A lithium battery energy storage adaptive management system

Abstract

The present invention belongs to the technical field of lithium battery energy storage management, and discloses a lithium battery energy storage adaptive management system; the system comprises: a real-time monitoring module, which is used for real-time monitoring of comprehensive battery data of each battery cell; pre-processing the comprehensive battery data to obtain comprehensive characteristic data; a central comprehensive module, which is used for taking the comprehensive characteristic data of each battery cell as input, and determining the balancing demand level corresponding to each battery cell through an adaptive control strategy based on fuzzy logic; a balancing management module, which divides all battery cells into M balancing groups based on the balancing demand level; for each balancing group, adopting an adaptive balancing strategy algorithm to balance the quantity of electricity; and connecting the modules through wired and / or wireless means, which significantly improves the limitations of traditional lithium battery management and greatly improves the performance and service life of the energy storage system.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery energy storage management, and more specifically, to a lithium battery energy storage adaptive management system. Background Art

[0002] The patent application with the publication number CN112721906A discloses a hybrid energy storage control system based on lithium battery power estimation and its energy management method, wherein the energy management method is: collecting the open circuit voltage, current and SOC of the lithium battery pack; according to the open circuit voltage and current of the lithium battery pack, performing parameter identification on the lithium battery pack, and estimating the maximum charge and discharge power of the current lithium battery pack; using the maximum charge and discharge power of the current lithium battery pack as the upper and lower limits of the adaptive control, and then using the adaptive droop control strategy to limit the allocated power of the lithium battery pack; according to the total required power of the hybrid energy storage system and the allocated power of the lithium battery pack, the power of the supercapacitor group is allocated; generating control signals of the DC / DC modules of the lithium battery pack and the supercapacitor group, using their respective output powers as their respective allocated powers, and comprehensively providing the total required power for the electric vehicle, which can protect the lithium battery pack from the influence of the peak charge and discharge power and improve the service life of the lithium battery pack.

[0003] However, the existing system's monitoring and data processing capabilities for battery cells are still limited. Traditional monitoring methods often only focus on some key parameters and cannot fully perceive the battery's operating status. Secondly, the battery balancing management of the existing system lacks flexibility and pertinence. Most of them use a unified balancing control method and cannot perform personalized balancing demand judgment and management based on the characteristics of different battery cells. This makes it difficult to maximize the personalized needs of each battery cell, affecting the overall balancing effect. Then there are certain limitations in balancing efficiency, which makes it difficult to maximize the overall balancing efficiency while ensuring the balancing effect, limiting the performance and service life of the energy storage system.

[0004] In view of this, the present invention proposes a lithium battery energy storage adaptive management system to solve the above problems. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solutions: A lithium battery energy storage adaptive management system, comprising: a real-time monitoring module, used to monitor the comprehensive battery data of each battery unit in real time; pre-processing the comprehensive battery data to obtain comprehensive characteristic data;

[0006] A central comprehensive module is used to take the comprehensive characteristic data of each battery cell as input and determine the balancing demand level corresponding to each battery cell through an adaptive control strategy based on fuzzy logic;

[0007] The balancing management module divides all battery cells into M balancing groups based on the balancing demand level; for each balancing group, an adaptive balancing strategy algorithm is used to balance the power; and each module is connected by wired and / or wireless means.

[0008] Furthermore, the comprehensive battery data includes voltage data, current data, temperature data, internal resistance data, capacity data and balancing data;

[0009] Voltage data includes battery charging cut-off voltage, battery discharging cut-off voltage and battery average voltage; current data includes maximum charging current and maximum discharging current; temperature data includes maximum operating temperature, minimum operating temperature and average operating temperature; internal resistance data includes real-time internal resistance value and internal resistance change rate; capacity data includes rated capacity and capacity attenuation rate; balancing data includes balancing current, balancing time and balancing energy transfer.

[0010] Furthermore, the method of preprocessing the comprehensive battery data includes:

[0011] Construct a missing value estimation model based on graph neural network, use the missing value estimation model to obtain the missing values ​​in the comprehensive battery data, fill the missing values ​​to obtain complete feature data; standardize / normalize the complete feature data to obtain comprehensive feature data.

[0012] Furthermore, the missing value estimation model is constructed in the following manner:

[0013] Collect the comprehensive battery data of the battery unit within a fixed historical period and record it as the historical comprehensive battery data; model the historical comprehensive battery data as a heterogeneous graph ;in, is a node set, is an edge set, and the nodes in the node set represent different feature values; the edge set represents the association relationship between features;

[0014] The graph convolution-based graph neural network architecture is used as the basic architecture of the missing value estimation model, and the missing value estimation model is trained using heterogeneous graphs;

[0015] The method of using heterogeneous graphs to train missing value estimation models includes:

[0016] For heterogeneous graphs For each node in , initialize a feature vector, which is a vector composed of original values;

[0017] For each node, collect the feature vector information of its neighboring nodes to obtain the updated feature vector; perform nonlinear transformation on the feature vector information, use the updated feature vector of the current layer node as the input of the next layer, and in the last layer, transform the updated feature vector of each node to obtain the estimated value of the corresponding missing value;

[0018] Define a loss function, calculate the value of the loss function for known battery characteristic values ​​and estimated values, calculate the gradient of the loss function with respect to the parameters of the missing value estimation model through the back propagation algorithm, use the optimization algorithm to update the parameters of the missing value estimation model according to the gradient, and complete the training of the missing value estimation model by iteratively minimizing the value of the loss function.

