Intelligent parameter estimation method for lithium battery model driven by hypernetwork and mechanism compensation
Through the dual-drive method of hypernetwork and mechanism compensation, combined with composite dynamic feature enhancement and multi-dimensional performance optimization regulator, the end-to-end estimation of the relationship between the lithium-ion battery model parameters and the state quantity function is achieved, solving the instability of parameter estimation in traditional methods and the instability of parameter estimation in traditional methods and The problem of limited open circuit voltage measurement improves the reliability and flexibility of the battery model.
Patent Information
- Application Number
- CN202411011416.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-26
AI Technical Summary
The prior art has the destructiveness, time-consuming and resource consumption of experimental methods in the estimation of lithium-ion battery model parameters, and the traditional methods rely on manual segmentation processing to be easily affected by subjective factors, resulting in unstable parameter estimation results, especially in low sampling. When the rate and open circuit voltage measurements are limited, it is difficult to effectively model multi-model battery stacks.
The dual-driven method of hypernetwork and mechanism compensation is adopted, and the relationship estimation of end-to-end model parameters and state quantity function is realized through the composite dynamic feature enhancement module, model equation networking module, hypernetwork parameter mapping module and multi-dimensional performance optimization regulator, and The open-circuit voltage compensation calculation module is used to perform refined estimation based on the model equation constraints.
It improves the dynamic processing capability of model parameters, reduces the dependence on open-circuit voltage measurement, enhances the reliability and flexibility of parameter estimation, and is suitable for virtual modeling of large-scale multi-model battery stacks.
Smart Images

Figure CN118536453B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of simulation of lithium-ion batteries, specifically to the technical field of estimation of the relationship between lithium-ion battery model parameters and state quantity functions, and in particular to a hyper-network and mechanism compensation dual-driven lithium battery model intelligent parameter estimation method. Background Art
[0002] Lithium-ion batteries have significant advantages such as low cost, environmental friendliness and high energy density, and are increasingly used in the field of energy storage technology. However, with the expansion of the scale of energy storage power stations, the challenges of lithium-ion battery stack safety and the technical complexity brought about by the diversity of models have become constraints on subsequent technology development. Therefore, it is necessary to build a high-precision and high-flexibility battery stack virtual simulation platform to simulate the behavior of batteries in the battery stack.
[0003] To ensure the effectiveness of battery models in practical applications, it is key to obtain reliable model parameter values. These parameters need to ensure that the model can map the electrochemical behavior and output characteristics of the real battery system. Although experimental methods are the traditional method for obtaining battery model parameters, they usually have problems such as being highly destructive, time-consuming, and resource-intensive, making it difficult to model large-scale, multi-model batteries in energy storage battery stacks. Therefore, compared to experimental methods, parameter estimation methods are a more reasonable way to obtain battery model parameters.
[0004] In the traditional method of battery model parameter estimation, determining the functional relationship between battery model parameters and state quantities usually relies on manual segmentation processing, which is easily affected by subjective factors. And when the battery system operates at a low sampling rate, the interval with sparse data may cause unstable parameter estimation results. In addition, as an important indicator reflecting the electrochemical characteristics of the battery, the open circuit voltage contributes significantly to the characterization of battery behavior. Traditional methods generally obtain the functional relationship between the battery open circuit voltage and the state quantity through experimental measurement. Under limited experimental conditions, traditional methods often adopt a single parameter estimation strategy, which confuses the influence of different battery model parameters on the battery output, which may cause the problem of parameter estimation failure. The above two problems limit the reliability of the battery model parameter estimation results and bring huge challenges to the virtual modeling of battery stacks. Summary of the invention
[0005] The purpose of the present invention is to provide a hyper-network and mechanism compensation dual-driven lithium battery model intelligent parameter estimation method in view of the deficiencies of the prior art. The present invention improves the ability to handle the dynamics of model parameters from the two perspectives of nonlinear enhancement and optimization target guidance, and realizes the end-to-end acquisition of the functional relationship between model parameters and state quantities; at the same time, based on the model equation constraints, the compensation mechanism is used to calculate the open circuit voltage, so as to achieve a refined estimation of the battery open circuit voltage.
[0006] The object of the present invention is achieved through the following technical scheme: a hypernetwork and mechanism compensation dual-driven lithium battery model intelligent parameter estimation method, which determines the parameters to be estimated according to the battery model; the method comprises the following steps:
[0007] (1) Collect the voltage and current data of the lithium battery under actual working conditions to initialize the battery's state of charge information initial value, historical terminal voltage, historical input current, open circuit voltage information list, real terminal voltage sequence and input current sequence, calculate the state of charge information sequence, and scale the input current sequence;
[0008] (2) Construct a composite dynamic feature enhancement module, a model equation networking module, a super-network parameter mapping module, a multi-dimensional performance optimization regulator, and an open-circuit voltage compensation calculation module;
[0009] (3) Sending the state of charge information sequence into the composite dynamic feature enhancement module to obtain an enhanced state of charge information sequence;
[0010] (4) Sending the enhanced state of charge information sequence into the hypernetwork parameter mapping module to generate a circuit element parameter estimation value sequence;
[0011] (5) The estimated value sequence of the circuit element parameters generated by the hypernetwork parameter mapping module, the estimated value sequence of the output voltage generated by the model equation networking module, and the real terminal voltage sequence are sent to the multidimensional performance optimization regulator to obtain the loss value. Based on the loss value, the parameters of the hypernetwork parameter mapping module are updated through the forward propagation stage of information and the reverse propagation stage of error;
[0012] (6) judging whether the cutoff condition is met according to the loss value and the number of iterations output by the multi-dimensional performance optimization regulator; if the cutoff condition is met, saving the hyper-network parameter mapping module and the open-circuit voltage information list at this time; otherwise, updating the open-circuit voltage information list by using the estimated value sequence of the circuit element parameters generated by the updated hyper-network parameter mapping module through the open-circuit voltage compensation calculation module, and then returning to step (4);
[0013] (7) Obtain the optimal circuit element parameter sequence, and establish a functional relationship between it and the state of charge input sequence, and use it together with the open circuit voltage information list as the optimal functional relationship between the battery model parameters and the state quantity finally obtained, and obtain the battery model parameters corresponding to any state of charge based on the optimal functional relationship.
[0014] Furthermore, the battery model is a second-order equivalent circuit model, which is composed of an ohmic polarization resistor, two RC loops and an ideal voltage source in series, and each RC loop is composed of a set of resistors and capacitors in parallel, and its discretization equation is:
[0015]
[0016] in, represents the terminal voltage value at the kth time point; represents the open circuit voltage at the kth time point; represents the input current value at the kth time point; , , , , Represents the model discretization equation parameters in the second-order equivalent circuit model, which are determined by the circuit element parameters, and their expressions are:
[0017]
[0018]
[0019]
[0020]
[0021]
[0022] Where T represents the sampling interval, represents the ohmic polarization resistance, represents the concentration polarization resistance, represents the activation polarization resistance, represents the concentration polarization capacitance, Stands for activated polarized capacitance.
