A control method based on an integrated energy storage battery cabinet

Abstract

The present application relates to the technical field of electrochemical energy storage, in particular to a control method based on an integrated energy storage battery cabinet to solve the technical problem of short board effect and invalid prediction model caused by static grouping of retired batteries lacking historical archives; the technical solution comprises: applying a pulse during system load rest period, extracting and analyzing to generate a transient feature vector without historical dependence; running the Euclidean distance clustering algorithm to minimize the distance of the same group module to output the grouping division; accordingly, the switch matrix reconstructs the topology; real-time monitoring of feature drift, abnormal module is separated as a grid-connected isolation state, through target DC bus voltage pre-synchronization closed-loop adjustment to realize non-impact grid connection, and according to the slope rate to smooth recovery; finally, based on the average value of each group feature, the differentiated cutoff voltage is issued; the present application has the beneficial effects of completely getting rid of the bondage of historical data, realizing seamless safe migration under electricity, and maximizing the available capacity of heterogeneous energy storage cabinet.

Description

Technical Field

[0001] This invention relates to the field of electrochemical energy storage technology, and more specifically, to a control method based on an integrated energy storage battery cabinet. Background Technology

[0002] Integrated energy storage battery cabinets are key equipment for recombining retired lithium-ion battery modules for use in microgrid peak shaving, backup power, and other scenarios. The cascade utilization of retired batteries is an important way to achieve closed-loop recycling of battery resources.

[0003] Existing cascade utilization control methods are mainly divided into two categories: The first category adopts static fixed grouping and passive balancing (such as matching capacity through offline sorting), but offline detection cannot cover the asynchronous aging of batteries under real operating conditions, and fixed series connection is prone to causing the weakest link effect, resulting in limited overall cabinet capacity; The second category adopts a model reconstruction scheme based on state of health (SOH) prediction. This type of scheme relies heavily on the historical operating records of batteries (such as historical internal resistance change rate, capacity decay curve, etc.) to predict future decay trends and thus optimize the topology.

[0004] However, retired batteries in reality come from a variety of sources and generally suffer from a complete lack of historical operating data and a high degree of uncertainty about their initial state. Existing degradation prediction models based on historical data will completely fail under these conditions and will be unable to calculate the degradation factor. Furthermore, traditional passive balancing cannot cope with sudden individual degradation during battery operation. This results in existing control methods being unable to achieve dynamic performance matching and isolation for batteries with completely unknown history. The system is highly susceptible to overcharging / over-discharging due to sudden degradation of local cells, leading to a sharp reduction in overall usable capacity and accelerating system failure.

[0005] Therefore, a control method based on an integrated energy storage battery cabinet is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a control method based on an integrated energy storage battery cabinet to solve the technical problems of the bottleneck effect and prediction model failure caused by static grouping of retired batteries lacking historical records.

[0007] To solve the above-mentioned technical problems, the present invention aims to provide a control method based on an integrated energy storage battery cabinet, executed by a central controller, comprising the following steps: S1: When the real-time output power and bus voltage fluctuation rate are both less than the corresponding set values, it is determined to be a system load rest period; during this period, feature recognition current pulses are applied to each battery module to obtain terminal voltage timing data, and internal resistance parameters, polarization parameters and time constants are analyzed to generate transient feature vectors based only on the current terminal voltage timing data. S2: Input all transient feature vectors into the Euclidean distance clustering algorithm to minimize the spatial distance of transient feature vectors within the same battery group, and output the grouping result; S3: Based on the grouping results, a switch status command is sent to the module switch matrix to connect the positive and negative terminals of the battery modules in the same battery group in series, and the two ends of the series link are connected to the corresponding cluster DC bus. Each cluster DC bus is connected to the system main bus via a cluster bidirectional DC converter. S4: Periodically update the transient feature vector during the system load rest period; when the Euclidean distance of the new transient feature vector of the first battery module from the target cluster center is less than the lower threshold and the Euclidean distance from the source cluster center is greater than the upper threshold, control the corresponding module switching matrix to disconnect it from the source cluster DC bus and mark it as an isolated state to be connected to the grid; S5: Obtain the open-circuit voltage of the first battery module, send a voltage regulation command to the cluster bidirectional DC-DC converter corresponding to the target battery group, and control the DC bus voltage of the target cluster to converge to the open-circuit voltage; when the absolute value of the voltage difference between the two is less than the voltage synchronization threshold, control the corresponding module switch matrix to connect the first battery module to the DC bus of the target cluster. S6: Extract the characteristic mean of each battery group, substitute it into the preset mapping function to calculate the charge / discharge cutoff voltage threshold, and send it to the corresponding cluster bidirectional DC-DC converter.

[0008] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This control method based on integrated energy storage battery cabinets eliminates the reliance on historical operation records of retired batteries. By applying current pulses during system downtime, it directly extracts transient physical response characteristics, achieving an instant snapshot of the current battery condition. Based on physical characteristics rather than trend prediction, it performs clustering and dynamic migration. When a battery experiences sudden degradation, it can achieve energized migration of faulty cells through a disconnection and DC bus voltage pre-synchronization mechanism. This eliminates grid connection impact and maximizes the release of the system's available capacity at the current moment through differentiated cutoff voltage control.

[0009] 2. In this control method based on integrated energy storage battery cabinet, micro-current pulses are generated by multiplexing through high-frequency chopper circuit, eliminating the need for additional high-power test signal sources and reducing hardware costs.

[0010] 3. The control method based on the integrated energy storage battery cabinet introduces a polynomial smoothing filter for pre-cleaning, which eliminates high-frequency electromagnetic noise caused by the operation of high-power power electronic switches, ensuring the absolute accuracy of internal resistance and polarization characteristic analysis.

[0011] 4. The control method based on integrated energy storage battery cabinet introduces system steady-state recovery logic with ramp rate after cross-cluster grid connection, which avoids power oscillation caused by voltage reference jump and ensures the stability of main bus power supply quality. Attached Figure Description

[0012] Figure 1 This is a flowchart of the overall method of the present invention. Detailed Implementation

[0013] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Examples, such as Figure 1 As shown, the purpose of this invention is to provide a control method based on an integrated energy storage battery cabinet, executed by a central controller, comprising the following steps: S1: When the real-time output power and bus voltage fluctuation rate are both less than the corresponding set values, it is determined to be a system load rest period; during this period, feature recognition current pulses are applied to each battery module to obtain terminal voltage timing data, and internal resistance parameters, polarization parameters and time constants are analyzed to generate transient feature vectors based only on the current terminal voltage timing data. S2: Input all transient feature vectors into the Euclidean distance clustering algorithm to minimize the spatial distance of transient feature vectors within the same battery group, and output the grouping result; S3: Based on the grouping results, a switch status command is sent to the module switch matrix to connect the positive and negative terminals of the battery modules in the same battery group in series, and the two ends of the series link are connected to the corresponding cluster DC bus. Each cluster DC bus is connected to the system main bus via a cluster bidirectional DC converter. S4: Periodically update the transient feature vector during the system load rest period; when the Euclidean distance of the new transient feature vector of the first battery module from the target cluster center is less than the lower threshold and the Euclidean distance from the source cluster center is greater than the upper threshold, control the corresponding module switching matrix to disconnect it from the source cluster DC bus and mark it as an isolated state to be connected to the grid; S5: Obtain the open-circuit voltage of the first battery module, send a voltage regulation command to the cluster bidirectional DC-DC converter corresponding to the target battery group, and control the DC bus voltage of the target cluster to converge to the open-circuit voltage; when the absolute value of the voltage difference between the two is less than the voltage synchronization threshold, control the corresponding module switch matrix to connect the first battery module to the DC bus of the target cluster. S6: Extract the characteristic mean of each battery group, substitute it into the preset mapping function to calculate the charge / discharge cutoff voltage threshold, and send it to the corresponding cluster bidirectional DC-DC converter.

