Method, system, device and storage medium for estimating health status of unknown lithium-ion batteries based on CMM0G model
Through the CMMOG model based on convolutional neural network and hybrid multi-gated cycle unit, combined with multiple charge and discharge data, a high-precision lithium-ion battery health status estimation method is constructed, which solves the problem of inaccurate estimation in the existing technology and realizes the Accurate health assessment of unknown batteries.
Patent Information
- Application Number
- CN202410452207.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-16
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-04-16
AI Technical Summary
Existing methods for estimating health status of lithium-ion batteries usually rely on a single electrical characteristics of the charge and discharge stage, and cannot adapt to the working conditions between different samples, resulting in inaccurate estimation.
Using a CMMOG model based on convolutional neural network and hybrid multi-gated cyclic unit, battery data is collected through multiple charge and discharge cycles, main and auxiliary aging indicators are determined, and high-precision health status estimation model is constructed, and aging characteristics are fusion is used for CNN layer. , GRU expert network processing features, sub-task dedicated network allocation weight, deep neural network completes health status estimation.
The high-precision health status estimation of unknown lithium-ion batteries is achieved, and the problems of traditional methods in sample differences and multitasking conflict are overcome, and the generalization ability and performance of the model are improved.
Smart Images

Figure CN118409216B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of lithium batteries, and in particular relates to a method, system, device and storage medium for estimating the health state of an unknown lithium-ion battery based on a CMMOG model. Background Art
[0002] In recent years, with the increasingly serious environmental pollution problem, electric vehicles, as a clean energy means of transportation to replace traditional fuel vehicles, have also received attention from the government and other relevant departments; lithium batteries, as the main power source of electric vehicles, have the advantages of high energy density, high conversion efficiency, and fast response time; the health status of lithium-ion batteries, as an important indicator to measure battery health and safety and ensure the smooth operation of electric vehicles, is an important part of the battery management system. It can ensure the safety and reliability of the battery system, extend battery life, improve energy utilization, and reduce maintenance costs.
[0003] However, the battery health status estimation method in the existing technology usually uses the electrical characteristics of a single charging and discharging stage as an aging indicator, without taking into account the actual vehicle usage conditions; in addition, the battery health status estimation model cannot adapt to the working condition differences between different samples and make accurate generalized estimates. Summary of the invention
[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and to provide a method, system, device and storage medium for estimating the health status of unknown lithium-ion batteries based on the CMMOG model, which can make high-precision estimates of the battery health status of random combination features and non-training operating condition samples.
[0005] The present invention provides the following technical solutions:
[0006] In a first aspect, a method for estimating the health state of an unknown lithium-ion battery based on a CMMOG model is provided, comprising:
[0007] All lithium-ion batteries in the data set are subjected to multiple charge and discharge cycles, and the battery data and health status data of each lithium-ion battery in each charge and discharge cycle are collected; the charge and discharge cycle includes a constant current charging stage, a constant voltage charging stage, and a pulse discharge stage;
[0008] Determine the main aging index and auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle;
[0009] Based on the main aging indicators, auxiliary aging indicators and health status data of lithium-ion batteries in the dataset, a CMMOG health status estimation model based on convolutional neural network and hybrid multi-gated recurrent unit was established;
[0010] The main aging indicators and auxiliary aging indicators of the unknown lithium-ion battery are input into the CMMOG health state estimation model to obtain the health state estimation value of the unknown lithium-ion battery.
[0011] Optionally, all lithium-ion batteries in the data set are subjected to multiple charge and discharge cycles, and battery data and health status data in each charge and discharge cycle are collected.
[0012] The battery health status H of the cth lithium-ion battery in the ith charge and discharge cycle c,i for:
[0013]
[0014] Among them, C c,i is the maximum discharge capacity of the lithium battery in the ith charge and discharge cycle of the cth battery, C c is the rated capacity of the cth lithium battery.
[0015] Optionally, in determining the main aging index and auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle,
[0016] The battery data includes battery voltage, battery current, battery capacity and battery surface temperature during the charging and discharging process;
[0017] The main aging indicators include the IC curve areas in the constant current charging stage and the constant voltage charging stage and the voltage capacity Euclidean distance in the pulse discharge stage; the auxiliary aging indicators include the fuzzy entropy value of the battery surface temperature sequence in the charge and discharge cycle.