[0019] Furthermore, the collection formula of the feature vector information is:

[0020] ;

[0021] in, Representation Node In the graph neural network architecture The feature vector of the layer, Indicates that in the graph neural network architecture The weight matrix of the layer; Representation Node For Node The attention weight, Representation Node In the graph neural network architecture The feature vector of the layer, represents the activation function, Representation Node The set of neighbor nodes of

[0022] in Ways to obtain include:

[0023] Defining the attention function , then we get , the attention function Through multi-layer perceptron parameterization, we get ;in, For Node Features, For Node Features, For Node and nodes edge features.

[0024] Furthermore, the method of filling the missing values ​​includes:

[0025] A binary mask vector will be defined , for non-missing values, the corresponding position The value is 1. For missing values, the corresponding position The value is 0, all battery comprehensive data, binary mask vectors and estimated values ​​are stacked into a 3D tensor, and a batch filling operation is performed on the 3D tensor; the batch filling operation is to scale the estimated value output by the missing value estimation model with the original data element by element.

[0026] The formula for element-by-element scaling is:

[0027] ;

[0028] in, Pair node representing the output of the missing value estimation model The estimated value of the missing value, Represents the original node The characteristic value of Represents the corresponding node in the binary mask vector The value of .

[0029] Furthermore, the method for determining the equilibrium demand level includes:

[0030] Define the input fuzzy variable set , the input fuzzy variables in the input fuzzy variable set correspond to the comprehensive feature data, and the output fuzzy variables are defined , represents the equilibrium demand level, for each input fuzzy variable, determine its language value set; for the output fuzzy variable , determine its language value set;

[0031] Collect historical data and construct IF-THEN fuzzy rules based on historical data. The form of fuzzy rules is:

[0032] IF x1 is A1 AND x2 is A2 AND ... AND xn is An THEN Y is Bk;

[0033] Where x1 is the first input fuzzy variable, x2 is the second input fuzzy variable, A1 is the language value of the first input, A2 is the language value of the second input, xn is the nth input fuzzy variable, and Bk is the language value set; for each input fuzzy variable, calculate its membership degree to each language value;

[0034] Based on the degree of membership, the fuzzy reasoning method is used to reason each fuzzy rule and obtain the fuzzy reasoning result of each fuzzy rule; the fuzzy reasoning results of all fuzzy rules are synthesized to obtain the comprehensive output fuzzy set B', and B' is mapped to the output space y* through the defuzzification method. y* is the fuzzy value of the balancing demand level corresponding to the battery unit, that is, the balancing demand level.

[0035] Furthermore, the degree of membership is calculated in the following manner:

[0036] From the comprehensive feature data, determine N1 feature sets that are related to the equilibrium demand and influence each other. The data in the feature sets are the relevant features.

[0037] The first function formula of membership is:

[0038] ;

[0039] in, For the The mean of the relevant features, For the The mean of the relevant features, For the The standard deviation of the relevant features, For the The standard deviation of the relevant features, For the The relevant features and The membership degree of the relevant features;

[0040] For each feature set, the value range of the data within it is divided into feature intervals; for each feature interval, the membership is calculated using the first function formula or the triangular function splicing method, and for the membership formula in each interval, its parameters are calculated, and the membership functions of different feature intervals are smoothly spliced ​​at the endpoints of the feature intervals.

[0041] Furthermore, the division method of the balancing group includes:

[0042] The battery cells with the same balancing requirement level are initially divided into a preliminary balancing group, a battery group topology diagram is constructed, and the battery cells are modeled as an undirected graph based on the physical adjacent relationship. ,in, is a collection of battery cells, is a battery edge set, and the battery edges in the battery edge set correspond to physical adjacent relationships;

[0043] For each battery cell , its equilibrium demand level is known , defines the level difference threshold of the battery cell , based on the preliminary balanced group, only Battery Cell and Classify into the same equilibrium group;

[0044] Define the objective function and constraints for balancing group partitioning; define is a 0-1 variable. When the battery cell and In the same equilibrium group, the objective function ;

[0045] The constraint condition is: for each battery cell , a special group must exist So that: ;

[0046] The objective function and constraints are constructed as an integer linear programming model, and the integer linear programming model is solved by a solver to obtain the optimal value. , represents the final grouping scheme, according to Get the value of the same group of battery cells ; That is, the division of the balancing group is completed.