[0015] The integrated energy storage battery cabinet of this embodiment includes: multiple retired battery modules; a module access unit configured for each module, containing a controlled module switch matrix; at least two independent cluster DC buses; each cluster DC bus is connected to a common system main bus via a cluster bidirectional DC converter; the main bus is connected to an energy storage converter; the central controller is connected to each module access unit, each cluster bidirectional DC converter, and voltage and current sensors via a communication bus, and is used to execute all the control steps of the present invention. Furthermore, each step is disclosed.

[0016] Considering that retired battery modules used in tiered applications generally lack complete and reliable historical operation records, and the differences in their initial factory condition and mileage and operating conditions during vehicle use cannot be traced, existing technologies attempt to assess the aging degree of each module by establishing a health status estimation model. However, the accuracy of the model parameters is highly dependent on the accumulation of historical data. When historical records are missing or incorrect, the model output is distorted, and the control decisions made based on this will exacerbate the inconsistency between modules. Therefore, step S1 is proposed, and the specific implementation method is as follows.

[0017] The central controller collects real-time data on the total output power and bus voltage fluctuation rate on the main bus of the system. When both conditions are met simultaneously—the real-time output power is less than the power resting setpoint and the bus voltage fluctuation rate is less than the fluctuation judgment setpoint—the central controller determines that the system has entered a load rest period. The power resting setpoint is 5% of the system's rated power, and the fluctuation judgment setpoint is 1% of the system's rated voltage. The reason for setting these two judgment conditions is that when the energy storage cabinet has a large power output, there is a large operating current on the system bus and each cluster of DC buses. If a test pulse is superimposed at this time, it will not only interfere with the normal charging and discharging process, but the voltage response caused by the pulse will also be mixed with the voltage change caused by the operating current, making it difficult to separate the response signal generated solely by the pulse excitation, resulting in a significant decrease in the accuracy of subsequent feature extraction. Only by detecting during the rest period when the external power is close to zero and the bus voltage is stable can pure terminal voltage response data unaffected by the operating current be obtained, laying the foundation for accurate analysis of subsequent feature parameters.

[0018] After determining that the load has entered a rest period, the central controller applies a characteristic identification current pulse to each battery module. The pulse amplitude is set to 0.5 times the nominal capacity of the battery module, i.e., 0.5C, and the duration is set to 2 seconds. The pulse form is a single constant current discharge pulse. After the discharge, the module remains in a static state with zero current for 3 seconds. The reason for choosing an amplitude of 0.5C is that this current level is sufficient to stimulate a significant and measurable voltage response from the ohmic resistance and polarization resistance inside the battery, while being much lower than the 1C to 3C rates that the module often withstands during automotive applications. It will not cause additional stress damage to the already aged retired modules; the 2-second duration is determined based on the electrochemical response characteristics of the battery: the voltage step caused by the ohmic internal resistance is completed within 10 milliseconds, while electrochemical polarization and concentration polarization fully develop and tend to stabilize within about 1.5 to 2 seconds. Shorter pulses cannot obtain complete polarization information, while longer pulses unnecessarily increase detection time and energy consumption; the 3-second resting time is sufficient for the polarization voltage to recover to more than 90% of the stable value, providing a sufficient observation window for subsequent extraction of steady-state recovery band features.

[0019] During the pulse application and subsequent resting process, the central controller synchronously acquires the terminal voltage of each battery module through the voltage sensor corresponding to each battery module at a sampling rate of no less than 100 Hz. The 100 Hz sampling rate ensures that at least one valid data point can be captured for the instantaneous step process, and hundreds of data points can be used for fitting analysis for the relaxation and recovery process, thus obtaining the terminal voltage timing data.

[0020] After obtaining the terminal voltage timing data, the central controller performs characteristic parameter analysis. The analysis process first calculates the voltage change rate sequence of the entire timing data, which is the voltage difference between adjacent sampling points divided by the sampling interval; then, the data is divided into three bands according to the magnitude of the change rate.

[0021] The first band is the instantaneous step band: at the start of the discharge pulse, the voltage drops sharply in a very short time, with the absolute value of the rate of change being much greater than in other time periods. By detecting the peak position of the rate of change, the starting and ending points of the step band are determined by extending forward and backward until the rate of change falls back to one-tenth of the peak value. The typical duration is 10 milliseconds. The difference between the voltage of the first sampling point after the pulse is applied and the voltage of the last sampling point before the pulse is applied within this band is extracted and recorded as the voltage change amplitude. The reason for choosing the initial moment of the step band is that the polarization effect of the battery has not yet been established at this time, and the voltage change is almost entirely caused by the ohmic internal resistance. Dividing the voltage change amplitude by the current amplitude of the feature identification current pulse yields the internal resistance parameter, which mainly characterizes the electrolyte ion conduction resistance, electrode material electronic conduction resistance, and contact resistance of each connection interface of the module.

[0022] The second band is the exponential relaxation band: immediately following the instantaneous step band, the voltage continues to decrease or slightly rebound at a gradually slowing rate, with the rate of change decreasing significantly but not yet approaching zero; this band begins approximately 0.1 seconds after the pulse is applied and lasts until 2 seconds after the pulse ends; the voltage change during this stage is mainly the process of the gradual establishment of the electrochemical polarization overpotential; the voltage change data over time within this band is fitted with a first-order exponential function, and the decay time coefficient obtained from the fitting is extracted and output as a time constant; the time constant reflects the combined relaxation rate of charge transfer on the electrode surface and solid-state diffusion of lithium ions inside the active material particles. The larger its value, the slower the electrode reaction kinetics and the deeper the aging of the module.

[0023] The third band is the steady-state recovery band: after the pulse ends and the current returns to zero, the voltage jumps instantaneously for a period of time, and then slowly recovers to a stable value; this stage begins after the pulse ends and covers the entire 3-second rest period; the average voltage of the last 0.5 seconds of the rest period is extracted as the steady-state extreme voltage; the difference between this steady-state extreme voltage and the initial voltage before the pulse is applied is calculated, and the output is the polarization parameter; the polarization parameter essentially represents the total overpotential accumulated inside the battery during the 2-second discharge pulse due to the combined effects of electrochemical polarization and concentration polarization.

[0024] The three parameters mentioned above are combined into a three-dimensional vector, which is the transient characteristic vector of the battery module. In order to eliminate the dimensional influence caused by the difference in nominal capacity and nominal voltage between different modules, the internal resistance parameter and polarization parameter are converted into dimensionless per-unit values ​​based on the nominal voltage of the module.

[0025] By utilizing the transient response of the terminal voltage under a single current pulse excitation, the multi-dimensional features reflecting the current health status of the battery module can be extracted quickly and accurately without relying on any historical data. This solves the technical problem of control failure caused by missing or incorrect historical data in cascaded batteries, enabling the system to obtain reliable module status information under any data conditions, and providing accurate input for subsequent grouping and dynamic management.

[0026] Considering that the feature identification current pulse is applied in step S1, if an independent dedicated pulse power supply module is used, an additional power generation circuit and its heat dissipation and control unit need to be added in the energy storage battery cabinet. This not only increases the hardware cost and the space occupied in the cabinet, but also introduces a new fault point. Considering that the existing cluster bidirectional DC-DC converter in the energy storage battery cabinet itself has a controllable high-frequency chopper circuit and has the ability to actively adjust the output current, in step S1, the feature identification current pulse is applied to each battery module during this period, including the following collaborative control logic: generating a perturbation injection command containing the pulse amplitude and duration; issuing the perturbation injection command to each cluster bidirectional DC-DC converter in the integrated energy storage battery cabinet, driving the high-frequency chopper circuit of the cluster bidirectional DC-DC converter to superimpose an alternating micro-current signal on the basic DC output, forming the feature identification current pulse to be injected into the downstream battery module. The specific implementation method is as follows.

