BMS hybrid battery management system based on AI machine cooperation

Abstract

The invention discloses a BMS hybrid battery management system based on AI machine collaboration, and relates to the technical field of battery management, and the system comprises a signal collection module which synchronously collects multi-dimensional signals, carries out the interpolation alignment, and generates a synchronous feature matrix; the graph generation module is used for calculating a correlation coefficient based on the feature matrix and filtering edges, and constructing a collaborative topological graph; the vector construction module is used for extracting collaborative potentials, historical responses, spectrum peak values and fold numbers, assembling multi-dimensional mark vectors and carrying out index storage on the multi-dimensional mark vectors; the monomer selection module is used for clustering according to density, constructing a multi-layer priority structure and sequentially selecting active and redundant monomers; and the collaborative management module is used for mapping environment load grids, fusing partition coefficients, distributing differentiated charging and discharging power and realizing accurate collaborative management. Integrated management of coupling relation revelation, hierarchical scheduling and differentiated power distribution is realized through construction of coupling topology and multi-level priority selection based on environmental load grid mapping and composite priority.

Description

Technical Field

[0001] The present invention relates to the field of battery management technology, and specifically to a BMS hybrid battery management system based on AI machine collaboration. Background Art

[0002] Existing battery management systems often rely on a single voltage or temperature channel for balancing and compensation control. They are unable to simultaneously perceive multi-dimensional status information such as internal resistance and power response rate, making it difficult to accurately identify cell differences and degradation trends. Furthermore, the systems have limited access to historical charge and discharge errors and aging patterns, and lack long-term assessment and adaptive correction of cell reliability, resulting in a sluggish response to performance drift or sudden anomalies in scheduling strategies.

[0003] The electrical and thermal coupling within large-scale battery packs is becoming increasingly complex, but traditional BMSs often treat them as simple series-parallel networks, ignoring the dynamic relationships between cells. This significantly increases the risk of local failure under high load or extreme temperature conditions. Furthermore, early warning and balancing mechanisms based primarily on static thresholds and linear control struggle to adapt to the ever-changing operating environment and cell performance, limiting online adaptive optimization capabilities.

[0004] Faced with massive amounts of high-frequency, multi-channel time-series data, traditional BMSs are inefficient in data screening and multivariate correlation analysis, making it difficult to meet the demands for real-time insight into battery pack operating status and rapid decision-making. These bottlenecks prohibit more refined collaborative management and more reliable operation and maintenance. A new generation of hybrid battery management technology urgently needs breakthroughs in multi-dimensional state synchronization, dynamic correlation topology construction, and adaptive optimization based on historical feedback. Summary of the Invention

[0005] Based on the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a BMS hybrid battery management system based on AI machine collaboration to solve the above-mentioned technical problems.

[0006] To achieve the above objectives, the present invention provides the following technical solutions: a BMS hybrid battery management system based on AI machine collaboration, comprising: Signal acquisition module: Synchronously collects the voltage gradient, temperature gradient, internal resistance response, and current fluctuation rate signals of each battery cell, performs polynomial interpolation processing according to a unified timestamp, and obtains the synchronous characteristic matrix of each cell; Graph generation module: Calculates nonlinear association weights between all monomers based on the synchronized feature matrix, and generates a topological collaborative graph consisting of monomer nodes and connectivity relationships according to a preset threshold. Vector construction module: This module obtains synergy potential, historical response scores, temperature spectrum peaks, and impedance spectrum breakpoints from the synchronization feature matrix and historical scheduling feedback, constructs a multi-dimensional tag vector, and stores it in a tag vector pool that supports multi-key indexing. Monomer selection module: All marker vectors in the marker vector pool are clustered in multidimensional space according to the density threshold, and a hierarchical priority structure is constructed within each cluster based on the synergy potential and historical response score, and active monomers and redundant monomers are selected in descending order; Collaborative management module: maps the selected active cells to the corresponding cells in the environmental load grid matrix, allocates the available charge and discharge power of each cell based on the preset partition coefficient, and sends it to each BMS controller for differentiated collaborative management of hybrid batteries.

[0007] The present invention is further configured such that the signal acquisition module includes: Record the original sampling signal in the ring buffer of each monomer and each channel, build a global equidistant aligned timestamp sequence, and unify the sampling time base of each channel; A set of adjacent sampling points is extracted from the buffer according to the neighborhood span factor preset by the channel, and a polynomial fitting with regularization constraints is applied to the extracted samples to obtain the local interpolation model coefficients; A local interpolation model is applied at the aligned timestamp to generate feature estimates for each monomer, and the generated feature estimates are arranged in channel and time order to form a synchronized feature matrix.

[0008] The present invention is further configured such that the graph generation module includes: Extract the voltage gradient sequence of each battery cell and calculate the voltage difference between cells. Couple the voltage difference based on the temperature gradient of each cell to obtain the temperature coupling factor. The voltage difference and the temperature coupling factor are nonlinearly fused to generate the initial correlation value; The initial correlation value is suppressed according to the difference in the internal resistance response inflection point of each monomer to obtain the correlation weight after suppression; The suppressed association weights are screened according to the preset association weight threshold to generate directed connected edges, and all battery cell nodes and directed connected edges are constructed into a topological collaborative graph.

