A multi-level networked energy storage control and protection system and method

Abstract

This invention discloses a multi-level network-type energy storage control and protection system and method, relating to the field of energy storage technology. The method includes: S1, performing feature modeling on the multi-source operating signals of the acquired energy storage system to construct a multi-source operating feature set; S2, performing time series rate of change analysis on the multi-source operating feature set to extract dynamic features of current direction change and power reversal; S3, performing energy disturbance amplitude analysis on the dynamic energy feature set to calculate an adaptive protection threshold set in real time; S4, comparing the real-time sampled signal with the adaptive protection threshold set to identify whether an over-limit state has been triggered; S5, when an over-limit state is detected to be triggered, extracting the real-time dynamic features of the over-limit event, matching the real-time dynamic features of the over-limit event with the energy backlash judgment model, determining the anomaly type of the over-limit event based on the matching result, and outputting a protection action command or dynamic balance control command corresponding to the anomaly type.

Description

Technical Field

[0001] This application relates to the field of energy storage technology, and more specifically to the field of control and protection of energy storage systems participating in power system power regulation. In particular, it relates to a multi-level grid-type energy storage control and protection system and method. Background Technology

[0002] With the widespread application of new energy vehicles and distributed renewable energy, charging piles with energy storage functions are becoming important nodes in the power distribution network. Energy storage systems are responsible not only for the temporary storage and release of electrical energy, but also for achieving rapid power regulation and dynamic balance. The reliability of their control and protection strategies directly affects the stable operation of the local power grid and the safe lifespan of the energy storage equipment itself. Therefore, developing intelligent protection technologies to adapt to these needs is of great significance.

[0003] In engineering practice, charging pile energy storage systems often face complex operating conditions such as rapid power fluctuations and sudden load changes, which can easily trigger dynamic phenomena such as instantaneous reverse flow of electrical energy. Currently, most common protection schemes rely on pre-set fixed thresholds for fault identification. Such methods are not adaptable enough to deal with highly dynamic and short-term energy disturbances, and it is difficult to accurately distinguish between normal transient processes and real fault states. This may lead to unnecessary protection malfunctions or protection blind spots, affecting the reliability and economy of continuous system operation.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a multi-level grid-type energy storage control and protection system and method to solve the above-mentioned technical problems.

[0006] This application provides a multi-level network-type energy storage control and protection system, comprising: a construction unit for performing feature modeling on the multi-source operating signals of the acquired energy storage system to construct a multi-source operating feature set; an extraction unit for performing time series rate of change analysis on the multi-source operating feature set to extract dynamic features of current direction change and power reversal; establishing an energy backlash judgment model based on the dynamic features to generate a dynamic energy feature set; a calculation unit for performing energy disturbance amplitude analysis on the dynamic energy feature set and calculating an adaptive protection threshold set in real time based on the energy backlash intensity and duration; an identification unit for comparing the real-time sampled signal with the adaptive protection threshold set to identify whether an over-limit state has been triggered; and an output unit for extracting the real-time dynamic features of the over-limit event when an over-limit state is detected, matching the real-time dynamic features of the over-limit event with the energy backlash judgment model, determining the anomaly type of the over-limit event based on the matching result, and outputting a protection action command or dynamic balance control command corresponding to the anomaly type to form a coordinated protection control command set.

[0007] This application provides a multi-level network-type energy storage control and protection method, including: S1, performing feature modeling on the multi-source operation signals of the collected energy storage system to construct a multi-source operation feature set; S2, performing time series rate of change analysis on the multi-source operation feature set to extract dynamic features of current direction change and power reversal; establishing an energy backlash judgment model based on the dynamic features to generate a dynamic energy feature set; S3, performing energy disturbance amplitude analysis on the dynamic energy feature set, and calculating an adaptive protection threshold set in real time based on the energy backlash intensity and duration; S4, comparing the real-time sampled signal with the adaptive protection threshold set to identify whether an over-limit state has been triggered; S5, when an over-limit state is detected to be triggered, extracting the real-time dynamic features of the over-limit event, matching the real-time dynamic features of the over-limit event with the energy backlash judgment model, determining the abnormal type of the over-limit event based on the matching result, and outputting a protection action command or dynamic balance control command corresponding to the abnormal type to form a coordinated protection control command set.

[0008] Based on the embodiments provided in this application, by collecting multi-source operating signals from the energy storage system and performing feature modeling, an operating feature set that comprehensively reflects the multi-dimensional state of the system is constructed. This enables protection and control strategies to be built on a more comprehensive and accurate system state profile, providing a reliable data foundation for subsequent analysis and thus improving the overall situational awareness capability of the protection system. By performing time-series rate-of-change analysis on the operating feature set and extracting dynamic features of current direction and power reversal, key transient behaviors in highly dynamic processes such as energy storage system charging and discharging switching and power surges can be captured and quantified in real time. The energy backlash judgment model established based on this can effectively identify dynamic processes such as short-term energy reverse flow that are difficult to distinguish with traditional fixed threshold protection, providing a core criterion for accurately distinguishing normal operating condition fluctuations from abnormal faults.

[0009] By analyzing the energy disturbance amplitude of the dynamic energy feature set and calculating the adaptive protection threshold in real time based on the intensity and duration of energy backlash, the protection threshold can be dynamically adjusted according to the actual operating conditions of the system. This overcomes the problem of poor adaptability of fixed threshold protection in complex dynamic environments, avoiding misjudgment of normal transient processes while ensuring sensitive response to real abnormal events. After identifying an over-limit state, the real-time dynamic characteristics of the over-limit event are matched with the pre-established energy backlash judgment model to determine the abnormal type of the over-limit event. Based on the matching result, the system can output more targeted protection action commands or dynamic balance control commands, which helps to maintain the continuous operation of the energy storage system to the maximum extent while ensuring safety. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a structural diagram of an optional multi-level grid-type energy storage control and protection system according to an embodiment of this application; Figure 2 This is a flowchart of an optional multi-level grid-type energy storage control and protection method according to an embodiment of this application; Figure 3 This is a flowchart of another optional multi-level grid-type energy storage control and protection method according to an embodiment of this application; Figure 4 This is a flowchart of another optional multi-level grid-type energy storage control and protection method according to an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an optional electronic device according to an embodiment of this application.

[0011] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0013] According to one aspect of the embodiments of this application, a multi-level grid-type energy storage control and protection system is also provided. For example... Figure 1 As shown, the system includes: The construction unit 101 is used to perform feature modeling on the multi-source operation signals of the collected energy storage system and construct a multi-source operation feature set; Extraction unit 102 is used to perform time series rate of change analysis on the multi-source operation feature set, extract dynamic features of current direction change and power reversal; and establish an energy backlash determination model based on the dynamic features to generate a dynamic energy feature set. The calculation unit 103 is used to perform energy disturbance amplitude analysis on the dynamic energy feature set and to calculate the adaptive protection threshold set in real time based on the energy recoil intensity and duration. The identification unit 104 is used to compare the real-time sampled signal with the adaptive protection threshold set to identify whether the over-limit state has been triggered. Output unit 105 is used to extract the real-time dynamic features of the over-limit event when an over-limit state is detected, match the real-time dynamic features of the over-limit event with the energy backlash determination model, determine the anomaly type of the over-limit event based on the matching result, and output the protection action command or dynamic balance control command corresponding to the anomaly type to form a coordinated protection control command set. According to one aspect of an embodiment of this application, as... Figure 2 As shown, this application provides a multi-level grid-type energy storage control and protection method, including: S1, perform feature modeling on the multi-source operation signals of the collected energy storage system to construct a multi-source operation feature set; Optionally, the multi-source operation feature set includes power change rate, energy change amplitude, and current direction change characteristic parameters; The multi-source operating signals acquired in step S1 specifically include: Battery-side signals: total current and total voltage of the battery cluster, with a sampling frequency of not less than 10kHz; temperature of each battery module; state of charge (SOC) estimated in real time by the battery management system (BMS).