[0027] The central controller first generates a perturbation injection command, which includes two key parameters: pulse amplitude and duration. The pulse amplitude corresponds to a 0.5C discharge current, and the duration corresponds to 2 seconds. The central controller then sends this perturbation injection command in parallel to each cluster of bidirectional DC-DC converters within the integrated energy storage battery cabinet via the communication bus. Each cluster of bidirectional DC-DC converters contains a high-frequency chopper circuit composed of insulated-gate bipolar transistors or metal-oxide-semiconductor field-effect transistors. During normal operation, this chopper circuit outputs a stable DC voltage. Upon receiving the perturbation injection command, the control unit of the chopper circuit superimposes a negative alternating micro-current signal onto the original basic DC output. The amplitude of this signal precisely corresponds to the 0.5C discharge current, and the duration is 2 seconds. After 2 seconds, the alternating signal automatically returns to zero, restoring the basic DC output. This forms a characteristic identification current pulse, which is injected into each downstream battery module via the DC bus.

[0028] The above technical solution utilizes the existing cluster bidirectional DC-DC converter hardware resources of the energy storage cabinet to generate characteristic identification current pulses, eliminating the need to add a separate pulse power supply module. This simplifies the hardware structure, reduces system complexity, material costs, and cabinet space occupation, while also reducing potential failure points and improving system reliability.

[0029] Considering that during the acquisition of terminal voltage time-series data in step S1, there are multiple high-frequency switching devices in the energy storage battery cabinet, including the chopper circuit of the cluster bidirectional DC-DC converter and the semiconductor switches of the module switch matrix, these devices will generate high-frequency electromagnetic interference when they operate. This interference enters the voltage sampling circuit through conduction and radiation coupling, and appears as random noise spikes superimposed on the real voltage signal in the original sampled data. If these noise spikes are used directly for subsequent feature parameter analysis without processing, they may be misidentified as voltage steps, leading to an overestimation or underestimation of the internal resistance parameter, which seriously affects the accuracy of the transient feature vector. Therefore, in step S1, before acquiring the terminal voltage time-series data and analyzing the internal resistance parameter, polarization parameter, and time constant, a pre-cleaning logic is performed on the terminal voltage time-series data. The pre-cleaning logic includes: extracting the original voltage data stream according to a preset sampling period; inputting the original voltage data stream into a polynomial smoothing filter; performing least-squares fitting on the data points within the extraction window; filtering out high-frequency electromagnetic interference noise data; and outputting the smoothed terminal voltage time-series data. The specific implementation method is as follows.

[0030] The central controller extracts the raw voltage data stream from the high-speed sampling data buffer according to a preset sampling period, i.e., an interval of 10 milliseconds per point. Then, the raw voltage data stream is input into a polynomial smoothing filter, which uses a Savitzky-Golay filter with a window width of 11 sampling points and a fitting order of 3. The filter performs least-squares polynomial fitting on the 11 data points within the window and replaces the original measured value with the value of the fitted curve at the center point of the window. The sliding window traverses the entire data sequence, and the value of each center point is replaced by the fitted value.

[0031] By employing the above technical solution, Savitzky-Golay polynomial smoothing filter effectively filters out high-frequency electromagnetic interference noise while preserving the transient change characteristics of the signal. The output is smoothed terminal voltage timing data. After pre-cleaning, the true voltage response waveform generated by the pulse excitation is completely preserved, while the local jitter caused by noise is suppressed, ensuring the accuracy and robustness of subsequent analysis of internal resistance parameters, polarization parameters, and time constants.

[0032] Since step S1 involves analyzing the internal resistance parameters, polarization parameters, and time constant from the terminal voltage time series data, a clear data processing logic is needed to separate the voltage response components from different physical sources. The voltage response of the battery module to the current pulse is the result of the superposition of multiple physical processes. The voltage change caused by the ohmic internal resistance is instantaneous; the voltage changes caused by electrochemical polarization and concentration polarization gradually build up and subside over time. If the time series data is not divided into bands and processed together, it will be impossible to accurately separate the characteristic parameters corresponding to each physical process. Therefore, step S1 involves acquiring the terminal voltage time series data and analyzing the internal resistance parameters, polarization parameters, and time constant, including the following data processing logic: The terminal voltage time series data is received and divided into an instantaneous step band, an exponential relaxation band, and a steady-state recovery band according to the voltage change rate. The voltage change amplitude of the instantaneous step band is extracted, and the ratio of the voltage change amplitude to the current amplitude of the feature identification current pulse is calculated. The ratio is output as the internal resistance parameter. An exponential function is fitted to the time series data of the exponential relaxation band, and the decay time coefficient in the fitted function is extracted and output as the time constant. The steady-state extreme voltage of the steady-state recovery band is extracted, and the difference between the steady-state extreme voltage and the initial voltage before the pulse is applied is calculated. The difference is output as the polarization parameter. The specific implementation method is as follows.

[0033] The central controller receives the smoothed terminal voltage timing data. First, it calculates the voltage change rate sequence of the entire timing data, which is the voltage difference between adjacent sampling points divided by the sampling interval. Then, it divides the data into three bands according to the magnitude of the change rate.

[0034] The first band is the instantaneous step band: at the start of the discharge pulse, the voltage drops sharply in a very short time, and the absolute value of the rate of change is much greater than in other periods. By detecting the peak position of the rate of change, the starting and ending points of the step band are determined by extending forward and backward until the rate of change falls back to one-tenth of the peak value. The typical duration is 10 milliseconds. The difference between the voltage of the first sampling point after the pulse is applied and the voltage of the last sampling point before the pulse is applied is extracted and recorded as the voltage change amplitude ΔV1. The reason for choosing the initial moment of the step band is that the polarization effect of the battery has not yet been established at this time, and the voltage change is almost entirely caused by the ohmic internal resistance. Dividing ΔV1 by the current amplitude I of the feature identification current pulse, the internal resistance parameter Ro is obtained. The unit of Ro is milliohms, which mainly characterizes the electrolyte ion conduction resistance, electrode material electronic conduction resistance, and contact resistance of each connection interface of the module.

[0035] The second band is the exponential relaxation band: immediately following the instantaneous step band, the voltage continues to decrease or slightly rebound at a gradually slowing rate, with the rate of change decreasing significantly but not yet approaching zero. This band begins approximately 0.1 seconds after the pulse is applied and lasts until 2 seconds after the pulse ends. The voltage change during this stage is mainly a process of gradual establishment of the electrochemical polarization overpotential. The voltage change data over time within this band is fitted with a first-order exponential function, with the fitting function form being ΔV2×[1-exp(-t / τ)], where ΔV2 is the polarization voltage amplitude, τ is the time constant, and t is time. The optimal τ value is solved using the Levenberg-Marquardt nonlinear least squares algorithm. The decay time coefficient τ obtained from the fitting is extracted and output as a time constant. The unit of τ is seconds, reflecting the combined relaxation rate of charge transfer on the electrode surface and solid-state diffusion of lithium ions inside the active material particles. The larger the τ value, the slower the electrode reaction kinetics and the deeper the aging of the module.

[0036] The third band is the steady-state recovery band: after the pulse ends and the current returns to zero, the voltage jumps instantaneously for a period of time, and then slowly recovers to a stable value. This stage begins after the pulse ends and covers the entire 3-second resting period. The average voltage value of the last 0.5 seconds of the resting period is extracted as the steady-state extreme voltage. The difference between this steady-state extreme voltage and the initial voltage before the pulse is applied is calculated, and the output is the polarization parameter Rp. Rp essentially represents the total overpotential accumulated inside the battery during the 2-second discharge pulse due to the combined effects of electrochemical polarization and concentration polarization. The unit of Rp is volts.

