reactive power output margin is allocated as a ratio to the total reactive power output margin of the entire station. Assume that in the baseline scenario, the rated capacity of each of the 10 devices is 5.5 MVA, with the current active power output of devices 1 to 5 each being 4 MW, and the current active power output of devices 6 to 10 each being 1 MW. Substituting into the formula, the reactive power output margin of devices 1 to 5 is... Rounded to two decimal places; numbers 6 through 10 are all... Round to two decimal places. The total reactive power output margin of the entire station is 5 × 3.77 + 5 × 5.41 = 45.90 Mvar. When the entire station needs to inject 30 Mvar of reactive power, the first 5 units each bear the load. The last 5 units each undertake Devices with smaller active power outputs have a larger PQ circle diagram coupling margin and therefore bear a larger share of reactive power.

[0090] Furthermore, the update of the split ratio in S121 includes sub-steps S1211-S1213.

[0091] S1211. Maintain a sliding statistical window for historical low-voltage ride-through events. Each historical low-voltage ride-through event includes the emergency reactive power capacity injected by the distributed energy storage system at the time of the event. The sample size of the sliding statistical window is the total number of low-voltage ride-through events recorded in the power grid area where the distributed energy storage system is located within the past 6 months, with a typical sample size of 20 to 50 events.

[0092] S1212. Based on the distribution of all emergency reactive power capacity values ​​within the sliding statistical window, the pre-set quantile value of the distribution is taken as the target value of emergency reactive power reserve for the next period. The method of selecting the quantile value supports one of several ranking methods, such as taking the integer part, i.e., taking the quantile after sorting. The sample value of the bit, where This represents the total number of samples. The quantile value is typically taken as the 95th quantile, ensuring that the emergency reactive power reserve can cover 95% of the capacity required for historical low voltage ride-through events.

[0093] S1213. Use the difference between the available reactive power capacity of the entire station and the target value as the normal reactive power capacity for the next cycle, and synchronize the target value and the normal reactive power capacity to the reactive power capacity classification parameters in the cross-mode shared data pool.

[0094] Construct a statistically driven adjustment condition. Assume a sliding statistical window contains 20 historical low-voltage ride-through events. The emergency reactive power capacity values ​​(in Mvar) of these events are sorted from smallest to largest as follows: 5, 6, 7, 8, 8, 9, 9, 10, 10, 10, 11, 11, 11, 12, 12, 12, 13, 13, 14, 15. The 95th percentile value is taken as the 19th value after sorting (20 × 0.95 = 19), which is 14 Mvar. S1212 outputs the target value of the emergency reactive power reserve for the next cycle as 14 Mvar. S1213 calculates the normal reactive power capacity for the next cycle as 30 Mvar - 14 Mvar = 16 Mvar, and writes 14 Mvar and 16 Mvar into the reactive power capacity classification parameters of the cross-mode shared data pool. The adjustable reactive power capacity range of the voltage regulation mode in the next cycle is correspondingly reduced to -16 Mvar to +16 Mvar, and the emergency reactive power reserve is expanded to 14 Mvar.

[0095] When the sample size is insufficient to support quantile estimation (typically less than 10 samples), the sample size is split according to a pre-set safety net ratio, typically the emergency reactive power reserve is 40% of the total available reactive power capacity of the station.

[0096] The frequency disturbance identification thresholds include the frequency deviation dead zone value, the rate of change threshold, and the duration threshold. Before performing priority arbitration, S2 performs frequency disturbance identification actions on the local station's frequency based on the frequency disturbance identification thresholds in the cross-mode shared data pool. The frequency disturbance identification action adopts a three-factor joint criterion, which includes the frequency deviation amplitude criterion, the frequency deviation rate of change criterion, and the frequency deviation duration criterion. The frequency deviation amplitude criterion is based on the frequency deviation dead zone value, the frequency deviation rate of change criterion is based on the rate of change threshold, and the frequency deviation duration criterion is based on the duration threshold. S2 outputs the valid identification result of the frequency response command only when all three criteria are met.

[0097] Frequency difference amplitude criterion is ,in This is the frequency of this site. It is the rated frequency of the power grid. It is the frequency dead zone value; the criterion for the rate of change of frequency is ,in It is the rate of change threshold; the frequency deviation duration criterion is the duration for which the frequency deviation amplitude criterion is continuously satisfied is greater than or equal to the duration threshold. .

[0098] The frequency difference duration criterion is determined based on a sliding sampling window of the local station frequency. The window length of the sliding sampling window is equal to the current effective value of the duration threshold, and the sampling period is equal to the frequency response scan period of the power conversion coordination control host. The frequency difference duration criterion is satisfied when all sampling points within the sliding sampling window satisfy the frequency difference amplitude criterion; if any sampling point does not satisfy the frequency difference amplitude criterion, the sliding sampling window is reset. In the baseline scenario, the frequency response scan period of the power conversion coordination control host is 10ms. If the current effective value of the duration threshold is 100ms, the sliding sampling window contains 10 sampling points, and the valid duration determination result is output only when all 10 sampling points satisfy the frequency difference amplitude criterion.

[0099] The frequency deviation amplitude criterion and frequency deviation duration criterion are determined using an amplitude-time linked threshold. This threshold makes the duration threshold inversely correlated with the frequency deviation amplitude. The first, second, and third amplitude intervals are all located above the frequency deviation dead zone value and are arranged in ascending order of amplitude. When the frequency deviation amplitude is in the first amplitude interval, the duration threshold takes a larger value; when it's in the second amplitude interval, it takes a middle value; and when it's in the third amplitude interval, it takes a smaller value. This ensures that large-amplitude short-duration disturbances meet the triggering condition, small-amplitude long-duration disturbances meet the triggering condition, and small-amplitude short-duration disturbances do not meet the triggering condition. Let the frequency deviation dead zone value be... The specific configuration of the amplitude-time linkage threshold is as follows: the first amplitude range is... The corresponding duration threshold is 200ms; the second amplitude range is... The corresponding duration threshold is 100ms; the third amplitude range is... The corresponding duration threshold is set to 30ms.