[0014] Power conversion side signals: DC bus voltage; AC side output voltage, current, and switching status of the bidirectional converter (PCS).

[0015] Grid-side signals: grid connection point voltage, frequency, and the instantaneous active and reactive power calculated from them.

[0016] All signals are time-aligned using a synchronous clock to ensure the timing consistency of the data analysis. After preprocessing these raw signals (such as filtering and differential calculation), a multi-source operating feature set is constructed. This set mainly includes: the first-order rate of change of current, the direction and rate of power change, real-time SOC, average battery temperature, and voltage / frequency deviation. S2, time-series rate of change analysis is performed on the multi-source operating feature set to extract the dynamic features of current direction change and power reversal; based on these dynamic features, an energy backlash determination model is established to generate a dynamic energy feature set. The dynamic energy feature set includes recoil duration, peak amplitude, and rate of change. S3 performs energy disturbance amplitude analysis on the dynamic energy feature set and calculates the adaptive protection threshold set in real time based on the energy recoil intensity and duration. The calculation of the adaptive protection threshold set in step S3 is not a simple table lookup, but a calculation process based on real-time operating conditions and event characteristics. The specific logic is as follows: This method first extracts the intensity (e.g., the ratio of peak current to rated current) and duration of the currently detected recoil event from the dynamic energy feature set output in step S2. Simultaneously, it acquires the current system's key states, such as SOC and battery temperature. Then, using these parameters (recoil intensity, duration, SOC, and temperature) as input, it dynamically adjusts the baseline protection threshold through a preset weighted calculation framework. For example, when a high recoil intensity is detected but the duration is extremely short, and the battery SOC is within the safe range and the temperature is normal, the algorithm outputs a temporary, moderately relaxed current threshold, allowing the transient impact to pass without triggering protection. This calculation framework can be a linear weighted, fuzzy inference, or model-predictive control algorithm; its core is to make the protection threshold a dynamic variable that matches the actual transient stress experienced by the system.

[0017] S4 compares the real-time sampled signal with the adaptive protection threshold set to identify whether an over-limit state has been triggered. When an over-limit state is identified in step S4, the real-time dynamic characteristics of the over-limit event are extracted. These characteristics include at least: peak amplitude, total event duration, the slope (rate of change) of the current or power waveform rise / fall, and the number of times the current direction reverses within the event period. These characteristics together constitute a digital profile of the transient event.

[0018] Subsequently, in step S5, these real-time dynamic features are matched with the energy recoil determination model established in step S2. This determination model stores feature templates of typical energy recoil events. The matching calculation is performed by calculating the Euclidean distance between the real-time feature vector and the template feature vector. The closer the distance, the higher the similarity, and the more likely it is to be determined as energy recoil. Two distance thresholds are preset: a smaller distance threshold as the recoil determination threshold, and a larger distance threshold as the fault determination threshold.

[0019] For handling uncertain matching results: When the calculated Euclidean distance falls between the recoil threshold and the fault threshold, it falls within the uncertainty range. In this case, the method triggers a pre-set conservative handling logic: immediately generating and executing a temporary "power reduction operation" command to limit the energy storage system's output power to a lower level, preventing potential risks from escalating. Simultaneously, the method adds a "to be analyzed" tag to this event and fully records all its characteristic data and system context, sending it to the subsequent online learning process for further processing. This design ensures the system's safety when facing fuzzy judgments.

[0020] S5, when an over-limit state is detected to be triggered, extract the real-time dynamic features of the over-limit event, match and calculate the real-time dynamic features of the over-limit event with the energy backlash judgment model, determine the abnormal type of the over-limit event based on the matching result, and output the protection action command or dynamic balance control command corresponding to the abnormal type to form a coordinated protection control command set.

[0021] The coordinated protection control instruction set output in step S5 is a collection of control commands at different levels, oriented towards specific physical execution units. Specifically, it includes: Battery Management System (BMS) level commands: such as "adjust the upper limit of the charge and discharge current of the battery cluster" and "disconnect the contactor of the specified battery module". These commands act directly on the battery cells to ensure battery safety.

[0022] Power converter (PCS) level instructions: such as "Switch control mode (e.g., from constant power mode to current limiting mode)" and "Inject a compensation current with a specified amplitude and phase angle". These level instructions are used to quickly smooth out power fluctuations and achieve dynamic energy balance.

[0023] System coordination controller-level commands: for example, "adjust the grid-connected power plan for the next few seconds" or "initiate power sharing with adjacent charging piles". These commands are scheduled from the perspective of overall system optimization.

[0024] When a genuine fault is identified, the instruction set will focus on emergency shutdown instructions at the BMS and PCS levels; when an energy backlash is identified, the instruction set will focus on dynamic adjustment instructions at the PCS level and planned adjustment instructions at the system level.

[0025] Exemplarily, the method of the present invention first performs step S1: feature modeling of the collected multi-source operating signals of the energy storage system to construct a multi-source operating feature set. Specifically, the multi-source operating signals include the current and voltage of the battery clusters, the DC bus voltage, the grid connection point voltage and frequency, the battery temperature, and the state of charge estimated in real time by the battery management system. These signals are collected synchronously at a sampling frequency of not less than 10 kHz to ensure the capture of millisecond-level dynamic changes. Feature modeling includes calculating the rate of change of current, the direction and magnitude of power change, and incorporating parameters such as battery temperature and state of charge to form a standardized feature set for subsequent analysis.

[0026] Next, step S2 is performed: a time-series rate-of-change analysis is conducted on the multi-source operating characteristic set to extract dynamic features of current direction change and power reversal. Based on these dynamic features, an energy recoil determination model is established. This model, by analyzing the similarity between characteristic waveforms and historical recoil events, can output a dynamic energy characteristic set describing recoil events. This characteristic set includes at least the recoil duration, peak amplitude, and rate of change.

[0027] Then, step S3 is executed: energy disturbance amplitude analysis is performed on the dynamic energy feature set, and an adaptive protection threshold set is calculated in real time based on the energy recoil intensity and duration reflected therein. This threshold set is not a fixed value, but a variable that is dynamically adjusted according to the severity of the recoil and the current state of the system.

[0028] Then, step S4 is executed: the real-time sampled signal is compared with the calculated set of adaptive protection thresholds to identify whether an over-limit state has been triggered.

[0029] Finally, step S5 is executed: When an over-limit state is detected, the real-time dynamic characteristics of the over-limit event are extracted and matched with the energy backlash determination model established in step S2. Based on the matching result, it is determined whether the over-limit event belongs to an energy backlash transient or a real fault. If it is determined to be a real fault, a protection action command is generated; if it is determined to be an energy backlash transient, a dynamic balance control command is generated. These commands together constitute the coordinated protection control command set.

[0030] The method of this invention solves the problem of instantaneous energy backlash caused by external disturbances (such as sudden load changes and inverter switching) in charging pile energy storage systems under highly dynamic operating environments such as fast charging and grid fluctuations.

[0031] In this invention, "energy backlash" refers to a transient power reversal or rapid change in current direction that occurs within a millisecond timescale due to external disturbances (such as sudden stop of an adjacent charging station, a sudden surge in grid voltage, or inverter switching). Such events occur frequently in high-power fast charging scenarios. Their key characteristics are: extremely short duration (typically 1 to 100 milliseconds), with the current or power amplitude momentarily exceeding normal limits, but without permanent damage or a persistent short circuit in the system. Traditional overcurrent / overvoltage protection devices based on fixed thresholds, whose protection settings are designed for persistent faults, cannot tolerate such short-term, high-amplitude dynamic shocks and are prone to malfunctions, leading to charging interruptions. This invention aims to solve this misjudgment problem.

[0032] refer to Figure 2 The multi-level grid-type energy storage control and protection method provided in this application relates to the field of control and protection of energy storage systems participating in power system power regulation.