[0009] The present invention is further configured to construct a pulse fusion tree based on the post-suppression association weights, and to perform rapid screening of the post-suppression association weights through the pulse fusion tree, including: Hash buckets are performed on the associated weight records according to predefined time periods, and a balanced binary search tree is constructed in each bucket, where the node is the time period identifier, and the node attributes include the decaying cumulative weight of the time period and the maximum weight of the node in the subtree to which it belongs; Update the maximum weight of the subtree of each ancestor node at each node in the balanced binary search tree; Based on the preset association weight threshold, depth-first pruning is performed from root to leaf in the balanced binary search tree, and the weight records corresponding to the nodes whose maximum weight of all subtrees is greater than the threshold are extracted and marked as candidate directed connected edges.

[0010] The present invention is further configured such that the vector building module includes: Extract the voltage gradient and temperature gradient time series data of each battery cell, and generate synergistic potential based on weighted logarithm and maximum internal resistance response suppression; Obtain the power execution error sequence of each unit from the historical scheduling feedback buffer and construct the historical response score through logarithmic exponential kernel accumulation; Perform fast Fourier transform on the temperature gradient time series and set the maximum spectrum component as the temperature spectrum peak; Perform second-order differential inflection point detection on the internal resistance response time series and count the number of inflection points to obtain the number of inflection points in the impedance spectrum; The synergistic potential, historical response score, temperature spectrum peak, and impedance spectrum breakpoint number are sequentially assembled into a multidimensional tag vector and inserted into a tag vector pool supporting multi-key indexing.

[0011] The present invention is further configured such that the monomer selection module includes: All labeled vectors in the labeled vector pool are divided based on kernel density estimation and a preset density threshold to form multiple disjoint clusters; For each tag vector within a cluster, a multi-level hierarchical structure is constructed based on the synergy potential and historical response score, and the ascending order of the hierarchy is mapped to the descending order of priority; All the marker vectors at the highest multiple levels are identified as active monomers, and the remaining marker vectors are identified as redundant monomers.

[0012] The present invention is further configured to construct a multi-level hierarchical structure for each marker vector within each cluster based on the synergy potential and historical response score, including: A composite priority index is calculated for each tag vector within each cluster based on synergy potential and historical response scores; Generate a multi-level hierarchical threshold sequence based on the maximum priority index within the cluster and the preset multiplication factor; Compare the composite priority index of each tag vector with the threshold sequence in sequence to determine the level to which it belongs; A chain tree structure is constructed for each level. Inside each level node, a small top heap is constructed with the monomer index as the key and the corresponding composite priority index as the weight to achieve fast access to high-weight monomers.

[0013] The present invention is further configured such that the collaborative management module includes: Map the selected active cells to the corresponding cells of the predefined environmental load grid matrix according to the ambient temperature and real-time load requirements of the cells; Obtain the partition coefficient corresponding to each grid cell from the preset partition coefficient lookup table, perform nonlinear fusion of the partition coefficient with the scheduling priority characteristics of each active cell, and generate the power allocation parameter; Calculate the available charge and discharge power of each active cell according to the power allocation parameters and the principle of proportional division of the total available power of the system; The available charge and discharge power instructions are sent to the BMS controller of the corresponding monomer to achieve differentiated collaborative charge and discharge management of the monomer.

[0014] The present invention is further configured to include: After each scheduling is completed, the charge and discharge execution errors and corresponding multi-dimensional tag vectors of each active monomer are obtained, packaged into a scheduling fingerprint, and stored in a ring buffer with a fixed depth; For each monomer in the buffer, the error sequence within the extraction period is subjected to wavelet packet decomposition to obtain the high-frequency energy ratio of each decomposition level; The monomer anomaly index is constructed based on the high-frequency energy ratio of each decomposition level, and the power amplification coefficient of the multi-level hierarchical structure and the correlation weight screening threshold of the topological collaborative map are updated based on the anomaly index; The updated amplification factor and screening threshold are sent to the monomer selection module and the graph generation module for hierarchical priority division and edge screening in the next scheduling cycle.

[0015] The present invention is further configured to map the latest multidimensional tag vector in the tag vector pool and the scheduling fingerprint sequence in the ring buffer, and generate a health topology map based on the fusion of multimodal similarity and original association weights; In the health topology graph, health centrality is calculated based on the similarity weights after fusion between nodes and the historical response scores of each monomer; The health centrality is compared with the preset threshold. For abnormal monomers below the threshold, they are removed from the active group and replaced with the monomer with the highest health centrality from the redundant group. The replacement results are updated to the active group and redundant group lists.

