Battery pack intelligent pairing and integrated optimization method, system, device and storage medium

Abstract

The present invention relates to the field of battery pack technology, and discloses a battery pack intelligent pairing and integrated optimization method, system, device and storage medium. The method comprises: performing parameter collection and orthogonal feature extraction on battery cells in a battery pack to obtain an orthogonal parameter characteristic matrix of the battery pack; performing feature deep learning to obtain a battery pack pairing prediction model; constructing a battery attribute similarity graph and a battery pack topology graph, and obtaining a battery pack intelligent pairing structure through a double-graph structure alignment operation; performing parallel multi-population particle swarm optimization calculation to obtain a battery pack optimization pairing scheme; inputting a hierarchical model prediction controller to perform energy storage boundary analysis to obtain upper-layer power allocation parameters and lower-layer charge and discharge current parameters; performing closed-loop control operations based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain battery pack pairing control optimization parameters. The present invention realizes real-time monitoring and dynamic adjustment of battery pack performance parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of battery packs, and in particular to a method, system, device and storage medium for intelligent pairing and integrated optimization of battery packs. Background Art

[0002] As a key component in the new energy field, the performance of the battery pack directly affects the reliability and service life of the entire system. Due to differences in manufacturing processes and material properties, there are inherent inconsistencies in performance parameters between battery cells, which can lead to problems such as uneven capacity utilization and reduced charging and discharging efficiency within the battery pack.

[0003] Traditional battery pack pairing methods mainly rely on manual experience and simple parameter matching. It is difficult to accurately capture the deep performance correlation between battery cells and cannot effectively cope with the dynamic changes in battery performance. At the same time, the existing battery pack management system lacks the ability to intelligently optimize and precisely control the battery pack pairing process, resulting in the inability to fully utilize the overall performance of the battery pack. In addition, battery packs face complex and changeable working environments and load conditions during actual operation, and a single pairing strategy is difficult to meet performance requirements under dynamic conditions. How to establish a method system that comprehensively considers battery performance feature extraction, intelligent pairing optimization, and integrated control to achieve continuous optimization of battery pack performance has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present invention provides a battery pack intelligent pairing and integrated optimization method, system, device and storage medium, which are used to realize real-time monitoring and dynamic adjustment of battery pack performance parameters.

[0005] In a first aspect, the present invention provides a battery pack intelligent pairing and integrated optimization method, the battery pack intelligent pairing and integrated optimization method comprising:

[0006] Perform parameter collection and orthogonal feature extraction on battery cells in the battery pack to obtain the orthogonal parameter characteristic matrix of the battery pack;

[0007] Inputting the battery pack orthogonal parameter characteristic matrix into a convolutional neural network optimized by a genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model;

[0008] Based on the battery pack pairing prediction model, a battery attribute similarity graph and a battery pack topology graph are constructed, and a battery pack intelligent pairing structure is obtained by a double-graph structure alignment operation;

[0009] Performing parallel multi-swarm particle swarm optimization calculations on the battery pack intelligent pairing structure to obtain an optimized battery pack pairing solution;

[0010] Inputting the battery pack optimization pairing scheme into a hierarchical model predictive controller to perform energy storage boundary analysis to obtain upper-layer power allocation parameters and lower-layer charge and discharge current parameters;

[0011] A closed-loop control operation is performed based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain battery pack pairing control optimization parameters.

[0012] In a second aspect, the present invention provides a battery pack intelligent pairing and integrated optimization system, the battery pack intelligent pairing and integrated optimization system comprising:

[0013] The acquisition module is used to acquire parameters and extract orthogonal features of battery cells in the battery pack to obtain an orthogonal parameter characteristic matrix of the battery pack;

[0014] A deep learning module, used for inputting the battery pack orthogonal parameter characteristic matrix into a convolutional neural network optimized by a genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model;

[0015] A construction module, used to construct a battery attribute similarity graph and a battery pack topology graph based on the battery pack pairing prediction model, and obtain a battery pack intelligent pairing structure through a double-graph structure alignment operation;

[0016] A calculation module, used for performing parallel multi-swarm particle swarm optimization calculation on the battery pack intelligent pairing structure to obtain an optimized pairing solution for the battery pack;

[0017] An analysis module is used to input the battery pack optimization pairing scheme into a hierarchical model predictive controller to perform energy storage boundary analysis to obtain upper power allocation parameters and lower charge and discharge current parameters;

[0018] The closed-loop control module is used to perform closed-loop control operations based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain battery pack pairing control optimization parameters.

[0019] The third aspect of the present invention provides a battery pack intelligent pairing and integrated optimization device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the battery pack intelligent pairing and integrated optimization device executes the above-mentioned battery pack intelligent pairing and integrated optimization method.

[0020] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned battery pack intelligent pairing and integrated optimization method.

[0021] In the technical solution provided by the present invention, by introducing orthogonal feature extraction and genetic algorithm optimized convolutional neural network, deep mining of battery cell performance characteristics is achieved, overcoming the problem that traditional methods are difficult to accurately describe the correlation of battery performance; based on the intelligent pairing mechanism constructed by the dual-graph structure neural network, an alignment optimization model of the battery attribute similarity graph and the topology graph is established, solving the dynamic optimization problem in the battery pack pairing process; a parallel multi-population particle swarm algorithm is used to optimize the pairing scheme, which significantly improves the pairing optimization efficiency, and an integrated optimization strategy based on hierarchical model predictive control is designed. Through energy storage boundary analysis and a double-layer control structure, the stable operation of the battery pack under dynamic conditions is guaranteed; a closed-loop control optimization mechanism is established to achieve real-time monitoring and dynamic adjustment of battery pack performance parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of a flow chart of a battery pack intelligent pairing and integrated optimization method provided in an embodiment of the present application;

[0024] Figure 2 A schematic block diagram of the structure of a battery pack intelligent pairing and integrated optimization system provided in an embodiment of the present application;

[0025] Figure 3 A schematic block diagram of the structure of the battery pack intelligent pairing and integrated optimization device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0028] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0029] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0030] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0031] See also Figure 1 , Figure 1 A flowchart of a battery pack intelligent pairing and integrated optimization method provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the battery pack intelligent pairing and integrated optimization method provided in the embodiment of the present application includes steps S100 to S600.

[0032] Step S100, collecting parameters and extracting orthogonal features of battery cells in the battery pack to obtain an orthogonal parameter characteristic matrix of the battery pack;

[0033] It is understandable that the execution subject of the present invention may be a battery pack intelligent pairing and integrated optimization system, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0034] Specifically, multi-dimensional parameters of the battery cells in the battery pack are collected in real time, including key performance parameters of each battery cell, such as voltage, internal resistance, capacity, state of charge (SOC) and operating temperature, etc. These parameters constitute the raw data matrix of the battery cell. The collected raw data matrix is ​​preprocessed to ensure the reliability and accuracy of the data. The preprocessing process includes noise elimination of the data, using filtering algorithms or other signal processing methods to eliminate random noise in the collected signal; at the same time, statistical methods and machine learning algorithms are used to eliminate abnormal values ​​that obviously deviate from the normal range to obtain effective characteristic parameters of the battery cell. Material characteristic analysis and parameter correlation calculation are performed on the effective characteristic parameters of the battery cell to analyze the intrinsic connection and mutual influence between different characteristics. For example, by analyzing the chemical properties of battery materials, the trend of internal resistance changing with state of charge, and the attenuation law of capacity with the number of cycles, the performance characteristic matrix of the battery pack is initially constructed to reflect the key performance indicators of the battery cell in actual work. However, due to the high dimension of the matrix, in order to avoid data redundancy and excessive calculation complexity, the performance characteristic matrix is ​​reduced in dimension. In the process of dimensionality reduction, principal component analysis or other feature selection algorithms are used to project the high-dimensional feature space into the low-dimensional space to extract the target orthogonal parameters that best represent the performance of the battery pack. The state of charge consistency, capacity matching, internal resistance balance and temperature distribution characteristics of the target orthogonal parameters of the battery pack are calculated to calculate the consistency and balance of the battery pack in different dimensions. By calculating the consistency of the state of charge (SOC), the difference in the current power distribution of each cell in the battery pack is evaluated; by calculating the capacity matching, the impact of the capacity difference of each cell in the battery pack on the overall performance is analyzed; by calculating the internal resistance balance, the uniformity of the current distribution of each cell is ensured; by calculating the temperature distribution characteristics, the balance of the heat distribution during the operation of the battery pack is evaluated, thereby reducing the risk of local overheating. These calculation results are expressed in the form of standardized feature vectors to form the standardized feature data of the battery pack. The standardized feature vectors are combined with the time series to generate the time series feature data of the battery pack. By introducing a dynamic feature update mechanism, the changes in the performance of the battery cells over time and working conditions are captured in real time to ensure that the extracted features can reflect the current true state of the battery pack. Based on the time series feature data, a structure-performance mapping model of the battery pack is constructed to associate the structural characteristics and performance of the battery cell. The mapping model is processed through orthogonal transformation operations to generate the battery pack orthogonal parameter characteristic matrix.