[0033] Furthermore, the energy recoil determination model established in S2 includes a reference parameter range, which is provided by a dynamic coupling knowledge base. The dynamic coupling knowledge base adopts a three-dimensional operating condition space index structure, which consists of the state of charge, battery temperature, and absolute value of the rate of change of current. Among them, the state of charge, battery temperature, and absolute value of the rate of change of current are obtained by real-time calculation of the multi-source operating signals collected in S1. When calculating the adaptive protection threshold set in S3, the adaptive protection threshold set is obtained by querying the asymmetric threshold parameters of the operating condition space cell determined by the current state of charge, battery temperature and absolute value of current change rate in the dynamic coupling knowledge base. The asymmetric threshold parameters include rising edge threshold and falling edge threshold, respectively calibrated based on electrochemical hysteresis effect.

[0034] In some embodiments, the implementation of the three-dimensional working condition space index structure includes: The three-dimensional operating space discretizes the continuous spatial parameters of SOC, temperature, and current change rate, forming a multi-dimensional query grid. In practical implementation: SOC dimension: divided into 10 intervals, such as [0%, 10%), [10%, 20%)...[90%, 100%].

[0035] Battery temperature dimension: divided into 5 ranges, such as <0°C, [0°C, 15°C), [15°C, 35°C), [35°C, 50°C), and ≥50°C.

[0036] Current change rate (|di / dt|) dimension: Based on the system rated current In, it is divided into 8 levels, such as [0,0.1In), [0.1In,0.5In)……[5In,∞).

[0037] The storage structure of the three-dimensional grid uses a three-dimensional array, and its index directly corresponds to a unique working condition space unit. During querying, the index number of the interval is calculated based on the real-time SOC, temperature, and |di / dt|. The query time complexity is O(1), which is much less than 1ms, meeting the real-time requirements.

[0038] Optionally, the dynamically coupled knowledge base is constructed using an offline calibration and initialization model, followed by online learning and updating. The initial data comes from: Factory test data: Hybrid pulse power characteristic (HPPC) test and charge-discharge cycle test at different temperature ranges are performed on the energy storage battery to obtain basic electrochemical response data.

[0039] Simulation data: A joint simulation system including battery model, converter model and power grid model is established to simulate various power grid disturbance and load change scenarios, and generate a large amount of sample data of energy backlash and faults.

[0040] Based on the analysis of the above data, an initial "asymmetric threshold parameter" is calibrated for each three-dimensional mesh cell. Each record in the knowledge base is formatted as: {SOC range, temperature range, current rate of change range, rising edge threshold, falling edge threshold}.

[0041] It's important to explain that the fundamental reason for the difference between the "rising edge threshold" and the "falling edge threshold" lies in the electrochemical hysteresis effect. Taking lithium-ion batteries as an example, during charging (lithium-ion insertion into the negative electrode), the diffusion rate of ions in the electrode material and the rate of interfacial reaction are asymmetrical with those during discharging (lithium-ion extraction), resulting in different tolerances to charging and discharging overcurrents. This difference can be quantified using electrochemical impedance spectroscopy (EIS): typically, the polarization resistance in the charging direction is slightly higher than that in the discharging direction. Therefore, based on EIS testing and temperature change experimental data, protection thresholds that better reflect the electrochemical characteristics of the charging and discharging processes under different operating conditions can be calibrated. The rising edge threshold (charging protection) is usually set more conservatively (i.e., smaller in value) than the falling edge threshold (discharging protection).

[0042] When a real-time parameter (e.g., SOC = 24.5%) falls precisely near the boundary between two grid cells, a trilinear interpolation method is used to obtain the final threshold. Specifically, based on the point's relative position in its corresponding SOC, temperature, and |di / dt| dimensions, a weighted average of the threshold parameters stored in the surrounding eight vertex grid cells is calculated. This results in a smooth and continuous threshold output, avoiding threshold jumps caused by grid discretization.

[0043] Furthermore, the dynamically coupled knowledge base achieves online iterative updates through a recoil event learning strategy. The results of these online iterative updates directly influence the generation of the adaptive protection threshold set. The recoil event learning strategy includes: When S5 determines that the abnormal type of the over-limit event is energy recoil transient and the determination result is consistent with the actual system response, the real-time dynamic characteristics of the over-limit event are used as positive samples to expand the recoil characteristic statistical boundary of the current working condition space unit. The recoil characteristic statistical boundary is used to update the reference parameter range. When S5 determines that the abnormal type of the over-limit event is a real fault and actually executes the protection action, if the over-limit state continues for more than the preset time, the threshold envelope of the current working condition space unit will be shrunk according to the preset shrinkage ratio. The shrinkage of the threshold envelope will act on the boundary value of the asymmetric threshold parameter in the dynamic coupling knowledge base to improve the conservatism of subsequent judgments in this area.

[0044] The learning strategy based on recoil events operates by post-hoc verification of historical decision results. The criterion for determining whether the result matches the actual system response is: If an event is determined to be an energy backlash transient, and after executing the dynamic balance control command, the relevant current and voltage signals return to the normal range within a very short time window (e.g., 20 milliseconds) without triggering any hardware protection relays, then the determination is considered correct.

[0045] For events determined to be genuine faults, if the voltage / current monitoring signal of the disconnected branch remains zero or extremely low for a preset duration (e.g., 100 milliseconds) after the protection action (e.g., disconnection) is executed, and other parts of the system return to stability, then the determination is considered correct.

[0046] Forward learning: Expanding the statistical boundary of recoil features, the learning strategy is activated when the determination of an energy recoil transient is verified as correct. The method uses the real-time dynamic features of the current event as positive samples to update the baseline parameter range of the corresponding operating condition spatial unit in the energy recoil determination model. Specifically, it calculates the relationship between each feature value of the event (e.g., peak value, duration) and the current model's statistical boundary (e.g., mean, standard deviation), and then expands the boundary according to preset rules (e.g., "if the new sample feature value exceeds the original boundary, then the boundary is expanded outward to 105% of the sample value"). This effectively expands the recognition range of legitimate recoil features under this operating condition.

[0047] Negative learning: The triggering and calculation of threshold envelope contraction, i.e., when a real fault determination is verified to be potentially inaccurate (i.e., the over-limit state persists after the protection action), negative learning is triggered. The determination logic is as follows: after the protection command is issued, the signal of the fault branch is continuously collected (sampling window length is 20ms). If, within this window, more than 80% of the sampled point values ​​still exceed the initially triggered protection threshold, it is determined that the over-limit state persists.

[0048] At this point, the learning strategy will shrink the threshold envelope (i.e., the boundary between the rising and falling edge thresholds) of the current operating condition spatial cell. The shrinkage is performed proportionally in all directions, with a preset shrinkage percentage typically chosen between 5% and 20%. For example, both the rising and falling edge thresholds can be multiplied by 0.9 (i.e., a 10% shrinkage). This makes the protection more sensitive under this operating condition.

[0049] The online learning module (which implements the above strategy) directly targets the threshold parameters stored in the dynamically coupled knowledge base and the feature template boundaries stored in the energy recoil determination model.

[0050] After positive or negative learning is completed, the new parameters and boundaries are written into the knowledge base and the decision model. In the next control cycle, when the method executes step S3 again, the adaptive protection threshold set is queried from the updated dynamically coupled knowledge base, thus obtaining the optimized threshold. During the matching calculation in step S5, the baseline parameter range in the referenced energy recoil decision model has also been updated, thus affecting the similarity calculation results.

[0051] This closed loop ensures that the system can continuously learn from actual operating experience, enabling protection strategies to adapt to specific operating environments and equipment aging conditions, demonstrating significant adaptability and dynamic evolution capabilities.

[0052] In one specific implementation, combined with Figure 3 The specific implementation process of the adaptive protection threshold closed-loop update procedure of the present invention will be described in detail. For example... Figure 3 As shown, the entire process forms a dynamic control loop that is interconnected and capable of self-evolution.

[0053] This process begins with the real-time acquisition of multi-source operating signals from the energy storage system. Specifically, these signals include the current and voltage of the battery clusters, the DC bus voltage, the voltage and frequency at the grid connection point, the temperature of each battery module, and the state of charge estimated in real time by the battery management system. All signals are synchronized by a high-precision clock and acquired at a sampling frequency of no less than 10 kHz to ensure complete capture of millisecond-level transients generated by the charging pile under dynamic operating conditions such as fast charging.