[0100] Three typical disturbance scenarios are constructed. Scenario 1 is a large-amplitude, short-duration disturbance with a frequency difference of 0.30Hz lasting for 50ms before falling back to the third amplitude range. The threshold requirement is 30ms, and the actual duration of 50ms satisfies the criterion, resulting in a valid identification result. Scenario 2 is a small-amplitude, long-duration disturbance with a frequency difference of 0.06Hz lasting for 250ms before falling back to the first amplitude range. The threshold requirement is 200ms, and the actual duration of 250ms satisfies the criterion, resulting in a valid identification result. Scenario 3 is also a small-amplitude, short-duration disturbance with a frequency difference of 0.06Hz lasting for 50ms before falling back to the first amplitude range. The threshold requirement is 200ms, but the actual duration of 50ms does not satisfy the criterion, so no valid identification result is output. Scenario 3 typically corresponds to high-instantaneous but short-duration pseudo-disturbances such as scheduling step transitions and grid connection / de-station transients.

[0101] A pre-filtering layer for a pseudo-disturbance fingerprint database is set before the three-factor joint criterion. The pseudo-disturbance fingerprint database stores the feature vector set of common pseudo-disturbance events of the distributed energy storage system and the upstream power grid. Pseudo-disturbance events include at least one of the following: power step events of the upper-level dispatch system, grid connection events or disconnection events of the upstream power grid, reactive power compensation equipment operation events of the system, and one-click power distribution events of the system. The pre-filtering layer of the pseudo-disturbance fingerprint database performs similarity matching between the feature vector of the current disturbance event and each feature vector in the pseudo-disturbance fingerprint database. Only when the similarity between the feature vector of the current disturbance event and all feature vectors in the pseudo-disturbance fingerprint database is lower than a preset similarity threshold, the current disturbance event is input into the three-factor joint criterion for further identification. The pseudo-disturbance fingerprint database is periodically updated by a self-tuning calculation thread of the frequency disturbance identification threshold based on historical pseudo-disturbance samples during idle computing time. The feature vector is constructed according to a pre-defined set of feature dimensions, which includes the initial amplitude of the disturbance, the time it takes for the disturbance to reach its peak, the duration of the disturbance, the correlation between frequency and active power during the disturbance, and whether the end of the disturbance is accompanied by local control actions. The typical similarity threshold is 0.85.

[0102] Specifically, S13 includes sub-steps S131-S132.

[0103] S131. During the idle calculation time between the active power scheduling cycle and the frequency response cycle in the normal low-priority mode, run the self-tuning calculation thread for the frequency disturbance identification threshold. The self-tuning process of S131 is described in subsequent blocks.

[0104] S132. Based on the output of the self-tuning calculation thread, the frequency difference dead zone value, rate of change threshold, and duration threshold are periodically synchronized to the cross-mode shared data pool.

[0105] Furthermore, S131 includes sub-steps S1311-S1313.

[0106] S1311. Maintain a sliding window for historical disturbance samples. The collection of historical disturbance samples is continuous under all operating modes. Each historical disturbance sample includes the disturbance amplitude, disturbance duration, the identification action result of S2, and the power tracking deviation after the identification action. The number of samples in the sliding window is determined based on the monthly average disturbance sample volume of the power grid area where the distributed energy storage system is located. The typical sample size is the total number of disturbance samples in the past month, approximately 200.

[0107] S1312. The objective function is to minimize the sum of the non-response true perturbation rate and the false response cost. Based on historical perturbation samples within a sliding window, it solves for the optimal combination of frequency difference dead zone, rate of change threshold, and duration threshold. The non-response true perturbation rate is the ratio of unidentified true perturbation samples within the sliding window, and the false response cost is the cumulative value of power tracking deviation after an action is identified within the sliding window. The optimal combination of the objective function in S1312 is solved using either a grid search or gradient descent approach: the grid search approach performs discrete grid sampling within a pre-defined range for each of the three thresholds at a pre-defined step size, calculates the objective function value for each sampling point, and selects the sampling point with the smallest objective function value as the optimal combination; the gradient descent approach treats the objective function as a continuous function with respect to the three thresholds, iteratively updating it along the opposite direction of the partial derivatives of the objective function with respect to each threshold at a pre-defined learning rate until the difference between the objective function values ​​between two consecutive iterations is less than a pre-defined convergence threshold. The objective function solution for S1312 only runs in the normal low-priority mode.

[0108] Let the objective function be... satisfy: in, It is the non-response true perturbation rate. It is the cost of a false response. , It is the corresponding cost weight, and . It equals the ratio of the number of samples marked as true perturbations but not identified by S2 in the sliding window to the total number of true perturbation samples in the sliding window; This is equal to the cumulative absolute value of the power tracking deviation after the action is identified within the sliding window. In the grid search format, the frequency difference dead zone ranges from 0.033Hz to 0.10Hz with a step size of 0.005Hz; the rate of change threshold ranges from 0.05Hz / s to 0.30Hz / s with a step size of 0.025Hz / s; and the duration threshold ranges from 30ms to 300ms with a step size of 10ms. After sampling the 3D grid, the grid point with the smallest objective function value is selected as the optimal combination.