[0035] Step S200, inputting the battery pack orthogonal parameter characteristic matrix into the convolutional neural network optimized by the genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model;

[0036] Specifically, the orthogonal parameter characteristic matrix of the battery pack is divided into a training set and a validation set according to the time sequence to ensure that the model can learn the dynamic change trend of battery performance. In the data preprocessing stage, the training set is batch normalized to standardize the data, eliminate the dimensional differences between different characteristic dimensions, improve the stability and convergence speed of training, and obtain standardized training data. In order to construct a convolutional neural network and optimize its structure, a genetic algorithm is used to encode the network parameters. The genetic algorithm defines the gene sequence of the network structure, including the size of the convolution kernel, the step size, the pooling parameter, and the number of neurons in the fully connected layer. This genetic encoding method can explore the optimal network structure in an evolutionary search method. Based on the network structure gene sequence generated by the genetic algorithm, a multi-layer convolutional neural network is constructed. The network contains three convolution modules, each of which consists of two convolution layers and a maximum pooling layer. The convolution layer uses the ReLU activation function to ensure the effective capture of nonlinear features. The feature extraction network can efficiently extract key deep features in battery performance. In order to enhance the generalization ability and training stability of the network, a batch normalization layer and a dropout layer are added to each convolution module to construct a regularized feature extraction network. Regularization can alleviate the overfitting problem and improve the robustness of the network to different battery performance parameters. On this basis, the standardized training data is input into the regularized feature extraction network. The data is sequentially subjected to the feature extraction operation of three convolution modules to obtain the deep features of the battery performance. These features contain the complex correlation information of the battery pack in different performance dimensions. The deep features of the battery performance are input into the fully connected layer for mapping. The fully connected layer consists of two hidden layers, where the first hidden layer contains 512 neurons and the second hidden layer contains 256 neurons. This design can gradually compress and fuse features, extract global information and eliminate redundancy. Through the calculation of the fully connected layer, the feature fusion result is finally obtained. Based on the feature fusion result, the output layer of pairing prediction is constructed. The output layer uses the softmax activation function for multi-classification prediction, and classifies the pairing of different monomers in the battery pack into multiple categories. The training at this stage uses the back propagation algorithm to optimize the network parameters through gradient descent to complete the training of the initial pairing prediction model. The initial model can provide relatively accurate pairing prediction results based on the input data. The validation set is input into the initial pairing prediction model for cross-validation to evaluate the performance of the model. The verification process can reveal the model's predictive ability on unseen data, and update the fitness function of the genetic algorithm by analyzing the verification results, thereby optimizing the network structure and parameters. The genetic algorithm guides the evolution of the network structure at this stage, and finally obtains the optimized battery pack pairing prediction model through multiple rounds of iterations.

[0037] Step S300: constructing a battery attribute similarity graph and a battery pack topology graph based on the battery pack pairing prediction model, and obtaining a battery pack intelligent pairing structure by aligning the two graph structures;

[0038] Specifically, the output of the battery pack pairing prediction model, including the pairing probability distribution between battery cells and their deep performance characteristics, is used to perform node mapping and generate a battery cell feature node set. In the node set, each node contains not only the pairing probability characteristics provided by the pairing prediction model, but also the deep features extracted from the battery performance data. These features together constitute a comprehensive description of the battery cell performance. Based on the battery cell feature node set, the similarity weights between nodes are calculated to construct a battery attribute similarity graph. The similarity weights are calculated using a variety of methods, such as cosine similarity, Euclidean distance or other feature correlation indicators to quantify the degree of characteristic similarity between each pair of nodes. At the same time, combined with the characteristics of the actual physical connection of the battery pack, the topological information is extracted from the structural layout of the battery pack to generate a battery pack topology graph. The topological graph reflects the actual connection relationship between battery cells, which is consistent with the attribute similarity. Figure 1 The two graph structures are formed at the same time. The battery attribute similarity graph is input into the dual graph structure neural network for feature processing. The design of the dual graph structure neural network includes two layers of graph attention network, the output dimension of the first layer is 64, and the output dimension of the second layer is 32. The graph attention network can dynamically adjust the information transfer weights between different nodes, so that the model pays more attention to the node connection relationship with important characteristics. After being processed by the two-layer graph attention network, the node representation features are output. The node representation features are input into the graph convolution layer to enhance the feature expression ability. The graph convolution layer improves the global representation ability of node features by aggregating the information of neighboring nodes and generates enhanced graph structure features. On this basis, the consistency of the graph structure is calculated by the structure-aware loss function to obtain the structure alignment parameter. The introduction of structure-aware loss can ensure that the enhanced graph structure features can reasonably align the attribute similarity and topological structure while retaining the original topological information, thereby improving the adaptability of the model on the dual graph structure. The structure alignment parameter is input into the message passing neural network to update the information between nodes. This process gradually integrates the characteristics and similarity information between nodes through multiple message passing iterations, so that the embedding features of each node more accurately reflect its position and role in the global graph structure. The updated node embedding is obtained through the node update of the message passing neural network. Based on the updated node embedding, the battery pack intelligent pairing structure is analyzed to generate the battery pack intelligent pairing structure matrix. This matrix contains the optimal pairing relationship between battery cells, and integrates the information of battery attribute similarity and topological structure alignment, which can fully reflect the performance optimization and structural matching of the battery pack.

[0039] Step S400, performing parallel multi-swarm particle swarm optimization calculation on the battery pack intelligent pairing structure to obtain a battery pack optimization pairing solution;

[0040] Specifically, the battery performance parameters and topological relationships in the intelligent pairing structure of the battery pack are converted into particle encoding. The particle encoding is expressed in the form of position vectors and velocity vectors, where the position vector describes the pairing scheme of the battery cells, including the relative position relationship of the battery cells in the intelligent pairing structure, and the velocity vector represents the adjustment direction and amplitude of the pairing scheme. This encoding method can intuitively transform the battery pairing problem into a problem to be solved in the particle swarm optimization algorithm, obtain the position matrix and velocity matrix of the initial population, and construct the initial conditions for optimization. After the initial population is generated, in order to improve the optimization efficiency and global search capability, the initial population position matrix is ​​divided into multiple sub-populations. Each sub-population corresponds to a group of battery cells to be paired, and the search space and iteration parameters are independently set for each sub-population to ensure that the optimization process can fully consider the specific performance requirements and constraints of each group of batteries. Through parallel sub-population division, the computational overhead caused by the population size is effectively reduced, and the adaptability to complex battery pack pairing problems in the optimization process is improved. In order to evaluate the fitness of each particle, a multi-objective fitness function is constructed based on multi-objective requirements such as battery performance consistency, charge balance, and battery pack service life. The fitness function comprehensively considers three indicators, among which the weight coefficient of battery performance parameter consistency is 0.4, the weight coefficient of charge balance is 0.3, and the weight coefficient of battery pack service life is 0.3. Weight allocation can balance the contradiction between battery performance and service life during the optimization process, while ensuring the operational stability and reliability of the battery pack. The fitness calculation model combines these weights with the current state of the particle to evaluate the degree of adaptation of each particle. After completing the fitness calculation, the fitness calculation model is input into the search module of the sub-population, and the position and speed are iteratively updated through the core formula of the particle swarm optimization algorithm. The position update formula determines the moving direction of the current position of the particle in the search space, while the speed update formula adjusts the exploration strategy of the particle according to the particle's own historical optimal solution and the local optimal solution of the population. After multiple rounds of iterations, each sub-population gradually converges to its local optimal solution. On this basis, the optimization information between each sub-population is interacted through the population migration operation, so that the population can break through the limitations of the local optimal solution and improve the global search capability. After completing the population migration, the result of the population migration is updated to the global optimal solution. The global optimal solution integrates the optimization results of each sub-population and takes into account the performance matching requirements of the overall battery pack. The global optimal pairing solution is input into the topology optimization module, with the optimization goal of minimizing the total internal resistance and voltage loss of the battery pack, adjusting the physical connection relationship of the battery pack, and generating an optimized topology structure, so that the operating efficiency and energy output performance of the battery pack reach the optimal level. Based on the optimal connection relationship, the pairing parameters of the battery pack are integrated to obtain the final battery pack optimization pairing solution.