[0037] The above three parameters are combined into a three-dimensional vector F=[Ro,Rp,τ], which is the transient characteristic vector of the battery module. In order to eliminate the dimensional influence caused by the difference in nominal capacity and nominal voltage between different modules, Ro and Rp are converted into dimensionless per-unit values ​​based on the nominal voltage of the module. Specifically, the per-unit value of Ro is equal to the actual value of Ro divided by the nominal voltage of the module and then multiplied by the nominal capacity of the module, and the per-unit value of Rp is equal to the actual value of Rp divided by the nominal voltage of the module.

[0038] It is important to emphasize that the entire calculation process described above only uses the terminal voltage timing data collected during the current pulse test, without referencing any previous operating records, charge / discharge curves, or health status estimates of the module. Therefore, the transient characteristic vector is a snapshot that completely reflects the current physical state of the module and is unrelated to historical records. This characteristic enables the method to function normally even when historical records are completely missing, solving the industry problem of unreliable control basis due to data loss in cascaded batteries.

[0039] After obtaining the transient characteristic vectors of each module, how to transform the quality information represented by these vectors into a battery module grouping scheme? The transient characteristic vectors of retired battery modules are discretely distributed in the three-dimensional feature space. Modules with good quality are clustered in the region of low internal resistance, low polarization, and short time constant, while modules with poor quality are scattered in the region of high internal resistance, high polarization, and long time constant. A method is needed to accurately identify this natural clustering structure in the feature space so that the grouping results can truly reflect the similarity and difference in electrical characteristics between modules, and ensure that the voltage changes of modules in the same group are highly synchronized when running in series. Step S2 is proposed, and the specific implementation is as follows.

[0040] The central controller constructs a sample set from the transient feature vectors of all online modules generated in step S1. Each sample is a three-dimensional vector. Since the dimensions of the three components have been unified through normalization, the feature space is a dimensionless space, and the values ​​of each dimension are within a comparable range.

[0041] Euclidean distance is chosen as the similarity measure. The Euclidean distance between two transient feature vectors is defined as the square root of the sum of the squares of the differences of the three components. In this embodiment, Euclidean distance is used instead of Mahalanobis distance or other weighted distances because: the normalized components Ro, Rp, and τ correspond to ohmic characteristics, polarization characteristics, and dynamic velocity characteristics, respectively. These three are physically independent processes with weak linear correlation. Euclidean distance directly measures the absolute distance of the vectors in the three-dimensional physical feature space, with a clear physical meaning. The smaller the distance, the closer the electrical response characteristics of the two modules are, and the smaller the voltage inconsistency caused by differences in internal resistance and polarization when they are connected in series. Mahalanobis distance, on the other hand, introduces the inverse transformation of the covariance matrix. When the feature components themselves are not strongly correlated, it may introduce unnecessary computational complexity. Furthermore, the estimation of the covariance matrix requires a sufficient sample size, which may not be reliable in small energy storage cabinets with a limited number of modules.

[0042] The objective function of the clustering algorithm is to minimize the sum of squared Euclidean distances from all modules within the same battery group to the cluster center of that group. This is equivalent to minimizing the spatial distance between transient feature vectors within the same group. The K-means algorithm is used to achieve this objective. The initial cluster centers are randomly selected, and the algorithm converges to a local optimum through iterative allocation and updating of the centers. The algorithm is repeated multiple times, and the result with the minimum objective function value is taken as the final result.

[0043] The initial value of the cluster number K is set to the number of cluster DC buses in the system hardware configuration. For example, the energy storage cabinet in this embodiment is configured with three independent cluster DC buses Bus-A, Bus-B, and Bus-C. Therefore, the initial value of K is 3. After clustering, each module is assigned a group label. Group A, group B, and group C each correspond to a cluster center vector. The cluster center of group A is located in the region of small Ro, small Rp, and short τ in the feature space. The cluster center of group C is located in the region of large Ro, large Rp, and long τ. Group B is in the middle. The output is the grouping result.

[0044] The above technical solution utilizes the Euclidean distance clustering algorithm to automatically group modules in the three-dimensional physical feature space, so that modules with similar electrical response characteristics are grouped into the same group. This results in high internal consistency within the group, small voltage differences during subsequent series operation, reduced workload of the equalization circuit, and improved energy utilization efficiency of the entire cluster.

[0045] In step S2, during the execution of the Euclidean distance clustering algorithm, the choice of the number of clusters K directly affects the grouping effect. If the value of K remains fixed, the overall degradation of the modules may deepen with long-term operation of the battery cabinet, and the dispersion between modules may change significantly. When the dispersion increases, a fixed, small K value will cause some modules with large differences in feature vectors to be mixed into some groups, resulting in decreased consistency within the groups. When the dispersion decreases, a fixed, large K value will cause the grouping to be too fine, with too few modules in each group, increasing the system control complexity without any practical benefit. Therefore, in step S2, during the execution of the Euclidean distance clustering algorithm, the following dynamic adjustment mechanism is included: calculating the global variance data of all transient feature vectors in the feature space; when it is determined that the global variance data is greater than the upper limit of the variance setting, increasing the number of grouping results output by the clustering algorithm; when it is determined that the global variance data is less than the lower limit of the variance setting, decreasing the number of grouping results output by the clustering algorithm. The specific implementation method is as follows.

[0046] In each clustering cycle, the central controller calculates the global variance data of all transient feature vectors in the feature space. The specific calculation method of the global variance is as follows: first, calculate the overall mean vector of all samples in three dimensions; then, calculate the squared Euclidean distance of each sample vector to the overall mean vector; and then calculate the arithmetic mean of this value for all samples. The larger the global variance, the higher the overall dispersion of the module and the greater the difference in physical properties between different modules.

[0047] When the global variance data exceeds the upper limit of the variance setting, it means that the currently set number of groups is insufficient to accommodate the differences between modules. Some groups may have already mixed in modules with large differences in feature vectors, leading to a decrease in consistency within the group. In this case, the number of groups output by the clustering algorithm is automatically increased, and the K value is incremented by one. Conversely, when the global variance data is less than the lower limit of the variance setting, it indicates that the modules as a whole are becoming more consistent, and a smaller number of groups can meet the consistency requirements. In this case, the K value is decremented by one. The upper and lower limit of the variance setting can be automatically calibrated after the system is first powered on and a certain amount of detection data is accumulated: the average global variance of the first ten detection cycles is taken, with the upper limit set at 150% of this average value and the lower limit set at 50%. The adjustment range of the K value is limited to a minimum of 2 and a maximum of 6 to avoid insufficient groups leading to a lack of consistency within the groups, or too many groups resulting in too few modules in each group, thus losing the meaning of serialization.

[0048] Through the above technical solution, the system can automatically adjust the management granularity according to the overall degradation trend of the modules in the battery cabinet by using a cluster number adaptive adjustment mechanism based on global variance. When the dispersion of the modules increases, the grouping is refined in time to ensure consistency within the group. When the modules tend to be consistent, the grouping is appropriately merged to simplify the system, and the optimal grouping effect is always maintained throughout the entire life cycle.

[0049] Considering that retired battery modules are physically installed in fixed positions inside the cabinet and cannot be manually moved and reconnected, a programmable hardware mechanism is needed to automatically, quickly, and reliably change the series and parallel connections between modules according to the grouping results, forming physical series branches corresponding to the logical grouping. Therefore, step S3 is proposed, and the specific implementation method is as follows.

[0050] Each battery module is equipped with a module access unit (MAU). The MAU contains a module switch matrix, which is composed of independently controlled first and second semiconductor switches. In this embodiment, both the first and second semiconductor switches are N-channel enhancement-mode metal-oxide-semiconductor field-effect transistors (MOSFETs). MOSFETs have the advantages of fast switching speed, low control power consumption, and easy parallel expansion. Each module switch matrix contains several pairs of MOSFETs, and each pair of MOSFETs is responsible for controlling the on / off state of one current path. The first semiconductor switch is used to disconnect the corresponding current path when a low-level turn-off signal is received, and the second semiconductor switch is used to close the corresponding current path when a high-level turn-on signal is received. The high level can be a typical value of 12 volts for the MOSFET gate drive voltage, and the low level can be zero volts.