[0054] The acquired raw signals are then processed through feature modeling and dynamic feature extraction. In this stage, the first-order rate of change of current, the direction and instantaneous rate of power change are calculated. Combined with real-time state of charge and average battery temperature, a multi-source operational feature set describing the instantaneous operating state of the system is constructed. Next, time-series analysis is performed on this feature set to extract key dynamic features characterizing energy mutation events, such as the rate of power reversal, the frequency of current direction changes, and the magnitude of energy mutations, thus forming a more refined dynamic energy feature set.

[0055] Next, the system proceeds to query the dynamically coupled knowledge base. This knowledge base is the core database for the system's adaptive capabilities, and it employs an index structure consisting of three dimensions: state of charge, battery temperature, and rate of change of current. Based on the currently extracted real-time features, the system quickly locates the corresponding operating unit in this three-dimensional space and reads the pre-stored baseline protection parameters.

[0056] Based on baseline parameters obtained from a knowledge base and information on the intensity and duration of the current event obtained from a dynamic energy feature set, an adaptive protection threshold calculation is performed. This step does not simply output a fixed value, but rather dynamically adjusts the set of protection thresholds to match the specific form of the event and the transient process. For example, for an impact with a high amplitude but an expected extremely short duration, the algorithm generates a set of temporarily relaxed thresholds that allow the impact to pass.

[0057] The calculated set of dynamic thresholds is immediately used for over-limit state identification. Real-time acquired signals such as current and voltage are continuously compared with their corresponding dynamic thresholds. Once the instantaneous value of a signal exceeds its dynamic threshold, an over-limit state is determined to have been triggered.

[0058] Once an out-of-limit condition is confirmed, the event is immediately locked, and its real-time dynamic characteristics are extracted. These characteristics include the peak value of the event waveform, the duration from occurrence to end, the steepness (rate of change) of the rising and falling edges, and the number of times the current direction changes within the event window.

[0059] Next, the process moves to the matching calculation stage. The real-time dynamic features extracted in the previous step are compared one by one with multiple typical energy recoil feature templates pre-established in the energy recoil determination model. By calculating the similarity between the two (e.g., calculating the Euclidean distance between feature vectors), the degree of similarity between the current event and the typical recoil pattern is quantified.

[0060] The matching calculation results are directly used for anomaly type determination. The system presets two determination thresholds: if the similarity is higher than the higher fault determination threshold, the current event is determined to be a real fault; if the similarity is lower than the lower recoil determination threshold, it is determined to be an energy recoil transient. Based on the determination results, the system will execute distinctly different control strategies: if it is a real fault, it generates and executes protection action commands, such as instructing the battery management system to disconnect the faulty battery cluster and instructing the power converter to block the output; if it is an energy recoil transient, it generates and executes dynamic balance control commands, such as instructing the power converter to fine-tune its power reference value to smoothly absorb or offset the transient impact.

[0061] After the control command is executed, the process enters the critical post-verification and learning phase. The system continuously monitors the actual response of the system after the control command is executed and evaluates whether the response is consistent with the previous judgment expectations. The specific verification criteria are as follows: if the system recovers stable operation within tens of milliseconds after a backlash is determined and balancing control is executed, without triggering other hardware protection, it is judged as "consistent"; if the abnormal signal of the branch continues for more than a preset duration (e.g., 100 milliseconds) after a fault is determined and protection is cut off, it is judged as "inconsistent".

[0062] The verification results will drive the online learning module to initiate the corresponding knowledge base update strategy. If the verification result is "consistent" and the event is judged as recoil, the event sample will be marked as a "positive sample," and the learning module will expand the acceptable range of recoil characteristics for the corresponding working condition unit in the dynamic coupling knowledge base accordingly. Conversely, if the verification result is "inconsistent," the event sample will be marked as a "negative sample," and the learning module will shrink the protection threshold boundary of the corresponding working condition unit, making future protection more sensitive under such working conditions.

[0063] The learning module updates the knowledge base in real time. This means that once the system processes an event and completes the learning process, the data obtained in the "query the dynamically coupled knowledge base" step in the next control cycle will already contain the optimized new parameters. Thus, the process forms a complete closed loop. This closed loop enables the protection system of this invention to continuously accumulate experience during actual operation, dynamically optimize its protection criteria and thresholds, thereby continuously improving its ability to distinguish between harmful faults and harmless transient impacts, ultimately achieving the comprehensive goals of ensuring safety, improving availability, and extending equipment lifespan.

[0064] Furthermore, the multi-source operating feature set constructed in S1 includes normalized electrochemical polarization states. ; Generated in real time by an electrochemical impedance spectroscopy (EIS), whose equation of state is: ; in, The observer attenuation coefficient (dimension 1 / [T]) is the dimensionless attenuation coefficient. η is the electrochemical coupling coefficient (dimension 1 / [I]), and η is the scaling factor (dimensionless). The preset critical state of charge threshold; It is the independent variable of the equation of state, representing the continuous time of operation of the electrochemical impedance spectroscopy. It is the rate of change of the current in the battery clusters of the energy storage system. It is the core dynamic input. The absolute value of the rate of change of current. It is in a charged state; The energy recoil determination model established in S2 includes, based on, The decision confidence correction logic; the energy backlash decision model output is used as the similarity decision threshold for matching calculation in S5; The confidence level correction logic is used to dynamically adjust the similarity judgment threshold, including: when When the preset polarization vulnerability threshold is exceeded, the similarity judgment threshold used for matching calculation is increased by a preset increase ratio, making the system more stringent in judging suspected recoil events under high electrochemical stress. and its change Used to construct subsequent generalized energy recoil eigenvectors.

[0065] Normalized electrochemical polarization state It is a dimensionless state quantity ranging from 0 to 1. Its physical significance lies in its quantification of the instantaneous intensity of electrochemical polarization within the battery caused by the obstruction of lithium-ion diffusion in the electrode materials and interfacial reactions. The closer the value is to 1, the more severe the internal polarization of the battery, the slower its response to changes in external current, and the more fragile it is.

[0066] This state variable is generated by an online, real-time estimation using an electrochemical impedance spectroscopy (EIS). The core of this EIS is a state equation, whose inputs are the absolute value of the rate of change of the cell cluster current and the current state of charge. The key parameters in the state equation and their determination methods are as follows: Observer attenuation coefficient It has the dimension of the reciprocal of time, and its reciprocal (1 / This corresponds to the time constant of the observer's dynamic response. This parameter is calibrated through a battery pulse power test: a standard current pulse is applied to the battery, and the relaxation response curve of its terminal voltage is observed. The dominant time constant of the voltage recovery process is determined by fitting the curve. The typical range of values ​​for this value results in a time constant between 5 milliseconds and 50 milliseconds. For example, for power-type lithium iron phosphate batteries, the time constant might be calibrated to approximately 10 milliseconds (i.e., ...). ≈100s -1 For energy-type ternary lithium batteries, the time constant may be calibrated to approximately 30 milliseconds (i.e., ...). ≈33s -1 ).

[0067] Electrochemical coupling coefficient Having the reciprocal dimension of current, this coefficient determines the strength of the effect of current changes on the polarization state. Its typical value ranges from 0.001 amperes to 0.1 amperes. This coefficient is closely related to the battery's chemical system; for example, lithium iron phosphate batteries, due to their relatively flat voltage plateau and high polarization resistance, The value is usually higher than that of ternary lithium batteries, which means it is more sensitive to changes in current.

[0068] Scaling factor Used to amplify or reduce the effect of charge state deviation on polarization state updates.

[0069] Critical state of charge threshold This is a preset threshold, typically set in the region where the battery's state of charge (SOC) is low and its polarization characteristics begin to change nonlinearly. A typical value is 20% or 30%. The setting is based on the battery's characteristic curve; when the SOC is below this value, the battery's diffusion polarization resistance increases significantly. The state equation is designed so that when the SOC is below... When, the change in current affects The excitation effect is nonlinearly enhanced, thus providing an earlier warning of polarization risks under low power conditions.