[0041] Step S500: inputting the battery pack optimization pairing scheme into the hierarchical model predictive controller to perform energy storage boundary analysis to obtain upper power allocation parameters and lower charge and discharge current parameters;

[0042] Specifically, the energy storage characteristics of the battery pack optimization pairing scheme are modeled. By analyzing the dynamic characteristics of the current, voltage, temperature, etc. of each battery cell in the battery pack, an energy storage model is constructed, which can reflect the energy storage behavior and performance changes of the battery pack under different operating conditions. At the same time, the energy storage potential is quantitatively analyzed, the boundary conditions of the battery pack during energy storage and release are identified, and the energy storage boundary constraints are generated. Based on the energy storage boundary constraints, a hierarchical model predictive controller is constructed, including an upper power controller and a lower current controller, which respectively handle the power distribution and current regulation tasks of the battery pack. Among them, the upper power controller adopts a model prediction algorithm to predict the power demand of the battery pack at multiple moments in the future through rolling time domain optimization, and generates corresponding power allocation strategies; while the lower current controller adopts a feedback control algorithm to perform accurate current regulation according to the output of the upper power controller to ensure that the operating state of each battery cell remains within a safe range. The hierarchical structure design enables the control system to achieve global optimization and local precise adjustment at the same time. In the upper power controller, the real-time status information of the battery pack is input into three power prediction units and one allocation optimization unit. The power prediction unit predicts the future power demand curve of the battery pack based on the energy storage model and historical operation data, while the allocation optimization unit dynamically adjusts the power allocation strategy based on the predicted power demand based on the rolling time domain optimization algorithm. The rolling optimization method can update the power allocation scheme in real time to ensure that the battery pack can still maintain efficient power output under dynamic operating conditions. The optimization target of the upper-level control is constructed based on the power allocation strategy. The optimization targets include power balance coefficient, voltage fluctuation coefficient and energy storage efficiency coefficient. By comprehensively optimizing these indicators, the power output stability of the battery pack is maximized, the voltage fluctuation inside the battery pack is reduced, and the overall energy storage efficiency is improved. By solving this optimization target, the upper-level power allocation parameters are obtained to provide guidance for the lower-level current control. After the upper-level power allocation parameters are input into the lower-level current controller, the lower-level controller generates the target current reference value through the current reference generation unit, and uses the feedback correction unit to dynamically adjust the deviation between the actual current and the reference value. Through this closed-loop control process, the precise regulation of the charging and discharging process of the battery pack is achieved, thereby effectively preventing the overcharging or over-discharging of the single battery and ensuring the safe operation of the entire battery pack. In the lower-level current controller, the optimization objectives of the lower-level control are constructed based on the charge and discharge regulation strategy. The lower-level optimization objectives include current balance, state of charge (SOC) consistency, and minimization of internal resistance loss. Through the joint optimization of these objectives, balanced operation within the battery pack can be achieved, energy loss caused by internal resistance differences can be reduced, and SOC consistency between each single cell can be maintained. This optimization objective is solved by quadratic programming to obtain the lower-level charge and discharge current parameters to ensure that the battery pack remains efficient and safe during operation.

[0043] Step S600: performing a closed-loop control operation based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain the battery pack pairing control optimization parameters.

[0044] Specifically, the upper power allocation parameters and the lower charge and discharge current parameters are checked for consistency. This check process analyzes the matching degree between the power allocation strategy and the charge and discharge current, constructs a mapping relationship matrix between power and current, reflects the dynamic relationship between the power output and current regulation of each single cell in the battery pack, and can identify potential inconsistencies or abnormal behaviors. Dual closed-loop control and performance evaluation are performed according to the mapping relationship matrix. The core of dual closed-loop control is to simultaneously adjust the power allocation and current output to achieve bidirectional optimization closed-loop regulation of power and current. The power allocation is controlled by the outer loop to keep the overall power output of the battery pack stable, while the inner loop adjusts the charge and discharge current to ensure that the current of each single cell operates within a safe range. The performance evaluation process forms a performance monitoring matrix by monitoring the key performance indicators of the battery pack under dual closed-loop control, including voltage fluctuation range, power output stability, and current balance. The performance monitoring matrix provides dynamic performance feedback for real-time optimization control, reflecting the operating status of the battery pack under the current control strategy. The performance monitoring matrix is ​​used for real-time optimization control. By analyzing the real-time operation data of the battery pack, the control strategy is dynamically adjusted to improve the overall performance of the battery pack, and the optimized control parameters are obtained and input into the multi-level controller. The multi-level controller adopts a hierarchical progressive design concept, which is divided into a voltage stability control level, a power balance control level, and a current balance control level. The voltage stability control level adjusts the voltage distribution of the single cell to reduce the unevenness of the internal voltage of the battery pack and ensure the stable operation of the entire system; the power balance control level optimizes the power distribution between each single cell based on the upper power distribution parameters to improve the overall output efficiency of the battery pack; the current balance control level is based on the lower layer of charge and discharge current parameters. Through detailed adjustment, the current difference between the single cells is reduced, thereby extending the service life of the battery pack and reducing internal consumption. Under the hierarchical progressive control of the multi-level controller, the controllers at all levels work together to achieve all-round performance improvement of the battery pack through gradual optimization. After optimization and adjustment of the three levels of voltage, power and current, the optimized parameters for battery pack pairing control are formed. These optimized parameters can not only dynamically adapt to the operating environment of the battery pack, but also significantly improve the operating efficiency and reliability of the battery pack in the energy storage system. Through this closed-loop control operation method, the upper-level power allocation and lower-level current regulation are organically combined, and a multi-level progressive control framework is introduced to achieve global optimization of battery pack performance.

[0045] In the embodiment of the present invention, by introducing orthogonal feature extraction and genetic algorithm optimized convolutional neural network, deep mining of battery cell performance characteristics is achieved, overcoming the problem that traditional methods are difficult to accurately describe the correlation of battery performance; based on the intelligent pairing mechanism constructed by the dual-graph structure neural network, an alignment optimization model of the battery attribute similarity graph and the topology graph is established, solving the dynamic optimization problem in the battery pack pairing process; a parallel multi-population particle swarm algorithm is used to optimize the pairing scheme, which significantly improves the pairing optimization efficiency, and an integrated optimization strategy based on hierarchical model predictive control is designed. Through energy storage boundary analysis and a double-layer control structure, the stable operation of the battery pack under dynamic conditions is guaranteed; a closed-loop control optimization mechanism is established to achieve real-time monitoring and dynamic adjustment of battery pack performance parameters.

[0046] In a specific embodiment, the process of executing step S100 may specifically include the following steps:

[0047] The multi-dimensional parameters of the battery cells in the battery pack are collected in real time to obtain the original data matrix of the battery cells, and the noise elimination and outlier removal operations are performed on the original data matrix of the battery cells to obtain the effective characteristic parameters of the battery cells;

[0048] Perform material property analysis and parameter correlation calculation on the effective characteristic parameters of the battery cells to obtain the battery pack performance characteristic matrix, and perform feature dimension reduction operation on the battery pack performance characteristic matrix to obtain the target orthogonal parameters of the battery pack;

[0049] The battery pack target orthogonal parameters are calculated for charge state consistency, capacity matching, internal resistance balance, and temperature distribution characteristics to obtain the standardized feature vector of the battery pack.