[0047] Furthermore, the method of performing power balancing includes:

[0048] For each battery cell in the balancing group, obtain the comprehensive vector consisting of its comprehensive characteristic data ,in, is the index of the data type in the comprehensive feature data; comprehensive vector It is composed of all types of data in the comprehensive feature data; define The imbalance index of each data type is the maximum value minus the minimum value of the corresponding data type, and the imbalance indexes of all data types are weighted and fused to obtain a comprehensive score. ;

[0049] Divide the balancing group into imbalance levels , for is less than t1, for is greater than or equal to t1 and less than t2, for Greater than or equal to t2; wherein t1 is the preset lower limit of the level threshold, and t2 is the preset upper limit of the level threshold;

[0050] The definition of balancing strategies includes passive balancing strategy, active balancing strategy and mixed balancing strategy; when the imbalance level of the balancing group When , the balancing group to which it belongs adopts a passive balancing strategy;

[0051] When the imbalance level of the balance group When , the balancing group to which it belongs adopts the active balancing strategy;

[0052] When the imbalance level of the balance group When , the balance group to which it belongs adopts a mixed balance strategy;

[0053] The definition of hybrid equilibrium strategy includes:

[0054] The optimization objectives of the hybrid balancing strategy are initially defined as the balancing time objective and the balancing loss objective; and the objective function is constructed based on the balancing time objective and the balancing loss objective. ;in, is the time weight coefficient, is the loss weight coefficient;

[0055] Establish the time loss model of passive equilibrium and active equilibrium, namely, the passive equilibrium time model and the active equilibrium time model;

[0056] The passive balancing time model is based on the imbalance level of the current balancing group and fits the time required to adopt the passive balancing strategy. The active balancing time model is based on the imbalance level of the current balancing group and estimates the time required to adopt the active balancing strategy. The passive balancing strategy is initially adopted. At each time step, the remaining passive balancing time and the remaining time required to switch to active balancing are calculated. If the weighted sum of the time loss when switching to active balancing is less than the weighted sum of the time loss when continuing passive balancing, the active balancing mode is switched to and a balancing current is determined. Otherwise, the passive balancing strategy is continued until the weighted sum of the time loss when switching to active balancing is less than the weighted sum of the time loss when continuing passive balancing.

[0057] The technical effects and advantages of the lithium battery energy storage adaptive management system of the present invention are as follows:

[0058] The present invention can realize comprehensive perception and intelligent preprocessing of battery cells, monitor various key parameters of each battery cell in real time, and fully grasp the operating status of the battery. At the same time, graph neural networks are used to preprocess data to effectively make up for data missing and improve data integrity and effectiveness, which lays a solid foundation for subsequent adaptive management. Secondly, an adaptive control strategy based on fuzzy logic is adopted, which can flexibly determine the balancing demand level according to the characteristics of different battery cells. By adaptively dividing the battery cells into different balancing groups and taking targeted balancing measures for each group, the personalized needs of each battery cell can be better met and the overall balancing effect can be improved. Finally, the hybrid balancing strategy dynamically weighs the balancing time and balancing loss, while ensuring the balancing effect, it maximizes the overall balancing efficiency, significantly improves the limitations of traditional lithium battery management, and greatly improves the performance and service life of the energy storage system. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 A schematic diagram of a lithium battery energy storage adaptive management system of the present invention;

[0060] Figure 2 It is a schematic diagram of a lithium battery energy storage adaptive management method of the present invention. DETAILED DESCRIPTION

[0061] 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 only 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.

[0062] Example 1

[0063] See also Figure 1 As shown, the lithium battery energy storage adaptive management system described in this embodiment includes:

[0064] The real-time monitoring module is used to monitor the comprehensive battery data of each battery unit in real time; the comprehensive battery data is pre-processed to obtain comprehensive characteristic data;

[0065] A central comprehensive module is used to take the comprehensive characteristic data of each battery cell as input and determine the corresponding balancing demand level of each battery cell through an adaptive control strategy based on fuzzy logic;

[0066] The balancing management module divides all battery cells into M balancing groups based on the balancing demand level; for each balancing group, an adaptive balancing strategy algorithm is used to balance the power; each module is connected through wired and / or wireless means to achieve data transmission between modules.

[0067] Furthermore, the comprehensive battery data includes voltage data, current data, temperature data, internal resistance data, capacity data and balancing data;

[0068] Voltage data includes battery charging cut-off voltage, battery discharging cut-off voltage and battery average voltage; current data includes maximum charging current and maximum discharging current; temperature data includes maximum operating temperature, minimum operating temperature and average operating temperature; internal resistance data includes real-time internal resistance value and internal resistance change rate; capacity data includes rated capacity and capacity attenuation rate; balancing data includes balancing current, balancing time and balancing energy transfer.

[0069] It should be noted that the battery charging cut-off voltage and the battery discharging cut-off voltage are both obtained through the manufacturer's manual. The voltage of the battery over a period of time (such as a charge and discharge cycle) is averaged to obtain the average voltage value. During the charging process, the battery charging current is monitored in real time and the maximum value, i.e., the maximum charging current, is recorded.