[0051] Based on the grouping results of step S2, the central controller calculates which DC bus cluster each module should be connected to, and its position order in the series link. The calculation principle is to connect the modules in the same group in a certain physical arrangement order (such as from bottom to top and from left to right according to the physical position in the cabinet) in series to minimize the length of the connecting cable and the wiring complexity. Then, it generates the switching state command sequence of each MOSFET in the switching matrix of each module.

[0052] For each module within group A, its positive terminal needs to be connected to the negative terminal of the preceding module via a switch, and its negative terminal needs to be connected to the positive terminal of the following module. The positive terminal of the first module in the series link is connected to the positive busbar of the DC busbar Bus-A via a switch, and the negative terminal of the last module is connected to the negative busbar of Bus-A. The switching state command to achieve this connection is as follows: the MOSFET corresponding to the positive series path receives a high-level turn-on signal, the MOSFET corresponding to the negative series path receives a high-level turn-on signal, and the MOSFETs of the remaining unused paths receive a low-level turn-off signal.

[0053] All switching status commands are sent in parallel to each module access unit (MAU) via the communication bus of the central controller. The drive circuits in each MAU synchronously drive the corresponding MOSFET gates according to the commands. The MOSFET turn-on and turn-off times are on the order of microseconds, and the entire topology reconstruction can be completed within tens of milliseconds.

[0054] After reconfiguration, all modules in group A form a complete series battery branch electrically. Both ends of this branch are directly connected to the cluster DC bus Bus-A. Similarly, groups B and C are connected to Bus-B and Bus-C respectively, thus forming three electrically isolated and independent series clusters. Each cluster DC bus is then connected to the input of an independent cluster bidirectional DC converter. The cluster bidirectional DC converter adopts a dual active bridge topology to achieve bidirectional power flow and electrical isolation between the low-voltage side (battery cluster side) and the high-voltage side (system main bus side). The high-voltage output sides of the three converters are connected in parallel to the common system main bus, which is then connected to the energy storage converter PCS.

[0055] The above technical solution utilizes the programmable switching of the module switch matrix to achieve rapid mapping from logical grouping to physical topology, enabling the internal topology of the battery cabinet to be automatically reconfigured as the module status changes. The electrical characteristics of the modules within each series cluster are consistent, and the charging and discharging synchronization is good. Different clusters are electrically isolated, and the operating parameters do not interfere with each other, providing a foundation for subsequent differentiated management and control.

[0056] Due to the effects of factory dispersion and asynchronous aging during automotive use, some modules deteriorate significantly faster than other modules in the same group. If these lagging modules cannot be identified in time and adjusted to a more suitable group, they will gradually become the bottleneck of the series cluster, severely dragging down the available capacity of the entire cluster, and eventually triggering overcharge or over-discharge protection, causing the entire cluster to shut down. Therefore, step S4 is proposed, and the specific implementation method is as follows.

[0057] Each time the system enters a load rest period, the central controller automatically executes a round of pulse detection in step S1 and clustering calculation in step S2 to update the transient feature vectors of each module. The update cycle is set to balance timeliness and detection cost: when the system runs frequently, an update is triggered once every ten equivalent full charge-discharge cycles; when the system is in a shallow charge-discharge standby state for a long time, an update is triggered at least once every twenty-four natural hours. This way, the system will not occupy available time due to excessively frequent detection, nor will it miss rapid changes in module status due to excessively long detection intervals.

[0058] After updating and obtaining the new transient feature vectors of each module, the central controller performs drift judgment on each module one by one. Taking the first battery module as an example, assuming that its current source battery group is group A, and the source cluster center vector is... Suppose there exists another target battery group B, whose cluster center vector is... .

[0059] Calculate the new transient feature vector of the first battery module To the target cluster center First Euclidean distance data and to the source cluster center Second Euclidean distance data .

[0060] When both conditions are met: Less than the lower limit threshold, and If the distance exceeds the upper limit threshold, it is considered that the first battery module has undergone fingerprint drift, that is, the real-time electrical characteristics have significantly deviated from the mainstream characteristics of group A and instead approached the feature space of group B.

[0061] The method for determining the lower and upper distance thresholds is as follows: After each clustering, the transient feature vectors of all modules within the target group B are calculated to... The Euclidean distances are used to form a distance sample set, and the mean of this sample set is calculated. and standard deviation The distance lower limit threshold is set to + This means that if arrive If the distance falls within twice the standard deviation of the normal dispersion within group B, then it can be considered to have statistical consistency with the members of group B. Similarly, the transient eigenvectors of all modules within the source group A can be calculated to... The mean of the Euclidean distance and standard deviation The distance upper limit threshold is set to + , arrive If the distance exceeds three standard deviations, it is statistically highly likely to be an outlier and can be confirmed to no longer belong to group A. This threshold setting method based on probability statistics avoids the subjectivity and lack of universality caused by manually setting fixed values.

[0062] To prevent misjudgment due to random noise or measurement error in a single detection, this embodiment also introduces a time-dimensional confirmation mechanism. Only when the above-mentioned dual threshold conditions are met in two consecutive update cycles will the central controller officially confirm that the first battery module has experienced a state drift and trigger the migration operation.

[0063] After confirming the drift, the central controller sends a disconnection command to the module switch matrix corresponding to the first battery module. Specifically, it sends a low-level turn-off signal to the gate of the currently closed MOSFET that connects the module to the source cluster DC bus Bus-A, causing the MOSFET to turn off. The module is disconnected from the series link of Bus-A and becomes electrically floating. At the same time, the central controller marks the operating state of the module as the grid-connected isolation state in the internal state management table and records context information such as its disconnection time, operating voltage before disconnection, and transient characteristic vector in the memory.

[0064] After the disconnection operation is completed, the number of series modules in the source battery group A is reduced by one. The central controller automatically adjusts the input voltage range parameters of the cluster bidirectional DC / DC-A converter to adapt to the new number of series modules, so that group A can continue to operate normally after disconnection and is not affected by the disconnection of a single module.

[0065] Through the above technical solution, the fingerprint drift recognition mechanism based on dual threshold distance criteria can automatically detect lagging modules during operation and safely disconnect them from the original group, preventing them from continuing to drag down the performance of the entire cluster. Marking them as isolated states to be connected to the grid provides a clear state anchor point for subsequent voltage synchronization and grid connection operations, making the migration process have clear stage divisions. The entire identification and isolation process is automatically completed during the system load rest period, without affecting the normal external power supply of the energy storage cabinet.

[0066] Considering that existing technologies typically employ hard switching methods such as disconnecting before connecting or connecting before disconnecting during topology reconfiguration, and lack active management of the voltage difference between the module to be connected and the target series link, if the open-circuit voltage of the module to be connected differs significantly from the current operating voltage of the target bus, a large surge current will be generated at the moment the switch closes, which may damage semiconductor switching devices, impact battery modules, or even cause arcing hazards. Therefore, step S5 is proposed, and the specific implementation method is as follows.

[0067] After the first battery module enters the grid-connected isolation state, the central controller first obtains its current open-circuit voltage through a high-impedance voltage detection circuit connected to the positive and negative terminals of the module. The input impedance of the high-impedance detection circuit is in the megaohm range, and it draws almost no current from the module during the detection process. Therefore, the measured open-circuit voltage can accurately reflect the current state of charge of the module.

[0068] Meanwhile, the central controller obtains the current operating voltage of the target cluster DC bus through a voltage sensor installed on the target cluster DC bus; then it calculates the voltage deviation between the current operating voltage of the target cluster DC bus and the open-circuit voltage of the first battery module; the central controller runs a voltage regulation control algorithm, generates a voltage regulation command based on the deviation, and sends it to the cluster bidirectional DC converter corresponding to the target battery group through the communication bus, controlling it to adjust the output voltage so that the target cluster DC bus voltage gradually converges to the open-circuit voltage; the entire regulation process is continuous and smooth, and the bus voltage gradually converges without any abrupt changes.