[0050] The standardized feature vector of the battery pack is dynamically updated to obtain the timing feature data of the battery pack, and a structure-performance mapping model is constructed based on the timing feature data of the battery pack. The orthogonal parameter characteristic matrix of the battery pack is obtained through orthogonal transformation operation.

[0051] Specifically, the key parameters of each battery cell in the battery pack are collected in real time, including voltage , Current 、Internal resistance ,capacity , state of charge (SOC) and temperature These parameters constitute the raw data matrix of the battery cell, recorded as ,in Indicates The first battery cell The matrix reflects the global operating state of the battery pack at a certain moment. The original data matrix is ​​subjected to noise elimination and outlier removal operations. The noise is eliminated by using a low-pass filter. , its transfer function is:

[0052] ;

[0053] in, is the cutoff frequency of the filter. Outlier removal is based on statistical methods, such as standardizing the data and removing outliers using the triple standard deviation principle, that is, for each ,if:

[0054] ;

[0055] It is believed that is an outlier, where and Respectively represent After this stage of processing, the effective characteristic parameter matrix of the battery cell is obtained. , which is used for subsequent analysis. Material characteristic analysis and parameter correlation calculation are performed on the effective characteristic parameters to reveal the influence of battery material characteristics on performance. Assume that the material characteristic vector is ,in Represents a physical or chemical property of a material, such as the specific capacity or ionic conductivity of an active substance. Parameter correlation is calculated using the correlation coefficient matrix to achieve this, where:

[0056] ;

[0057] in, is a parameter and The covariance of and are their standard deviations respectively. Through this process, the battery pack performance characteristic matrix is ​​generated , indicating the key performance indicators of each battery cell. The target orthogonal parameters are extracted by feature dimension reduction methods (such as principal component analysis PCA). Let the orthogonal parameter matrix after dimension reduction be , satisfying the following orthogonality conditions:

[0058] ;

[0059] in is the identity matrix. The most important performance features in the data are retained. Analyze and extract the standardized feature vector of the battery pack through calculation of charge state consistency, capacity matching, internal resistance balance and temperature distribution characteristics. For example, the SOC consistency is calculated using the following formula:

[0060] ;

[0061] Where SOC is the average state of charge of the battery pack. The capacity matching is calculated as:

[0062] ;

[0063] in, is the number of battery cells, is the average capacity of the battery pack. The calculation of internal resistance balance and temperature distribution characteristics adopts similar standardization methods to ensure that each feature dimension has the same dimension and range. After generating the standardized feature vector s, the time series feature is updated in combination with the dynamic operation data. The dynamic behavior of the feature over time is described by the time series analysis method (such as the autoregressive moving average model ARMA). Assume that the time series feature is , and its update equation is:

[0064] ;

[0065] in is the state transition matrix, is Gaussian white noise. Based on these time series features, a structure-performance mapping model is constructed, and the feature space is mapped into the final orthogonal parameter characteristic matrix through orthogonal transformation operation. The orthogonal transformation formula is:

[0066] ;

[0067] in is an orthogonal transformation matrix, ensuring Preserve key features and satisfy constraints.

[0068] In a specific embodiment, the process of executing step S200 may specifically include the following steps:

[0069] The orthogonal parameter characteristic matrix of the battery pack is divided into a training set and a validation set according to the time sequence, and the training set is batch normalized to obtain standardized training data;

[0070] Genetic algorithm encoding is performed on the convolutional neural network structure to obtain the network structure gene sequence, where the genetic algorithm encoding includes the convolution kernel size, step size, pooling parameters, and the number of neurons in the fully connected layer;

[0071] A multi-layer convolutional neural network is constructed based on the network structure gene sequence. The multi-layer convolutional neural network includes three convolutional modules. Each convolutional module contains two convolutional layers and a maximum pooling layer. The convolutional layer uses the ReLU activation function to obtain a feature extraction network.

[0072] Add batch normalization layers and dropout layers to each convolutional module of the feature extraction network to obtain a regularized feature extraction network. Then input the standardized training data into the regularized feature extraction network, and obtain the deep features of battery performance through sequential calculations of three convolutional modules.

[0073] The deep features of battery performance are mapped to a fully connected layer, which includes two hidden layers. The number of neurons in the first hidden layer is 512, and the number of neurons in the second hidden layer is 256, to obtain the feature fusion result;

[0074] Based on the feature fusion results, the pairing prediction output layer is constructed, and the softmax activation function is used for multi-classification prediction. The network parameters are optimized through the back propagation algorithm to obtain the initial pairing prediction model.

[0075] The validation set is input into the initial pairing prediction model for cross-validation to obtain the validation result, and the fitness function of the genetic algorithm is updated and iteratively optimized based on the validation result to obtain the battery pack pairing prediction model.

[0076] Specifically, the orthogonal parameter characteristic matrix of the battery pack is preprocessed. is a set of key parameters with orthogonal characteristics extracted from the battery pack. Each row represents the characteristics of a battery cell, and each column corresponds to an independent parameter dimension. In order to build a deep learning model, Divide into training sets according to time series and validation set The partitioning strategy is based on a fixed time window, such as The data is used for training, and then To avoid the impact of parameter differences on model training, Perform batch normalization. The batch normalization formula is:

[0077] ;

[0078] in, and Respectively represent the mean and standard deviation of each parameter in the training set. Through normalization, standardized training data is obtained In the design stage of deep learning models, genetic algorithms are introduced to automatically optimize the structure of convolutional neural networks (CNNs). Genetic algorithms encode network structures by simulating the natural selection process and encode them into gene sequences. ,in It is The size of the convolution kernel of the layer, is the step length, is the pooling parameter, is the number of neurons in the fully connected layer. The initial gene sequence is randomly generated and gradually optimized during the evolution process to find the optimal network structure. A multi-layer convolutional neural network is constructed based on the gene sequence generated by the genetic algorithm. The network includes three convolutional modules, each of which contains two convolutional layers and a maximum pooling layer. The convolutional layer uses the ReLU activation function, and the formula is:

[0079] ;

[0080] This activation function can effectively capture the nonlinear characteristics of the input data. Each maximum pooling layer is calculated by the formula:

[0081] ;

[0082] in is the size of the pooling window, which compresses local information into a higher-level feature representation. After three convolution modules, a feature extraction network is formed to extract deep features of battery performance. To enhance the generalization and robustness of the network, a batch normalization layer and a dropout layer are added to each convolution module. The batch normalization layer standardizes each batch of input data to ensure that the input distribution of each layer is more stable; the dropout layer randomly discards a part of neurons to prevent overfitting. After input into the regularized feature extraction network, the deep features of battery performance are generated through calculation by the convolution module. Assumptions The dimension is ,in is the number of battery cells, is the feature dimension. Input the fully connected layer for feature fusion. The fully connected layer contains two hidden layers. The number of neurons in the first hidden layer is 512, and the number of neurons in the second hidden layer is 256. The fully connected layer calculates the output using the following formula:

[0083] ;

[0084] in is the weight matrix, is the bias term, is the activation function (such as ReLU), It is The input of the layer. After being mapped by the fully connected layer, the feature fusion result is obtained Based on the fusion features, a pair prediction output layer is constructed, and the output layer uses a softmax activation function to achieve multi-classification prediction. The formula of the softmax function is:

[0085] ;

[0086] in Yes Category The linear score of is predicted as class The probability of . The network parameters are optimized by the back propagation algorithm, and the loss function value is gradually reduced, such as the cross entropy loss:

[0087] ;

[0088] in It is The actual category labels of samples. After completing the initial model training, the validation set The initial pairing prediction model is input for cross-validation to evaluate the performance of the model, including classification accuracy, precision, recall rate, and F1 score. According to the verification results, the fitness function of the genetic algorithm is adjusted, and the network structure with high accuracy and strong generalization ability is preferentially selected for multiple rounds of iterative optimization. After multiple rounds of genetic algorithm iterations, the optimized battery pack pairing prediction model is obtained.