[0070] During the discharge process, the discharge current of the battery is monitored in real time, and the maximum value, i.e., the maximum discharge current, is recorded. The temperature of the battery during operation is monitored by a temperature sensor, and the highest temperature value, i.e., the highest operating temperature, is recorded. The temperature of the battery during operation is monitored by a temperature sensor, and the lowest temperature value, i.e., the lowest operating temperature, is recorded. The temperature data of the battery over a period of working time is averaged to obtain the average operating temperature.

[0071] By measuring the terminal voltage and current of the battery in different working states (such as charging, discharging, etc.), the real-time internal resistance value is calculated according to the internal resistance model, and the change of the battery internal resistance value over a period of time is monitored to calculate the change rate of the internal resistance value, that is, the internal resistance change rate.

[0072] The rated capacity is the rated capacity value of the battery given by the battery manufacturer. The capacity attenuation rate is calculated by performing a cycle charge and discharge test on the battery and monitoring the capacity attenuation of the battery over a period of time.

[0073] During active balancing, when the battery pack is balanced through the balancing circuit, the value of the balancing current is monitored and recorded, and the time of each balancing process, i.e., the balancing time, is recorded; based on the balancing current and the balancing time, the energy transferred between the battery packs during the balancing process is calculated.

[0074] Furthermore, the method of preprocessing the comprehensive battery data includes:

[0075] Construct a missing value estimation model based on graph neural network, use the missing value estimation model to obtain the missing values ​​in the comprehensive battery data, fill the missing values ​​to obtain complete feature data; standardize / normalize the complete feature data to obtain comprehensive feature data; unify data of different dimensions to the same order of magnitude to facilitate subsequent processing.

[0076] The missing value estimation model is constructed in the following manner:

[0077] Collect the comprehensive battery data of the battery unit within a fixed historical period and record it as the historical comprehensive battery data; model the historical comprehensive battery data as a heterogeneous graph ;in, is a node set, It is an edge set, and the nodes in the node set represent different feature values, such as voltage data, current data, etc.; the edge set represents the association between features, which is built based on correlation and causality; for example, there is correlation between comprehensive battery data, such as there is a certain correlation between voltage, temperature and internal resistance, and the correlation coefficients between different feature data are calculated, such as the Pearson correlation coefficient, Spearman correlation coefficient, etc.; if the correlation coefficient of two feature data exceeds the preset threshold, an undirected edge is added between them in the heterogeneous graph to indicate that there is a correlation between them.

[0078] A graph convolution-based graph neural network architecture is used as the basic architecture of the missing value estimation model. The graph neural network architecture can directly process graph structured data; heterogeneous graphs are used to train the missing value estimation model.

[0079] The method of using heterogeneous graphs to train missing value estimation models includes:

[0080] For heterogeneous graphs For each node in , initialize a feature vector, which is a vector composed of original values;

[0081] For each node, the feature vector information of its neighboring nodes is collected to obtain the updated feature vector. The collection formula of the feature vector information is:

[0082] ;

[0083] in, Representation Node In the graph neural network architecture The feature vector of the layer, Indicates that in the graph neural network architecture The weight matrix of the layer is used to linearly combine the feature vectors of neighboring nodes; Representation Node For Node The attention weight is used to adjust the contribution of neighbor nodes in the aggregation. Representation Node In the graph neural network architecture The feature vector of the layer, Represents the activation function, which is used to introduce nonlinear transformation; Representation Node The set of neighbor nodes of

[0084] in Ways to obtain include:

[0085] Defining the attention function , then we get , the attention function Through multi-layer perceptron parameterization, we get ;in, For Node Features, For Node Features, For Node and nodes edge features.

[0086] Perform nonlinear transformation on the feature vector information and introduce nonlinear expression ability. The nonlinear transformation is implemented using the LeakyReLU activation function. The updated feature vector of the current layer node is used as the input of the next layer. In the last layer, the updated feature vector of each node is transformed to obtain the estimated value of the corresponding missing value. A loss function is defined, and the value of the loss function is calculated for the known battery feature values ​​and estimated values. The gradient of the loss function to the parameters of the missing value estimation model is calculated through the back propagation algorithm. The optimization algorithm (such as Adam) is used to update the parameters of the missing value estimation model according to the gradient. The loss function is minimized through iteration to complete the training of the missing value estimation model.

[0087] Furthermore, the method of filling the missing values ​​includes:

[0088] A binary mask vector mask will be defined. For non-missing values, the mask value at the corresponding position is 1, and for missing values, the mask value at the corresponding position is 0. The estimated value output by the missing value estimation model will be scaled element-by-element with the original data. All battery comprehensive data, binary mask vectors and estimated values ​​will be stacked into a 3D tensor, and a batch filling operation will be performed on the 3D tensor.

[0089] The formula for element-by-element scaling is:

[0090] ;

[0091] in, Pair node representing the output of the missing value estimation model The estimated value of the missing value, Represents the original node The characteristic value of Represents the corresponding node in the binary mask vector The value of is used to indicate whether the node feature is known.

[0092] Furthermore, the method for determining the equilibrium demand level includes:

[0093] Define the input fuzzy variable set , the input fuzzy variables in the input fuzzy variable set correspond to the comprehensive feature data, and the output fuzzy variables are defined , represents the equilibrium demand level, and for each input fuzzy variable, determines its language value set, such as {very low, lower, medium, higher, very high};

[0094] Output fuzzy variables , determine its language value set, such as {no demand, low demand, medium demand, high demand}.