[0018] Optionally, the IC curve area CCArea of the cth battery in the constant current charging stage of the i-th charge and discharge cycle c,i for:
[0019]
[0020] Among them, u is the number of the sampling point in the constant current charging stage, Q c,i,u+1 is the battery state of charge of the cth battery at the i-th charge-discharge cycle u+1 sampling point, Q c,i,u is the battery state of charge of the cth battery at the uth sampling point in the ith charge-discharge cycle;
[0021] The IC curve area CVArea of the cth battery in the constant voltage charging stage of the i-th charge and discharge cycle c,i for:
[0022]
[0023] Among them, m is the number of the sampling point in the constant voltage charging stage, Qc,i,m+1 is the battery state of charge of the cth battery at the m+1th sampling point of the ith charge-discharge cycle, Q c,i,m is the battery state of charge of the cth battery at the mth sampling point of the i-th charge and discharge cycle;
[0024] The voltage capacity Euclidean distance ED of the cth battery in the pulse discharge phase of the i-th charge and discharge cycle c,i for:
[0025]
[0026] Among them, q is the initial sequence number of the pulse discharge, t is the data sampling interval during charge and discharge, z is the number of sampling times, Q c,i,q×t is the initial battery state of charge of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle; U c,i,q×t is the initial voltage of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle; Q c,i,(q+1)×t is the battery state of charge of the cth battery at the i-th charge-discharge cycle (q+1)×t sampling time, U c,i,(q+1)×t is the voltage of the cth battery at the i-th charge / discharge cycle (q+1)×t sampling time, Q c,i,z×t is the final battery state of charge of the cth battery at the pulse discharge stage of the i-th charge and discharge cycle, U c,i,z×t is the termination voltage of the cth battery in the pulse discharge stage of the ith charge and discharge cycle.
[0027] Optionally, in the CMMOG health state estimation model based on convolutional neural network and hybrid multi-gated cyclic unit is established based on the main aging indicators, auxiliary aging indicators and health state data of lithium-ion batteries in the data set,
[0028] The CMMOG health status estimation model includes a CNN layer, a GRU expert network layer and a subtask-specific network layer;
[0029] The CNN layer is used to fuse the input aging features of the lithium-ion battery into a high-dimensional battery health feature sequence; the aging features are the main aging indicators and auxiliary aging indicators of the lithium-ion battery;
[0030] The GRU expert network layer has at least two GRU expert network models, each of which is used to process the input high-dimensional battery health features and output an output result corresponding to the subtask;
[0031] The subtask-specific network layer is used to use the gating network corresponding to the subtask to allocate the mapping weights of all expert network models in each subtask; and determine the final output result of each subtask based on the mapping weights of all expert network models in each subtask and the output result corresponding to the subtask output by each expert network model; in the training process of the CMMOG health status estimation model, the final output result of the subtask is the health status value of the lithium-ion battery corresponding to the subtask, and in the testing process of the CMMOG health status estimation model, based on the final output result of each subtask, the subtask module length is mapped through a deep neural network to complete the health status estimation of the unknown lithium-ion battery.
[0032] Optionally, the method is used to fuse the input aging characteristics of the lithium-ion battery into a high-dimensional battery health feature sequence,
[0033] The aging features are fused into a high-dimensional battery health feature sequence through 5×5 convolution, maximum pooling, 3×3 convolution, and average pooling operations in the CNN layer;
[0034] The final output result of each subtask is determined according to the mapping weights of all expert network models in each subtask and the output result corresponding to the subtask output by each expert network model.
[0035] The final output result y of the mth subtask m for:
[0036] y m =Liner m (f m (x)
[0037]
[0038] g m (x) = softmax(W gm x)
[0039] Among them, Liner m (·) is the linear mapping network output of the mth subtask, f m (x) is the weighted output of all GRU expert networks for the mth subtask, GRU(·) o is the oth GRU computing unit, G is the total number of GRU computing units, g m (x) is the mapping weight of all expert network models in the mth subtask, W gm is the weight of the gating network of the mth subtask, softmax(·) is the normalized exponential function, and x is the high-dimensional battery health feature sequence output by the CNN layer.
[0040] Optionally, in the CMMOG health state estimation model based on convolutional neural network and hybrid multi-gated cyclic unit is established based on the main aging indicators, auxiliary aging indicators and health state data of lithium-ion batteries in the data set,
[0041] The data set is divided into a training set and a test set. When establishing the CMOG health state estimation model, the training set is used for training, and the test set is used to test the trained CMOG health state estimation model.