[0016] The present invention provides a BMS hybrid battery management system based on AI machine collaboration, including a signal acquisition module: synchronously collects the voltage gradient, temperature gradient, internal resistance response, and current fluctuation rate signals of each battery cell, performs polynomial interpolation processing according to a unified timestamp, and obtains a synchronous feature matrix of each cell; a spectrum generation module: calculates nonlinear association weights between all cells based on the synchronous feature matrix, and generates a topological collaborative spectrum consisting of cell nodes and connectivity relationships according to a preset threshold; a vector construction module: obtains the collaborative potential, historical response score, temperature spectrum peak, and impedance spectrum inflection point number from the synchronous feature matrix and historical scheduling feedback, constructs a multidimensional tag vector, and stores it in a tag vector pool supporting multi-key indexing; a cell selection module: clusters all tag vectors in the tag vector pool according to the density threshold in the multidimensional space, and constructs a hierarchical priority structure within each cluster based on the collaborative potential and historical response score, and selects active cells and redundant cells in descending order; and a collaborative management module: maps the selected active cells to the corresponding cells of the environmental load grid matrix, allocates the available charge and discharge power of each cell based on the preset partition coefficient, and sends it to each BMS controller to perform differentiated collaborative management of hybrid batteries, resulting in the following beneficial effects: 1. Coupling correlation topology construction: The graph generation module uses nonlinear correlation weights and a pulse fusion tree for rapid screening to generate a topological synergy graph, enabling in-depth understanding of the electrical and thermal coupling relationships between monomers. 2. Hierarchical priority management: Through the kernel density clustering and weight ladder tree stratification strategies in the vector construction module and the monomer selection module, multi-level priority division and rapid selection of monomers are achieved, improving scheduling efficiency and decision-making accuracy; 3. Differentiated power allocation: Through the collaborative management module combined with environmental load grid mapping and composite priority index, precise charge and discharge allocation is achieved for monomers with different chemical systems and health states, optimizing energy utilization and thermal balance.

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without inventive efforts. In the drawings: Figure 1The figure is a schematic structural diagram of a BMS hybrid battery management system based on AI machine collaboration, showing an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following describes the embodiments of the present invention with reference to the accompanying drawings and preferred embodiments. Those skilled in the art will readily appreciate the other advantages and benefits of the present invention from the disclosure herein. The present invention may also be implemented or applied through various other specific embodiments, and the various details in this specification may be modified or altered based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are intended only to illustrate the present invention and are not intended to limit the scope of protection of the present invention.

[0020] It should be noted that the illustrations provided in the following embodiments are merely schematic illustrations of the basic concept of the present invention. Therefore, the illustrations only show components related to the present invention and are not drawn according to the number, shape, and size of components in actual implementation. In actual implementation, the type, quantity, and proportion of each component may be changed arbitrarily, and the component layout may also be more complex.

[0021] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the embodiments of the present invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring the embodiments of the present invention.

[0022] A BMS hybrid battery management system based on AI machine collaboration, such as Figure 1 Shown, including: Signal acquisition module: Synchronously collects the voltage gradient, temperature gradient, internal resistance response, and current fluctuation rate signals of each battery cell, performs polynomial interpolation processing according to a unified timestamp, and obtains the synchronous characteristic matrix of each cell; Graph generation module: Calculates nonlinear association weights between all monomers based on the synchronized feature matrix, and generates a topological collaborative graph consisting of monomer nodes and connectivity relationships according to a preset threshold. Vector construction module: This module obtains synergy potential, historical response scores, temperature spectrum peaks, and impedance spectrum breakpoints from the synchronization feature matrix and historical scheduling feedback, constructs a multi-dimensional tag vector, and stores it in a tag vector pool that supports multi-key indexing. Monomer selection module: All marker vectors in the marker vector pool are clustered in multidimensional space according to the density threshold, and a hierarchical priority structure is constructed within each cluster based on the synergy potential and historical response score, and active monomers and redundant monomers are selected in descending order; Collaborative management module: maps the selected active cells to the corresponding cells in the environmental load grid matrix, allocates the available charge and discharge power of each cell based on the preset partition coefficient, and sends it to each BMS controller for differentiated collaborative management of hybrid batteries.

[0023] Specifically, the voltage gradient reflects the unevenness of the potential distribution between different pole pieces or electrode areas of the battery cell, and is used to reflect the imbalance state of the internal electrochemical reaction rate and charge transfer. The acquisition logic is: multiple segmented voltage sampling points are arranged between the positive and negative poles of the cell, and the voltage value of each sampling point is synchronously collected in real time with a high-precision analog-to-digital converter (ADC), and the gradient is calculated based on the voltage difference between adjacent sampling points divided by the sampling interval. Through timed pulse or continuous synchronous sampling, a voltage gradient timing curve across the entire electrode surface can be obtained, which is used to evaluate the risk of uneven discharge or local overload inside the battery; the temperature gradient indicates the degree of change in heat distribution between different positions on the surface or inside of the battery cell, and is used to reflect local heating, heat dissipation imbalance and potential thermal runaway hazards. The acquisition logic is: multiple temperature sensors, including thermocouples or thermistors, are arranged along the cell shell, and the gradient is calculated based on the distance between the sensors and the measured temperature difference. The sensor uses a synchronous sampling mode to generate a temperature gradient time series, providing a basis for active thermal management. The internal resistance response characterizes the impedance characteristics of the cell in response to an applied current surge, revealing the electrochemical blockage, interfacial impedance, and aging of the battery's active materials. The acquisition logic involves periodically applying short charge and discharge pulses to the cell, simultaneously sampling the voltage and current changes before and after the pulses. The internal resistance response is calculated using impedance spectroscopy (EIS) or a two-point transient method. Methods used include multi-frequency AC excitation or pulse step excitation, combined with Fourier transform or time-domain difference quotient calculations, to achieve high-precision detection of impedance components and their inflection points. The current fluctuation rate reflects the instantaneous fluctuation amplitude of the current sourced or absorbed by the cell under a given load change and is used to assess the stability of the control strategy and the consistency of the cell response. The acquisition logic involves connecting a high-bandwidth current sensor, such as a Hall effect sensor or a shunt resistor, in series with the main current path collected by the BMS controller. Current time series data is acquired at high speed, and the absolute value of the current difference between adjacent sampling points is normalized to generate a fluctuation rate curve. A sliding window algorithm is then used to calculate the maximum fluctuation amplitude over a short period of time, which is used to detect sudden load surges or cell mismatch.