[0109] S1313. Based on the optimal combination obtained in S1312, an adaptive load factor correction is added. When the local load factor of the distributed energy storage system is lower than a pre-set load factor threshold, the frequency dead zone value is reduced; when the local load factor is higher than the load factor threshold, the frequency dead zone value is increased. The local load factor is the ratio of the current actual output power of the distributed energy storage system to its rated power. The frequency dead zone value after the adaptive load factor correction satisfies: in, It is the frequency deviation dead zone value after adaptive correction of load rate. It is the frequency difference dead zone value output by S1312. This is the station's load rate. It is the load factor threshold. It is a correction factor. hour If it is negative, then... Distributed energy storage systems prioritize sensitivity. hour For positive, make Distributed energy storage systems prioritize battery life. The typical load factor threshold is 0.5. Let... It is 0.05Hz. The value is 0.5. When the station's load rate is 0.3, it is below the threshold. Substituting this into the formula, we get... When the station's load factor is 0.8, it is higher than the threshold. Substituting this into the formula, we get... The Hz value is 0.058Hz, rounded to three decimal places.

[0110] After the self-tuning calculation thread of the frequency disturbance identification threshold generates a new threshold, the new threshold does not directly overwrite the currently effective value in the cross-mode shared data pool. Instead, it first enters the shadow identification channel. The shadow identification channel uses the new threshold to identify real-time frequency disturbance samples, but the identification result does not trigger any action. After the shadow identification channel runs for a preset grayscale duration, it compares the difference between the shadow identification result and the formal identification result. When the difference rate falls within a preset tolerance range, the new threshold is submitted for human-in-the-loop approval. When the difference rate exceeds the tolerance range, the new threshold is discarded, and the difference is fed back as a counterexample to the objective function of the next cycle of the self-tuning calculation thread. The typical grayscale duration is 24 hours. The difference rate is defined as the ratio of the number of samples whose identification results of the shadow channel and the formal channel are inconsistent within the grayscale duration to the total number of samples within that duration. The upper limit of the tolerance range is typically 5%. When the difference rate does not exceed 5%, the new threshold is submitted for human-in-the-loop approval; when the difference rate exceeds 5%, the new threshold is discarded; when the difference rate is 0%, the identification results of the old and new thresholds are completely consistent for all samples. Since the self-tuning output of the weighted coefficients cannot be verified through the shadow recognition channel, a simplified path of direct submission and in-loop approval is adopted, bypassing the shadow recognition channel.

[0111] Each parameter in the cross-mode shared data pool maintains two sets of values: the currently effective value and the self-tuning recommended value. The self-tuning recommended value is available for operators to view through the monitoring interface and switch to the currently effective value via the approval interface. The switching action is recorded in the version log of the cross-mode shared data pool. When the approval interface is triggered during the high-priority functional mode, the parameter version switch is delayed until the state machine returns to the normal low-priority mode. The version log records the parameter name, old value, new value, approver identifier, and switch timestamp for each switch in chronological order, making each parameter change traceable.

[0112] During the high-priority functional mode, the power allocation weight vector, reactive capacity classification parameters, and frequency disturbance identification threshold in the cross-mode shared data pool are not recalculated on-site. However, the sample acquisition thread belonging to the normal low-priority mode continues to collect the latest state of charge, latest temperature, latest available power, and historical disturbance samples of each power conversion device under all operating modes. The latest collected data is stored in the offline observation area of ​​the cross-mode shared data pool. The latest data in the offline observation area is only read by the weight recalculation of the next cycle S111-S113 during the normal low-priority mode transition phase after the high-priority functional mode ends. Under extreme conditions with a 5-minute main timeout duration and a maximum of 3 extensions, the emergency interaction mode can last up to 20 minutes. During this period, the offline observation area receives approximately 1200 second-level sampling points, allowing the subsequent S111-S113 recalculation to be corrected based on the latest data rather than on the old values ​​before the event was triggered.

[0113] S2. Receive concurrent control commands for multiple functional modes within the distributed energy storage system, and perform mutual exclusion priority arbitration on the concurrent control commands based on a pre-set functional mode priority sequence to obtain the target operating mode. The mutual exclusion priority arbitration ensures that only one functional mode is active at any given time, and the functional mode priority sequence sets the priority of emergency interaction commands to the highest, the priority of rapid response commands to the second highest, and the priority of normal scheduling commands corresponding to the normal low-priority mode to the lowest. At the instant the target operating mode switches from the normal low-priority mode to the high-priority functional mode, the target power value, target voltage value, and power allocation weight vector corresponding to the normal low-priority mode before the switch are fixed as a mode snapshot.

[0114] In its physical implementation, the mode snapshot is an independent memory region within a cross-mode shared data pool, physically isolated from the currently effective parameters and self-tuning recommended values. At the moment of mode switching, the state machine copies the currently effective target power value, target voltage value, and power allocation weight vector to the mode snapshot region. This copying operation is completed atomically. In the baseline scenario of 50MW / 10 power conversion devices, the mode snapshot contains a target power floating-point value, a target voltage floating-point value, and a power allocation weight floating-point array of length 10, with a total data volume of approximately 96 bytes and an atomic write time of no more than 1ms.

[0115] Routine dispatching commands include voltage dispatching commands for buses of different voltage levels. Each voltage level bus dispatching command corresponds to a pre-set control step size and control dead zone, with the control step size and control dead zone increasing sequentially from low to high voltage level. Specifically, the configuration is as follows: 0.2kV control step size and 0.1kV control dead zone for 10kV buses; 0.3kV control step size and 0.15kV control dead zone for 35kV buses; and 0.8kV control step size and 0.3kV control dead zone for 110kV buses. Low-voltage level buses have finer voltage regulation granularity, while high-voltage level buses have a wider stable regulation range.