[0042] The objective optimization function of the CMM0G health status estimation model during training is:
[0043]
[0044] Among them, Loss(Sample c-train ,label,σ) is the homoscedasticity loss between the battery health state value and the health state estimation value, N C is the total number of samples, label is the health status value of the cth battery sample in the training set, CMMOG (Sample c-train ) is the health status estimation value of the c-th battery sample input to the CMMOG health status estimation model, σ task is the noise regularization term of the task.
[0045] In a second aspect, a health status estimation system for an unknown lithium-ion battery based on a CMMOG model is provided, comprising:
[0046] An aging feature acquisition module is used to obtain the main aging indicators and auxiliary aging indicators of unknown lithium-ion batteries;
[0047] An estimation module, used for inputting the main aging index and the auxiliary aging index of the unknown lithium-ion battery into the CMMOG health state estimation model output by the model building module to obtain the health state estimation value of the unknown lithium-ion battery;
[0048] The model building module includes an acquisition unit, a determination unit and a building unit;
[0049] An acquisition unit is used to perform multiple charge and discharge cycles on all lithium-ion batteries in the data set, and collect battery data and health status data of each lithium-ion battery in each charge and discharge cycle; the charge and discharge cycle includes a constant current charging stage, a constant voltage charging stage, and a pulse discharge stage;
[0050] A determination unit, used to determine the main aging index and the auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle;
[0051] A construction unit is used to establish a CMMOG health state estimation model based on a convolutional neural network and a hybrid multi-gated cyclic unit based on the main aging indicators, auxiliary aging indicators and health state data of lithium-ion batteries in the data set.
[0052] In a third aspect, a computer device is provided, comprising a processor and a memory; wherein, when the processor executes a computer program stored in the memory, the steps of the method for estimating the health status of an unknown lithium-ion battery based on the CMMOG model described in any one of the first aspects are implemented.
[0053] In a fourth aspect, a computer-readable storage medium is provided for storing a computer program; when the computer program is executed by a processor, the steps of the method for estimating the health state of an unknown lithium-ion battery based on the CMMOG model described in any one of the first aspects are implemented.
[0054] Compared with the prior art, the present invention has the following beneficial effects:
[0055] (1) The present invention proposes to construct a CMM0G multi-task learning model, and obtains model weights by training a set of battery data for health state estimation of unknown lithium-ion batteries. The CMM0G model of the present invention first uses CNN to construct a high-dimensional feature extractor as a shared layer for multi-task learning, and then introduces a multi-gated network to deploy multiple GRU expert networks to achieve the best mapping weights and form a task-specific layer. Finally, a deep neural network is used to map the sub-task module length to complete the state estimation.
[0056] (2) The present invention proposes a fuzzy entropy of the battery surface temperature sequence that is weakly correlated with the battery health status as an auxiliary feature. The fuzzy entropy of the battery surface temperature sequence in each charging and discharging stage is calculated to reflect the complexity of the battery surface temperature change. Unlike traditional electrical features, it can periodically add disturbances to the model, which greatly improves the multi-task processing capability of CMMOG.
[0057] (3) Unlike traditional multi-task learning that uses hard connections, which may cause conflicts between tasks and degrade model performance due to different sample distributions, the present invention introduces a multi-gated network with a soft connection mechanism for multi-task learning, and dynamically adjusts the output weights of the expert network for each group of subtasks to alleviate this problem.
[0058] (4) In order to solve the problem that the use of the average loss function for multi-task differences leads to a decrease in model performance, the present invention constructs a homoscedasticity loss function to collaboratively process the output losses of multiple regression tasks. The homoscedasticity loss function adds a noise regularization term and brings it into the model training to dynamically adjust the regularization term. While reducing the risk of overfitting in the model, the loss output of each model is reasonably allocated. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] Figure 1 It is a flow chart of a method for estimating the health status of an unknown lithium-ion battery based on a CMMOG model of the present invention;
[0060] Figure 2 It is the IC curve of batteries with different rates during constant current charging stage;
[0061] Figure 3 It is the IC curve of batteries at different rates during constant voltage charging stage;
[0062] Figure 4 is the aging characteristic value of each cycle area extracted during the constant current charging stage;
[0063] Figure 5 is the aging characteristic value of each cycle area extracted during the constant voltage charging stage;
[0064] Figure 6 is the aging characteristic value of the Euclidean distance of each cycle extracted during the discharge pulse stage;
[0065] Figure 7 yes Figure 6 A partial enlarged view of
[0066] Figure 8 is the characteristic value of temperature entropy aging in each cycle during the constant current charging stage;
[0067] Fig. 9 is the characteristic value of temperature entropy aging in each cycle during the constant voltage charging stage;
[0068] Fig.10 is the characteristic value of temperature entropy aging in each cycle during the discharge pulse stage;
[0069] Fig.11 This is a comparison chart of the estimated effects of CMMOG and other existing models under 0.2C charging and 1C discharging;
[0070] Fig.12 This is a partial enlarged view of the 50% charge and discharge cycle of CMM0G and other existing models under 0.2C charge and 1C discharge;
[0071] Fig.13 This is a comparison chart of the generalization performance test of non-training samples with 0.3C charging and 1C discharging under different models;
[0072] Fig.14 This is a comparison chart of the generalization performance test of the non-training samples of 0.3C charging and 1C discharging in the last 50% of the charging and discharging cycles under different models;
[0073] Fig.15 This is the theoretical network structure diagram of the CMMOG health status estimation training model;
[0074] Fig.16is a network structure diagram of the CMMOG health status estimation model during the test of this embodiment; DETAILED DESCRIPTION
[0075] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.