[0024] The present invention is further configured such that the signal acquisition module includes: Record the original sampling signal in the ring buffer of each cell and each channel, build a global equidistant aligned timestamp sequence, and unify the sampling time base of each channel; specifically, each battery cell ( and each of its channels (voltage gradient, temperature gradient, internal resistance response, current fluctuation rate) through the ring buffer Recording time series sampling pairs , and construct a global equidistant timestamp sequence: ,in, is the total number of battery cells, is the channel index, To align the starting point, To unify the sampling interval, The total number of timestamps after alignment is used to eliminate the sampling time base differences of each channel and achieve synchronization alignment of all channel data; According to the preset neighborhood span factor of the channel, a set of adjacent sampling points is extracted from the buffer, and a polynomial fitting with regularization constraints is applied to the extracted samples to obtain the local interpolation model coefficients; specifically, for each point to be interpolated , select the front and back The sampling points constitute a neighborhood set in, , is an integer, ranging from 2 to 5; As a reference, construct Taylor basis polynomials of order: , whose coefficient vector Determined by a regularized least squares problem: ,in, is an integer, and its value is [2,4]. , is a high-order difference matrix, Control overfitting suppression, the value range is [10 -3 ,10 -1 ], the above logic has both local fitting ability and noise robustness; Apply the local interpolation model at the alignment timestamp to generate the estimated feature values ​​of each monomer, and arrange the generated feature estimates in channel and time order to form a synchronized feature matrix; specifically, at the alignment time At , directly take the polynomial constant term as the estimated value: , for all channels and time of Arrange in order of channel and time to form Synchronous feature matrix of monomers , as a unified input for subsequent correlation analysis and scheduling.

[0025] The present invention is further configured such that the graph generation module includes: Extract the voltage gradient sequence of each battery cell and calculate the voltage difference between cells. Based on the temperature gradient of each cell, couple the voltage difference to obtain the temperature coupling factor. Specifically, extract the voltage gradient sequence of each cell and calculate the voltage difference between cells. Hedi Voltage gradient timing of the monomer and , calculate the time The absolute difference in: , extract the Hedi Temperature gradient timing of monomers and , applying a coupling map to the voltage difference: ,in, For the Hedi The temperature coupling factor of the monomer, is the temperature coupling mapping power, the value range is [1.2, 2.0], is the temperature attenuation sensitivity coefficient, which is a positive real number and is obtained through calibration. The exponential mapping converts the temperature difference into an attenuation factor, weakening the influence of high temperature difference on the correlation; The voltage difference and the temperature coupling factor are nonlinearly fused to generate the initial correlation value. Specifically, the power amplification of the voltage difference is multiplied by the temperature coupling factor to form the initial correlation value: ,in, For the Hedi The initial association value of the monomer, The power of voltage difference amplification is in the range of [1.5, 3.0], which strengthens the contribution of large voltage difference to the correlation; The initial correlation value is suppressed according to the difference in the internal resistance response inflection point of each monomer to obtain the correlation weight after suppression; specifically, let the monomer and The internal resistance inflection points are , the difference is ; for all times Perform power accumulation on the initial correlation value of , and suppress high breakpoint differences: ,in, is the post-suppression association weight, It is the cumulative amplification power of the association, used to amplify continuous high association, and its value range is [1.5, 3.0]. The internal resistance suppression power is used to suppress monomer pairs with large degradation differences. The value range is [1.0, 2.0]. The suppressed association weights are screened according to the preset association weight threshold, and directed connected edges are generated, and all battery cell nodes and directed connected edges are constructed into a topological collaborative graph; specifically, the association weight threshold is obtained. , is a positive real number, and according to the robustness requirement, all the satisfied Directed pairs As a graph connection edge; as a monomer set And the above directed edge set constructs a topological collaborative graph , used to describe the synergistic relationship between monomers.