[0069] During the adjustment process, the central controller continuously monitors the change in the absolute value of the voltage difference. When the absolute value of the voltage difference is less than the voltage synchronization threshold, it is determined that the grid connection conditions have been met. The voltage synchronization threshold is set to 2 volts. The basis for selecting 2 volts is that the battery module in this embodiment is a retired lithium iron phosphate module with a nominal voltage of about 48 volts. The voltage difference of 2 volts accounts for about four percent of the nominal voltage. When the switch is closed under this voltage difference, the peak value of the equalization current generated at the moment of switching does not exceed one-tenth of the module's rated current, which is far lower than the safe current limit of semiconductor switching devices and the battery's allowable inrush current. At the same time, 2 volts is not an overly stringent setting, and synchronization can be achieved within a few seconds under most operating conditions without indefinitely prolonging the migration waiting time.

[0070] Once the grid connection conditions are met, the central controller immediately sends a grid connection command to the module switch matrix corresponding to the first battery module, controls the corresponding semiconductor switching devices to turn on, and connects the first battery module to the target cluster DC bus.

[0071] The above technical solution utilizes a migration strategy of first disconnecting the module, then synchronizing the voltage, and finally connecting it to the grid. This strategy decomposes the module migration process into three stages: safe disconnection, active voltage synchronization, and low-dropout grid connection. Before grid connection, the target bus voltage is actively adjusted to synchronize with the module's open-circuit voltage, keeping the voltage difference at the moment of closing within the safe threshold of 2 volts. This eliminates the current surge risk present in traditional hard switching schemes, ensures the safety of semiconductor switching devices and battery modules, and improves the reliability of dynamic migration.

[0072] Considering that the target cluster DC bus may simultaneously experience charging and discharging current fluctuations from other modules in the same group during the adjustment process, voltage regulation needs to have good anti-disturbance capability and steady-state accuracy. Therefore, in step S5, a voltage regulation command is issued to the cluster bidirectional DC converter corresponding to the target battery group to control the target cluster DC bus voltage to converge to the open-circuit voltage. This includes the following feedback control logic: obtaining the current operating voltage of the target cluster DC bus from the voltage sensor; calculating the voltage deviation between the current operating voltage and the open-circuit voltage; inputting the voltage deviation as an error compensation term into the proportional-integral control loop to calculate the duty cycle adjustment data; and issuing the duty cycle adjustment data as the voltage regulation command to the cluster bidirectional DC converter to drive its output stage voltage to approach the open-circuit voltage. The specific implementation method is as follows.

[0073] The central controller calculates the voltage deviation between the current operating voltage of the target cluster DC bus and the open-circuit voltage of the first battery module. This deviation can be positive or negative. A positive value indicates that the bus voltage is higher than the open-circuit voltage, and a negative value indicates that the bus voltage is lower than the open-circuit voltage.

[0074] The central controller internally operates a proportional-integral (PI) control loop, using the aforementioned voltage deviation as an error compensation input. The PI controller consists of a proportional element and an integral element connected in parallel. The proportional element outputs a control quantity proportional to the error, used for rapid response to voltage deviation. The integral element outputs a control quantity proportional to the integral of the error over time, used to eliminate steady-state error. The proportional coefficient is set to 0.5, and the integral time constant is set to 0.2 seconds. These two parameters were determined during the system commissioning phase through step response testing: the proportional coefficient of 0.5 ensures that the controller output duty cycle adjustment does not exceed 10% under a typical 20V deviation, avoiding overshoot oscillation; the integral time constant of 0.2 seconds ensures that more than 90% of the steady-state error can be eliminated within 1 second.

[0075] The proportional-integral controller outputs duty cycle adjustment data; the duty cycle refers to the proportion of the conduction time of the switching transistor in the high-frequency chopper circuit of the cluster bidirectional DC-DC converter to the entire switching cycle; for a dual active bridge converter, adjusting the phase shift angle between the primary and secondary bridge arms is equivalent to adjusting the duty cycle, thereby controlling the output voltage amplitude; the central controller uses the duty cycle adjustment data as a voltage regulation command and sends it to the cluster bidirectional DC / DC-B corresponding to the target battery group via the communication bus.

[0076] After receiving a command, the DC / DC-B adjusts the timing of the drive signals for its internal switching transistors to change the output voltage. When the bus voltage is higher than the open-circuit voltage, the duty cycle is reduced to lower the output voltage; when the bus voltage is lower than the open-circuit voltage, the duty cycle is increased to raise the output voltage. Through this closed-loop feedback, the output voltage of the DC / DC-B is driven to approach the open-circuit voltage. The entire adjustment process is continuous and smooth, with the bus voltage gradually converging without any abrupt changes.

[0077] During the adjustment process, the central controller continuously monitors the change in the absolute value of the voltage difference. When the absolute value of the voltage difference is detected to be less than the voltage synchronization threshold, it is determined that the grid connection conditions have been met. The voltage synchronization threshold is set to 2 volts. The basis for choosing 2 volts is as follows: In this embodiment, the battery module is a retired lithium iron phosphate module with a nominal voltage of approximately 48 volts. The voltage difference of 2 volts is about four percent of the nominal voltage. Under this voltage difference, when the switch is closed, the peak value of the equalization current generated at the moment of switching does not exceed one-tenth of the module's rated current, which is far lower than the safe operating area limit of the MOSFET and the allowable inrush current of the battery. At the same time, the setting of 2 volts is not too stringent, and synchronization can be achieved within a few seconds under most operating conditions without indefinitely prolonging the migration waiting time.

[0078] Once the grid connection conditions are met, the central controller immediately sends a grid connection command to the module switch matrix corresponding to the first battery module, that is, sends a high-level conduction signal to the gate of the target MOSFET that should be closed. The MOSFET turns on, and the positive and negative terminals of the first battery module are connected to the corresponding positions of the target series link, thus completing the grid connection.

[0079] During the process of controlling the DC bus voltage of the target cluster to converge to the open-circuit voltage in step S5, unexpected situations may occur, such as the bidirectional DC converter of the target cluster failing to respond to the adjustment command due to a fault, or the voltage of a module in the target group being abnormal, limiting the adjustable range of the bus voltage. If the system does not have a timeout handling mechanism and waits indefinitely, the migration process will be blocked indefinitely. Not only will the module to be connected to the grid be in a suspended state for a long time, but it may also affect the normal periodic detection and maintenance plan of the system. Therefore, in step S5, during the process of controlling the DC bus voltage of the target cluster to converge to the open-circuit voltage, the following abnormal blocking mechanism is also included: a synchronous cumulative timer is started at the same time as the voltage adjustment command is issued; when the current value of the synchronous cumulative timer is greater than the timeout threshold, and it is determined that the absolute value of the voltage difference between the two is still greater than the voltage synchronization threshold, a migration interruption command is generated; the migration interruption command is issued to keep the module switch matrix of the first battery module in the open state, and a synchronous abnormal data packet containing the identification information of the first battery module is generated and written to the local log node. The specific implementation is as follows.

[0080] While issuing the voltage regulation command, the central controller starts a synchronous cumulative timer with a time base of 10 milliseconds, counting upwards from zero. When the current value of the synchronous cumulative timer is greater than the timeout threshold, and the absolute value of the voltage difference between the current operating voltage of the target cluster DC bus and the open circuit voltage of the first battery module is still greater than the voltage synchronization threshold, the central controller determines that the voltage synchronization has failed and generates a migration interrupt command. The timeout threshold is set to 5 seconds, which is based on experimental data that voltage synchronization is usually completed within 3 seconds under normal operating conditions. Approximately twice the normal time is taken as the timeout criterion, which provides the system with sufficient response margin without waiting indefinitely.