[0076] like Figure 1 As shown, a method for estimating the health state of an unknown lithium-ion battery based on a CMM0G model is provided, comprising the following steps:
[0077] S1: All lithium-ion batteries in the data set are subjected to multiple charge and discharge cycles, and the battery data and health status data of each lithium-ion battery in each charge and discharge cycle are collected; the charge and discharge cycle includes a constant current charging stage, a constant voltage charging stage, and a pulse discharge stage.
[0078] Specifically, the battery data includes battery voltage, battery current, battery capacity and battery surface temperature during the charging and discharging process.
[0079] The charge and discharge voltage data of the i-th charge and discharge cycle of the c-th battery is the sequence U c,i,0 ,U c,i,t ,...,U c,i,k×t ,...,U c,i,q×t ,...U c,i,z×t , the charge and discharge current data of the i-th charge and discharge cycle is sequence I c,i,0 ,I c,i,t ,...,I c,i,k×t ,...,I c,i,q×t ...,I c,i,z×t , the charge and discharge capacity data is a sequence Q c,i,0 ,Q c,i,t ,...,Q c,i,k×t ,...Q c,i,q×t ...,Q c,i,z×t , charging and discharging battery surface temperature series T c,i,0 ,T c,i,t ,...,T c,i,k×t ,...T c,i,q×t ...,T c,i,z×t .U c,i,0 ,U c,i,k×t ,U c,i,q×t ,U c,i,z×t are the initial voltage of the constant current stage, the initial voltage of the constant voltage stage, the initial voltage of the discharge pulse, and the termination voltage of the discharge pulse of the i-th charge and discharge cycle of the c-th battery respectively; I c,i,0 ,I c,i,k×t ,I c,i,q×t ,I c,i,z×tare the initial current of the constant current stage, the initial current of the constant voltage stage, the initial current of the discharge pulse, and the termination current of the discharge pulse of the i-th charge and discharge cycle of the c-th battery; Q c,i,0 ,Q c,i,k×t ,Q c,i,q×t ,Q c,i,z×t is the initial capacity of the constant current stage, the initial capacity of the constant voltage stage, the initial capacity of the discharge pulse, and the termination capacity of the discharge pulse of the i-th charge and discharge cycle of the c-th battery; T c,i,0 ,T c,i,k×t ,T c,i,q×t ,T c,i,z×t are the initial surface temperature of the constant current stage, the initial surface temperature of the constant voltage stage, the initial surface temperature of the discharge pulse, and the termination surface temperature of the i-th charge and discharge cycle of the c-th battery; k is the initial serial number of the constant voltage stage, q is the initial serial number of the discharge pulse, t is the data sampling interval during charge and discharge, z is the number of sampling times, and z×t is the total sampling time, which is also the total charging time.
[0080] The battery health status H of the cth lithium-ion battery in the ith charge and discharge cycle c,i for:
[0081]
[0082] Among them, C c,i is the maximum discharge capacity of the lithium battery in the i-th (i=1,2,...,n) charge-discharge cycle of the c-th (c=1,2,...,N)-th battery, C c is the rated capacity of the cth lithium battery, N is the total number of batteries, and n is the number of charge and discharge cycles.
[0083] S2: Determine the main aging index and the auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle.
[0084] The main aging indicators include the IC curve areas in the constant current charging stage and the constant voltage charging stage and the voltage capacity Euclidean distance in the pulse discharge stage; the auxiliary aging indicators include the fuzzy entropy value of the battery surface temperature sequence in the charge and discharge cycle.