[0026] The present invention is further configured to construct a pulse fusion tree based on the post-suppression association weights, and to perform rapid screening of the post-suppression association weights through the pulse fusion tree, including: The associated weight records are hashed into buckets according to predefined time periods, and a balanced binary search tree is constructed in each bucket, where the node is the time period identifier, and the node attributes include the decaying cumulative weight of the time period and the maximum weight of the node in the subtree to which it belongs; specifically, the global time domain is divided into equal-length time periods , each segment length is , define the hash function and number of barrels , for each suppressed association weight record Calculate the bucket index: ; In each bucket Initialize an empty balanced binary search tree, whose node identifiers correspond to the time period index , node attributes include: cumulative weight of time period ; Maximum weight of subtree ; New record Calculation period index: , if the node already exists in the tree , then update its cumulative weight using the exponential decay model: ,in is the time difference since the last update of this node, is the time series attenuation coefficient, and its value range is [0.01, 1.0]. If it does not exist, insert a new node and assign it a value , ; At each node in the balanced binary search tree, update the maximum weight of the subtree of each ancestor node; specifically, from the node Backtrack to the root node, press Update the subtree maximum weight field of each ancestor node to ensure that each node knows the highest weight it can provide in its subtree; Based on the preset association weight threshold, a depth-first pruning filter is performed from root to leaf in the balanced binary search tree, and the weight records corresponding to the nodes whose maximum weight of all subtrees is greater than the threshold are extracted and marked as candidate directed connected edges; specifically, the association threshold is read. , is a positive real number, set according to the robustness requirement, and performs a depth-first traversal starting from the root node in its balanced tree for each bucket in parallel: if the current node of , then skip the entire subtree and no further visit is required; otherwise, check the node's own cumulative weight :like All monomer pair records corresponding to the time period are marked as candidate directed connected edges; then the left and right child nodes are recursively checked, and finally the candidate edge lists in all buckets are summarized as the input for subsequent topology construction.

[0027] The present invention is further configured such that the vector building module includes: Extract the voltage gradient and temperature gradient time series data of each battery cell, and generate synergistic potential based on weighted logarithm and maximum internal resistance response suppression; specifically, from the Extracting Voltage Gradient Sequence from Synchronous Characteristic Matrix of Monomers Temperature gradient sequence ; The two sequences are fused in a weighted logarithmic sum manner and then subjected to exponential mapping, while responding with maximum internal resistance. Inhibit decay and gain synergistic potential: , For synergistic potential, is the synergistic potential normalization factor, which is a positive real number. For the The maximum internal resistance response value of a single unit, The power factor is the internal resistance suppression power, and its value range is [0.5, 2.0]. The dimension difference is eliminated by logarithm. are the voltage and temperature weight powers, respectively, with a value range of [1.0, 3.0], which strengthen the high output state and then suppress the internal resistance attenuation with the exponent and denominator; The power execution error sequence of each monomer is obtained from the historical scheduling feedback buffer, and the historical response score is constructed through logarithmic exponential kernel accumulation; specifically, continuous Power execution error sequence of cycles ; Use log-exponential kernel accumulation to emphasize peak deviations and damp the overall magnitude, and calculate the historical response score: ; is the error amplification index, and its value range is [1.0,3.0]. is the logarithmic damping coefficient, which is a positive real number. Control damping, Amplify abnormal fluctuations, with both sensitivity and robustness; Perform fast Fourier transform on the temperature gradient time series and set the maximum spectrum component as the temperature spectrum peak. Specifically, Apply Fast Fourier Transform (FFT) to obtain the amplitude spectrum ; Select the maximum spectrum amplitude as the temperature spectrum peak: ,This peak reflects the main frequency response intensity of the body temperature fluctuation; Perform second-order differential inflection point detection on the internal resistance response time series and count the number of inflection points to obtain the number of inflection points of the impedance spectrum; specifically, from the internal resistance response time series Calculate second-order differences ; Count the number of inflection points , It is the cardinality of counting or collection, that is, the number of all positions that meet the conditions, which is used to quantify the non-smooth turning caused by aging or interface changes on the impedance curve; The synergistic potential, historical response score, temperature spectrum peak and impedance spectrum breakpoint number are assembled into a multi-dimensional tag vector in sequence and inserted into the tag vector pool that supports multi-key indexing; specifically, the obtained and Assemble into four-dimensional vectors in a fixed order: ;Will Insertion supports arbitrary dimensions Indexed vector pool for clustering and monomer selection modules to call.