[0070] The energy recoil determination model established in step S2 includes the following: The logic for correcting the confidence level is as follows. The core of this logic is to dynamically adjust the similarity threshold used in the matching calculation during step S5.

[0071] First, the system presets a baseline value for similarity judgment. This baseline value can be obtained based on historical data statistics; for example, setting an energy recoil threshold when the Euclidean distance is less than 0.5. Second, a preset polarization vulnerability threshold is set, for example... >0.7 or >0.8. This threshold indicates that the battery is in a highly polarized state and its ability to withstand stress is reduced.

[0072] When real-time estimation When the aforementioned polarization vulnerability threshold is exceeded, correction logic is triggered. The system will increase the similarity judgment threshold by a preset increase percentage. For example, the threshold may be increased by 20% or 30%. This means that when the battery is vulnerable, the system requires the matching calculation results to have a higher similarity (i.e., a smaller Euclidean distance) to be judged as a relatively harmless energy backlash; otherwise, the system will tend to make a more conservative judgment. For example, the base value is 0.5, which becomes 0.6 after a 20% increase; and 0.65 after a 30% increase. This mechanism effectively prevents the underestimation of the risk of impact events under high polarization conditions.

[0073] Furthermore, the dynamic energy feature set generated in S2 includes the generalized energy recoil feature vector. ; in, The virtual angular frequency deviation is used to characterize the synchronization stability of the power grid. for The change in; Indicates matrix transpose; ; The power integral term characterizing the depth of electro-mechanical energy coupling, Power deviation; because The physical dimensions of each component are different (angular frequency, dimensionless number, and energy, respectively). To facilitate subsequent unified analysis and stability criterion calculation, they need to be projected onto a dimensionless state space. Specifically, this is achieved through a dimension normalization matrix. To achieve this. Matrix It is a diagonal matrix whose diagonal elements are the system's fixed rated angular frequencies. Factory-calibrated polarization reference and energy benchmark The reciprocal, specifically: Dimensionally normalized matrix Projected onto a dimensionless state space, where To fix the rated angular frequency of the system, and The polarization and energy references are factory-calibrated. It should be noted that the notation here... Meaning to construct a This is a diagonal matrix with diagonal elements. Through operations... This yields a normalized eigenvector in which each component is transformed into a dimensionless number. This eliminates the influence of dimensions, allowing subsequent analyses based on vector magnitude or Lyapunov exponents to have accurate mathematical and physical meaning.

[0074] Identifying out-of-limit states in S4 includes: based on the generalized energy recoil feature vector. For discrete time series data, the Wolf algorithm is used to estimate the Lyapunov exponent. ;when When the negative stability margin threshold is less than the preset threshold, the abnormal type of the out-of-limit event is determined to be an energy recoil transient; when When the value exceeds the preset positive instability threshold, the abnormal type of the out-of-limit event is determined to be a real fault; when... When the value is between the negative stability margin threshold and the positive instability threshold, the hierarchical model predictive control architecture is activated, using the generalized energy recoil eigenvector. Using multi-source operational feature sets as input, the abnormal type of the limit-breaking event is re-determined; The preset negative stability margin threshold and positive instability threshold are determined based on the rated capacity of the energy storage system and the grid support strength requirements.

[0075] Constructed generalized energy recoil eigenvector This vector is used to comprehensively describe disturbance events from a system energy perspective. It contains three components: (Virtual angular frequency deviation): In the grid-type control mode, the converter simulates the rotor motion equation of the synchronous generator to generate a virtual angular frequency. This refers to the instantaneous deviation between the virtual angular frequency and the rated angular frequency of the power grid (such as 314.16 rad / s corresponding to 50 Hz), which is estimated in real time using algorithms such as phase-locked loops.

[0076] That is, the aforementioned normalized electrochemical polarization state The amount of change within the event time window, for example, within a 10-millisecond window before and after the event is triggered. The difference.

[0077] (Electrical-Electrical Energy Coupling Depth): This is a power integral term, calculated over a time window (e.g., 10 milliseconds) to measure the power deviation. With polarization state change The integral of the product. This component physically reflects the coupling energy between grid-side power disturbances and changes in electrochemical stress within the battery. Discrete summation approximates the integral in the calculation.

[0078] The above three components with different physical dimensions are used to form a vector. Then, a dimensionally normalized matrix is ​​used. Project it onto a dimensionless space. Matrix The method for determining the baseline value in the data is as follows: (System fixed rated angular frequency): The preset angular frequency is the angular frequency corresponding to the power grid frequency, such as 2π×50rad / s or 2π×60rad / s.

[0079] and Factory calibration benchmark: Measured under standard test conditions (e.g., ambient temperature 25°C, SOC=50%, charge / discharge test at 1C rate). Typical fluctuation amplitude as And, measurements yielded typical Value as .

[0080] To determine the nature of events from a system dynamics perspective, this method is based on eigenvectors. For discrete time series data, the Wolf algorithm is used to estimate its maximum Lyapunov exponent. This algorithm performs calculations by reconstructing the phase space of the time series. During implementation, the embedding dimension m (e.g., m=3 or 5) and time delay τ (e.g., τ=1 millisecond) must be preset. The calculation requires a data sequence of a certain length, such as 1000 consecutive sampling points (corresponding to a 100-millisecond data window), and can be updated every 10 milliseconds.

[0081] Calculated It is a key indicator for judging whether a system is stable under disturbances: A negative value indicates that the disturbance has decayed and the system is stable; a positive value indicates that the disturbance has diverged and the system is unstable. Therefore, the system presets two key thresholds: Negative stability margin threshold: for example <-0.1. When the estimated value is less than this threshold, the system is determined to have sufficient stability margin, and the current event is a recoverable energy recoil transient.

[0082] Positive instability threshold: for example >0.05. When the estimated value is greater than this threshold, the system is considered to be trending towards instability, and the current event is a dangerous "real failure".

[0083] These two thresholds are set based on the specific requirements for damping ratio and small-disturbance stability of energy storage systems during grid-connected operation. When When the estimated value is between these two thresholds (e.g., between -0.1 and 0.05), the system stability is in a fuzzy zone. At this point, the method will activate the hierarchical model predictive control architecture, which uses its multi-layer model to perform more refined joint diagnosis and re-determine the event type.

[0084] Furthermore, in S5, when based on the Lyapunov index... The anomaly type of the out-of-limit event is determined to be an energy recoil transient, and based on the current... When the determined similarity threshold is met, and the matching calculation results still support the determination (i.e., Lyapunov determination and similarity threshold), the similarity threshold is met. When the corrected judgment conclusions are consistent, if the magnitude of the generalized energy recoil eigenvector is... If the high-confidence trigger threshold is exceeded, a dynamic balancing control command is generated; the dynamic balancing control command includes: Electrochemical stress wave canceller command, which is based on the S2-predicted arrival time of the recoil stress wave. ,exist A reverse micro-pulse current is constantly injected through the switching matrix within the battery cluster; the amplitude of this reverse micro-pulse current... Based on the recoil energy intensity provided by the dynamic energy characteristic set and preset current injection coefficient (Dimension [I] / [Energy]) is determined, that is Its pulse width is a preset fixed pulse width. ;in, This is the preset time lead time; The pre-charge sub-command for grid support temporarily stores the residual recoil energy that is not completely offset by the reverse micro-pulse current in the DC bus capacitor, and releases the residual recoil energy to the grid in the form of increasing virtual inertia output in several power frequency cycles in the subsequent preset period through virtual synchronous machine control. The high-confidence trigger threshold is based on the current grid frequency deviation and The threshold value is dynamically adjusted; the more fragile the power grid or the higher the electrochemical stress, the higher the threshold value will be.