[0081] The central controller issues a migration interrupt command to force the module switch matrix of the first battery module to remain in the open state, ensuring that it will not be accidentally closed due to subsequent misoperation; at the same time, it generates a synchronization exception data packet containing the identification information of the first battery module, the current open circuit voltage value, the target bus voltage value, the cumulative waiting time and the failure reason code, and writes it to the local log node, that is, the system log partition in the non-volatile memory of the central controller. By utilizing the above technical solution and the timeout-triggered abnormal blocking mechanism, a clear abnormal exit is set for the voltage synchronization process. Once a fault occurs that cannot be completed within the specified time, the system will not wait indefinitely, but will actively interrupt the migration process, lock the safe state, and record complete abnormal information for maintenance personnel to troubleshoot, thus ensuring the robustness and maintainability of the system.

[0082] After the control module switch matrix connects the first battery module to the target cluster DC bus in step S5, the voltage of the target cluster DC bus is currently at a level synchronized with the open-circuit voltage of the module. This synchronized voltage level is temporarily adjusted to meet the safe grid connection conditions and is usually not equal to the target operating voltage reference value that the target battery group should operate at according to the differentiated management strategy. If the bus voltage is not actively adjusted back to the optimized operating point after grid connection is completed, the target group will continue to operate at a suboptimal voltage level. The energy release of the high-quality modules will be insufficient, and the safety margin of the suboptimal modules will not be guaranteed, affecting the overall energy utilization efficiency and safety of the system. Therefore, in step S5, when controlling the corresponding module... After the switching matrix connects the first battery module to the target cluster DC bus, the central controller executes the system steady-state recovery logic for the target battery group. Specifically, it acquires the transient feature vectors of all battery modules in the target battery group after the first battery module is connected, and recalculates and generates updated feature mean values; it substitutes the updated feature mean values ​​into the preset mapping function to calculate the target operating voltage reference value; it generates a voltage recovery command containing the ramp-up rate and ramp-down rate, and drives the cluster bidirectional DC converter corresponding to the target battery group to smoothly adjust the voltage of the target cluster DC bus from the open-circuit voltage to the target operating voltage reference value. The specific implementation method is as follows.

[0083] The central controller first acquires the transient feature vectors of all battery modules in the target battery group after the first battery module is connected. Due to the addition of the new module, the composition of the group has changed, and its overall characteristics are different from those before the migration. The arithmetic mean of all transient feature vectors in the group is recalculated to generate the updated feature mean.

[0084] Then, the updated feature mean is substituted into the preset mapping function, namely the charge and discharge cutoff voltage threshold calculation function used in step S6, to calculate the target operating voltage reference value applicable to the new formation. This reference value comprehensively reflects the current optimal upper limit of the charging cutoff voltage and the lower limit of the discharging cutoff voltage of the formation.

[0085] Finally, the central controller generates a voltage recovery command containing either a ramp-up rate or a ramp-down rate. If the current bus voltage is lower than the target operating voltage reference value, a positive ramp-up rate command is generated; if it is higher than the reference value, a negative ramp-down rate command is generated. The ramp rate is set to 1 volt per second. This rate is neither too slow, causing the system to remain in an optimized state for an extended period, nor too fast, causing a step-like voltage stress on the battery modules. The voltage recovery command is sent to the cluster bidirectional DC-DC converter corresponding to the target battery group, driving it to smoothly adjust the voltage of the target cluster DC bus from the open-circuit voltage to the target operating voltage reference value.

[0086] Through the above technical solution, by utilizing the system steady-state recovery logic after grid connection, the system can automatically and smoothly transition the target group from the temporary synchronization voltage point to the optimal operating voltage point after each dynamic migration, so that the new group can quickly recover to the optimal working range set by the differentiated control strategy, ensuring that the entire battery cabinet always operates in the state of highest efficiency and best safety throughout its entire life cycle.

[0087] Considering that after grouping and dynamic migration, the physical properties of the modules within each battery group are relatively similar, but there are still significant differences between groups, step S6 is proposed, and the specific implementation method is as follows.

[0088] For each battery pack, the central controller extracts the arithmetic mean of the transient feature vectors of all modules within the pack to obtain the feature mean. This feature mean is then substituted into a pre-calibrated mapping function to calculate the upper limit of the charging cutoff voltage and the lower limit of the discharging cutoff voltage for that pack. The mapping function is pre-calibrated based on the allowable operating range and safety boundaries of the battery chemistry system. Its input is the feature mean, and its output is the corresponding safe charging and discharging voltage boundaries. The central controller packages the upper limit of the charging cutoff voltage and the lower limit of the discharging cutoff voltage into a set of charging and discharging cutoff voltage thresholds and sends its dedicated threshold parameters to the cluster bidirectional DC-DC converter corresponding to each battery pack.

[0089] During subsequent operation, each cluster of bidirectional DC-DC converters writes the received threshold parameters into its own control register; during charging, when the battery cluster voltage is detected to reach the upper limit of the charging cutoff voltage, it automatically switches from constant current mode to constant voltage mode or stops charging; during discharging, when the battery cluster voltage is detected to drop to the lower limit of the discharging cutoff voltage, it automatically cuts off the discharging circuit.

[0090] Through the above technical solution, by using differentiated charge and discharge cutoff voltage settings based on the average characteristics of the groups, high-quality groups can obtain a wider charge and discharge window to fully release energy; suboptimal groups can obtain a more conservative window, always operating within the safe range to prevent overcharging and over-discharging; each group operates independently within its own optimal voltage boundary without hindering each other, and the total available capacity provided by the whole is the algebraic sum of the available capacities of all groups, which improves the economic benefits of cascaded energy storage cabinets and extends the service life of the entire cabinet.

[0091] There is a clear physical correlation between the usable safe voltage window of a retired battery module and its internal resistance parameter: the higher the internal resistance, the more severe the ohmic heating during charging and discharging, the more active the side reactions, and the shallower the depth of charge and discharge that can be safely withstood. If the mapping rule cannot accurately reflect this negative and positive correlation, it may lead to the window of a high-quality group being too conservative and wasting capacity, or the window of a suboptimal group being too aggressive and causing safety risks. Therefore, in step S6, the characteristic mean of each battery group is extracted and substituted into a preset mapping function to calculate the charge and discharge cutoff voltage threshold, including the following calculation rules: extract the average internal resistance parameter from the characteristic mean; call the first linear mapping function to map the average internal resistance parameter to the upper limit data of the charging cutoff voltage, the first linear mapping function exhibiting negative correlation characteristics; call the second linear mapping function to map the average internal resistance parameter to the lower limit data of the discharging cutoff voltage, the second linear mapping function exhibiting positive correlation characteristics; package the upper limit data of the charging cutoff voltage and the lower limit data of the discharging cutoff voltage into the charge and discharge cutoff voltage threshold, the specific implementation is as follows.

[0092] The central controller extracts the average internal resistance parameter from the feature mean, which is the arithmetic mean of the internal resistance parameters of all battery modules within the same battery group. It then calls the first linear mapping function, which is in the form that the upper limit of the charging cut-off voltage is equal to the reference charging upper limit voltage minus the product of the mapping coefficient and the average internal resistance parameter. The reference charging upper limit voltage is taken as the highest safe charging voltage allowed by the battery chemical system. The mapping coefficient is calibrated through cycle life experiments of modules with different aging levels, and it reflects the magnitude of the decrease in the safe charging upper limit for each unit increase in internal resistance. This function exhibits a negative correlation characteristic, that is, the higher the average internal resistance, the lower the upper limit of the charging cut-off voltage. The physical basis for this is that modules with high internal resistance generate more heat internally at the end of charging, and lowering the charging upper limit can effectively suppress side reactions and ensure safety.