[0028] The present invention is further configured such that the monomer selection module includes: All the marker vectors in the marker vector pool are divided based on kernel density estimation and preset density threshold to form multiple non-overlapping clusters; specifically, for each marker vector ( , calculate the local density based on the Gaussian kernel in the vector pool: , where the kernel width is a positive real number, adjusted according to the data distribution, used to adjust the size of the receptive field. Density amplification index, ranging from [1.0, 3.0], is used to amplify high-density differences based on the preset density threshold , all satisfied The points are taken as core points, and according to the accessibility between the core points ( Aggregate the core points into several disjoint clusters, and then assign the non-core points to adjacent core clusters; For each tag vector in a cluster, a multi-level hierarchical structure is constructed based on the synergy potential and historical response score, and the ascending order of the levels is mapped to the descending order of priority; for each tag vector in a cluster, a multi-level hierarchical structure is constructed based on the synergy potential and historical response score, including: calculating the composite priority index of each tag vector in the cluster based on the synergy potential and historical response score; generating a multi-level hierarchical threshold sequence based on the maximum priority index in the cluster and the preset multiplication factor; comparing the composite priority index of each tag vector with the threshold sequence in turn to determine the level to which it belongs; constructing a chain tree structure for each level, and inside each level node, using the monomer index as the key and the corresponding composite priority index as the weight, constructing a small top heap for fast access to high-weight monomers. Specifically, for each cluster within , based on synergistic potential and historical response scoring Calculate the composite priority index: ,in, is the potential power magnification coefficient, with a value range of [1.2, 2.5], which is used to amplify the potential difference. is the historical response mapping coefficient, with a value range of [0.5, 1.5], used to enhance the impact of historical responses; Set the maximum index within the cluster and layering ratio ,generate Level threshold: ;Layering ratio The value range is [1.2, 2.0], with the number of levels As it increases, the threshold decays exponentially; For each , find the minimum in ascending order Make , and mark it as layer; initialize a tree with a depth of The weight ladder tree (chain hierarchical structure), where the first Layer nodes mount all nodes marked as layers The single index is maintained by using a small top heap in each layer. Descending access of monomers; Identify all marker vectors at the highest multiple levels as active monomers, and identify the remaining marker vectors as redundant monomers; specifically, define the upper limit of the active level ; From the 1st floor to the Layer by layer: Extraction of the small top pile before monomers ( The number of active monomers can be set in proportion or fixedly. If the total number of active monomers reaches the global number, the extraction is terminated. The remaining unextracted members of each layer are uniformly marked as redundant monomers.

[0029] The present invention is further configured such that the collaborative management module includes: According to the ambient temperature and real-time load requirements of the monomer, the selected active monomer is mapped to the corresponding unit of the predefined environmental load grid matrix; specifically, the environmental load grid maps each active monomer Get the ambient temperature from the environmental monitor and load demand , and according to the predefined number of grid rows and columns Calculate the mapping coordinates: , , is the upper temperature limit benchmark, The above mapping ensures that the monomer is allocated to the corresponding temperature load condition unit in the environmental load matrix, unifying the subsequent partition processing; The partition coefficient corresponding to each grid unit is obtained from the preset partition coefficient lookup table, and the partition coefficient is nonlinearly fused with the scheduling priority characteristics of each active monomer to generate the power allocation parameter; specifically, the partition coefficient lookup table and the nonlinear fusion are obtained from the preset lookup table. Extract the grid cells Partition coefficient: , is the calibration value in the partition coefficient table, determined by historical data or simulation; the coefficient is compared with the monomer scheduling priority characteristic Generate success rate distribution parameters through power interaction fusion: , where the power coefficient Amplify the impact of priority, Strengthen environmental adaptability; According to the power allocation parameters and the total available power of the system, the available charge and discharge power of each active monomer is calculated according to the proportional division principle; specifically, the normalized power allocation ratio is used for all active monomers. Apply power normalized accumulation: ,in is a collection of active monomers, Used to suppress extreme imbalance and ensure that the total ratio sum is 1. The value range is [1.2,2]; The available charge and discharge power instructions are sent to the BMS controller of the corresponding monomer to realize the differentiated collaborative charge and discharge management of the monomer. Specifically, the total available power of the system is set to , calculate the available power of a single cell: , and Issued to The single BMS controller realizes differentiated charging and discharging.

[0030] The present invention is further configured to include: After each scheduling is completed, the charge and discharge execution errors of each active monomer and the corresponding multi-dimensional tag vector are obtained, packaged to form a scheduling fingerprint, and stored in a ring buffer with a fixed depth; specifically, after each scheduling is completed, the first Active monomer charge and discharge execution error sequence and the corresponding multidimensional labeled vector , and package the two into a scheduling fingerprint: ,Will Insertion depth is If the buffer is full, the oldest fingerprint will be overwritten to ensure that the buffer always retains the most recent fingerprint. Period data; For each monomer in the buffer, the error sequence within the period is extracted and decomposed by wavelet packet to obtain the high-frequency energy ratio of each decomposition level; specifically, for each monomer in the buffer, The latest monomer Period error series Perform Level-L wavelet packet decomposition to obtain the coefficient set ,in , subband index ; For each layer Increase the subband index set , calculate the high frequency energy ratio: ,The high-frequency energy ratio reflects the relative strength of the mutation component of the error sequence; The monomer anomaly index is constructed based on the high-frequency energy ratio of each decomposition level. The power amplification coefficient of the multi-level hierarchical structure and the associated weight screening threshold of the topological collaborative graph are updated based on the anomaly index. Specifically, the power interaction is applied to the ratio of each layer and the logarithmic scaling is taken to generate the monomer anomaly index: in, is the level weight, and its value range is (0,1], , is the power amplification coefficient, which is used to amplify the contribution of high-frequency anomalies to the index. Its value range is [1.5, 3.0]. Normalized scaling of common effects; based on Power coefficients for multi-level stratification , Spectrum association threshold and layering ratio Make adaptive adjustments: , , ,in 、 and is the online gain coefficient, with a value range of (0, 1], which is used to ensure the balance between the sensitivity and stability of each parameter to abnormal response; The updated amplification factor and screening threshold are sent to the monomer selection module and the graph generation module for hierarchical priority division and edge screening in the next scheduling cycle; specifically, the updated It is sent to the monomer selection module and the graph generation module for priority stratification and association screening in the next scheduling cycle, forming the closed-loop adaptive optimization capability of the present invention.