[0116] Fast-response commands include frequency response commands and dynamic reactive power response commands. The execution parameters of frequency response commands include response delay, response arrival time, inequality rate, and settling time. The response delay does not exceed a preset upper limit, typically not exceeding 100ms; the response arrival time is the time it takes for the output power to climb from the trigger of the detection action to the power value corresponding to the preset arrival ratio, typically 90% and with a typical response arrival time not exceeding 400ms; the inequality rate is the proportionality coefficient between frequency deviation and power response, typically settable from 0.001% to 0.1%; the settling time is the duration the frequency response process sustains, typically settable from 1 minute to 3 minutes. All execution parameters can be configured via an engineering-based monitoring interface, allowing the frequency response capability of the distributed energy storage system to be flexibly configured according to the specific parameter requirements of the grid dispatcher. The performance constraints of the dynamic reactive power response mode include that the time from voltage drop triggering to the reactive power output reaching the target reactive power regulation amount does not exceed the preset transient response time limit, with a typical response rate of 90% and a typical transient response time limit of no more than 30ms.

[0117] Emergency interaction commands include emergency interaction response commands and load shedding level commands. Performance constraints for emergency interaction mode include a pre-set response rate from the triggering of the emergency interaction response command to the output power reaching full power response level, with the time not exceeding a pre-set upper limit for emergency response duration. A typical response rate is 90%, and the typical upper limit for emergency response duration is no more than 100ms. When S2 receives a load shedding level command, it performs a load shedding error prevention judgment before executing mode switching: Based on a pre-stored hierarchy-frequency interval mapping table, it determines the target frequency interval corresponding to the load shedding level command; it uses a three-factor joint criterion to determine the station's frequency, obtaining the station's frequency determination result; it only executes mode switching corresponding to the load shedding level command when the station's frequency falls within the target frequency interval and the station's frequency determination result is valid; when the station's frequency determination result is valid but the station's frequency does not fall within the target frequency interval, it sends a criterion inconsistency alarm back to the control center and temporarily suspends mode switching; when the station's frequency falls within the target frequency interval but the station's frequency determination result is invalid, it executes mode switching with half the action amount corresponding to the load shedding level command and continues monitoring the station's frequency. The specific content of the hierarchy and frequency range mapping table is as follows: the first level of load shedding corresponds to the frequency range of 49.0Hz to 49.25Hz, the second level of load shedding corresponds to the frequency range of 48.7Hz to 49.0Hz, and the third level of load shedding corresponds to the frequency range below 48.7Hz. The higher the level, the more serious the frequency deviation.

[0118] Concurrent control commands also include group control commands. Group control commands include panoramic monitoring commands, one-click power issuance commands, and veto signals for individual power conversion devices. Panoramic monitoring commands are used to acquire real-time operating status parameters and battery data for each power conversion device. Operating status parameters include active power output, reactive power output, bus voltage, current, and power factor. Battery data includes SOC, temperature, voltage, and current. One-click power issuance commands are used to issue batch power commands to all or some power conversion devices in normal low-priority mode, allowing operators to issue unified power commands to multiple power conversion devices at once. The use of veto signals is explained in the subsequent local veto block.

[0119] Emergency interaction commands include emergency interaction response commands, and high-priority function modes include emergency interaction modes triggered by emergency interaction response commands. S2 also includes sub-steps S21-S23 when the target operating mode is emergency interaction mode. The timing of the emergency interaction mode activation and deactivation process is shown in Figure 4.

[0120] S21. After the emergency interaction mode takes effect, start the main timeout timer and initialize the extension count counter to zero. During the pre-set secondary confirmation window period before the main timeout timer reaches the pre-set main timeout duration, collect the local frequency and voltage samples of the distributed energy storage system and perform matching determination between the local frequency and voltage samples and the normal range of the power grid. When all local frequency and voltage samples fall into the normal range of the power grid within the secondary confirmation window period, exit the emergency interaction mode when the main timeout timer reaches the main timeout duration. When any local frequency or voltage sample does not fall into the normal range of the power grid within the secondary confirmation window period, reset the main timeout timer and increment the extension count counter by 1. When the extension count counter reaches the pre-set maximum extension count, report a continuous abnormal alarm and return the decision to exit the emergency interaction mode to the operators.

[0121] The primary timeout duration is 5 minutes under the baseline scenario, and the secondary confirmation window is the last 30 seconds before the primary timeout ends. During the secondary confirmation window, the sampling period for the station's frequency and voltage samples is 1 second, with a total of 30 frequency samples and 30 voltage samples collected. The maximum number of extensions is 3. The extension scenario is constructed as follows: the emergency interaction mode activates at second 0, the primary timeout timer starts counting, and the secondary confirmation window is from second 270 to second 300. If the 18th sample out of the 30 frequency samples within the window measures a station frequency of 49.92Hz (below the lower limit of the normal grid range of 49.95Hz), the emergency interaction mode is not exited at second 300; instead, the primary timeout timer is reset to second 0, and the extension count counter increments from 0 to 1. A new primary timeout begins counting again. If the extension count counter reaches 3 after 3 rounds of extensions and secondary confirmation is still not passed, a continuous abnormality alarm is reported, and the operator manually decides whether to exit the emergency interaction mode via the monitoring interface.

[0122] S22. After exiting the emergency interaction mode, a transition state is entered. In the transition state, the output power in the emergency interaction mode is reduced back to the target power value in the mode snapshot using a slope-limited ramp function. The slope of the ramp function is adaptively selected based on the deviation of the local station frequency and voltage from the normal range of the power grid; the closer the deviation is to the normal range of the power grid, the larger the slope. When an emergency interaction response command is received again during the transition state, the slope function's reduction process is interrupted, and the output power of the power conversion equipment is restored to the full-power response level corresponding to the emergency interaction mode.