[0095] Collect historical data, which is the comprehensive feature data within a fixed period of time in history. Construct IF-THEN fuzzy rules based on historical data. The form of fuzzy rules is:

[0096] IF x1 is A1 AND x2 is A2 AND ... AND xn is An THEN Y is Bk;

[0097] Among them, x1 is the first input fuzzy variable, x2 is the second input fuzzy variable, A1 is the language value of the first input, A2 is the language value of the second input, xn is the nth input fuzzy variable, and Bk is the language value set; expressed in IF-THEN form, it specifies the relationship between the language value of the input fuzzy variable and the language value of the output fuzzy variable.

[0098] For each comprehensive feature data, calculate its membership degree to each language value. The membership degree calculation method includes:

[0099] From the comprehensive feature data, N1 feature sets that are related to the balancing demand and influence each other are determined. The data in the feature set are relevant features. For example, features such as voltage, temperature, and internal resistance are correlated and can jointly influence the balancing demand.

[0100] The first function formula of membership is:

[0101] ;

[0102] in, For the The mean of the relevant features, For the The mean of the relevant features, For the The standard deviation of the relevant features, For the The standard deviation of the relevant features, For the The relevant features and The membership degree of the relevant features.

[0103] For each feature set, the value range of the data in it is divided into feature intervals. The nodes of the interval division can be determined based on historical data distribution or expert experience. For example, the voltage interval is divided into [2.5V, 3.2V], [3.2V, 3.6V], [3.6V, 4.2V], etc.

[0104] For each characteristic interval, the membership is calculated by using the first function formula or the triangular function splicing method. For the membership formula in each interval, its parameters are calculated, and the membership functions of different characteristic intervals are smoothly spliced ​​at the interval endpoints; multivariate expansion is performed according to actual needs to improve the modeling ability of the fuzzy system.

[0105] Based on the degree of membership, a fuzzy reasoning method (such as the Mamdani reasoning method) is used to reason each fuzzy rule and obtain the fuzzy reasoning result of each fuzzy rule; the fuzzy reasoning results of all fuzzy rules are synthesized to obtain a comprehensive output fuzzy set B', and B' is mapped to the output space y* through a defuzzification method (such as the centroid method, the average maximum method, etc.), where y* is the fuzzy value of the balancing demand level corresponding to the battery unit, that is, the balancing demand level.

[0106] By using fuzzy rule base and fuzzy reasoning, the balancing requirement level corresponding to the battery unit is calculated from the comprehensive characteristic data of the battery, providing support for the adaptive control strategy.

[0107] Furthermore, the division method of the balancing group includes:

[0108] The battery cells with the same balancing requirement level are initially divided into a preliminary balancing group, a battery group topology diagram is constructed, and the battery cells are modeled as an undirected graph based on the physical adjacent relationship. ,in, is a collection of battery cells, is a battery edge set, and the battery edges in the battery edge set correspond to physical adjacent relationships.

[0109] For each battery cell , its equilibrium demand level is known , defines the level difference threshold of the battery cell , based on the preliminary balanced group, only Battery Cell and are classified into the same equilibrium group.

[0110] Define the objective function and constraints for balancing group partitioning;

[0111] definition is a 0-1 variable. When the battery cell and In the same equilibrium group, the objective function , that is, minimize the number of adjacent unit pairs in the same group and reduce the complexity of the equalization line.

[0112] The constraint condition is: for each battery cell , a special group must exist So that: ;

[0113] The objective function and constraints are constructed as an integer linear programming (ILP) model, and the integer linear programming (ILP) model is solved using a solver (such as Gurobi, CPLEX, etc.) to obtain the optimal value. , represents the final grouping scheme, according to Get the value of the same group of battery cells ; That is, the division of the balancing group is completed.

[0114] The topology-based optimization process can prioritize adjacent cells with similar balancing requirements into the same group while taking into account the physical layout of the battery pack, thereby reducing the complexity of the balancing circuit and improving the efficiency of balancing control.

[0115] Furthermore, the method of performing power balancing includes:

[0116] For each battery cell in the balancing group, obtain the comprehensive vector consisting of its comprehensive characteristic data ,in, is the index of the data type in the comprehensive feature data; comprehensive vector It is composed of all types of data in the comprehensive feature data; define the The imbalance index of each data type is the maximum value minus the minimum value of the corresponding data type, and the imbalance indexes of all data types are weighted and fused to obtain a comprehensive score. ,The weight coefficients of imbalance indicators of different data types are determined through optimization.

[0117] Divide the balancing group into imbalance levels , (slight) is less than t1, (Medium) is greater than or equal to t1 and less than t2, (Severe) Greater than or equal to t2; wherein t1 is the preset lower limit of the level threshold, and t2 is the preset upper limit of the level threshold;

[0118] It should be noted that the construction method of the lower limit of the level threshold includes: constructing a multi-objective optimization model, which includes model objective functions of balancing time, balancing loss, and battery health, t1 and t2 as optimization variables, and using an evolutionary algorithm (such as NSGA-II) to globally search for the optimal t1 and t2 combination that can balance multiple objectives.