[0023] Further, the calculating of the state of charge information sequence specifically includes:
[0024] According to the initial value of the state of charge information and the input current sequence, the state of charge information at time k is calculated using the ampere-hour integration method, and the calculation formula is:
[0025]
[0026] in, represents the state of charge information at time k, Indicates battery capacity;
[0027] Substitute the initial value of the state of charge information and the input current sequence into the above formula to calculate The state of charge information at each time is arranged in an increasing manner according to the time k into a column vector with a length of N, which is the state of charge information sequence;
[0028] The scaling process of the input current sequence specifically includes:
[0029] According to battery capacity Determine the scaling factor m: where the battery capacity The unit is Ah. When , the zoom ratio m is set to 1; when When , the zoom ratio m is set to 2; when When Round down to the value An integer is randomly selected from the range of rounded-up values as the scaling factor m;
[0030] The input current sequence is divided by the scaling factor m to obtain the scaled input current sequence.
[0031] Furthermore, the construction of the composite dynamic feature enhancement module specifically includes:
[0032] The specific form of the composite dynamic feature enhancement module is a vector value function including trigonometric function terms, linear terms of polynomial functions and constant terms, and its mathematical expression is:
[0033]
[0034] in, is the output vector of the composite dynamic feature enhancement module, is the input value of the composite dynamic feature enhancement module, the trigonometric function term In represents the frequency multiplication of trigonometric functions, Represents the highest frequency multiplier of trigonometric functions.
[0035] Furthermore, the construction of the model equation networking module specifically includes:
[0036] A model equation network module is constructed based on the discretized equation of the second-order equivalent circuit model to convert the discretized second-order equivalent circuit model into a historical information contribution term. , Input current contribution Open circuit voltage calibration item cumulative form;
[0037] Among them, the historical information contribution item It is composed of system matrix and state quantity, and its mathematical expression is:
[0038]
[0039] In the formula, is the state quantity at time k-1, ; W is the system matrix, ;
[0040] The input current contribution It is composed of input matrix and input current, and its mathematical expression is:
[0041]
[0042] In the formula, is the input current of the battery; U is the input matrix, ;
[0043] The open circuit voltage calibration item It is composed of a constant vector, a system matrix and a calibration quantity, and its mathematical expression is:
[0044]
[0045] In the formula, the calibration quantity is the battery open circuit voltage at the kth moment, , and are all constant vectors, , , ;
[0046] The historical information contribution item , the input current contribution term The open circuit voltage calibration item The accumulated result is the state quantity in the historical information contribution item The first-order advance term , its mathematical expression is:
[0047]
[0048] Combining the above mathematical expressions, we get the mathematical expression of the model equation network module:
[0049]
[0050] Where W is the system matrix, is the state quantity at time k-1, U is the input matrix, is the battery input current, , and are all constant vectors, is the open circuit voltage of the battery at the kth moment.
[0051] Furthermore, the construction of the hypernetwork parameter mapping module specifically includes:
[0052] The hypernetwork parameter mapping module includes a feature extraction conversion module, a feature recommendation module and a clipping module, wherein the feature extraction conversion module includes multiple groups of three-layer structures, the three-layer structure includes a fully connected layer, an activation function layer and a regularization layer connected in sequence, wherein the number of groups of the three-layer structure is determined according to the width of each network layer, the feature dimension of the enhanced state of charge information sequence and the number of types of circuit element parameters; the feature recommendation module includes 5 groups of parallel feature augmentation layers, self-attention mechanism layers, feature reduction layers and layer normalization layers; the clipping module includes 5 groups of parallel clipping units, each clipping unit includes a fully connected layer and a clipping function layer, and the mathematical expression of the clipping function layer is:
[0053]
[0054] in, represents the minimum value of the i-th circuit element parameter, Represents the maximum value of the i-th circuit element parameter; sigmoid() represents the activation function; the input of the limiting function layer is the output of the fully connected layer in the clipping unit at the kth moment of the i-th circuit element parameter; the output of the clipping function layer is the estimated value of the parameter of the ith circuit element at the kth moment, and the estimated value sequence of the parameter of the ith circuit element is formed according to the estimated values of the parameter of the ith circuit element at all moments; Corresponding to the circuit element parameters .
[0055] Furthermore, the construction of the multi-dimensional performance optimization regulator specifically includes:
[0056] The multi-dimensional performance optimization regulator includes three parts: difference evaluation item, fluctuation evaluation item and output guidance item. The mathematical expression of the multi-dimensional performance optimization regulator is obtained by weighting the above three parts:
[0057]
[0058] in, represents the loss, which is the output value of the multi-dimensional performance optimization regulator; and are the weight coefficients of the fluctuation assessment item and the output guidance item respectively; represents the difference evaluation item, represents the volatility assessment item, Represents the output boot item;
[0059] The fluctuation evaluation term is quantified by the mean square error, expressed as:
[0060]
[0061] in, represents the real terminal voltage sequence, represents a sequence of output voltage estimation values;
[0062] The mathematical expression of the fluctuation evaluation item is a piecewise function. The first section uses the mean of the sum of the squares of the first-order differences of the output voltage estimation value sequence to measure the continuity of the output. The second section uses the mean of the squares of the first-order differences of the difference between the output voltage estimation value sequence and the real terminal voltage sequence to measure the volatility of the output, which is expressed as:
[0063]
[0064] in, is the set threshold; is the mask vector; is the difference operator;
[0065] The output guidance term uses the inverse quantization parameter of the coefficient of variation to estimate the discretization degree of the network output data, which is expressed as:
[0066]
[0067] in, represents the mean value calculated from the estimated value sequence of the i-th circuit element parameter, represents the standard deviation calculated from the sequence of estimated values of the i-th circuit element parameter.
[0068] Furthermore, the state of charge information sequence is sent to the composite dynamic feature enhancement module to obtain an enhanced state of charge information sequence, specifically comprising: sending the state of charge information sequence calculated in step (1) The value of each moment in the composite dynamic feature enhancement module is sent to the composite dynamic feature enhancement module. After being processed by the vector value function of the composite dynamic feature enhancement module, the enhanced state of charge information vector corresponding to each moment is obtained. , for all times the enhanced state of charge information vector Sort by time k in increasing order to obtain an enhanced state of charge information sequence ;in represents the enhanced state of charge information vector at time k in the enhanced state of charge information sequence, and N is the sequence length;
[0069] The step of sending the enhanced state of charge information sequence into the hypernetwork parameter mapping module to generate a sequence of estimated values of circuit element parameters specifically includes: sending the enhanced state of charge information sequence According to the time k, the rows are sorted into a matrix, and the dimension size of the matrix is , each row corresponds to the enhanced state of charge information vector at time k, and each column corresponds to each dimension of the vector value function; after inputting this matrix into the hypernetwork parameter mapping module, the estimated value sequence of circuit element parameters is obtained ;
[0070] The method of updating the parameters of the hypernetwork parameter mapping module through the forward propagation phase of information and the reverse propagation phase of error specifically includes: in the forward propagation phase of information, firstly, the circuit element parameters are discretized according to the discretization equation of the second-order equivalent circuit model. The corresponding estimated value sequence is converted into the model discretization equation parameters The corresponding sequence is input into the model equation network module at time k as the parameter of the model equation network module; then the initial value of the state quantity , Input current sequence after scaling With open circuit voltage information sequence The estimated value sequence of the circuit element parameters, the estimated value sequence of the output voltage and the real terminal voltage sequence are input into the multidimensional performance optimization regulator to obtain the output loss value. In the back propagation stage of the error, the loss value output by the multidimensional performance optimization regulator is used to generate update information according to the gradient descent method to correct the parameters of each layer in the hypernetwork parameter mapping module. After completing a complete forward propagation of information and back propagation of errors, the update of the parameters of each layer in the hypernetwork parameter mapping module is completed.