[0123] The slope of the ramp function is adaptively selected based on the deviation of the station frequency from the center 50Hz of the normal grid range: a larger slope and a transition time of 500ms are used when the deviation is less than 0.02Hz; a medium slope and a transition time of 1000ms are used when the deviation is between 0.02Hz and 0.05Hz; and a smaller slope and a transition time of 2000ms are used when the deviation is between 0.05Hz and 0.10Hz. Assuming the output power in emergency interaction mode is full power 50MW, and the target power value in the mode snapshot is 30MW, the transition requires a drop of 20MW. In the baseline scenario, if the station frequency is 49.97Hz when the transition state starts, with a deviation of 0.03Hz from the center frequency of 50Hz, a medium slope is used, the transition time is 1000ms, and the slope is 20MW / s. The output power drops from 50MW to 30MW linearly within 1000ms.

[0124] A secondary intrusion interruption scenario is constructed. 600ms after the transition state is initiated, the output power has dropped from 50MW to approximately 38MW. At this point, an emergency interactive response command is received again, and the ramp function is immediately interrupted. The output power is then increased from the current 38MW back to the full power level of 50MW in the emergency interactive mode. The secondary increase is executed at the ramp rate of the upper limit of the emergency response duration, enabling the distributed energy storage system to continue providing grid support under secondary emergency events.

[0125] S23. After the transition state ends, the current operating mode of the distributed energy storage system is marked as the normal low-priority mode, and the target power value, target voltage value, and power allocation weight vector corresponding to the normal low-priority mode are restored based on the mode snapshot. After the mode snapshot restoration operation is completed, the mode snapshot area is released, the cross-mode shared data pool returns to normal operation, and a new snapshot is written during the next mode switch.

[0126] S3. During the period when the target operating mode is in effect, execution instructions are issued to the multiple power conversion devices based on the parameters in the cross-mode shared data pool corresponding to the target operating mode, and the parameters in the cross-mode shared data pool are not recalculated during the period when the high-priority function mode is in effect; after the response process corresponding to the high-priority function mode ends, the preempted normal low-priority mode is resumed based on the mode snapshot, and the resumption does not depend on the re-issuance of external host computer instructions.

[0127] The mapping relationship between the target operating mode and the corresponding parameters in the cross-mode shared data pool is as follows: when the active power scheduling mode, frequency response mode, and emergency interaction mode are active, the power allocation weight vector is read; when the voltage regulation mode and dynamic reactive power response mode are active, the reactive power capacity classification parameters are read; the effective identification action of the frequency response command and the load shedding error prevention judgment action are read, along with the frequency disturbance identification threshold. Each mode generates and issues execution commands according to the corresponding parameters, and the targets of these commands are each power conversion device.

[0128] Construct a scenario where the active power dispatch mode takes effect. Under the baseline operating condition of a target power value of 30MW, S3 reads the power allocation weight vector (values ​​are 0.087, 0.090, 0.093, 0.096, 0.099, 0.101, 0.104, 0.107, 0.110, and 0.113 respectively), calculates the active power output commands of the 1st to 10th power conversion devices according to the weights, which are 2.61MW, 2.70MW, 2.79MW, 2.88MW, 2.97MW, 3.03MW, 3.12MW, 3.21MW, 3.30MW, and 3.39MW respectively, and sends them to each power conversion device for execution.

[0129] Construct a scenario where the frequency response mode takes effect. After the frequency response command is triggered, S3 reads the power allocation weight vector (the value is consistent with the active power dispatch mode mentioned above), and recalculates the weights without recalculating them in the frequency response mode. Multiplying the total power of the frequency response by the weight vector yields frequency response increments of 0.87MW, 0.90MW, 0.93MW, 0.96MW, 0.99MW, 1.01MW, 1.04MW, 1.07MW, 1.10MW, and 1.13MW for each power conversion device. These increments are then superimposed on the original active power output and sent out for execution.

[0130] After the emergency interaction event ends, S3 performs a follow-up operation based on the mode snapshot, restoring the target power value, target voltage value, and power allocation weight vector in the snapshot to the current effective area of ​​the cross-mode shared data pool. This allows the distributed energy storage system to continue operating according to the normal operating parameters before the event was triggered immediately after the transition state ends, without waiting for the upper-level dispatch system to reissue any instructions.

[0131] Concurrent control instructions also include group control instructions. When S3 issues execution instructions to the power conversion device, it is also subject to the local veto power of the group control instructions. Specifically, the application of the local veto power includes sub-steps S31-S33.

[0132] S31. Receive a rejection signal from a group control command for a single power conversion device. The rejection signal is triggered when the state of charge or temperature of a single power conversion device exceeds the corresponding safe range. The safe range for the state of charge is typically 0.10 to 0.95, and the safe range for the temperature is typically -10°C to 50°C. When either parameter exceeds the safe range, the group control command sends a rejection signal to S3 for that power conversion device.

[0133] S32. Set the share of a single power conversion device in the power response allocation base to zero without affecting the overall effectiveness of the target operating mode, and redistribute the share to other available power conversion devices in the distributed energy storage system according to the nearest transfer rule.