[0119] The definition of balancing strategies includes passive balancing strategy, active balancing strategy and mixed balancing strategy; when the imbalance level of the balancing group When , the balancing group to which it belongs adopts a passive balancing strategy;

[0120] When the imbalance level of the balance group When , the balancing group to which it belongs adopts the active balancing strategy;

[0121] When the imbalance level of the balance group , the balancing group it belongs to adopts a mixed balancing strategy.

[0122] The passive balancing strategy is suitable for slightly unbalanced states. The principle is to use the natural characteristics of battery self-discharge to achieve spontaneous transfer of electricity. Specifically, for battery cells with higher electricity in the balancing group, a small resistance is added to discharge, and the discharge amount is dynamically adjusted according to the voltage / capacity difference of the battery cells.

[0123] The active balancing strategy is applicable to severe unbalanced conditions. It uses an external balancing circuit to force the transfer of power between battery cells and determines the appropriate balancing circuit topology and balancing current according to the degree of unbalance.

[0124] The definition of hybrid equilibrium strategy includes:

[0125] The optimization objectives of the hybrid balancing strategy are initially defined as the balancing time objective and the balancing loss objective; and the objective function is constructed based on the balancing time objective and the balancing loss objective. ;in, is the time weight coefficient, is the loss weight coefficient, which is adjusted according to actual conditions.

[0126] Establish the time loss model of passive equilibrium and active equilibrium, namely, the passive equilibrium time model and the active equilibrium time model;

[0127] The passive balancing time model is based on the imbalance level of the current balancing group, and fits the time required to adopt the passive balancing strategy; the active balancing time model is based on the imbalance level of the current balancing group, and estimates the time required to adopt the active balancing strategy; both the passive balancing time model and the active balancing time model are obtained by fitting historical data, and the passive balancing strategy is initially adopted. At each time step, the remaining passive balancing time and the remaining time required to switch to active balancing are calculated; if the weighted sum of the time loss when switching to active balancing is less than the weighted sum of the time loss when continuing passive balancing, then switch to active balancing mode and determine a balancing current, otherwise, continue to adopt the passive balancing strategy until the weighted sum of the time loss when switching to active balancing is less than the weighted sum of the time loss when continuing passive balancing.

[0128] In addition, in the process of adopting the active balancing strategy, the balancing current is dynamically adjusted according to the real-time status to further minimize the weighted sum of time losses. The dynamic adjustment of the balancing current is solved by an intelligent optimization algorithm, such as particle swarm optimization or simulated annealing.

[0129] It can dynamically determine the switching timing between passive balancing and active balancing according to the real-time balancing status, and seek the best balance between balancing time and balancing loss. Through the process of hierarchical selection and parameter adaptation, it can flexibly select the most appropriate balancing strategy according to the actual unbalanced state of the balancing group, and seek the best balance between balancing time and balancing loss, thereby achieving efficient and accurate power balancing.

[0130] This embodiment can realize comprehensive perception and intelligent preprocessing of battery cells, monitor various key parameters of each battery cell in real time, and fully grasp the operating status of the battery. At the same time, graph neural network is used to preprocess data to effectively make up for data missing and improve data integrity and effectiveness, which lays a solid foundation for subsequent adaptive management. Secondly, the adaptive control strategy based on fuzzy logic can flexibly determine the balancing demand level according to the characteristics of different battery cells. By adaptively dividing the battery cells into different balancing groups and taking targeted balancing measures for each group, the personalized needs of each battery cell can be better met and the overall balancing effect can be improved. Finally, the hybrid balancing strategy dynamically weighs the balancing time and balancing loss, while ensuring the balancing effect, it maximizes the overall balancing efficiency, significantly improves the limitations of traditional lithium battery management, and greatly improves the performance and service life of the energy storage system.

[0131] Example 2

[0132] See also Figure 2 As shown, the part not described in detail in this embodiment is described in Example 1, which provides a lithium battery energy storage adaptive management method, including:

[0133] S1. Real-time monitoring of comprehensive battery data of each battery cell; pre-processing the comprehensive battery data to obtain comprehensive characteristic data;

[0134] S2, taking the comprehensive characteristic data of each battery cell as input, and determining the balancing demand level corresponding to each battery cell through an adaptive control strategy based on fuzzy logic;

[0135] S3. Based on the balancing requirement level, all battery cells are divided into M balancing groups; for each balancing group, an adaptive balancing strategy algorithm is used to perform power balancing.

[0136] Example 3

[0137] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned lithium battery energy storage adaptive management method is implemented.

[0138] Since the electronic device introduced in this embodiment is an electronic device used to implement a lithium battery energy storage adaptive management method in the embodiment of this application, based on the lithium battery energy storage adaptive management method introduced in the embodiment of this application, the technical personnel of this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application is not described in detail here. As long as the technical personnel of this field implement the electronic device used in the lithium battery energy storage adaptive management method in the embodiment of this application, it belongs to the scope of protection of this application.