[0031] The present invention is further configured to map the latest multidimensional tag vector in the tag vector pool to the scheduling fingerprint sequence in the ring buffer, and generate a health topology map based on the multimodal similarity and the original association weight fusion; specifically, extract the latest multidimensional tag vector of the i-th monomer from the tag vector pool ; Extract the corresponding monomer anomaly index from the ring buffer For any two monomers i and j, the multimodal similarity is constructed based on the difference in label vectors and the difference in anomaly index: , where is the p-order norm, parameter Control sensitivity to vector and outlier differences, respectively, is the power of the tag similarity mapping, and its value range is [1.5, 3.0]. is the abnormal difference mapping power, with a value range of [1.0, 2.0]. This similarity quantifies the similarity between monomers in terms of performance and health. According to the suppressed association weight in the original topology map Similarity with the previous step Fusion generates health graph edge weights: ,in The power of association weight amplification is used to amplify the impact of historical association on health interconnection, and the value range is [1.2, 2.0]. All battery cells are nodes and fusion weights are Construct a weighted directed health topology for edges ; In the health topology graph, health centrality is calculated based on the similarity weights after fusion between nodes and the historical response scores of each monomer; specifically, in the health graph For each node Evaluate its health importance based on the generalization centrality metric: ,in is the cumulative power of health centrality, which is used to amplify the influence of strong connectivity. Its value range is [1.5, 3.0]. The historical response score in the denominator Suppress the centrality of abnormal nodes; Compare the health centrality with the preset threshold. For abnormal monomers below the threshold, remove them from the active group and select the monomer with the highest health centrality from the redundant group for replacement. Update the replacement results to the active group and redundant group lists. Specifically, read the health centrality threshold ; For all active group monomers, if a monomer satisfy , then: select the monomer with the highest centrality from the redundant group, where Belong to redundant monomer; Remove from active group, Join the active group and update both lists.

[0032] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0033] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. A and B can be singular or plural. Furthermore, the character " / " as used herein generally indicates an "or" relationship between the associated objects, but it may also indicate an "and / or" relationship. For specific understanding, please refer to the context.

[0034] In this application, "at least one" means one or more, and "plurality" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0035] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0036] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0037] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0038] In the several embodiments provided in this application, it should be understood that the disclosed system can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0039] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0040] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0041] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0042] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A BMS hybrid battery management system based on AI machine collaboration, characterized in that: include: Signal acquisition module: Synchronously collects the voltage gradient, temperature gradient, internal resistance response, and current fluctuation rate signals of each battery cell, performs polynomial interpolation processing according to a unified timestamp, and obtains the synchronous characteristic matrix of each cell; Graph generation module: Calculates nonlinear association weights between all monomers based on the synchronized feature matrix, and generates a topological collaborative graph consisting of monomer nodes and connectivity relationships according to a preset threshold. Vector construction module: This module obtains synergy potential, historical response scores, temperature spectrum peaks, and impedance spectrum breakpoints from the synchronization feature matrix and historical scheduling feedback, constructs a multi-dimensional tag vector, and stores it in a tag vector pool that supports multi-key indexing. Monomer selection module: All marker vectors in the marker vector pool are clustered in multidimensional space according to the density threshold, and a hierarchical priority structure is constructed within each cluster based on the synergy potential and historical response score, and active monomers and redundant monomers are selected in descending order; Collaborative management module: maps the selected active cells to the corresponding cells in the environmental load grid matrix, allocates the available charge and discharge power of each cell based on the preset partition coefficient, and sends it to each BMS controller for differentiated collaborative management of hybrid batteries.

2. The BMS hybrid battery management system based on AI machine collaboration according to claim 1 is characterized in that: The signal acquisition module includes: Record the original sampling signal in the ring buffer of each monomer and each channel, build a global equidistant aligned timestamp sequence, and unify the sampling time base of each channel; A set of adjacent sampling points is extracted from the buffer according to the neighborhood span factor preset by the channel, and a polynomial fitting with regularization constraints is applied to the extracted samples to obtain the local interpolation model coefficients; A local interpolation model is applied at the aligned timestamp to generate feature estimates for each monomer, and the generated feature estimates are arranged in channel and time order to form a synchronized feature matrix.

3. The BMS hybrid battery management system based on AI machine collaboration according to claim 1 is characterized in that: The graph generation module includes: Extract the voltage gradient sequence of each battery cell and calculate the voltage difference between cells. Couple the voltage difference based on the temperature gradient of each cell to obtain the temperature coupling factor. The voltage difference and the temperature coupling factor are nonlinearly fused to generate the initial correlation value; The initial correlation value is suppressed according to the difference in the internal resistance response inflection point of each monomer to obtain the correlation weight after suppression; The suppressed association weights are screened according to the preset association weight threshold to generate directed connected edges, and all battery cell nodes and directed connected edges are constructed into a topological collaborative graph.