[0089] In a specific embodiment, the process of executing step S300 may specifically include the following steps:

[0090] Perform node mapping on the pairing probability distribution and battery performance deep features output by the battery pack pairing prediction model to obtain a battery cell feature node set, where each node contains pairing probability features and performance deep features;

[0091] Based on the battery cell feature node set, the similarity weights between nodes are calculated to obtain the battery attribute similarity graph, and the topological structure of the actual physical connection state of the battery pack is extracted to obtain the battery pack topology graph;

[0092] The battery attribute similarity graph is input into a dual-graph structure neural network for processing. The dual-graph structure neural network includes a first-layer graph attention network with an output dimension of 64 and a second-layer graph attention network with an output dimension of 32 to obtain node representation features.

[0093] The node representation features are input into the graph convolution layer for feature enhancement to obtain the enhanced graph structure features, and the structure-aware loss is calculated for the enhanced graph structure features to obtain the structure alignment parameters;

[0094] The structural alignment parameters are input into the message passing neural network to update the node information to obtain the updated node embedding, and the battery pack intelligent pairing structure analysis is performed based on the updated node embedding to obtain the battery pack intelligent pairing structure matrix.

[0095] Specifically, the pairing probability distribution output by the battery pack pairing prediction model and the deep characteristics of battery performance are node mapped. The output of the pairing prediction model includes the pairing probability distribution between each pair of battery cells. and battery performance deep feature matrix ,in Indicates and The probability of successful pairing of battery cells, It is The deep feature vector of each battery. This information is integrated into a battery cell feature node set through node mapping. Each node It consists of pairing probability features and performance deep features, and its feature vector is expressed as Each node in the node set can reflect the pairing potential of the battery cell and its own performance characteristics. After obtaining the node set, the feature vector of the node is and Calculate the similarity weight between nodes , generate battery attribute similarity graph ,in is a collection of nodes, is the edge set of the similarity graph, is the edge weight matrix. Similarity weight Calculated by cosine similarity:

[0096] ;

[0097] This formula measures the similarity between the feature vectors of two nodes, ranging from [-1, 1], where 1 indicates complete similarity and -1 indicates complete dissimilarity. At the same time, the topology is extracted from the actual physical connection state of the battery pack to generate a battery pack topology map. ,in is the edge set of the actual connection relationship of the battery, is the corresponding connection weight matrix. The topology graph reflects the physical layout and connection constraints of the battery pack, and forms a dual graph structure with the attribute similarity graph. Input the dual-graph structure neural network for feature processing. The dual-graph structure neural network consists of two layers of graph attention networks, where the output dimension of the first layer is 64 and the output dimension of the second layer is 32. The core of each layer of graph attention network is to update the node representation features by aggregating the features of adjacent nodes. The formula is:

[0098] ;

[0099] in, is the updated node feature, is the activation function, is a trainable weight matrix, is the attention coefficient, indicating that the node and The importance relationship between them. The attention coefficient is calculated by softmax regularization:

[0100] ;

[0101] in is a trainable attention vector, and || represents a feature connection operation. After the two-layer graph attention network is calculated, the node representation feature matrix is ​​obtained. Then Input the graph convolution layer to enhance feature expression. The calculation formula of graph convolution is:

[0102] ;

[0103] in is the enhanced graph structure feature, is the adjacency matrix of the graph, is the degree matrix, is a trainable weight. The enhanced features are input into the structure-aware loss function for calculation to optimize the alignment and consistency of the graph. The structure-aware loss is defined as:

[0104] ;

[0105] in and are the feature matrices of the similarity graph and the topology graph, respectively. is the Frobenius norm. The optimized structural alignment parameters are input into the message passing neural network to update the node information. The update process of the message passing neural network includes two parts: message generation and state update. The formula is as follows:

[0106] ;

[0107] in are messages aggregated from neighboring nodes, is the message generation function, is the state update function, is the edge feature. After multiple rounds of iterations, an updated node embedding matrix is ​​generated. Based on the updated node embedding, the battery pack intelligent pairing structure analysis is performed. By analyzing the distribution and relationship of node embeddings, the battery pack intelligent pairing structure matrix is ​​generated. ,in Indicates and The final pairing score of each battery cell.

[0108] In a specific embodiment, the process of executing step S400 may specifically include the following steps:

[0109] The battery performance parameters and topological relationships in the battery pack intelligent pairing structure are converted into particle codes, where the position vector of each particle contains the pairing scheme of the battery cell, and the velocity vector represents the adjustment direction of the pairing scheme, thereby obtaining the initial population position matrix and velocity matrix;

[0110] The initial population position matrix is ​​divided into sub-populations, each sub-population corresponds to a group of batteries to be paired, and an independent search space and iteration parameters are set for each sub-population to obtain multiple parallel optimization sub-populations;

[0111] A multi-objective fitness function is constructed based on the parallel optimization subpopulation. The fitness function comprehensively considers the battery performance parameter consistency weight coefficient of 0.4, the charge balance weight coefficient of 0.3, and the battery pack service life weight coefficient of 0.3 to obtain a fitness calculation model.

[0112] The fitness calculation model is input into the sub-population search module, and the position update formula and the speed update formula are used for iterative calculation to obtain the local optimal solution of the sub-population, and the population migration operation is performed on the local optimal solution of the sub-population to obtain the population migration result;

[0113] The population migration results are updated with the global optimal solution to obtain the global optimal pairing solution, which is then input into the topology optimization module. The topology structure is optimized by minimizing the optimization objective function of the total internal resistance and voltage loss of the battery pack to obtain the optimal connection relationship. Based on the optimal connection relationship, the pairing parameters of the battery pack are integrated to obtain the optimized pairing solution for the battery pack.

[0114] Specifically, the battery performance parameters and topological relationships in the battery pack intelligent pairing structure are converted into particle encoding, which provides a basis for particle swarm optimization. Each particle is represented by a position vector and a velocity vector, where the position vector Indicates the current pairing scheme. It is The paired target index of each battery cell; the velocity vector Indicates the direction and magnitude of the pairing scheme adjustment, It is The position matrix of the initial population is obtained by converting the particle encoding of the battery pack intelligent pairing structure. and the velocity matrix ,in is the number of particles. In order to speed up the search and improve the optimization efficiency, the initial population position matrix Subpopulations are divided. Each subpopulation corresponds to a group of battery cells to be paired, grouped according to topological constraints and performance parameters, such as partitioning by performance similarity or physical connection relationship. Each subpopulation is assigned an independent search space and a set of iteration parameters, including inertia weight , Individual learning factor and group learning factor The subpopulation division allows the optimization to be performed simultaneously in different subspaces, forming multiple parallel optimization subpopulations. ,in is the number of sub-populations. Based on the parallel optimization of sub-populations, a multi-objective fitness function is constructed to evaluate the quality of each particle. The fitness function comprehensively considers the three objectives of battery performance parameter consistency, charge balance and battery pack service life, and assigns weight coefficients , and The specific formula is as follows:

[0115] Consistency SOC_Balance Lifetime ;

[0116] Among them, Consistency The consistency of battery performance parameters can be expressed by the following formula:

[0117] ;

[0118] in, Is with particles A set of paired battery performance parameters; SOC_Balance(x) measures the balance of the state of charge (SOC) and is defined as:

[0119] ;

[0120] in is the standard deviation of SOC, SOC is the mean value of SOC; The expected life of the battery pack is measured and estimated based on the current load uniformity and the distribution of charge and discharge rates. After the fitness calculation model is input into the search module of the subpopulation, the position and speed are iteratively updated through the core formula of particle swarm optimization. The position update formula is:

[0121] ;

[0122] The speed update formula is:

[0123] ;

[0124] in is the number of iterations, It is a particle The best historical position, is the global optimal position of the current population, and are two independent random numbers. After multiple rounds of iterations, each subpopulation gradually converges to a local optimal solution. In order to avoid the limitations of the local optimal solution, a population migration operation is performed on the local optimal solution of each subpopulation, and the interaction information is updated through migration to form a new population distribution. The migrated population is reintegrated, and the population results are further optimized through a global search strategy to obtain a global optimal pairing solution. The global optimal pairing solution is input into the topology optimization module, and the optimization objective function is constructed by minimizing the total internal resistance and voltage loss of the battery pack. For example, the total internal resistance optimization objective is expressed as:

[0125] ;

[0126] in It is a battery cell and The connection internal resistance, is the edge set of the pairing scheme. By optimizing the internal resistance and voltage loss, the optimal connection relationship is generated, and the pairing parameters of the battery pack are integrated based on the relationship to finally obtain the optimized pairing scheme.