[0139] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters and thresholds in the formula are set by technicians in this field according to actual conditions.

[0140] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technical users in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.

Claims

1. A lithium battery energy storage adaptive management system, characterized in that: include: The real-time monitoring module is used to monitor the comprehensive battery data of each battery unit in real time; the comprehensive battery data is pre-processed to obtain comprehensive characteristic data; A central comprehensive module is used to take the comprehensive characteristic data of each battery cell as input and determine the corresponding balancing demand level of each battery cell through an adaptive control strategy based on fuzzy logic; The balancing management module divides all battery cells into M balancing groups based on the balancing demand level; for each balancing group, an adaptive balancing strategy algorithm is used to balance the power; each module is connected by wire and / or wireless means; The method for determining the equilibrium demand level includes: Define an input fuzzy variable set X, the input fuzzy variables in the input fuzzy variable set correspond to comprehensive characteristic data, define an output fuzzy variable Y, which represents the equilibrium demand level, and for each input fuzzy variable, determine its language value set; for the output fuzzy variable Y, determine its language value set; Collect historical data and construct IF-THEN fuzzy rules based on historical data. The form of fuzzy rules is: IF x1 is A1 AND x2 is A2 AND...AND xn is An THEN Y is Bk; Where x1 is the first input fuzzy variable, x2 is the second input fuzzy variable, A1 is the language value of the first input, A2 is the language value of the second input, xn is the nth input fuzzy variable, and Bk is the language value set; for each input fuzzy variable, calculate its membership degree to each language value; Based on the degree of membership, the fuzzy reasoning method is used to reason each fuzzy rule and obtain the fuzzy reasoning result of each fuzzy rule; the fuzzy reasoning results of all fuzzy rules are synthesized to obtain the comprehensive output fuzzy set B', and B' is mapped to the output space y* through the defuzzification method. y* is the fuzzy value of the balancing demand level corresponding to the battery unit, that is, the balancing demand level; The calculation method of the membership degree includes: From the comprehensive feature data, determine N1 feature sets that are related to the equilibrium demand and influence each other. The data in the feature sets are the relevant features. The first function formula of membership is: Among them, μ x is the mean of the xth related feature, μy is the mean of the yth related feature, σ x is the standard deviation of the xth related feature, σ y is the standard deviation of the y-th related feature, λ(x,y) is the membership degree of the x-th related feature and the y-th related feature; For each feature set, the value range of the data within it is divided into feature intervals; for each feature interval, the membership is calculated using the first function formula or the triangular function splicing method, and for the membership formula in each interval, its parameters are calculated, and the membership functions of different feature intervals are smoothly spliced ​​at the endpoints of the feature intervals.

2. The lithium battery energy storage adaptive management system according to claim 1, characterized in that: The battery comprehensive data includes voltage data, current data, temperature data, internal resistance data, capacity data and balancing data; Voltage data includes battery charge cut-off voltage, battery discharge cut-off voltage and battery average voltage; current data includes maximum charge current and maximum discharge current; Temperature data includes the maximum operating temperature, the minimum operating temperature and the average operating temperature; internal resistance data includes the real-time internal resistance value and the internal resistance change rate; capacity data includes the rated capacity and capacity attenuation rate; The balancing data includes balancing current, balancing time and balancing energy transfer.

3. The lithium battery energy storage adaptive management system according to claim 2, characterized in that: The method of preprocessing the comprehensive battery data includes: Construct a missing value estimation model based on graph neural network, use the missing value estimation model to obtain the missing values ​​in the comprehensive battery data, fill the missing values ​​to obtain complete feature data; standardize / normalize the complete feature data to obtain comprehensive feature data.

4. The lithium battery energy storage adaptive management system according to claim 3, characterized in that: The missing value estimation model is constructed in the following manner: Collect comprehensive battery data of battery cells within a fixed historical time period and record it as historical comprehensive battery data; model the historical comprehensive battery data as a heterogeneous graph G = (V, E); wherein V is a node set, E is an edge set, and the nodes in the node set represent different feature values; the edge set represents the association relationship between the features; The graph convolution-based graph neural network architecture is used as the basic architecture of the missing value estimation model, and the missing value estimation model is trained using heterogeneous graphs; The method of using heterogeneous graphs to train missing value estimation models includes: For each node in the heterogeneous graph G, initialize a feature vector, which is a vector composed of original values; For each node, collect the feature vector information of its neighboring nodes to obtain the updated feature vector; perform nonlinear transformation on the feature vector information, use the updated feature vector of the current layer node as the input of the next layer, and in the last layer, transform the updated feature vector of each node to obtain the estimated value of the corresponding missing value; Define a loss function, calculate the value of the loss function for known battery characteristic values ​​and estimated values, calculate the gradient of the loss function with respect to the parameters of the missing value estimation model through the back propagation algorithm, use the optimization algorithm to update the parameters of the missing value estimation model according to the gradient, and complete the training of the missing value estimation model by iteratively minimizing the value of the loss function.