4. The BMS hybrid battery management system based on AI machine collaboration according to claim 3 is characterized in that: A pulse fusion tree is constructed based on the post-suppression association weights, and the post-suppression association weights are quickly screened through the pulse fusion tree, including: Hash buckets are performed on the associated weight records according to predefined time periods, and a balanced binary search tree is constructed in each bucket, where the node is the time period identifier, and the node attributes include the decaying cumulative weight of the time period and the maximum weight of the node in the subtree to which it belongs; Update the maximum weight of the subtree of each ancestor node at each node in the balanced binary search tree; Based on the preset association weight threshold, depth-first pruning is performed from root to leaf in the balanced binary search tree, and the weight records corresponding to the nodes whose maximum weight of all subtrees is greater than the threshold are extracted and marked as candidate directed connected edges.

5. The BMS hybrid battery management system based on AI machine collaboration according to claim 1 is characterized in that: Vector building blocks include: Extract the voltage gradient and temperature gradient time series data of each battery cell, and generate synergistic potential based on weighted logarithm and maximum internal resistance response suppression; Obtain the power execution error sequence of each unit from the historical scheduling feedback buffer and construct the historical response score through logarithmic exponential kernel accumulation; Perform fast Fourier transform on the temperature gradient time series and set the maximum spectrum component as the temperature spectrum peak; Perform second-order differential inflection point detection on the internal resistance response time series and count the number of inflection points to obtain the number of inflection points in the impedance spectrum; The synergistic potential, historical response score, temperature spectrum peak, and impedance spectrum breakpoint number are sequentially assembled into a multidimensional tag vector and inserted into a tag vector pool supporting multi-key indexing.

6. The BMS hybrid battery management system based on AI machine collaboration according to claim 1 is characterized in that: The monomer selection module includes: All labeled vectors in the labeled vector pool are divided based on kernel density estimation and a preset density threshold to form multiple disjoint clusters; For each tag vector within a cluster, a multi-level hierarchical structure is constructed based on the synergy potential and historical response score, and the ascending order of the hierarchy is mapped to the descending order of priority; All the marker vectors at the highest multiple levels are identified as active monomers, and the remaining marker vectors are identified as redundant monomers.

7. The BMS hybrid battery management system based on AI machine collaboration according to claim 6 is characterized in that: For each cluster marker vector, a multi-level hierarchical structure is constructed based on the synergy potential and historical response score, including: A composite priority index is calculated for each tag vector within each cluster based on synergy potential and historical response scores; Generate a multi-level hierarchical threshold sequence based on the maximum priority index within the cluster and the preset multiplication factor; Compare the composite priority index of each tag vector with the threshold sequence in sequence to determine the level to which it belongs; A chain tree structure is constructed for each level. Inside each level node, a small top heap is constructed with the monomer index as the key and the corresponding composite priority index as the weight to achieve fast access to high-weight monomers.

8. The BMS hybrid battery management system based on AI machine collaboration according to claim 1 is characterized in that: The collaborative management module includes: Map the selected active cells to the corresponding cells of the predefined environmental load grid matrix according to the ambient temperature and real-time load requirements of the cells; Obtain the partition coefficient corresponding to each grid cell from the preset partition coefficient lookup table, perform nonlinear fusion of the partition coefficient with the scheduling priority characteristics of each active cell, and generate the power allocation parameter; Calculate the available charge and discharge power of each active cell according to the power allocation parameters and the principle of proportional division of the total available power of the system; The available charge and discharge power instructions are sent to the BMS controller of the corresponding monomer to achieve differentiated collaborative charge and discharge management of the monomer.

9. The BMS hybrid battery management system based on AI machine collaboration according to claim 1, characterized in that: Also includes: After each scheduling is completed, the charge and discharge execution errors and corresponding multi-dimensional tag vectors of each active monomer are obtained, packaged into a scheduling fingerprint, and stored in a ring buffer with a fixed depth; For each monomer in the buffer, the error sequence within the extraction period is subjected to wavelet packet decomposition to obtain the high-frequency energy ratio of each decomposition level; The monomer anomaly index is constructed based on the high-frequency energy ratio of each decomposition level, and the power amplification coefficient of the multi-level hierarchical structure and the correlation weight screening threshold of the topological collaborative map are updated based on the anomaly index; The updated amplification factor and screening threshold are sent to the monomer selection module and the graph generation module for hierarchical priority division and edge screening in the next scheduling cycle.

10. The BMS hybrid battery management system based on AI machine collaboration according to claim 9, characterized in that: Map the latest multi-dimensional tag vector in the tag vector pool and the scheduling fingerprint sequence in the ring buffer, and generate a health topology map based on the fusion of multimodal similarity and original association weights; In the health topology graph, health centrality is calculated based on the similarity weights after fusion between nodes and the historical response scores of each monomer; The health centrality is compared with the preset threshold. For abnormal monomers below the threshold, they are removed from the active group and replaced with the monomer with the highest health centrality from the redundant group. The replacement results are updated to the active group and redundant group lists.

Images