[0127] In a specific embodiment, the process of executing step S500 may specifically include the following steps:

[0128] Model the energy storage characteristics of the battery pack optimization pairing scheme, establish an energy storage model that includes the dynamic characteristics of battery pack current, voltage, and temperature, and quantitatively analyze the battery pack energy storage potential to obtain energy storage boundary constraints;

[0129] A hierarchical model predictive controller is constructed based on energy storage boundary constraints. The hierarchical model predictive controller includes an upper power controller and a lower current controller. The upper power controller adopts a model prediction algorithm, and the lower current controller adopts a feedback control algorithm.

[0130] The real-time status information of the battery pack is input into the upper power controller, which includes three power prediction units and one allocation optimization unit. The power allocation strategy is obtained through rolling time domain optimization calculation.

[0131] Based on the power allocation strategy, the upper-level control optimization target is constructed. The upper-level control optimization target includes power balance coefficient, voltage fluctuation coefficient and energy storage efficiency coefficient, and the upper-level power allocation parameters are obtained by solving.

[0132] The upper power allocation parameters are input into the lower current controller, which includes a current reference generation unit and a feedback correction unit, and a charge and discharge regulation strategy is obtained through current closed-loop control;

[0133] The lower-level control optimization objectives are constructed based on the charge and discharge regulation strategy. The lower-level control optimization objectives include current balance, charge state consistency and minimization of internal resistance loss. The lower-level charge and discharge current parameters are obtained through quadratic programming.

[0134] Specifically, the energy storage characteristics of the battery pack optimization pairing scheme are modeled, and an energy storage model including the dynamic characteristics of battery pack current, voltage, and temperature is established. Each single cell is composed of The state of the state variable Indicates that , and They are The current, voltage and temperature of each battery. In this model, the dynamic behavior of the battery is described by the following equation:

[0135] ;

[0136] in, is the open circuit voltage, is the internal resistance, is the battery capacity, is the time step. By modeling the characteristics of all battery cells, an energy storage model describing the behavior of the entire battery pack is obtained. Based on this model, the energy storage potential of the battery pack is quantitatively analyzed and its energy storage boundary constraints are calculated. For example, the energy storage boundary is defined by the following constraints:

[0137] ;

[0138] in, and is the minimum and maximum output power of the battery pack. This condition ensures that the battery pack operates within a safe power range. Based on the energy storage boundary constraints, a hierarchical model predictive controller is constructed. The controller includes an upper power controller and a lower current controller. The upper layer is responsible for global power allocation, and the lower layer is responsible for specific current regulation. In the upper layer control, the model prediction algorithm calculates the power allocation strategy through rolling time domain optimization. Suppose the total power demand of the battery pack is , the upper control goal is to reasonably allocate this demand to each single battery. The upper power controller contains three power prediction units, each of which predicts short-term, medium-term and long-term power demand respectively, and generates an optimized power allocation strategy through the allocation optimization unit ,satisfy:

[0139] ;

[0140] Based on this strategy, the optimization objectives of the upper-level control are constructed, including the power balancing coefficient , Voltage fluctuation coefficient and energy storage efficiency coefficient The objective function is expressed as:

[0141] ;

[0142] in, is the weight coefficient, which is defined as:

[0143] ;

[0144] in, and are the average values ​​of power and voltage, respectively. It is The output of the upper control is the power allocation parameter , which is input to the lower-level current controller to generate a charge and discharge regulation strategy. The lower-level control adopts a feedback control algorithm, including a current reference generation unit and a feedback correction unit. The current reference generation unit calculates the target current of each battery by the following formula:

[0145] ;

[0146] The feedback correction unit monitors the actual current of the battery in real time , correction deviation:

[0147] ;

[0148] Based on this regulation strategy, the optimization objectives of the lower-level control are constructed, including current balance , Charge state consistency and minimize internal resistance losses The objective function is expressed as:

[0149] min

[0150] in, is the weight coefficient, which is defined as:

[0151] ;

[0152] in, and are the standard deviations of current and SOC, and is its average value. Solving the above objective function through quadratic programming, we get the lower layer charge and discharge current parameter and feed it back to the battery pack control system to complete closed-loop control.

[0153] In a specific embodiment, the process of executing step S600 may specifically include the following steps:

[0154] Perform consistency check on the upper power allocation parameters and the lower charge and discharge current parameters to obtain the mapping relationship matrix between power and current;

[0155] Based on the mapping relationship matrix between power and current, double closed-loop control and performance evaluation are performed to obtain a performance monitoring matrix;

[0156] Real-time optimization control is performed based on the performance monitoring matrix to obtain optimized control parameters, which are then input into a multi-stage controller. The multi-stage controller includes a voltage stabilization control stage, a power balancing control stage, and a current balancing control stage. The optimized control parameters for battery pack pairing are obtained through hierarchical progressive control.

[0157] Specifically, the upper power allocation parameters and the lower charge and discharge current parameters are checked for consistency, and the two are associated to form a power and current mapping matrix. The upper power distribution parameter is , the lower layer charge and discharge current parameters are The mapping relationship between power and current is defined by the following formula:

[0158] ;

[0159] in It is The distributed power of each battery, is its operating voltage, is its charge and discharge current. By constructing the mapping relationship of all battery cells, a mapping relationship matrix between power and current is formed. , whose elements Indicates the power and current in the The time step is The dynamic relationship of the cells. The construction process ensures consistency verification of upper and lower control parameters, detecting possible deviations through the following conditions:

[0160] ;

[0161] in, Is the power deviation. If its absolute value exceeds the set threshold, it is considered inconsistent and needs to be adjusted. Based on the mapping relationship matrix between power and current , and conduct dual closed-loop control and performance evaluation. Dual closed-loop control includes power closed-loop and current closed-loop. The goal of the power closed-loop is to ensure that the total power output of the battery pack meets the demand. The control equation is:

[0162] ;

[0163] in is the total power demand of the system; the goal of the current closed loop is to make the actual current of the battery cell With reference current Keeping the same, the adjustment formula is:

[0164] ;

[0165] The power closed loop and current closed loop work together to ensure global consistency by adjusting the power and current parameters. Based on the dual closed loop control, the performance of the battery pack is evaluated in real time to generate a performance monitoring matrix . Elements of the performance monitoring matrix It is calculated through the following performance indicators:

[0166] Power Deviation: A measure of the balance of power distribution:

[0167] ;

[0168] Voltage fluctuation: reflects the unevenness of voltage within the battery pack:

[0169] ;

[0170] Current balance: measures the balance of current distribution between battery cells:

[0171] ;

[0172] in, is the standard deviation of the current, is the average value of the current. Performance Monitoring Matrix Summarize multiple performance indicators to provide data support for real-time optimization control. Based on the performance monitoring matrix , perform real-time optimization control to generate optimized control parameters. The optimization control objectives include minimizing power deviation, voltage fluctuation and current imbalance, and its comprehensive objective function is:

[0173] ;

[0174] in, , and is the weight coefficient, which reflects the importance of different performance indicators. By optimizing the objective function, the optimal control parameters are obtained. ,in Indicates The optimized adjustment value of each battery. The optimized control parameters are input into the multi-stage controller to complete the hierarchical progressive control. The multi-stage controller includes a voltage stability control stage, a power balance control stage, and a current balance control stage. The voltage stability control stage ensures the operating stability of the battery pack by adjusting the battery voltage distribution. Its control equation is:

[0175] ;

[0176] The power balance control stage optimizes the control parameters Dynamically adjust the power allocation strategy to make the power output reach the global optimum. The control formula is:

[0177] ;

[0178] The current balancing control stage reduces the load difference between battery cells by correcting the current distribution, and finally adjusts the current reference value to:

[0179] ;

[0180] Under the hierarchical progressive control of the multi-level controller, each layer of control works together to finally generate the optimal parameters for battery pack pairing control. , used to guide the actual operation of the battery pack.