[0085] Regarding the dynamic adjustment of the high-confidence trigger threshold, the prerequisite for implementing this control is that the judgment result has a high confidence level. To this end, a threshold based on the generalized feature vector magnitude can be set. A high-confidence trigger threshold. This threshold is not a fixed value, but is dynamically adjusted: First, set a base value, for example... >0.8. Then adjust according to the real-time system status: When the grid frequency deviation Δf is large (e.g., exceeding 0.5Hz), it indicates an urgent need for grid support. In this case, the trigger threshold should be appropriately lowered (e.g., reduced to 0.7) to allow for more proactive measures to stabilize the grid. When the electrochemical polarization state... A higher threshold (e.g., above 0.75) indicates a fragile battery. In this case, the trigger threshold should be appropriately increased (e.g., to 0.9) to avoid applying unnecessary active control stress when the battery is in a vulnerable state.

[0086] When the high-confidence triggering condition is met, an electrochemical stress wave cancellation sub-instruction is generated first.

[0087] Predicted peak arrival time of recoil stress wave Based on the rate of change of current or power provided by the dynamic energy feature set, the time for the current disturbance signal to reach its expected peak value is estimated by simple prediction methods such as linear extrapolation.

[0088] Determine the reverse micropulse current parameters: a "time advance" before the predicted peak arrival time. "Emits a reverse pulse." The preset value must be less than the rise time of the stress wave, for example, set to 2 milliseconds.

[0089] Amplitude of reverse micro-pulse current The maximum energy intensity of the predicted recoil event and a current injection coefficient A joint decision. The dimension of is current per unit energy, and its typical value can be calibrated through the pulse power withstand capability test of the battery cell, for example, the range is 0.3A / kJ to 1.0A / kJ. For a reverse pulse with an estimated energy of 5kJ, if If we take 0.5 A / kJ, then The amplitude is 2.5A. The width of the pulse is... It is a preset fixed value, which should be selected to be smaller than the main electrochemical time constant of the battery to avoid creating new persistent stress, for example, set to 1 millisecond.

[0090] The aforementioned reverse pulse may not completely cancel out all recoil energy; the remaining portion is defined as residual recoil energy. This energy is handled by constructing a mesh to support the precharge sub-command: Energy storage: The control temporarily stores residual energy in the DC bus capacitor. This requires the DC-side capacitor of the system to have a certain energy buffering capacity; for example, its capacity design must be able to absorb surge energy of several milliseconds in extreme cases without causing overvoltage.

[0091] Energy Release: Over subsequent power frequency cycles (e.g., 5 cycles, or 100 milliseconds), the temporarily stored energy is gradually released back to the grid by adjusting the converter's "virtual synchronous machine" control algorithm, in the form of a smooth increase in "virtual inertia" power output. This avoids energy waste and achieves beneficial energy feedback by supporting the grid frequency. The release rate is controlled to ensure that it does not cause secondary impacts on the grid.

[0092] Furthermore, in S3, a hierarchical model predictive control architecture is used to calculate the adaptive protection threshold set. The hierarchical model predictive control architecture includes a battery cluster layer predictive model, a converter layer predictive model, and a system layer predictive model. The objective function of each layer predictive model embeds the discrimination uncertainty penalty term of the output of other layer predictive models. The uncertainty penalty term is calculated based on the matching residuals of the feature vector of the current out-of-limit event and the recoil template in each model of each layer. The larger the matching residual, the higher the penalty weight is applied. Through iterative optimization (executed by the network support priority arbitrator), the hierarchical prediction results are driven to converge towards the direction of low uncertainty, so as to obtain the optimal consistent estimate of whether the anomaly type of the current out-of-limit event is an energy recoil transient or a real fault.

[0093] The core of using a hierarchical model predictive control architecture to calculate adaptive protection thresholds lies in eliminating judgment discrepancies between different control levels through a collaborative mechanism to form the optimal decision.

[0094] The hierarchical model predictive control architecture comprises three layers: the battery cluster layer predictive model, the converter layer predictive model, and the system layer predictive model. Each layer analyzes and predicts current out-of-limit events from its own perspective (such as electrochemical stress, power balance, and grid stability), determining whether they are energy backlashes or actual faults. To ensure consistency in prediction results across different layers, an uncertainty penalty term is embedded in the objective function of each predictive model.

[0095] The penalty term is calculated based on the matching residual. Specifically, each layer compares the feature vector of the current out-of-limit event with the typical energy recoil template stored in its own model library, calculating the difference between the two. For example, it calculates the square root of the sum of the squares of the differences in each dimension of the feature vector to obtain the matching residual in the form of Euclidean distance. The larger the residual, the higher the uncertainty of the prediction result (prone to recoil or failure) of the layer model. The penalty term takes this residual as input, calculates the penalty value through a weight function (e.g., the weights are exponentially positively correlated with the residual), and adds it to the objective function of the layer model. This means that if the prediction result of a certain layer differs greatly from that of other layers (resulting in a large residual calculated by itself with the common template), it will be subject to a stronger penalty during the optimization process, thereby driving it to adjust its internal state and move closer to the results of other layers.

[0096] The main body responsible for collaborative optimization is the network support priority arbitrator. This arbitrator collects preliminary, uncertain predictions from each layer of the model and initiates an iterative negotiation process. In each iteration, each layer of the model recalculates its optimal prediction based on the results from other layers and feedback from penalty terms. The arbitrator can employ a game-theoretic strategy based on weighted voting to evaluate convergence; for example, iteration stops when the consensus rate among the layers of the model in judging the event type (recoil / failure) exceeds 95%, or when the number of iterations reaches a preset limit (e.g., 5 times).

[0097] After iterative convergence, the arbitrator outputs the best consistent estimate for the current out-of-bounds event. This output is not a simple label, but a quantitative judgment that integrates the confidence levels of each layer. For example, it might output a probability value: "There is an 85% probability that the event is an energy recoil transient, and a 15% probability that it is a real failure." This fusion result is obtained by weighting the predicted probabilities of each layer model according to their final uncertainty weights, for example, using a Bayesian fusion framework.

[0098] Furthermore, the hierarchical model predicts the time domain of the control architecture. The code is dynamically compiled and generated by an electrochemical relaxation time estimator, with online code generation performed every 50 milliseconds to update the objective function matrix and constraint matrix. satisfy: ; in, The electrochemical relaxation time is estimated in real time based on the characteristic roots of the equation of state of the electrochemical impedance observer. To control the cycle, The fixed number of steps required for network support; when When estimation timeout or compilation failure occurs, the hierarchical model predictive control architecture falls back to the pre-compiled conservative prediction time domain. . It represents the smallest integer not less than x (i.e., rounded up).

[0099] The dynamic generation mechanism in the prediction time domain enables model predictive control to adapt to the dynamic characteristics of the battery.

[0100] The length of the predicted time domain ( The electrochemical relaxation time is not fixed, but dynamically generated by the "electrochemical relaxation time estimator". This reflects the timescale required for the internal polarization state of the battery to return to equilibrium after a disturbance. The estimator is based on the state equation of the aforementioned electrochemical impedance spectroscopy, and estimates the timescale by analyzing the eigenvalues ​​of the dominant modes in the equation in real time. The basic principle is that the relaxation time is related to the reciprocal of the absolute value of the real part of the eigenvalue. The estimation process has strict time limits; for example, the estimation calculation must be completed within 5 milliseconds. If the calculation times out or an illegal value (such as a negative value or infinity) is encountered, it is judged as an estimation timeout or compilation failure.

[0101] The control system operates with a fixed control cycle. Its typical value is set according to the controller hardware performance, such as 1 millisecond (for high-speed FPGAs) or 5 milliseconds (for high-performance DSPs).

[0102] Prediction Time Domain The calculation formula is: round up. .in, This is a fixed number of steps required for grid support, used to ensure that the forecast can cover the necessary transient processes of the power grid. Typical values ​​can be 5 or 10 steps. For example, if estimating... It is 15 milliseconds. It is 5 milliseconds. If the value is 5, then the calculation is as follows: =ceil(15 / 5)+5=3+5=8 steps.