5. The lithium battery energy storage adaptive management system according to claim 4, characterized in that: The collection formula of the feature vector information is: in, represents the feature vector of node i in the l+1th layer of the graph neural network architecture, W l Represents the weight matrix of the lth layer of the graph neural network architecture; α ij represents the attention weight of node i to node j, represents the feature vector of node i in the lth layer of the graph neural network architecture, σ represents the activation function, and N(i) represents the set of neighbor nodes of node i; where α ij Ways to obtain include: Define the attention function att, then we get α ij = att(h i ,h j ,e ij ), the attention function att is parameterized by a multi-layer perceptron to obtain α ij =MLP([h i ||,h j ||,e ij ]); where h i is the feature of node i, h j is the feature of node j, e ij is the edge feature between node i and node i.

6. The lithium battery energy storage adaptive management system according to claim 5, characterized in that: The method of filling missing values ​​includes: A binary mask vector mask will be defined. For non-missing values, the mask value at the corresponding position is 1, and for missing values, the mask value at the corresponding position is 0. All battery comprehensive data, binary mask vectors, and estimated values ​​are stacked into a 3D tensor, and a batch filling operation is performed on the 3D tensor. The batch filling operation is to scale the estimated value output by the missing value estimation model with the original data element by element. The formula for element-by-element scaling is: in, represents the estimated value of the missing value of node i output by the missing value estimation model, x i represents the eigenvalue of the original node i, mask i Represents the value of the corresponding node i in the binary mask vector.

7. The lithium battery energy storage adaptive management system according to claim 6, characterized in that: The division method of the balancing group includes: The battery cells with the same balancing requirement level are preliminarily divided into a preliminary balancing group, and a battery group topology diagram is constructed. The battery cells are modeled as an undirected graph F = (D, S) according to the physical adjacent relationship, where D is the set of battery cells, S is the battery edge set, and the battery edges in the battery edge set correspond to the physical adjacent relationship; For each battery unit I, its balancing requirement level L is known I , define the level difference threshold ΔL of the battery cells, and only allow |L I -L J Battery cells I and J with |≤ΔL are classified into the same balancing group; Define the objective function and constraints for the balanced group partition; define U IJ is a 0-1 variable. IJ =1, indicating that battery cells I and J are in the same balancing group, then the objective function FL = min∑( I,J ) ∈S U IJ ; The constraint is that for each battery cell I, there must be a special group g I So that: The objective function and constraints are constructed as an integer linear programming model, and the integer linear programming model is solved by a solver to obtain the optimal value U IJ ', represents the final grouping scheme, according to U IJ ', obtain the set of battery cells in the same group LK = {g I |U IJ '=1}; that is, the division of the balancing group is completed.

8. The lithium battery energy storage adaptive management system according to claim 7, characterized in that: The method of performing power balancing includes: For each battery cell in the balancing group, obtain the comprehensive vector R consisting of its comprehensive characteristic data z , where z is the index of the data type in the comprehensive feature data; the comprehensive vector R z It is composed of the concatenation of all types of data in the comprehensive feature data; the imbalance index of the zth data type is defined as the maximum value minus the minimum value of the corresponding data type, and the imbalance indexes of all data types are weighted and fused to obtain a comprehensive score B; The balancing group is divided into imbalance levels LP, LP=1 means B is less than t1, LP=2 means B is greater than or equal to t1 and less than t2, and LP=3 means B is greater than or equal to t2; wherein t1 is the preset lower limit of the level threshold, and t2 is the preset upper limit of the level threshold; The defined balancing strategies include passive balancing strategy, active balancing strategy and mixed balancing strategy; when the imbalance level LP of the balancing group is 1, the balancing group adopts the passive balancing strategy; When the imbalance level LP of the balancing group is 3, the balancing group adopts the active balancing strategy; When the imbalance level LP of the balancing group is 2, the balancing group to which it belongs adopts a mixed balancing strategy; The definition of hybrid equilibrium strategy includes: The optimization objectives of the hybrid balancing strategy are initially defined as the balancing time objective and the balancing loss objective; and the objective function HU=min(β1·TL+β2·EL) is constructed based on the balancing time objective and the balancing loss objective; where β1 is the time weight coefficient and β2 is the loss weight coefficient; Establish the time loss model of passive equilibrium and active equilibrium, namely, the passive equilibrium time model and the active equilibrium time model; The passive balancing time model is based on the imbalance level of the current balancing group and fits the time required to adopt the passive balancing strategy. The active balancing time model is based on the imbalance level of the current balancing group and estimates the time required to adopt the active balancing strategy. The passive balancing strategy is initially adopted. At each time step, the remaining passive balancing time and the remaining time required to switch to active balancing are calculated. If the weighted sum of the time loss when switching to active balancing is less than the weighted sum of the time loss when continuing passive balancing, the active balancing mode is switched to and a balancing current is determined. Otherwise, the passive balancing strategy is continued until the weighted sum of the time loss when switching to active balancing is less than the weighted sum of the time loss when continuing passive balancing.

Images