[0093] Simultaneously, the second linear mapping function is invoked. The function form is that the lower limit of discharge cutoff voltage is equal to the reference lower limit of discharge voltage plus the product of the mapping coefficient and the average internal resistance parameter. The reference lower limit of discharge voltage is taken as the lowest safe discharge voltage allowed by the battery chemical system, and the mapping coefficient is also calibrated through cycle life test. This function exhibits a positive correlation characteristic, that is, the larger the average internal resistance, the higher the lower limit of discharge cutoff voltage. The reason is that the module with high internal resistance has a faster voltage drop at the end of discharge and a higher risk of over-discharge. Raising the lower limit can leave a safety margin.

[0094] The central controller packages the calculated upper limit data of charging cutoff voltage and the lower limit data of discharging cutoff voltage into a charging and discharging cutoff voltage threshold.

[0095] The above technical solution utilizes a bilinear mapping function based on average internal resistance to establish a quantitative correspondence between the physical characteristics of the modules and their safe operating boundaries. Negatively correlated charging mapping and positively correlated discharging mapping together achieve precise tailoring of the voltage windows for each module, fully utilizing the available capacity of high-quality modules while preserving sufficient safety margins for suboptimal modules, thus achieving an optimal balance between overall cabinet capacity utilization and safety.

[0096] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A control method based on an integrated energy storage battery cabinet, executed by a central controller, characterized in that, Includes the following steps: S1: When the real-time output power and bus voltage fluctuation rate are both less than the corresponding set values, it is determined to be a system load rest period; during this period, feature recognition current pulses are applied to each battery module to obtain terminal voltage timing data, and internal resistance parameters, polarization parameters and time constants are analyzed to generate transient feature vectors based only on the current terminal voltage timing data. S2: Input all transient feature vectors into the Euclidean distance clustering algorithm to minimize the spatial distance of transient feature vectors within the same battery group, and output the grouping result; S3: Based on the grouping results, a switch status command is sent to the module switch matrix to connect the positive and negative terminals of the battery modules in the same battery group in series, and the two ends of the series link are connected to the corresponding cluster DC bus. Each cluster DC bus is connected to the system main bus via a cluster bidirectional DC converter. S4: Periodically update the transient feature vector during the system load rest period; when the Euclidean distance of the new transient feature vector of the first battery module from the target cluster center is less than the lower threshold and the Euclidean distance from the source cluster center is greater than the upper threshold, control the corresponding module switching matrix to disconnect it from the source cluster DC bus and mark it as an isolated state to be connected to the grid; S5: Obtain the open-circuit voltage of the first battery module, send a voltage regulation command to the cluster bidirectional DC-DC converter corresponding to the target battery group, and control the DC bus voltage of the target cluster to converge to the open-circuit voltage; when the absolute value of the voltage difference between the two is less than the voltage synchronization threshold, control the corresponding module switch matrix to connect the first battery module to the DC bus of the target cluster. S6: Extract the characteristic mean of each battery group, substitute it into the preset mapping function to calculate the charge / discharge cutoff voltage threshold, and send it to the corresponding cluster bidirectional DC-DC converter.

2. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S1, during this period, a feature recognition current pulse is applied to each battery module, including the following collaborative control logic: Generate perturbation injection commands containing pulse amplitude and duration; The micro-perturbation injection command is sent to each cluster of bidirectional DC-DC converters in the integrated energy storage battery cabinet, driving the high-frequency chopper circuit of the cluster of bidirectional DC-DC converters to superimpose an alternating micro-current signal on the basic DC output, forming the feature recognition current pulse to be injected into the downstream battery module.

3. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S1, before acquiring the terminal voltage timing data and parsing the internal resistance parameters, polarization parameters, and time constant, a pre-cleaning logic is performed on the terminal voltage timing data. The pre-cleaning logic includes: Extract the raw voltage data stream according to the preset sampling period; The original voltage data stream is input into a polynomial smoothing filter, and the data points within the extraction window are subjected to least squares fitting to filter out high-frequency electromagnetic interference noise data, and the smoothed terminal voltage time series data is output.

4. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S1, the terminal voltage timing data is acquired, and the internal resistance parameters, polarization parameters, and time constant are analyzed, including the following data processing logic: The terminal voltage timing data is received, and the terminal voltage timing data is divided into instantaneous step band, exponential relaxation band and steady-state recovery band according to the voltage change rate. Extract the voltage jump amplitude of the instantaneous step band, calculate the ratio of the voltage jump amplitude to the current amplitude of the feature identification current pulse, and output the ratio as the internal resistance parameter; An exponential function is fitted to the time series data of the exponential relaxation band, and the decay time coefficient in the fitted function is extracted and output as the time constant. Extract the steady-state extreme voltage of the steady-state recovery band, calculate the difference between the steady-state extreme voltage and the initial voltage before pulse application, and output the difference as the polarization parameter.

5. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S2, the following dynamic adjustment mechanism is included during the execution of the Euclidean distance clustering algorithm: Calculate the global variance data of all transient eigenvectors in the feature space; When the global variance data is determined to be greater than the upper limit of variance, the number of grouping results output by the clustering algorithm is increased. When the global variance data is determined to be less than the lower limit of variance, the number of grouping results output by the clustering algorithm is reduced.

6. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S5, a voltage regulation command is sent to the cluster bidirectional DC-DC converter corresponding to the target battery group to control the target cluster DC bus voltage to converge to the open-circuit voltage, including the following feedback control logic: Obtain the current operating voltage of the target cluster DC bus from the voltage sensor; Calculate the voltage deviation between the current operating voltage and the open-circuit voltage; The voltage deviation is input as an error compensation term into the proportional-integral control loop to calculate the duty cycle adjustment data. The duty cycle adjustment data is then sent as the voltage regulation command to the cluster bidirectional DC-DC converter to drive its output stage voltage to approach the open-circuit voltage.

7. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S5, during the process of controlling the DC bus voltage of the target cluster to converge to the open-circuit voltage, the following abnormal blocking mechanism is also included: Simultaneously with issuing the voltage adjustment command, a synchronous cumulative timer is started; When the current value of the synchronous cumulative timer is greater than the timeout threshold, and it is determined that the absolute value of the voltage difference between the two is still greater than the voltage synchronization threshold, a migration interrupt instruction is generated. The migration interrupt command is issued to keep the module switch matrix of the first battery module in the off state, and a synchronization exception data packet containing the identification information of the first battery module is generated and written to the local log node.

8. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S5, after controlling the corresponding module switch matrix to connect the first battery module to the target cluster DC bus, the central controller executes the system steady-state recovery logic for the target battery group, specifically: Obtain the transient feature vectors of all battery modules in the target battery group after the first battery module is connected, and recalculate and generate the updated feature mean. Substitute the updated feature mean into the preset mapping function to calculate the target operating voltage reference value; A voltage recovery command containing the ramp-up rate and ramp-down rate is generated to drive the cluster bidirectional DC-DC converter corresponding to the target battery group to smoothly adjust the voltage of the target cluster DC bus from the open-circuit voltage to the target operating voltage reference value.

9. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, In step S6, the characteristic mean value of each battery group is extracted and substituted into a preset mapping function to calculate the charge / discharge cutoff voltage threshold, including the following calculation rules: Extract the average internal resistance parameter from the mean of the features; The first linear mapping function is invoked to map the average internal resistance parameter to the upper limit data of the charging cutoff voltage. The first linear mapping function exhibits negative correlation characteristics. The second linear mapping function is invoked to map the average internal resistance parameter to the lower limit data of the discharge cutoff voltage. The second linear mapping function exhibits positive correlation characteristics. The upper limit of the charging cutoff voltage and the lower limit of the discharging cutoff voltage are packaged into the charging and discharging cutoff voltage threshold.

10. The control method based on an integrated energy storage battery cabinet according to claim 1, characterized in that, The module switch matrix is ​​composed of independently controlled first semiconductor switching devices and second semiconductor switching devices; The method of sending switch status commands to the module switch matrix includes: controlling the first semiconductor switch device to open by sending a low-level turn-off signal, and controlling the second semiconductor switch device to close by sending a high-level turn-on signal.

Images