[0071] Further, the open circuit voltage compensation calculation module updates the open circuit voltage information list using the estimated value sequence of the circuit element parameters generated by the updated hypernetwork parameter mapping module, specifically including:
[0072] First, the scaled input current sequence, the estimated value sequence of circuit element parameters generated by the updated hypernetwork parameter mapping module, the estimated value sequence of output voltage generated by the model equation networking module, the real terminal voltage sequence, the open circuit voltage information list, and the open circuit voltage information sequence calculated based on the state of charge information sequence and the open circuit voltage information list are processed. Input to the open circuit voltage compensation calculation module;
[0073] Then the open circuit voltage compensation calculation module calculates the voltage compensation term and the current compensation term through the above input information, and their expressions are:
[0074]
[0075]
[0076] in, is the voltage compensation term, is the current compensation term;
[0077] Then, the open circuit voltage at each moment in the updated open circuit voltage information sequence is calculated based on the voltage compensation term and the current compensation term, and the expression is:
[0078]
[0079] in, represents the open circuit voltage value at time k in the open circuit voltage information sequence before updating, represents the open circuit voltage value at the kth moment after update, are the weight coefficients of the voltage compensation term and the current compensation term respectively;
[0080] Secondly, Divide segment, of which The value of needs to be given in advance based on experience. Conditional k-time The cumulative update value of the open circuit voltage information list is:
[0081]
[0082] in, Indicates the value at position j in the open circuit voltage information list; Indicates that the open circuit voltage value at the kth moment in the state of charge information sequence belongs to scope;
[0083] Recalculate the compliance The number of conditional points , and calculate the open circuit voltage information list update value according to the following formula:
[0084]
[0085] in, represents the updated value of position j in the open circuit voltage information list;
[0086] Finally, the values of each position in the open circuit voltage information list are limited to obtain the updated open circuit voltage information list:
[0087]
[0088] in, Indicates the updated value of position j in the open circuit voltage information list after limiting, and They represent the lower and upper limits of the open circuit voltage respectively.
[0089] Furthermore, the step (7) specifically includes the following sub-steps:
[0090] (7.1) The battery state of charge is divided evenly from 0% to 100% equal parts, arranged in ascending order to form a state of charge estimation sequence, where Determined by the level of detail required in parameter estimation;
[0091] (7.2) using the state of charge estimation sequence as an input of a hypernetwork parameter mapping module, and the hypernetwork parameter mapping module outputs a sequence of estimated values of circuit element parameters;
[0092] (7.3) The estimated value sequence of the circuit element parameters is processed according to the scaling factor m, specifically: the ohmic polarization resistance , concentration polarization resistance , Activation Polarization Resistance The corresponding estimated value is divided by the scaling factor m to convert the concentration polarization capacitance , activated polarized capacitor The corresponding estimated value is multiplied by the scaling factor m to obtain the optimal circuit element parameter sequence;
[0093] (7.4) The optimal circuit element parameter sequence is matched with the state of charge input sequence to establish a functional relationship between the two, and together with the open circuit voltage information list, it is used as the optimal functional relationship between the battery model parameters and the state quantity finally obtained. Based on this optimal functional relationship, the battery model parameters corresponding to any state of charge are obtained.
[0094] Compared with the prior art, the present invention has the following beneficial effects:
[0095] (1) The present invention effectively processes the functional relationship between lithium battery model parameters and state quantities by designing a model equation networking module and a hypernetwork parameter mapping module. When the same state of charge information is input, the parameter information of the model equation networking module is only related to the output of the hypernetwork parameter mapping module, thereby simulating the mapping relationship of the function, so that the present invention is able to capture the dynamics of the model parameters; at the same time, the present invention innovatively improves the ability of the parameter estimation method to process the dynamics of model parameters from two perspectives: nonlinear enhancement and optimization target guidance through a composite dynamic feature enhancement module and a multi-dimensional performance optimization regulator; the present invention achieves the goal of end-to-end estimation of the functional relationship between battery model parameters and state quantities for the first time through the collaborative work between the above modules.
[0096] (2) The present invention calculates the open circuit voltage parameters of the battery through an open circuit voltage compensation calculation module, and for the first time uses the compensation mechanism to specifically adjust the estimated value of the battery open circuit voltage, rather than using the parameter estimation method to search in the parameter space in an untargeted manner. In the absence of an open circuit voltage measurement experiment, the effectiveness of the parameter estimation method of the present invention can be guaranteed, effectively reducing the dependence on the open circuit voltage measurement experiment.
[0097] (3) The present invention utilizes an iterative strategy and uses equivalent circuit model constraints to improve the reliability of parameter estimation. By separating the estimation of circuit element parameters from the estimation of open-circuit voltage parameters and performing parameter estimation in an iterative manner, the parameter space can be reduced in dimension and the difficulty of parameter estimation can be reduced. For the estimation of circuit element parameters, the present invention uses a model equation networking module based on model equation transformation as a constraint condition. For the estimation of open-circuit voltage, the present invention adopts a strategy of using the model equation constraint as a benchmark and a compensation mechanism for calculation. Both of these allow the entire parameter estimation process to be carried out under the constraint of the model equation, thereby improving the reliability of parameter estimation. BRIEF DESCRIPTION OF THE DRAWINGS
[0098] Figure 1 It is a flow chart of the intelligent parameter estimation method of lithium battery model driven by hypernetwork and mechanism compensation of the present invention;
[0099] Figure 2 It is a schematic diagram of the overall structure of the hypernetwork parameter mapping module of the present invention;
[0100] Figure 3 It is a schematic diagram of information flow between various modules in the present invention;
[0101] Figure 4 This is a diagram showing the effect of capturing the dynamics of ohmic polarization resistance of the present invention;
[0102] Figure 5 This is a diagram showing the effect of capturing the dynamics of concentration polarization resistance according to the present invention;
[0103] Figure 6 This is a diagram showing the effect of capturing the dynamic properties of the activated polarization resistance of the present invention;
[0104] Figure 7 This is a diagram showing the effect of capturing the dynamics of the concentration polarization capacitor according to the present invention;
[0105] Figure 8 This is a diagram showing the effect of capturing the dynamic properties of activated polarized capacitors according to the present invention;
[0106] Fig. 9 This is a diagram showing the test results of the battery 1 according to the present invention;
[0107] Fig.10This is a diagram showing the test results of the present invention on battery 2;
[0108] Fig.11 This is a diagram showing the test results of the present invention on battery 3;
[0109] Fig.12 This is a diagram showing the test results of the present invention on battery 4. DETAILED DESCRIPTION
[0110] This embodiment uses input and output data of a real lithium-ion energy storage battery to perform parameter estimation, and simulates the parameter estimation results under real working conditions to verify the effectiveness of this method. The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.
[0111] In this embodiment, the service data of the battery module for 5 consecutive days is collected to construct the training set and the test set. For each battery, the total number of sampling points is 7200. The first 6400 sampling points are used as the training set, and the last 800 sampling points are used as the test set. The second-order equivalent circuit model is selected as the model body for parameter estimation. There are six parameters that need to be estimated, namely, ohmic polarization resistance , concentration polarization resistance , Activation Polarization Resistance , concentration polarization capacitance , activated polarized capacitor With open circuit voltage .