[0103] Online code generation is performed every 50 milliseconds to update the objective function matrix and constraint matrices in the model predictive controller. This process can be achieved by integrating a code generation toolchain (such as MATLAB / SimulinkCoder), which will update the code based on the latest... The optimization problem model generated with other parameters is automatically compiled into efficient code that can run on hardware. If the aforementioned relaxation time estimation fails, the system will automatically fall back to a pre-compiled, conservative prediction time domain, for example... =10 steps to ensure the continuity of control functions and basic security.

[0104] Furthermore, S5 also includes a network support priority arbitration mechanism, which addresses the uncertainty of electrochemical stress in the battery cluster layer. Uncertainty of power mutation in converter layer and the grid frequency deviation at the system level As a state input, a dynamic programming strategy is used to output the weight allocation of each level of control in the coordinated protection control instruction set; Optionally, the dynamic programming strategy is implemented through offline strategy table pre-computation and online table lookup; Value function of dynamic programming strategy Used to assess the state Take action below The long-term benefit is equal to the weighted sum of grid support benefits and equipment stress costs, with weighting factors... With generalized energy recoil eigenvector The modulus length is positively correlated, and the action selection is optimized. To maximize this benefit.

[0105] Its expression is: ; in, Indicates by , , The state variables constituted This indicates the adjustment action for the control weights at each level. The reward function characterizes the revenue supported by the power grid. The reward function characterizes the stress cost of equipment. This is the dynamic recoil confidence weight, and its value is related to the generalized energy recoil eigenvector. The modulus is positively correlated with the length of the model. This indicates that the action is adjusted by optimizing the selection. This maximizes the value of the expression within the square brackets; The value function of the dynamic programming strategy is used to evaluate the long-term benefit of taking action a in state s. This benefit is determined by the weighted difference between grid support rewards and equipment stress penalties, with weighting coefficients... The magnitude of the generalized energy recoil eigenvector is dynamically adjusted.

[0106] when When the value is between the negative stability margin threshold and the positive instability threshold, The value was increased to prioritize equipment safety. When the power grid frequency deviation When the preset vulnerability threshold is exceeded, the arbitration mechanism prioritizes the converter-level judgment result and relaxes the equipment-side protection threshold to prioritize grid support functions; when the electrochemical stress uncertainty is high... When the preset risk threshold is exceeded, the arbitration mechanism will prioritize the judgment result of the battery cluster layer and tighten the protection threshold to prioritize the safety of the energy storage unit. The arbitration result reverses the electrochemical impedance spectroscopy by adjusting the weight allocation. This forms a complete algorithmic recursive closed loop from the decision-making stage to the perception stage.

[0107] The grid support priority arbitration mechanism is a core decision-making unit that dynamically balances different control objectives (protecting the power grid vs. protecting equipment).

[0108] The arbitration mechanism receives three key status inputs: Electrochemical stress uncertainty This value reflects the confidence level in the estimate of the battery's current polarization state. For example, it can be based on recent data. It is calculated using the variance of the observed values; the larger the variance, the better. The larger the value, the more difficult it is to accurately assess the battery's condition.

[0109] Power mutation uncertainty This value reflects the error level of the converter layer in predicting power surge events. For example, it can be obtained by comparing the predicted power value with the actual measured value and calculating its root mean square error.

[0110] Grid frequency deviation This refers to the difference between the actual frequency of the power grid and the rated frequency.

[0111] The arbitration mechanism employs a dynamic programming strategy, the core of which is a value function. This function is used to evaluate the performance of a given system in a specific system state s (as defined by...). , , Under the given conditions, the long-term comprehensive benefit that can be obtained by taking a certain weight allocation action a.

[0112] The value function consists of two reward functions: Power grid support revenue reward function This function quantifies the positive benefits of action 'a' in mitigating grid frequency deviation and enhancing grid stability. Its construction principle is... The smaller the rate of change, the higher the reward value. For example, it can be constructed as a negative [variable]. The form related to square terms.

[0113] Equipment stress cost reward function This function quantifies the additional electrochemical stress risk cost that action 'a' may impose on devices such as batteries. Its construction principle is... The larger the value, the lower the reward (and the higher the cost). For example, it can be constructed as a negative [value]. The form related to square terms.

[0114] These two reward functions are passed through a dynamic weight. To synthesize. This is called the dynamic recoil confidence weight, and its value is related to the generalized energy recoil eigenvector. The modulus is positively correlated. For example, when When it is large (e.g., >0.8), The value is 0.7; when When the value is small (e.g., <0.3), The value is 0.3. This means that when a strong energy recoil characteristic is detected, the decision will be more inclined to consider the safety costs of the equipment.

[0115] The arbitration mechanism executes the following rules based on preset thresholds: When the power grid frequency deviation When the voltage exceeds a preset vulnerability threshold (e.g., 0.5Hz or 0.8Hz), it indicates that the power grid urgently needs support. In this case, the arbitration mechanism will prioritize the judgment result of the converter layer (fast power regulation) and relax the protection threshold on the equipment side to prioritize the protection function of the power grid.

[0116] When electrochemical stress uncertainty When the risk threshold is exceeded (e.g., 0.6 or 0.7), it indicates that the battery state risk is high and difficult to predict. In this case, the arbitration mechanism will prioritize the judgment result of the battery cluster layer and tighten the protection threshold to ensure the safety of the energy storage unit.

[0117] The final output of the arbitration is the weights assigned to the control commands at the battery cluster layer, converter layer, and system layer. This decision also forms a closed loop: based on the final weight allocation, the attenuation coefficient β in the electrochemical impedance observer is fine-tuned in reverse. For example, if the decision favors equipment protection (high weight for the battery cluster layer), the β value can be slightly increased by 1%, making the observer's estimation of polarization state more robust; if the decision favors grid support (high weight for the converter layer), the β value can be slightly decreased by 1%, making the observer respond faster. This correction process is performed at a low frequency (e.g., every 1 second), achieving recursive closed-loop optimization from high-level decision-making to low-level perception, giving the entire system overall adaptive capabilities.

[0118] Optionally, the multi-level network-type energy storage control and protection method can be implemented on an embedded heterogeneous computing platform (DSP runs the observer + FPGA performs the comparison + coprocessor solves the MPC).

[0119] In one specific implementation, combined with Figure 4 This paper provides a detailed description of specific implementations of the hierarchical collaborative decision-making and control system architecture of the present invention. For example... Figure 4 As shown, when an abnormal event is detected (i.e., an out-of-limit state is triggered), all relevant information of the event is synchronously distributed to three independently running hierarchical prediction models for parallel deep analysis.

[0120] The bottom-level battery cluster prediction model primarily assesses the battery from an electrochemical safety perspective. Its input focuses on the battery's internal state characteristics, such as state variables reflecting polarization and their fluctuation uncertainties. The core function of this model is to assess internal risk; that is, based on the battery's current chemical kinetics, it determines whether the impact may cause irreversible damage and outputs an estimate of the level of electrochemical stress risk.

[0121] The intermediate converter layer prediction model analyzes the power conversion and balancing from the perspective of power conversion. Its input information includes instantaneous power deviation, current mutation rate, and prediction uncertainty. The core function of this model is to assess power surges, that is, to analyze whether the power converter has the ability to quickly smooth out the fluctuation under the current operating conditions, and output an estimate of the strength of the system's power balancing capability.

[0122] The highest-level system-level prediction model assesses the situation from a macroscopic perspective of grid interaction and overall stability. Its input information primarily consists of grid state characteristics, such as frequency and voltage deviations at grid connection points. The core function of this model is to evaluate stability impacts, specifically determining the potential influence of the event on the synchronization stability and power quality of the local grid, and outputting an estimate of the urgency of grid stability support requirements.

[0123] The evaluation results of the three-level models are simultaneously uploaded to the grid support priority arbitrator. This arbitrator is the core unit for making the final decision. It receives the evaluation results from each level and makes a ruling by comprehensively considering the current global state of the system (such as real-time grid frequency deviation and overall battery health).