[0181] In this embodiment, a closed-loop control operation is performed based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain the battery pack pairing control optimization parameters, which also includes: inputting the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters into a self-learning feedback controller, and constructing a feedback matrix including power balance terms and current compensation terms to obtain an initial feedback control sequence; performing error analysis on the initial feedback control sequence, and obtaining a control error matrix by establishing a multi-dimensional error vector including voltage deviation, current deviation, and temperature deviation; inputting the control error matrix into a deep reinforcement learning network, the deep reinforcement learning network includes a dual-channel recurrent neural network structure, each channel includes three layers of LSTM units, and an intelligent compensation strategy is obtained; adjusting the control parameters based on the intelligent compensation strategy, and dynamically optimizing the control gain through the Actor-Critic algorithm to obtain the optimal control gain matrix; perform stability analysis on the optimal control gain matrix, verify the asymptotic stability of the system through the Lyapunov function, and obtain the stability boundary conditions; input the stability boundary conditions into the adaptive controller, generate the compensation control quantity through the model reference adaptive control algorithm, and obtain the adaptive control parameters; construct a multi-time scale control structure based on the adaptive control parameters, and obtain the time-sharing control strategy by setting the fast response layer and the slow optimization layer respectively; input the time-sharing control strategy into the intelligent coordinator, and obtain the coordinated control instructions by establishing a switching mechanism based on fuzzy rules; conduct online evaluation of the coordinated control instructions, and obtain the performance evaluation results by constructing a comprehensive evaluation function including dynamic performance indicators and steady-state performance indicators; perform online iterative optimization of the control parameters based on the performance evaluation results, update the control law parameters through the gradient projection algorithm, and obtain the battery pack pairing control optimization parameters.

[0182] See also Figure 2 , Figure 2 A schematic block diagram of the structure of the battery pack intelligent pairing and integrated optimization system 200 provided in an embodiment of the present application, such as Figure 2 As shown, the battery pack intelligent pairing and integrated optimization system 200 includes:

[0183] The acquisition module 210 is used to acquire parameters and extract orthogonal features of battery cells in the battery pack to obtain an orthogonal parameter characteristic matrix of the battery pack;

[0184] A deep learning module 220 is used to input the battery pack orthogonal parameter characteristic matrix into a convolutional neural network optimized by a genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model;

[0185] A construction module 230 is used to construct a battery attribute similarity graph and a battery pack topology graph based on the battery pack pairing prediction model, and obtain a battery pack intelligent pairing structure through a double-graph structure alignment operation;

[0186] A calculation module 240 is used to perform parallel multi-swarm particle swarm optimization calculations on the battery pack intelligent pairing structure to obtain an optimized battery pack pairing solution;

[0187] An analysis module 250 is used to input the battery pack optimization pairing scheme into the hierarchical model predictive controller to perform energy storage boundary analysis to obtain upper power allocation parameters and lower charge and discharge current parameters;

[0188] The closed-loop control module 260 is used to perform closed-loop control operations based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain the battery pack pairing control optimization parameters.

[0189] Through the collaborative cooperation of the above-mentioned components, by introducing orthogonal feature extraction and convolutional neural network optimized by genetic algorithm, deep mining of battery cell performance characteristics is achieved, overcoming the problem that traditional methods are difficult to accurately describe the correlation of battery performance; based on the intelligent pairing mechanism constructed by dual-graph structure neural network, an alignment optimization model of battery attribute similarity graph and topology graph is established, which solves the dynamic optimization problem in the battery pack pairing process; the parallel multi-population particle swarm algorithm is used to optimize the pairing scheme, which significantly improves the pairing optimization efficiency, and an integrated optimization strategy based on hierarchical model predictive control is designed. Through energy storage boundary analysis and double-layer control structure, the stable operation of the battery pack under dynamic conditions is guaranteed; a closed-loop control optimization mechanism is established to realize real-time monitoring and dynamic adjustment of battery pack performance parameters.

[0190] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of a battery pack intelligent pairing and integrated optimization device 300 provided in an embodiment of the present application. The battery pack intelligent pairing and integrated optimization device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0191] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned battery pack intelligent pairing and integration optimization methods.

[0192] The processor 301 is used to provide computing and control capabilities to support the operation of the entire battery pack intelligent pairing and integrated optimization device 300.

[0193] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned battery pack intelligent pairing and integrated optimization methods.

[0194] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the battery pack intelligent pairing and integrated optimization device 300 involved in the scheme of the present application. The specific battery pack intelligent pairing and integrated optimization device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0195] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0196] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the battery pack intelligent pairing and integrated optimization device 300 described above can refer to the corresponding process of the aforementioned battery pack intelligent pairing and integrated optimization method, and will not be repeated here.

[0197] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the battery pack intelligent pairing and integrated optimization method provided in the embodiment of the present application.

[0198] The computer-readable storage medium may be an internal storage unit of the battery pack intelligent pairing and integrated optimization device 300 of the aforementioned embodiment, such as a hard disk or memory of the battery pack intelligent pairing and integrated optimization device 300. The computer-readable storage medium may also be an external storage device of the battery pack intelligent pairing and integrated optimization device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped with the battery pack intelligent pairing and integrated optimization device 300.

[0199] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0200] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or 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, including several instructions to enable a computer device (which can be a personal computer, server, or 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: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0201] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A battery pack intelligent pairing and integrated optimization method, characterized in that: include: Perform parameter collection and orthogonal feature extraction on battery cells in the battery pack to obtain the orthogonal parameter characteristic matrix of the battery pack; Inputting the battery pack orthogonal parameter characteristic matrix into a convolutional neural network optimized by a genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model; Based on the battery pack pairing prediction model, a battery attribute similarity graph and a battery pack topology graph are constructed, and a battery pack intelligent pairing structure is obtained by a double-graph structure alignment operation; Performing parallel multi-swarm particle swarm optimization calculations on the battery pack intelligent pairing structure to obtain an optimized battery pack pairing solution; The battery pack optimization pairing scheme is input into a hierarchical model predictive controller for energy storage boundary analysis to obtain upper power allocation parameters and lower charge and discharge current parameters; specifically including: energy storage characteristic modeling of the battery pack optimization pairing scheme, establishing an energy storage model including the dynamic characteristics of battery pack current, voltage, and temperature, and quantitatively analyzing the battery pack energy storage potential to obtain energy storage boundary constraints; constructing a hierarchical model predictive controller based on the energy storage boundary constraints, the hierarchical model predictive controller includes an upper power controller and a lower current controller, wherein the upper power controller adopts a model prediction algorithm and the lower current controller adopts a feedback control algorithm; inputting the real-time status information of the battery pack into the upper power controller, the upper power controller It includes three power prediction units and one allocation optimization unit, and obtains the power allocation strategy through rolling time domain optimization calculation; constructs the upper-level control optimization target based on the power allocation strategy, and the upper-level control optimization target includes the power balance coefficient, the voltage fluctuation coefficient and the energy storage efficiency coefficient, and obtains the upper-level power allocation parameters by solving; inputs the upper-level power allocation parameters into the lower-level current controller, and the lower-level current controller includes a current reference generation unit and a feedback correction unit, and obtains the charge and discharge regulation strategy through current closed-loop control; constructs the lower-level control optimization target based on the charge and discharge regulation strategy, and the lower-level control optimization target includes the current balance, the consistency of the state of charge and the minimization of the internal resistance loss, and obtains the lower-level charge and discharge current parameters by solving the quadratic programming; A closed-loop control operation is performed based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain battery pack pairing control optimization parameters.