[0112] The hypernetwork and mechanism compensation dual-driven lithium battery model intelligent parameter estimation method of the present invention first determines the parameters to be estimated according to the battery model, and then performs the following steps (1) to (7) to achieve the estimation of the battery model parameters. Figure 1 shown.
[0113] In this embodiment, the battery model is a second-order equivalent circuit model, which is composed of an ohmic polarization resistor, two RC loops and an ideal voltage source in series. Each RC loop is composed of a set of resistors and capacitors in parallel. The discretization equation is
[0114] in, represents the terminal voltage value at the kth time point; represents the open circuit voltage at the kth time point; represents the input current value at the kth time point; , , , , Represents the model discretization equation parameters in the second-order equivalent circuit model, which are determined by the circuit element parameters, and their expressions are:
[0115]
[0116]
[0117]
[0118]
[0119]
[0120] Where T represents the sampling interval, represents the ohmic polarization resistance, represents the concentration polarization resistance, represents the activation polarization resistance, represents the concentration polarization capacitance, Stands for activated polarized capacitance. , , , and These five parameters are collectively referred to as circuit element parameters, and together with the open circuit voltage, serve as parameters to be estimated for the lithium battery model described in the present invention.
[0121] See also Figure 1 , the method specifically comprises the following steps:
[0122] (1) Collect the voltage and current data of the lithium battery under actual working conditions to initialize the battery's state of charge information initial value, historical terminal voltage, historical input current, open circuit voltage information list, real terminal voltage sequence and input current sequence, calculate the state of charge information sequence, and scale the input current sequence.
[0123] Furthermore, the voltage data and current data of the lithium battery under actual working conditions are collected to initialize the initial value of the battery's state of charge information, historical terminal voltage, historical input current, open circuit voltage information list, real terminal voltage sequence and input current sequence. The length of the real terminal voltage sequence and input current sequence is represented by N, which is specifically determined according to the voltage data and current data collected in the sampling interval. In this embodiment, the value of N is 7200. The real terminal voltage sequence and the input current sequence are divided into two subsequences, whose lengths are respectively and , is the sampling length of the input data used for training, is the sampling length of the input data used for testing. The value of is 6400. The historical terminal voltage is divided into a first-order historical terminal voltage and a second-order historical terminal voltage. The first-order historical terminal voltage corresponds to the lithium battery voltage sampling value at one moment before the initial sampling moment, and the second-order historical terminal voltage corresponds to the lithium battery voltage sampling value two moments before the initial sampling moment. The historical input current is divided into a first-order historical input current and a second-order historical input current. The first-order historical input current corresponds to the lithium battery current sampling value at one moment before the initial sampling moment, and the second-order historical input current corresponds to the lithium battery current sampling value two moments before the initial sampling moment. It should be noted that when the lithium battery voltage or current sampling value at one moment before the initial sampling moment or the first two moments is unknown, the voltage sampling value and current sampling value at the initial sampling moment are directly used instead. The open circuit voltage information list is initialized to a column vector of length s, and the value of each position in the column vector is set to the value 3. The s selected in this embodiment is 101. It should be understood that the value of s can be pre-specified based on experience. The initial value of the state of charge information is assigned using the ratio of the initial power of the lithium battery to the nominal capacity of the battery. The initial value of the state of charge information is a constant value.
[0124] Further, calculating the state of charge information sequence specifically includes:
[0125] According to the initial value of the state of charge information and the input current sequence, the state of charge information at time k is calculated using the ampere-hour integration method, and the calculation formula is:
[0126]
[0127] in, represents the state of charge information at time k, Indicates the battery capacity;
[0128] Substitute the initial value of the state of charge information and the input current sequence into the above formula to calculate The state of charge information at each moment is arranged in an increasing manner according to the time k into a column vector with a length of N, which is a state of charge information sequence.
[0129] It should be noted that the state of charge information sequence may also be calculated using existing methods such as the Kalman filter method and the particle filter method.
[0130] Furthermore, the input current sequence is scaled, which specifically includes the following sub-steps:
[0131] (1.1) According to battery capacity Determine the scaling factor m: where the battery capacity The unit is Ah. When , the zoom ratio m is set to 1; when When , the zoom ratio m is set to 2; when When Round down to the value An integer is randomly selected from the range of rounded-up values as the scaling factor m, which can be expressed as:
[0132]
[0133] For example, in this embodiment, the battery capacity The value of the scaling factor m is 280Ah. According to the above expression, the range of the scaling factor m can be determined to be [175, 233]. For example, 200 is selected as the scaling factor in this embodiment, that is, m=200.
[0134] (1.2) The input current sequence is divided by the scaling factor m to obtain the scaled input current sequence.
[0135] (2) Construct a composite dynamic feature enhancement module, a model equation networking module, a super-network parameter mapping module, a multi-dimensional performance optimization regulator, and an open-circuit voltage compensation calculation module, such as Figure 3 shown.
[0136] In this embodiment, the construction of the composite dynamic feature enhancement module specifically includes: the specific form of the composite dynamic feature enhancement module is a vector value function including trigonometric function terms, linear terms of polynomial functions and constant terms, and its mathematical expression is:
[0137]
[0138] in, is the output vector of the composite dynamic feature enhancement module, is the input value of the composite dynamic feature enhancement module, the trigonometric function term In represents the frequency multiplication of trigonometric functions, Represents the highest frequency multiplication of trigonometric functions, is an integer that needs to be pre-set based on experience. Set to 7. At the same time, the first-order term of the input value is added to the output vector And the constant term To capture the overall trend of change.
[0139] In this embodiment, the construction of the model equation network module specifically includes: constructing the model equation network module according to the discretized equation of the second-order equivalent circuit model to convert the discretized second-order equivalent circuit model into a historical information contribution term , Input current contribution Open circuit voltage calibration item Cumulative form.
[0140] Furthermore, historical information contribution It is composed of system matrix and state quantity, and its mathematical expression is:
[0141]
[0142] in, is the state quantity at time k-1, ; W is the system matrix, .
[0143] Furthermore, the input current contribution It is composed of input matrix and input current, and its mathematical expression is:
[0144]
[0145] in, is the input current of the battery; U is the input matrix, .
[0146] Furthermore, the open circuit voltage calibration term It is composed of a constant vector, a system matrix and a calibration quantity, and its mathematical expression is:
[0147]
[0148] Among them, the calibration quantity is the battery open circuit voltage at the kth moment, , and are all constant vectors, , , .
[0149] Furthermore, historical information contribution , Input current contribution Open circuit voltage calibration item The accumulated result is the state quantity in the historical information contribution item The first-order advance term , its mathematical expression is:
[0150]
[0151] In summary, by combining the above mathematical expressions, we can get the mathematical expression of the model equation network module:
[0152]
[0153] Where W is the system matrix, is the state quantity at time k-1, U is the input matrix, is the battery input current, , and are all constant vectors, is the open circuit voltage of the battery at the kth moment. With the input matrix In The input of the model equation network module includes the initial value of the state quantity , input current sequence With open circuit voltage information sequence Among them, the initial value of the state quantity The specific expression is , is the first-order historical terminal voltage, is the second-order history terminal voltage, is the first-order historical input current, is the second-order historical input current.