[0124] The arbitrator operates internally using a dynamic programming strategy. Its core logic lies in performing multi-objective trade-offs: finding the current optimal balance between the sometimes conflicting objectives of "ensuring grid stability" and "ensuring equipment safety." The arbitrator calculates the benefits of grid support and the costs of equipment losses under different decision weights, ultimately generating an optimal weight allocation scheme and a consistent event type determination conclusion.

[0125] Based on the weighting scheme output by the arbitrator, the system generates a coordinated protection control command set. This command set is not a single command, but a set of coordinated tasks assigned to different actuators: according to the battery layer weight, it issues commands to the battery management system to adjust the cell current limit or disconnect a specific module; according to the converter layer weight, it issues commands to the power converter to switch control modes or inject precise compensation current; according to the system layer weight, it issues commands to the higher-level coordinating controller to adjust the grid-connected power plan or initiate mutual assistance with adjacent units. Each actuator (battery manager, power converter, grid dispatcher) acts synchronously according to the commands it receives, achieving a system-level coordinated response.

[0126] It's important to note that this architecture includes a closed-loop feedback path from the decision-making layer to the perception layer. The final decision weight of the arbitrator serves as a feedback signal, propagating back to the underlying feature perception layer. For example, if the arbitration decision favors protecting device safety in this event (giving the battery layer a higher weight), the system may fine-tune the internal parameters of the battery state observer, slightly increasing its sensitivity to similar features in the future. This feedback mechanism enables the entire system not only to make collaborative decisions in a single event but also to adaptively optimize the coordination between its "senses" and "brain" over long-term operation, forming a continuously self-adjusting intelligent whole. This allows for precise protection and control in the complex and ever-changing charging pile operating environment.

[0127] It should be noted that the embodiments implemented on the side of the multi-level grid-type energy storage control and protection system in this application can be referenced with the embodiments implemented on the side of the multi-level grid-type energy storage control and protection method, and will not be described in detail in this application.

[0128] According to another aspect of the embodiments of this application, an electronic device for implementing the above-described multi-level grid-type energy storage control and protection method is also provided. This electronic device may be... Figure 5 The terminal device or server shown. This embodiment uses this electronic device as an example of a server. Figure 5 As shown, the electronic device includes a memory 402, a processor 404, and a transmission device 406. The memory 402 stores a computer program, and the processor 404 is configured to execute the steps in any of the above method embodiments through the computer program.

[0129] Optionally, in this embodiment, the aforementioned electronic device may be located in at least one of a plurality of network devices in a computer network.

[0130] Optionally, the transmission device 406 is used to receive or send data via a network. Specific examples of the network described above may include wired and wireless networks. In one example, the transmission device 406 includes a Network Interface Controller (NIC), which can be connected to other network devices and a router via a network cable to communicate with the Internet or a local area network. In another example, the transmission device 406 is a radio frequency (RF) module used to communicate with the Internet wirelessly. Furthermore, the electronic device also includes a display 408 and a connection bus 410, which connects the various module components within the electronic device.

[0131] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A multi-level grid-type energy storage control and protection system, characterized in that, include: The construction unit is used to perform feature modeling on the multi-source operating signals of the collected energy storage system and construct a multi-source operating feature set; The extraction unit is used to perform time series rate of change analysis on the multi-source operating feature set and extract the dynamic features of current direction change and power reversal. An energy recoil determination model is established based on the aforementioned dynamic characteristics to generate a dynamic energy feature set; The calculation unit is used to perform energy disturbance amplitude analysis on the dynamic energy feature set and to calculate the adaptive protection threshold set in real time based on the energy recoil intensity and duration. The identification unit is used to compare the real-time sampled signal with the set of adaptive protection thresholds to identify whether the over-limit state has been triggered. The output unit is used to extract the real-time dynamic features of the over-limit event when an over-limit state is detected, match the real-time dynamic features of the over-limit event with the energy backlash determination model, determine the abnormal type of the over-limit event based on the matching result, and output the protection action command or dynamic balance control command corresponding to the abnormal type to form a coordinated protection control command set.

2. A multi-level grid-type energy storage control and protection method, characterized in that, include: S1, perform feature modeling on the multi-source operation signals of the collected energy storage system to construct a multi-source operation feature set; S2, perform time series rate of change analysis on the multi-source operation feature set to extract dynamic features of current direction change and power reversal; An energy recoil determination model is established based on the aforementioned dynamic characteristics to generate a dynamic energy feature set; S3, perform energy disturbance amplitude analysis on the dynamic energy feature set, and calculate the adaptive protection threshold set in real time based on the energy recoil intensity and duration; S4, compare the real-time sampled signal with the set of adaptive protection thresholds to identify whether the over-limit state has been triggered; S5. When an over-limit state is detected to be triggered, the real-time dynamic features of the over-limit event are extracted, and the real-time dynamic features of the over-limit event are matched and calculated with the energy backlash determination model. Based on the matching result, the abnormal type of the over-limit event is determined, and the protection action command or dynamic balance control command corresponding to the abnormal type is output to form a coordinated protection control command set.

3. The multi-level grid-type energy storage control and protection method according to claim 2, characterized in that, The energy recoil determination model established in S2 includes a reference parameter range, which is provided by a dynamic coupling knowledge base. The dynamic coupling knowledge base adopts a three-dimensional operating condition space index structure, and the three-dimensional operating condition space consists of the state of charge, battery temperature, and absolute value of the rate of change of current. The state of charge, battery temperature, and absolute value of current change rate are obtained in real time from the multi-source operating signals collected in S1.

4. The multi-level grid-type energy storage control and protection method according to claim 3, characterized in that, When calculating the adaptive protection threshold set in S3, the adaptive protection threshold set is obtained by querying the asymmetric threshold parameters of the operating condition space unit determined by the current state of charge, battery temperature and absolute value of current change rate in the dynamic coupling knowledge base; the asymmetric threshold parameters include rising edge threshold and falling edge threshold respectively calibrated based on electrochemical hysteresis effect.

5. The multi-level grid-type energy storage control and protection method according to claim 3, characterized in that, The dynamically coupled knowledge base achieves online iterative updates through a recoil event learning strategy; the recoil event learning strategy includes: When the abnormal type of the over-limit event is determined to be energy recoil transient and the determination result is consistent with the actual system response, the real-time dynamic characteristics of the over-limit event are used as positive samples to expand the recoil characteristic statistical boundary of the current operating condition space unit. The recoil characteristic statistical boundary is used to update the reference parameter range.

6. The multi-level grid-type energy storage control and protection method according to claim 5, characterized in that, The recoil event learning strategy also includes: When the abnormal type of the out-of-limit event is determined to be a real fault and the out-of-limit state persists for more than a preset time after the protection action is executed, the threshold envelope of the current working condition space unit is shrunk according to a preset shrinkage ratio. The shrinkage of the threshold envelope is applied to the boundary values ​​of the asymmetric threshold parameters in the dynamic coupling knowledge base.

7. The multi-level grid-type energy storage control and protection method according to claim 2, characterized in that, The multi-source operating feature set constructed in S1 includes normalized electrochemical polarization states. ; It is generated in real time by an electrochemical impedance spectroscopy (EIS) observer.

8. The multi-level grid-type energy storage control and protection method according to claim 7, characterized in that, The energy recoil determination model established in S2 includes models based on... The determination confidence correction logic; the energy backlash determination model outputs a similarity determination threshold for matching calculation in S5; The confidence correction logic is used to dynamically adjust the similarity determination threshold.

9. The multi-level grid-type energy storage control and protection method according to claim 2, characterized in that, In S3, a hierarchical model predictive control architecture is used to calculate the adaptive protection threshold set. The hierarchical model predictive control architecture includes a battery cluster layer predictive model, a converter layer predictive model, and a system layer predictive model.

10. The multi-level grid-type energy storage control and protection method according to claim 9, characterized in that, The objective function of each prediction model incorporates a penalty term for the discrimination uncertainty output of other prediction models.

Images