[0134] Construct an abnormal operating condition for a single power converter's State of Charge (SOC). In active power dispatch mode, the SOC of the third power converter drops to 0.08 due to continuous discharge, falling below the lower limit of the safe range (0.10). Group control commands issue a rejection signal. S32 sets the weight of the third converter in the power response allocation base from 0.093 to zero, and its share is redistributed to the physically adjacent second and fourth converters according to the nearest transfer rule, with each receiving 0.0465. The active power output command of the third converter is set to 0, and the remaining nine converters continue to execute the original active power dispatch mode according to the adjusted weights. The overall target operating mode remains unaffected.

[0135] S33. When the target operating mode is emergency interaction mode, the local veto power is applied only to a single power conversion device, with its share set to zero. When the nearest transfer rule causes the total response depth of the distributed energy storage system to decrease beyond a pre-set emergency depth threshold due to the limited available power of other available power conversion devices, a response depth limitation signal is sent back to the upper-level dispatch system, and the application of subsequent veto signals is temporarily suspended. The emergency depth threshold is a pre-set percentage of the full-power response level corresponding to the emergency interaction mode, typically 5% of the full-power response level. In a 50MW full-power response level benchmark scenario, the emergency depth threshold is 2.5MW. When multiple consecutive power conversion devices trigger veto signals in emergency interaction mode, the nearest transfer rule cannot absorb the transfer share due to the limited available power of other available devices, or the actual response depth of the distributed energy storage system is less than 47.5MW, a response depth limitation signal is sent back, and the application of subsequent veto signals is temporarily suspended, ensuring that the distributed energy storage system prioritizes grid support in emergency events.

[0136] Reference Figure 1 The power regulation method based on large-scale distributed energy storage systems comprises an energy storage monitoring system and a power conversion coordination control host. The energy storage monitoring system receives and arbitrates routine dispatch and group control commands on minute and second timescales, while the power conversion coordination control host identifies and issues frequency response and dynamic reactive power response commands on millisecond timescales. Both systems jointly receive and execute emergency interactive response commands. The primary copy of the cross-mode shared data pool is maintained by the energy storage monitoring system, while the power conversion coordination control host maintains a local copy. The local copy is periodically updated from the primary copy through the routine parameter synchronization channel in the hierarchical heartbeat synchronization mechanism.

[0137] The energy storage monitoring system maintains a second-level state machine. The state space of this second-level state machine includes active power dispatch mode, voltage regulation mode, group control batch issuance mode, and second-level idle mode. State transition trigger conditions for the second-level state machine include the receipt of normal dispatch commands, the receipt of group control commands, and the end signal of the current mode's response process. The power conversion coordination control host maintains a millisecond-level state machine. The state space of this millisecond-level state machine includes frequency response mode, dynamic reactive power response mode, emergency interaction mode, and millisecond-level idle mode. State transition trigger conditions for the millisecond-level state machine include the valid identification result of frequency disturbance identification actions, the receipt of dynamic reactive power response commands, the receipt of emergency interaction response commands, and the end signal of the current mode's response process. The second-level and millisecond-level state machines make state switching decisions independently. When the millisecond-level state machine preempts a normal dispatch function mode, it does not wait for a response from the second-level state machine but directly executes the response based on the currently active parameters in the cross-mode shared data pool. Within the state machines of the same host, mutex locks ensure that only one function mode is active at any given time. The order of acquiring the mutex locks is ordered from high to low priority according to the function mode priority sequence. The second-level state machine and the millisecond-level state machine are asynchronously synchronized through the state version number in the cross-mode shared data pool. The mode snapshot at the moment of mode switching is written to the cross-mode shared data pool by the preemption initiator and then read asynchronously by the other host. When any host detects that the state version number maintained by its own host is inconsistent with the state version number of the other host, the host's decision is based on the latest version number of the millisecond-level state machine, and the current decision of the second-level state machine is put on hold until the state version number synchronization is completed.

[0138] Construct millisecond-level preemption scenarios. The second-level state machine currently in active power scheduling mode, with the state version number... The value is 100; the millisecond-level state machine is currently in millisecond-level idle mode, and the state version number is 100. It is 50. Among them, It is the state version number of the state machine, which is measured in seconds. This is the state version number of the millisecond-level state machine. When the frequency disturbance recognition action outputs a valid recognition result, the millisecond-level state machine immediately switches itself to frequency response mode, and the state version number is updated to [value missing]. It equals 51.

[0139] This switch does not wait for a second-level state machine response; the millisecond-level state machine directly reads the currently active power allocation weight vector maintained by the second-level state machine from the cross-mode shared data pool, and then... The execution frequency response is equal to 51 states. The difference in state version number is asynchronously transmitted to the second-level state machine via the state version synchronization channel, and the second-level state machine detects this. After changing from 50 to 51, the local decision is postponed to a millisecond-level state machine, returning to a millisecond-level idle mode, and the state version number returns to its previous state. The current decision is to restore the active power scheduling mode after 52.

[0140] The energy storage monitoring system and the power conversion coordination control host maintain consistency in the state of the cross-mode shared data pool, mode snapshots, and state machine through a hierarchical heartbeat synchronization mechanism. The hierarchical heartbeat synchronization mechanism is divided into three channels according to the data lifecycle: normal parameter synchronization channel, transient snapshot synchronization channel, and state version synchronization channel. The normal parameter synchronization channel synchronizes the cross-mode shared data pool at a second-level cycle, using a version number-based differential transmission method, and only transmits fields in the cross-mode shared data pool that have changed compared to the previous synchronization. The transient snapshot synchronization channel synchronizes mode snapshots in an event-triggered manner, transmitting the complete mode snapshot at the moment of mode switching. The state version synchronization channel synchronizes the state version number at a millisecond-level cycle, and each heartbeat message only contains the state version number.