[0154] Furthermore, the method for obtaining the open circuit voltage information sequence is as follows: the k-th moment value in the state of charge information sequence and the open circuit voltage information list are interpolated to calculate the battery open circuit voltage at the k-th moment, and the open circuit voltage information sequence with a length of N is arranged in an increasing manner according to the moment k. Among them, the interpolation method includes existing interpolation methods such as piecewise cubic Hermite interpolation method and cubic spline interpolation method. In this embodiment, the piecewise cubic Hermite interpolation method is used to interpolate and calculate the open circuit voltage of the battery.
[0155] In this embodiment, the construction of the hypernetwork parameter mapping module specifically includes: the hypernetwork parameter mapping module includes a feature extraction conversion module, a feature recommendation module and a limiting module, such as Figure 2 As shown. Among them, the feature extraction conversion module includes multiple groups of three-layer structures, and the three-layer structure includes a fully connected layer, an activation function layer and a regularization layer connected in sequence, wherein the number of groups of the three-layer structure is determined according to the width of each network layer, the feature dimension of the enhanced state of charge information sequence and the number of types of circuit element parameters, and is set in advance according to experience; the activation function layer selected in this embodiment is the Mish layer, the regularization layer is the dropout layer, and the number of groups of the three-layer structure is 3. The feature recommendation module includes 5 groups of parallel feature augmentation layers, self-attention mechanism layers, feature reduction layers and layer normalization layers. The limiting module includes 5 groups of parallel limiting units, each of which includes a fully connected layer and a limiting function layer, and the mathematical expression of the limiting function layer is:
[0156]
[0157] in, represents the minimum value of the i-th circuit element parameter, represents the maximum value of the i-th circuit element parameter. The upper and lower limits of each circuit element parameter are shown in Table 1; sigmoid() represents the activation function; the input of the limiting function layer is the output of the fully connected layer in the clipping unit at the kth moment of the i-th circuit element parameter; the output of the clipping function layer is the estimated value of the parameter of the ith circuit element at the kth moment, and the estimated value sequence of the parameter of the ith circuit element is formed according to the estimated values of the parameter of the ith circuit element at all moments; Corresponding to the circuit element parameters .
[0158] Table 1: Upper and lower limits of parameters of various circuit components in the limiting module
[0159]
[0160] From the above content, it can be seen that the parameters of the hypernetwork parameter mapping module are the network parameters in the feature extraction conversion module, the feature recommendation module and the limiting module, the input is the output value of the composite dynamic feature enhancement module, and the output is the estimated value sequence of the circuit element parameters.
[0161] In this embodiment, the construction of the multi-dimensional performance optimization regulator specifically includes: the multi-dimensional performance optimization regulator includes a difference evaluation item , Fluctuation Assessment Item and output boot items Three parts, by weighting the above three parts to obtain the mathematical expression of the multi-dimensional performance optimization regulator:
[0162]
[0163] in, represents the loss, which is the output value of the multi-dimensional performance optimization regulator; and are weight coefficients of the fluctuation assessment item and the output guidance item, respectively, which can be preset based on experience. In this embodiment, , .
[0164] Furthermore, the fluctuation evaluation term is quantified by the mean square error, expressed as:
[0165]
[0166] in, represents the real terminal voltage sequence, Represents a sequence of output voltage estimates.
[0167] Furthermore, the mathematical expression of the fluctuation evaluation term is a piecewise function. The first section uses the mean of the sum of the squares of the first-order differences of the output voltage estimation sequence to measure the continuity of the output. The second section uses the mean of the squares of the first-order differences of the difference between the output voltage estimation sequence and the real terminal voltage sequence to measure the output volatility, which is expressed as:
[0168]
[0169] in, is the threshold value to be set, which can be set in advance based on experience. In this embodiment, The value of is 5.5; is a mask vector, whose length is the length of the input data N minus 1, which sets the mutation moments in the charge and discharge phase and the static period to 0, and sets other positions to 1; is a difference operator used to calculate the first-order difference sequence of the sequence.
[0170] Furthermore, the output guidance term uses the inverse quantization parameter of the coefficient of variation to estimate the discretization degree of the network output data, expressed as:
[0171]
[0172] in, represents the mean value calculated from the estimated value sequence of the i-th circuit element parameter, represents the standard deviation calculated from the sequence of estimated values of the i-th circuit element parameter.
[0173] (3) Send the state of charge information sequence into the composite dynamic feature enhancement module to obtain the enhanced state of charge information sequence .
[0174] Specifically, the state of charge information sequence calculated in step (1) The value of each moment in the composite dynamic feature enhancement module is sent to the composite dynamic feature enhancement module, and after being processed by the vector value function of the composite dynamic feature enhancement module, the enhanced state of charge information vector corresponding to each moment is obtained. , the vector The length is The row vector of the embodiment has a length of 16 in the medium term. Sort by time k in increasing order to obtain an enhanced state of charge information sequence ;in represents the enhanced state of charge information vector at time k in the enhanced state of charge information sequence, and N is the sequence length.
[0175] (4) Send the enhanced state of charge information sequence to the hypernetwork parameter mapping module to generate a sequence of estimated values of circuit element parameters ,in represents the parameters of the i-th circuit element.
[0176] Specifically, the enhanced state of charge information sequence The matrix is sorted in rows according to the time k. The first dimension of the matrix is N. In this embodiment, N is 6400 during the training process and 800 during the test phase. Each row of the matrix corresponds to the enhanced state of charge information vector at the time k. The second dimension of the matrix is , the second dimension of the matrix in this embodiment is 16. Each column of the matrix corresponds to each dimension of the vector value function. After the matrix is input into the hypernetwork parameter mapping module, the estimated value sequence of the circuit element parameters is obtained. .
[0177] (5) The estimated value sequence of circuit element parameters generated by the hypernetwork parameter mapping module, the estimated value sequence of output voltage generated by the model equation networking module, and the real terminal voltage sequence are sent to the multidimensional performance optimization regulator to obtain the loss value. Based on the loss value, the parameters of the hypernetwork parameter mapping module are updated through the forward propagation stage of information and the backward propagation stage of error.
[0178] Specifically, in the forward propagation stage of information, the circuit element parameters are firstly converted into The corresponding estimated value sequence is converted into the model discretization equation parameters The corresponding sequence is input into the model equation network module at time k as the parameter of the model equation network module; then the initial value of the state quantity , Input current sequence after scaling With open circuit voltage information sequence Input into the model equation network module to obtain the estimated value sequence of the output voltage; then input the estimated value sequence of the circuit element parameters, the estimated value sequence of the output voltage and the real terminal voltage sequence into the multidimensional performance optimization regulator to obtain the output loss value. In the back propagation stage of the error, the loss value output by the multidimensional performance optimization regulator is used to generate update information according to the gradient descent method to correct the parameters of each layer in the hypernetwork parameter mapping module. After completing a complete forward propagation of information and back propagation of errors, the update of the parameters of each layer in the hypernetwork parameter mapping module is completed.
[0179] (6) Determine whether the cutoff condition is met based on the loss value and the number of iterations output by the multi-dimensional performance optimization regulator. If the cutoff condition is met, save the hypernetwork parameter mapping module and the open-circuit voltage information list at this time; otherwise, update the open-circuit voltage information list using the estimated value sequence of the circuit element parameters generated by the updated hypernetwork parameter mapping module through the open-circuit voltage compensation calculation module, and then return to step (4).