2. The battery pack intelligent pairing and integrated optimization method according to claim 1, characterized in that: The parameter collection and orthogonal feature extraction of the battery cells in the battery pack to obtain the orthogonal parameter characteristic matrix of the battery pack includes: Performing real-time multi-dimensional parameter acquisition on battery cells in a battery pack to obtain a battery cell raw data matrix, and performing noise elimination and outlier removal operations on the battery cell raw data matrix to obtain effective characteristic parameters of the battery cell; Performing material property analysis and parameter correlation calculation on the effective characteristic parameters of the battery cell to obtain a battery pack performance characteristic matrix, and performing feature dimension reduction operation on the battery pack performance characteristic matrix to obtain a battery pack target orthogonal parameter; Performing state of charge consistency calculation, capacity matching degree calculation, internal resistance balance degree calculation and temperature distribution characteristic calculation on the target orthogonal parameters of the battery pack to obtain a standardized feature vector of the battery pack; Dynamically update the standardized feature vector of the battery pack to obtain time series feature data of the battery pack, and build a structure-performance mapping model based on the time series feature data of the battery pack, and obtain the orthogonal parameter characteristic matrix of the battery pack through orthogonal transformation operation.

3. The battery pack intelligent pairing and integrated optimization method according to claim 2, characterized in that: The method of inputting the battery pack orthogonal parameter characteristic matrix into a convolutional neural network optimized by a genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model includes: Dividing the battery pack orthogonal parameter characteristic matrix into a training set and a validation set according to time sequence, and performing batch normalization processing on the training set to obtain standardized training data; Genetic algorithm encoding is performed on the convolutional neural network structure to obtain the network structure gene sequence, where the genetic algorithm encoding includes the convolution kernel size, step size, pooling parameters, and the number of neurons in the fully connected layer; A multi-layer convolutional neural network is constructed based on the network structure gene sequence, wherein the multi-layer convolutional neural network includes three convolutional modules, each of which includes two convolutional layers and one maximum pooling layer, and the convolutional layer uses a ReLU activation function to obtain a feature extraction network; Adding a batch normalization layer and a dropout layer to each convolutional module of the feature extraction network to obtain a regularized feature extraction network, and inputting the standardized training data into the regularized feature extraction network to obtain deep features of battery performance through sequential calculations of three convolutional modules; Performing full connection layer mapping on the deep characteristics of the battery performance, wherein the full connection layer includes two hidden layers, the number of neurons in the first hidden layer is 512, and the number of neurons in the second hidden layer is 256, to obtain a feature fusion result; Based on the feature fusion results, a pairing prediction output layer is constructed, a softmax activation function is used for multi-classification prediction, and the network parameters are optimized by a back propagation algorithm to obtain an initial pairing prediction model; The validation set is input into the initial pairing prediction model for cross validation to obtain a validation result, and based on the validation result, the fitness function of the genetic algorithm is updated and iteratively optimized to obtain a battery pack pairing prediction model.

4. The battery pack intelligent pairing and integrated optimization method according to claim 3, characterized in that: The method of constructing a battery attribute similarity graph and a battery pack topology graph based on the battery pack pairing prediction model, and obtaining a battery pack intelligent pairing structure by a double-graph structure alignment operation, includes: Performing node mapping on the pairing probability distribution and the deep characteristics of battery performance output by the battery pack pairing prediction model to obtain a battery cell feature node set, wherein each node contains a pairing probability feature and a deep performance feature; Calculating similarity weights between nodes based on the battery cell feature node set to obtain a battery attribute similarity graph, and extracting a topological structure of an actual physical connection state of the battery pack to obtain a battery pack topology graph; Input the battery attribute similarity graph into a dual-graph structure neural network for processing, wherein the dual-graph structure neural network includes a first-layer graph attention network with an output dimension of 64 and a second-layer graph attention network with an output dimension of 32, to obtain node representation features; Inputting the node representation features into the graph convolution layer for feature enhancement to obtain enhanced graph structure features, and performing structure-aware loss calculation on the enhanced graph structure features to obtain structure alignment parameters; The structural alignment parameters are input into a message passing neural network to update node information to obtain updated node embeddings, and a battery pack intelligent pairing structure analysis is performed based on the updated node embeddings to obtain a battery pack intelligent pairing structure matrix.

5. The battery pack intelligent pairing and integrated optimization method according to claim 4, characterized in that: The performing of parallel multi-swarm particle swarm optimization calculations on the battery pack intelligent pairing structure to obtain a battery pack optimization pairing solution includes: The battery performance parameters and topological relationships in the battery pack intelligent pairing structure are converted into particle codes, wherein the position vector of each particle contains the pairing scheme of the battery cell, and the velocity vector represents the adjustment direction of the pairing scheme, thereby obtaining an initial population position matrix and a velocity matrix; Dividing the initial population position matrix into sub-populations, each sub-population corresponds to a group of batteries to be paired, and setting an independent search space and iteration parameters for each sub-population to obtain multiple parallel optimization sub-populations; A multi-objective fitness function is constructed based on the parallel optimization subpopulation, wherein the fitness function comprehensively considers the battery performance parameter consistency weight coefficient of 0.4, the charge balance weight coefficient of 0.3, and the battery pack service life weight coefficient of 0.3 to obtain a fitness calculation model; Inputting the fitness calculation model into the subpopulation search module, performing iterative calculations through the position update formula and the speed update formula to obtain a local optimal solution of the subpopulation, and performing a population migration operation on the local optimal solution of the subpopulation to obtain a population migration result; The population migration result is updated with a global optimal solution to obtain a global optimal pairing solution, and the global optimal pairing solution is input into a topology optimization module. The topology structure is optimized by minimizing the optimization objective function of the total internal resistance and voltage loss of the battery pack to obtain an optimal connection relationship. Based on the optimal connection relationship, the pairing parameters of the battery pack are integrated to obtain an optimized pairing solution for the battery pack.

6. The battery pack intelligent pairing and integrated optimization method according to claim 1, characterized in that: The closed-loop control operation is performed based on the upper-layer power allocation parameter and the lower-layer charge and discharge current parameter to obtain the battery pack pairing control optimization parameter, including: Performing consistency check on the upper layer power allocation parameters and the lower layer charge and discharge current parameters to obtain a mapping relationship matrix between power and current; Based on the mapping relationship matrix between power and current, double closed-loop control and performance evaluation are performed to obtain a performance monitoring matrix; Based on the performance monitoring matrix, real-time optimization control is performed to obtain optimized control parameters, and the optimized control parameters are input into a multi-stage controller. The multi-stage controller includes a voltage stabilization control stage, a power balancing control stage and a current balancing control stage. The battery pack pairing control optimization parameters are obtained through hierarchical progressive control.

7. A battery pack intelligent pairing and integrated optimization system, characterized in that: The method for performing intelligent pairing and integrated optimization of a battery pack according to any one of claims 1 to 6 comprises: The acquisition module is used to collect parameters and extract orthogonal features of battery cells in the battery pack to obtain an orthogonal parameter characteristic matrix of the battery pack; A deep learning module, used for inputting the battery pack orthogonal parameter characteristic matrix into a convolutional neural network optimized by a genetic algorithm to perform feature deep learning to obtain a battery pack pairing prediction model; A construction module, used to construct a battery attribute similarity graph and a battery pack topology graph based on the battery pack pairing prediction model, and obtain a battery pack intelligent pairing structure through a double-graph structure alignment operation; A calculation module, used for performing parallel multi-swarm particle swarm optimization calculation on the battery pack intelligent pairing structure to obtain an optimized battery pack pairing solution; An analysis module is used to input the battery pack optimization pairing scheme into a hierarchical model predictive controller to perform energy storage boundary analysis to obtain upper power allocation parameters and lower charge and discharge current parameters; The closed-loop control module is used to perform closed-loop control operations based on the upper-layer power allocation parameters and the lower-layer charge and discharge current parameters to obtain battery pack pairing control optimization parameters.

8. A battery pack intelligent pairing and integrated optimization device, characterized in that: The battery pack intelligent pairing and integrated optimization device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the battery pack intelligent pairing and integrated optimization device to execute the battery pack intelligent pairing and integrated optimization method as described in any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the battery pack intelligent pairing and integrated optimization method as described in any one of claims 1-6 is implemented.

Images