[0141] All three synchronization channels are established on an industrial Ethernet link between the two hosts and ensure message delivery through a message sequence number-based acknowledgment mechanism. When the number of consecutive timeouts for any synchronization channel exceeds a pre-set synchronization timeout threshold, a fault degradation strategy for the corresponding channel is triggered. The fault degradation strategy includes at least one of the following: extension of the second-level period, local caching of transient snapshots for later retransmission after recovery, and degradation of state version synchronization to single-host local decision. The typical value for the synchronization timeout threshold is three consecutive timeouts. When the normal parameter synchronization channel times out three times consecutively, the second-level period is extended from the original 1 second to 5 seconds; when the transient snapshot synchronization channel times out three times consecutively, the transient snapshot is cached locally by the preemptive initiator and retransmitted to the other host in timestamp order after the communication link is restored; when the state version synchronization channel times out three times consecutively, the millisecond-level state machine decision is executed only based on the local effective parameters, and the second-level state machine decision is executed only based on the local effective parameters, enabling the two hosts to maintain their local control capabilities during communication interruptions.

[0142] This application also provides a computer device, which includes one or more processors, a memory, and one or more application programs. The one or more application programs are stored in the memory and configured to be executed by the one or more processors. The one or more application programs are configured to perform the aforementioned power regulation method based on a large-scale distributed energy storage system. The processors are connected to the memory via a bus. The processors load the instructions of the application programs from the memory and execute them, enabling the computer device to implement all the steps of the aforementioned method. In a specific implementation scenario, the computer device is deployed in an energy storage monitoring system and a power conversion coordination control host. The processors load application programs corresponding to the responsibilities of their respective hosts from their respective host memories to collaboratively complete the aforementioned power regulation method based on a large-scale distributed energy storage system.

[0143] This application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed to implement the above-described method. The computer-readable storage medium stores at least one instruction, at least one program segment, a code set, or an instruction set. The at least one instruction, the at least one program segment, the code set, or the instruction set is loaded and executed by the processor to implement the above-described power regulation method based on a large-scale distributed energy storage system. The computer-readable storage medium can be any non-volatile storage medium commonly found in the art, such as a read-only memory, random access memory, disk, optical disk, magnetic tape, or flash memory. After the computer program is loaded into the memory by the processor, the processor executes the instructions contained in the computer program to complete all the steps of the above-described method.

[0144] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0145] 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for power conditioning based on large scale distributed energy storage system, characterized in that, Includes the following steps: S1. Maintain a cross-mode shared data pool for a distributed energy storage system, wherein the distributed energy storage system includes multiple power conversion devices connected in parallel to the power grid, and the cross-mode shared data pool includes power allocation weight vectors for each power conversion device, reactive capacity classification parameters of the distributed energy storage system, and frequency disturbance identification thresholds. The cross-mode shared data pool is periodically updated by the normal low-priority mode of the distributed energy storage system during normal operation. S2. Receive concurrent control commands for multiple functional modes within the distributed energy storage system, and perform mutual exclusion priority arbitration on the concurrent control commands based on a pre-set functional mode priority sequence to obtain a target operating mode. The mutual exclusion priority arbitration ensures that only one functional mode is active at any given time, and the functional mode priority sequence sets the priority of emergency interaction commands to the highest, the priority of rapid response commands to the second highest, and the priority of normal scheduling commands corresponding to the normal low-priority mode to the lowest. At the instant the target operating mode switches from the normal low-priority mode to a high-priority functional mode, the target power value, target voltage value, and power allocation weight vector corresponding to the normal low-priority mode before the switch are fixed as a mode snapshot. S3. During the period when the target operating mode is in effect, based on the parameters in the cross-mode shared data pool corresponding to the target operating mode, an execution instruction is issued to the multiple power conversion devices, and the parameters in the cross-mode shared data pool are not recalculated during the period when the high-priority function mode is in effect; After the response process corresponding to the high-priority function mode ends, the normal low-priority mode that was preempted is resumed based on the mode snapshot, and the resumption does not depend on the re-issuance of external host computer instructions.

2. The method of claim 1, wherein, S1 includes the following sub-steps: S11. In the normal low-priority mode, the power allocation weight vector is updated periodically; S12. In the normal low-priority mode, the reactive power capacity classification parameters are updated periodically; S13. In the normal low-priority mode, the frequency disturbance identification threshold is updated periodically.

3. The method of claim 2, wherein, S11 includes the following sub-steps: S111. Collect the state of charge, temperature, and available power of each power conversion device; S112. Perform a three-factor weighting on each of the power conversion devices based on the state of charge, the normalized term of the temperature, and the normalized term of the available power to obtain an initial weight vector; S113. The initial weight vector is corrected to meet the station-level injected power accuracy through station-level closed-loop correction to obtain the power allocation weight vector; Furthermore, when the target operating mode switches from the normal low-priority mode to the high-priority functional mode, the high-priority functional mode directly reuses the power allocation weight vector obtained in S113 as the power response allocation basis in the high-priority functional mode, and does not re-execute S111 to S113 in the high-priority functional mode; in response to any power conversion device being unable to assume the share corresponding to any power conversion device in the power response allocation basis in the high-priority functional mode, the share is redistributed to other available power conversion devices in the distributed energy storage system according to a pre-set nearest-neighbor transfer rule.

4. The power regulation method based on a large-scale distributed energy storage system according to claim 3, characterized in that, S113 includes the following sub-steps: S1131. Calculate the deviation between the current actual station-level injected power of the distributed energy storage system and the target power value corresponding to the normal low-priority mode; S1132. In response to the deviation being greater than a preset accuracy threshold, the initial weight vector is proportionally corrected according to the relative direction of the deviation to obtain the power allocation weight vector; S1133. In response to the deviation being less than or equal to the accuracy threshold, the initial weight vector is directly determined as the power allocation weight vector; Furthermore, in response to the nearby transfer rule causing the total response depth of the distributed energy storage system to decrease beyond a preset depth threshold, a response depth limitation signal is sent back to the upper-level scheduling system, and the depth threshold is synchronously injected into the proportional correction constraint of S1132 in the next cycle.