[0180] It should be understood that the cutoff condition can be set according to actual needs, such as the loss value being less than a preset loss threshold or being greater than a preset number of iterations.
[0181] Further, the open circuit voltage information list is updated by the open circuit voltage compensation calculation module using the estimated value sequence of the circuit element parameters generated by the updated hypernetwork parameter mapping module, which specifically includes the following sub-steps:
[0182] (6.1) The scaled input current sequence, the estimated value sequence of the circuit element parameters generated by the updated hypernetwork parameter mapping module, the estimated value sequence of the output voltage generated by the model equation networking module, the real terminal voltage sequence, the open circuit voltage information list, and the open circuit voltage information sequence calculated based on the state of charge information sequence and the open circuit voltage information list are combined. Input to the open circuit voltage compensation calculation module, where the information flow between each module is as follows Figure 3 shown.
[0183] (6.2) The open circuit voltage compensation calculation module calculates the voltage compensation term and the current compensation term through the above input information, and their expressions are:
[0184]
[0185]
[0186] in, is the voltage compensation term, is the current compensation term.
[0187] (6.3) The open circuit voltage at each moment in the updated open circuit voltage information sequence is calculated based on the voltage compensation term and the current compensation term, and its expression is:
[0188]
[0189] in, represents the open circuit voltage value at time k in the open circuit voltage information sequence before updating, represents the open circuit voltage value at the kth moment after update, are the weight coefficients of the voltage compensation term and the current compensation term, respectively, which can be given in advance based on experience. In this embodiment, , .
[0190] (6.4) Divide segment, of which The value of needs to be given in advance based on experience. Conditional k-time The cumulative update value of the open circuit voltage information list is:
[0191]
[0192] in, Indicates the value at position j in the open circuit voltage information list; Indicates that the open circuit voltage value at the kth moment in the state of charge information sequence belongs to scope.
[0193] (6.5) Calculation complies with The number of conditional points , and calculate the open circuit voltage information list update value according to the following formula:
[0194]
[0195] in, Indicates the updated value of position j in the open circuit voltage information list.
[0196] (6.6) Finally, the values of each position in the open circuit voltage information list are limited to obtain the updated open circuit voltage information list:
[0197]
[0198] in, Indicates the updated value of position j in the open circuit voltage information list after limiting, and They represent the lower limit and upper limit of the open circuit voltage respectively. In this embodiment, the lower limit is set to 2.2V, and the upper limit is set to 3.5V.
[0199] (7) Obtain the optimal circuit element parameter sequence, and establish a functional relationship between it and the state of charge input sequence, and use it together with the open circuit voltage information list as the optimal functional relationship between the battery model parameters and the state quantity finally obtained, and obtain the battery model parameters corresponding to any state of charge based on the optimal functional relationship.
[0200] (7.1) The battery state of charge is divided evenly from 0% to 100% equal parts, arranged in ascending order to form a state of charge estimation sequence, where Determined by the level of detail required for parameter estimation, in this embodiment The value of is 1000.
[0201] (7.2) The state of charge estimation sequence is used as the input of the hypernetwork parameter mapping module, and the hypernetwork parameter mapping module outputs a sequence of estimated values of circuit element parameters.
[0202] (7.3) The estimated value sequence of the circuit element parameters is processed according to the scaling factor m, specifically: the ohmic polarization resistance , concentration polarization resistance , Activation Polarization Resistance The corresponding estimated value is divided by the scaling factor m to convert the concentration polarization capacitance , activated polarized capacitor The corresponding estimated value is multiplied by the scaling factor m to obtain the optimal circuit element parameter sequence.
[0203] (7.4) The optimal circuit element parameter sequence is matched with the state of charge input sequence to establish a functional relationship between the two, and together with the open circuit voltage information list, it is used as the optimal functional relationship between the battery model parameters and the state quantity finally obtained. Based on this optimal functional relationship, the battery model parameters corresponding to any state of charge are obtained.
[0204] It should be understood that by establishing the optimal functional relationship between the battery model parameters and the state quantities, the battery model parameters under different state quantities can be obtained, and the parameter estimation of the battery model can be realized.
[0205] Exemplarily, the parameter estimation method of the present invention is used to perform parameter estimation and testing on the data of four batteries respectively to illustrate the effect of the present invention.
[0206] The dynamics of four battery parameters are captured, among which the ohmic polarization resistance The dynamics of the estimate captures the effect of Figure 4 As shown, the concentration polarization resistance The dynamics of the estimate captures the effect of Figure 5 As shown, the activation polarization resistance The dynamics of the estimate captures the effect of Figure 6 As shown, the concentration polarization capacitance The dynamics of the estimate captures the effect of Figure 7 As shown, the activation polarization capacitance The dynamics of the estimate captures the effect of Figure 8 shown. Figure 4-Figure 8 In the figure, the solid line is the parameter corresponding to battery 1, the dashed line is the parameter corresponding to battery 2, the dotted line is the parameter corresponding to battery 3, and the dot-dash line is the parameter corresponding to battery 4.
[0207] Four batteries were tested to illustrate the generalizability of the parameter estimation method of the present invention. The test results for battery 1 are shown in FIG. Fig. 9 As shown, the test results of battery 2 are as follows Fig.10 As shown, the test results of battery 3 are as follows Fig.11 As shown, the test results of battery 4 are as follows Fig.12 As shown. Figure 9-12 It can be seen that the test set data of four batteries are used for verification. The solid line is the actual operation situation, and the dotted line is the test result. The mean square error is within 0.05%, and the maximum error does not exceed 0.05V.
[0208] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. 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 invention.
Claims
1. A hyper-network and mechanism compensation dual-driven lithium battery model intelligent parameter estimation method, characterized in that: Determine the parameters to be estimated according to the battery model; the method comprises the following steps: (1) Collecting the voltage and current data of the lithium battery under actual working conditions to initialize the battery's state of charge information initial value, historical terminal voltage, historical input current, open circuit voltage information list, real terminal voltage sequence and input current sequence, and calculate the state of charge information sequence, and scale the input current sequence; (2) Construct a composite dynamic feature enhancement module, a model equation networking module, a super-network parameter mapping module, a multi-dimensional performance optimization regulator, and an open-circuit voltage compensation calculation module; (3) sending the state of charge information sequence into the composite dynamic feature enhancement module to obtain an enhanced state of charge information sequence; (4) sending the enhanced state of charge information sequence to the hypernetwork parameter mapping module to generate a circuit element parameter estimation value sequence; (5) sending the estimated value sequence of the circuit element parameters generated by the hypernetwork parameter mapping module, the estimated value sequence of the output voltage generated by the model equation networking module, and the real terminal voltage sequence to the multidimensional performance optimization regulator to obtain the loss value, and based on the loss value, updating the parameters of the hypernetwork parameter mapping module through the forward propagation stage of information and the reverse propagation stage of error; (6) judging whether the cut-off condition is met according to the loss value and the number of iterations output by the multi-dimensional performance optimization regulator; if the cut-off condition is met, saving the hypernetwork parameter mapping module and the open-circuit voltage information list at this time; otherwise, updating the open-circuit voltage information list by using the estimated value sequence of the circuit element parameters generated by the updated hypernetwork parameter mapping module through the open-circuit voltage compensation calculation module, and then returning to step (4); (7) Obtain an optimal circuit element parameter sequence, and establish a functional relationship between it and the state of charge input sequence, and use it together with the open circuit voltage information list as the optimal functional relationship between the battery model parameters and the state quantity finally obtained, and obtain the battery model parameters corresponding to any state of charge based on the optimal functional relationship.
2. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 1 is characterized in that: The battery model is a second-order equivalent circuit model, which consists of an ohmic polarization resistor, two RC loops and an ideal voltage source in series. Each RC loop consists of a set of resistors and capacitors in parallel. Its discretization equation is: <h2 style=";text-align:left;direction:ltr">U<h2 style=";text-align:left;direction:ltr"> L <h2 style=";text-align:left;direction:ltr"> (k) = (1+b1+b2)U<h2 style=";text-align:left;direction:ltr"> ocv <h2 style=";text-align:left;direction:ltr"> (k)-b1U<h2 style=";text-align:left;direction:ltr"> L <h2 style=";text-align:left;direction:ltr"> (k-1)-b2U<h2 style=";text-align:left;direction:ltr"> L <h2 style=";text-align:left;direction:ltr"> (k-2)-b3I(k)-b4I(k-1)-b5I(k-2) Among them, U L (k) represents the terminal voltage value at the kth time point; U ocv (k) represents the open circuit voltage at the kth time point; I(k) represents the input current value at the kth time point; b1, b2, b3, b4, b5 represent the model discretization equation parameters in the second-order equivalent circuit model, which are determined by the circuit element parameters, and their expressions are: Wherein, T represents the sampling interval, R0 represents the ohmic polarization resistance, R1 represents the concentration polarization resistance, R2 represents the activation polarization resistance, C1 represents the concentration polarization capacitance, and C2 represents the activation polarization capacitance.
3. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 1 is characterized in that: The calculating of the state of charge information sequence specifically includes: According to the initial value of the state of charge information and the input current sequence, the state of charge information at time k is calculated using the ampere-hour integration method, and the calculation formula is: Wherein, SoC(k) represents the state of charge information at time k, and Cap represents the battery capacity; Substitute the initial value of the state of charge information and the input current sequence into the above formula, calculate the state of charge information at time k=1, 2, ..., N, and arrange the state of charge information at all times in an increasing manner according to the time k in the form of a column vector with a length of N, which is the state of charge information sequence; The scaling process of the input current sequence specifically includes: Determine the scaling factor m according to the battery capacity Cap: The unit of the battery capacity Cap is Ah. When Cap < 2, set the scaling factor m to 1; when 2 < Cap < 3.6, set the scaling factor m to 2; when Cap > 3.6, Round down to the value An integer is randomly selected from the range of rounded-up values as the scaling factor m; The input current sequence is divided by the scaling factor m to obtain the scaled input current sequence.
4. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 1 is characterized in that: The construction of the composite dynamic feature enhancement module specifically includes: The specific form of the composite dynamic feature enhancement module is a vector value function including trigonometric function terms, linear terms of polynomial functions and constant terms, and its mathematical expression is: y=[1,sin(ωx),cos(ωx),x],ω=1,2,...,Ω Among them, y is the output vector of the composite dynamic feature enhancement module, x is the input value of the composite dynamic feature enhancement module, ω in the trigonometric function terms sin(ωx), cos(ωx) represents the frequency multiplier of the trigonometric function, and Ω represents the highest frequency multiplier of the trigonometric function.
5. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 2 is characterized in that: The construction of the model equation networking module specifically includes: A model equation network module is constructed based on the discretized equation of the second-order equivalent circuit model to convert the discretized second-order equivalent circuit model into a historical information contribution term. Input current contribution Open circuit voltage calibration item cumulative form; Among them, the historical information contribution item It is composed of system matrix and state quantity, and its mathematical expression is: In the formula, s(k-1) is the state quantity at time k-1, s(k-1) = [U L (k-1) I(k-1) U L (k-2) I(k-2)] T ; W is the system matrix, The input current contribution It is composed of input matrix and input current, and its mathematical expression is: Where I(k) is the input current of the battery; U is the input matrix, The open circuit voltage calibration item It is composed of a constant vector, a system matrix and a calibration quantity, and its mathematical expression is: In the formula, the calibration quantity U ocv (k) is the open circuit voltage of the battery at the kth moment, c1, c2 and c3 are all constant vectors, c1 = [1000] T , c2=[-1010] T , c3 = [1010] T ; The historical information contribution item The input current contribution The open circuit voltage calibration item The result of the accumulation is the first-order leading term s(k) of the state quantity s(k-1) in the historical information contribution term, and its mathematical expression is: s(k)=[U L (k)I(k)U L (k-1)I(k-1)] T Combining the above mathematical expressions, we get the mathematical expression of the model equation network module: Where W is the system matrix, s(k-1) is the state quantity at time k-1, U is the input matrix, I(k) is the input current of the battery, c1, c2 and c3 are all constant vectors, and U is ocv (k) is the open circuit voltage of the battery at the kth moment.
6. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 1 is characterized in that: The construction of the hypernetwork parameter mapping module specifically includes: The hypernetwork parameter mapping module includes a feature extraction conversion module, a feature recommendation module and a clipping module, wherein the feature extraction conversion module includes multiple groups of three-layer structures, the three-layer structure includes a fully connected layer, an activation function layer and a regularization layer connected in sequence, wherein the number of groups of the three-layer structure is determined according to the width of each network layer, the feature dimension of the enhanced state of charge information sequence and the number of types of circuit element parameters; the feature recommendation module includes 5 groups of parallel feature augmentation layers, self-attention mechanism layers, feature reduction layers and layer normalization layers; the clipping module includes 5 groups of parallel clipping units, each clipping unit includes a fully connected layer and a clipping function layer, and the mathematical expression of the clipping function layer is: in, represents the minimum value of the i-th circuit element parameter, Represents the maximum value of the i-th circuit element parameter; sigmoid() represents the activation function; the input of the limiting function layer is the output of the fully connected layer in the clipping unit at the kth moment of the i-th circuit element parameter; the output ECM of the clipping function layer i (k) is the estimated value of the i-th circuit element parameter at the k-th moment, and the estimated value sequence of the i-th circuit element parameter is constructed based on the estimated values of the i-th circuit element parameter at all moments; i=1, 2, 3, 4, 5 correspond to circuit element parameters R0, R1, R2, C1, C2 respectively.
7. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 1 is characterized in that: The construction of the multi-dimensional performance optimization regulator specifically includes: The multi-dimensional performance optimization regulator includes three parts: difference evaluation item, fluctuation evaluation item and output guidance item. The mathematical expression of the multi-dimensional performance optimization regulator is obtained by weighting the above three parts: loss=MSE_loss+λSMOOTH_loss+μCOV_loss Among them, loss represents loss, which is the output value of the multi-dimensional performance optimization regulator; λ and μ are the weight coefficients of the fluctuation evaluation item and the output guidance item respectively; MSE_loss represents the difference evaluation item, SMOOTH_loss represents the fluctuation evaluation item, and COV_loss represents the output guidance item; The fluctuation evaluation term is quantified by the mean square error, expressed as: in, represents the real terminal voltage sequence, represents a sequence of output voltage estimation values; The mathematical expression of the fluctuation evaluation item is a piecewise function. The first section uses the mean of the sum of the squares of the first-order differences of the output voltage estimation value sequence to measure the continuity of the output. The second section uses the mean of the squares of the first-order differences of the difference between the output voltage estimation value sequence and the real terminal voltage sequence to measure the volatility of the output, which is expressed as: Among them, δ is the set threshold; MASK is the mask vector; is the difference operator; The output guidance term uses the inverse quantization parameter of the coefficient of variation to estimate the discretization degree of the network output data, expressed as: COV_loss=∑ i (s i / m i ) -1 ,i=1,2,3,4,5 Among them, μ i represents the mean value calculated from the estimated value sequence of the i-th circuit element parameter, σ i represents the standard deviation calculated from the sequence of estimated values of the i-th circuit element parameter.
8. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 4 is characterized in that: The step of sending the state of charge information sequence into the composite dynamic feature enhancement module to obtain an enhanced state of charge information sequence specifically includes: sending the value of each moment in the state of charge information sequence SoC(k), k=1, 2, ..., N, calculated in step (1) into the composite dynamic feature enhancement module, and obtaining the enhanced state of charge information vector SoC corresponding to each moment after being processed by the vector value function of the composite dynamic feature enhancement module. * (k), the enhanced state of charge information vector SoC for all times * (k) Sort by time k in increasing order to obtain the enhanced state of charge information sequence SOC * (k), k=1,2,...,N; where SoC * (k) represents the enhanced state of charge information vector at time k in the enhanced state of charge information sequence, and N is the sequence length; The step of sending the enhanced state of charge information sequence into the hypernetwork parameter mapping module to generate a sequence of estimated values of circuit element parameters specifically includes: sending the enhanced state of charge information sequence SOC * (k), k=1,2,...,N are sorted into a matrix in ascending order according to the time k. The dimension of the matrix is N*2(Ω+1), each row corresponds to the enhanced state of charge information vector at the time k, and each column corresponds to each dimension of the vector value function. After inputting the matrix into the hypernetwork parameter mapping module, the estimated value sequence ECM of the circuit element parameters is obtained. i (k),k=1,2,...,N,i=1,2,3,4,5; The method of updating the parameters of the hypernetwork parameter mapping module through the forward propagation phase of information and the reverse propagation phase of error specifically includes: in the forward propagation phase of information, firstly, according to the discretization equation of the second-order equivalent circuit model, the estimated value sequence corresponding to the circuit element parameters R0, R1, R2, C1, C2 is converted into a sequence corresponding to the model discretization equation parameters b1, b2, b3, b4, b5, and is input into the model equation network module in sequence at time k as the parameters of the model equation network module; then, the initial value s(0 + ), the scaled input current sequence I(k), k=1,2,...,N and the open circuit voltage information sequence U ocv (k), k=1,2,...,N are input into the model equation networking module to obtain the estimated value sequence of the output voltage; then the estimated value sequence of the circuit element parameters, the estimated value sequence of the output voltage and the real terminal voltage sequence are input into the multidimensional performance optimization regulator together to obtain the output loss value; in the back propagation stage of the error, the loss value output by the multidimensional performance optimization regulator is used to generate update information according to the gradient descent method to correct the parameters of each layer in the super network parameter mapping module; when a complete information forward propagation and error back propagation are completed, the update of the parameters of each layer in the super network parameter mapping module is completed.
9. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 2 is characterized in that: The updating of the open circuit voltage information list by the open circuit voltage compensation calculation module using the estimated value sequence of the circuit element parameters generated by the updated hyper-network parameter mapping module specifically includes: First, the scaled input current sequence, the estimated value sequence of circuit element parameters generated by the updated hypernetwork parameter mapping module, the estimated value sequence of output voltage generated by the model equation networking module, the real terminal voltage sequence, the open circuit voltage information list, and the open circuit voltage information sequence U calculated based on the state of charge information sequence and the open circuit voltage information list are converted into the scaled input current sequence, the estimated value sequence of circuit element parameters generated by the updated hypernetwork parameter mapping module, the estimated value sequence of output voltage generated by the model equation networking module, the real terminal voltage sequence, the open circuit voltage information list, and the open circuit voltage information sequence U calculated based on the state of charge information sequence and the open circuit voltage information list. ocv (k), k=1,2,...,N is input to the open circuit voltage compensation calculation module; Then the open circuit voltage compensation calculation module calculates the voltage compensation term and the current compensation term through the above input information, and their expressions are: I com (k)=b3I(k)+b4I(k-1)+b5I(k-2),k=1,2,...,N Among them, U com (k), k=1,2,...,N is the voltage compensation term, I com (k), k=1,2,...,N is the current compensation term; Then, the open circuit voltage at each moment in the updated open circuit voltage information sequence is calculated based on the voltage compensation term and the current compensation term, and the expression is: in, represents the open circuit voltage value at time k in the open circuit voltage information sequence before updating, represents the open-circuit voltage value at the kth moment after update, ξ and η are the weight coefficients of the voltage compensation term and the current compensation term respectively; Secondly, divide SoC∈[0,100] into θ segments, where the value of θ needs to be given in advance based on experience. j ,SoC j+1 ],j=1,2,...,θ at time k The cumulative update value of the open circuit voltage information list is: in, Indicates the value at position j in the open circuit voltage information list; SoC(k)∈[SoC j ,SoC j+1 ] indicates that the open circuit voltage value at the kth moment in the state of charge information sequence belongs to [SoC j ,SoC j+1 ]scope; Then calculate the SoC(k)∈[SoC j ,SoC j+1 ]The number of points N of the condition j , and calculate the open circuit voltage information list update value according to the following formula: in, represents the updated value of position j in the open circuit voltage information list; Finally, the values of each position in the open circuit voltage information list are limited to obtain the updated open circuit voltage information list: in, Indicates the updated value of position j in the open circuit voltage information list after limiting, and They represent the lower and upper limits of the open circuit voltage respectively.
10. The method for intelligent parameter estimation of lithium battery model driven by dual drive of hypernetwork and mechanism compensation according to claim 1, characterized in that: The step (7) specifically includes the following sub-steps: (7.1) The state of charge of the battery is evenly divided into γ equal parts from 0% to 100%, and arranged in an increasing order to form a state of charge estimation sequence, where γ is determined by the required level of detail of the parameter estimation; (7.2) using the state of charge estimation sequence as input to a hypernetwork parameter mapping module, and the hypernetwork parameter mapping module outputs a sequence of estimated values of circuit element parameters; (7.3) Processing the estimated value sequence of the circuit element parameters according to the scaling factor m, specifically: dividing the estimated values corresponding to the ohmic polarization resistance R0, the concentration polarization resistance R1, and the activation polarization resistance R2 by the scaling factor m, and multiplying the estimated values corresponding to the concentration polarization capacitance C1 and the activation polarization capacitance C2 by the scaling factor m, to obtain the optimal circuit element parameter sequence; (7.4) The optimal circuit element parameter sequence is matched with the charge state input sequence to establish a functional relationship between the two, and together with the open circuit voltage information list, it is used as the optimal functional relationship between the battery model parameters and the state quantity finally obtained. Based on this optimal functional relationship, the battery model parameters corresponding to any charge state are obtained.