5. The power regulation method based on a large-scale distributed energy storage system according to claim 2, characterized in that, S12 includes the following sub-steps: S121. The total available reactive power capacity of the distributed energy storage system is pre-divided into normal reactive power capacity and emergency reactive power reserve. The ratio of the normal reactive power capacity to the emergency reactive power reserve is periodically updated based on the statistical results of historical low voltage ride-through events of the power grid. S122. In the normal low-priority mode, the emergency reactive power reserve is blocked, so that the voltage regulation mode triggered by the voltage dispatch command only performs reactive power regulation within the normal reactive power capacity; S123. When the target operating mode is switched from the voltage regulation mode to the dynamic reactive power response mode triggered by the dynamic reactive power response command, the emergency reactive power reserve is unlocked, so that the dynamic reactive power response mode performs reactive power regulation within the range of the sum of the normal reactive power capacity and the emergency reactive power reserve, and the emergency reactive power reserve is re-locked after the low voltage ride-through event corresponding to the dynamic reactive power response mode ends.

6. The power regulation method based on a large-scale distributed energy storage system according to claim 2, characterized in that, The fast response type instruction includes a frequency response instruction, and S13 includes the following sub-steps: S131. During the idle calculation time between the active power scheduling cycle and the frequency response cycle in the normal low priority mode, the self-tuning calculation thread of the frequency disturbance identification threshold is run. S132. The frequency disturbance identification threshold includes a frequency deviation dead zone value, a rate of change threshold, and a duration threshold. A three-factor joint criterion is applied to the local station frequency. The three-factor joint criterion includes a frequency deviation amplitude criterion, a frequency deviation rate of change criterion, and a frequency deviation duration criterion. The frequency deviation amplitude criterion is based on the frequency deviation dead zone value, the frequency deviation rate of change criterion is based on the rate of change threshold, and the frequency deviation duration criterion is based on the duration threshold. S133. Only when all of the frequency difference amplitude criterion, the frequency difference change rate criterion, and the frequency difference duration criterion are satisfied, S13 outputs the valid identification result of the frequency response command.

7. The power regulation method based on a large-scale distributed energy storage system according to claim 1, characterized in that, The emergency interaction commands include emergency interaction response commands, and the high-priority function modes include emergency interaction modes triggered by the emergency interaction response commands. When the target operating mode is the emergency interaction mode, step S2 further includes the following sub-steps: S21. After the emergency interaction mode takes effect, start the main timeout timer. During the pre-set secondary confirmation window period before the main timeout timer reaches the pre-set main timeout duration, collect the local frequency sample and local voltage sample of the distributed energy storage system, and perform a matching determination between the local frequency sample and the local voltage sample and the normal range of the power grid. In response to all the local frequency sample and all the local voltage sample falling into the normal range of the power grid within the secondary confirmation window period, exit the emergency interaction mode when the main timeout timer reaches the main timeout duration. In response to any local frequency sample or any local voltage sample not falling into the normal range of the power grid within the secondary confirmation window period, start the extension timer and reset the main timeout timer. In response to the cumulative triggering number of the extension timer reaching the pre-set maximum extension number, report a continuous abnormal alarm and return the exit decision of the emergency interaction mode to the operators. S22. After exiting the emergency interaction mode, a transition state is entered. In the transition state, the output power of the emergency interaction mode is reduced back to the target power value in the mode snapshot using a slope-limited ramp function. The slope of the ramp function is adaptively selected based on the deviation of the local station frequency and local station voltage relative to the normal range of the power grid. The closer the deviation is to the normal range of the power grid, the larger the slope. In response to receiving the emergency interaction response command again during the transition state, the reduction process of the ramp function is interrupted, and the output power of the power conversion device is increased back to the full power response level corresponding to the emergency interaction mode. S23. After the transition state ends, the current operating mode of the distributed energy storage system is marked as the normal low-priority mode, and the target power value, the target voltage value and the power allocation weight vector corresponding to the normal low-priority mode are restored based on the mode snapshot.

8. The power regulation method based on a large-scale distributed energy storage system according to claim 3, characterized in that, The concurrent control instructions also include group control instructions. When S3 issues the execution instruction to the power conversion device, it is also subject to the local veto power of the group control instructions. The application of the local veto power includes the following sub-steps: S31. Receive the rejection signal of the group control command for a single power conversion device, the rejection signal being triggered when the state of charge or temperature of the single power conversion device exceeds the corresponding safe range; S32. Set the share of the individual power conversion device in the power response allocation base to zero without affecting the overall effectiveness of the target operating mode, and redistribute the share to other available power conversion devices in the distributed energy storage system according to the nearest transfer rule; S33. In response to the target operating mode being an emergency interaction mode, the local veto power is used to set the share to zero only for the individual power conversion device, and the total response depth of the distributed energy storage system is attenuated to no more than a preset emergency depth threshold due to the application of the local veto power; in response to the attenuation exceeding the emergency depth threshold, a response depth limited signal is sent back to the upper-level scheduling system and the application of subsequent veto signals is suspended.

9. A computer device, characterized in that, It includes: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: perform the power regulation method based on a large-scale distributed energy storage system according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by a processor to implement: the power regulation method based on a large-scale distributed energy storage system as described in any one of claims 1 to 8.

Images