[0085] Step S2.1, determining the IC curve area during the constant current charging stage.
[0086] Specifically, Figure 2 As shown, the IC curve CCIC of the cth battery at the uth sampling point in the i-th charge and discharge cycle during the constant current stage is calculated. c,i,u for:
[0087]
[0088] Q c,i,u =Ic,i,0 ×t+I c,i,t ×t+...+I c,i,u×t ×t
[0089] Wherein, u=1,2,...,k-1 is the number of the sampling point in the constant current stage; Q c,i,u is the battery state of charge of the cth battery at the uth sampling point in the ith charge and discharge cycle.
[0090] like Figure 4 As shown, the IC curve area CCArea of the cth battery in the constant current charging stage of the i-th charge and discharge cycle c,i for:
[0091]
[0092] Among them, u is the number of the sampling point in the constant current stage, Q c,i,u+1 is the battery state of charge of the cth battery at the i-th charge-discharge cycle u+1 sampling point, Q c,i,u is the battery state of charge of the cth battery at the uth sampling point in the ith charge and discharge cycle.
[0093] 0.2C1D, 0.2C0.5D, 0.5C1D, and 0.5C0.5D represent four lithium battery samples with charging rates of 0.2C and discharging at 1C, charging rates of 0.2C and discharging at 0.5C, charging rates of 0.5C and discharging at 1C, and charging rates of 0.5C and discharging at 0.5C, respectively.
[0094] Step S2.2, determining the IC curve area during the constant voltage charging stage.
[0095] Specifically, Figure 3 As shown, the IC curve CVIC of the cth battery at the mth sampling point of the i-th charge and discharge cycle in the constant voltage stage is calculated. c,i,m for:
[0096]
[0097] Q c,i,m =Q c,i,k-1 +I c,i,k×t ×t+I c,i,(k+1)×t ×t+...+I c,i,m×t ×t
[0098] Where m=k, k+1,...,q-1 is the number of the sampling point in the constant pressure stage, Q c,i,m is the battery state of charge of the cth battery at the mth sampling point in the ith charge and discharge cycle.
[0099] like Figure 5 As shown, the IC curve area CVArea of the cth battery in the constant voltage charging stage of the i-th charge and discharge cyclec,i for:
[0100]
[0101] Among them, m is the number of the sampling point in the constant current stage, Q c,i,m+1 is the battery state of charge of the cth battery at the m+1th sampling point of the ith charge-discharge cycle, Q c,i,m is the battery state of charge of the cth battery at the mth sampling point in the ith charge and discharge cycle.
[0102] Step S2.3, determining the voltage-capacity Euclidean distance during the pulse discharge phase.
[0103] Specifically, Figure 6 and 7 As shown ( Figure 7 for Figure 6 The local enlarged schematic diagram of the cycle number of 200-550) shows the voltage capacity Euclidean distance ED of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle. c,i for:
[0104]
[0105] Among them, q is the initial sequence number of the pulse discharge, t is the data sampling interval during charge and discharge, z is the number of sampling times, Q c,i,q×t is the initial battery state of charge of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle; U c,i,q×t is the initial voltage of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle; Q c,i,(q+1)×t is the battery state of charge of the cth battery at the i-th charge-discharge cycle (q+1)×t sampling time, U c,i,(q+1)×t is the voltage of the cth battery at the i-th charge / discharge cycle (q+1)×t sampling time, Q c,i,z×t is the final battery state of charge of the cth battery at the pulse discharge stage of the i-th charge and discharge cycle, U c,i,z×t is the termination voltage of the cth battery in the pulse discharge stage of the ith charge and discharge cycle.
[0106] Step S2.4, determining the fuzzy entropy value of the battery surface temperature sequence of the charge and discharge cycle.
[0107] like Figure 8-10 As shown, the existing temperature series T c,i,0 ,T c,i,t ,...,T c,i,k×t ,...T c,i,q×t ...,T c,i,z×t , divided into constant current charging stage T c,i,0 ,T c,i,t ,...,T c,i,(k-1)×t, Constant voltage charging stage T c,i,k×t ,T c,i,(k+1)×t ...,T c,i,(q-1)×t , pulse discharge stage T c,i,q×t ,T c,i,(q+1)×t ...,T c,i,z×t , the temperature fuzzy entropy of each stage is calculated as follows.
[0108] The temperature fuzzy entropy calculation method of each stage is as follows:
[0109] Step S2.4.1: Given an L-dimensional temperature sequence [T(1), T(2), …, T(L)], L is the length of the temperature sequence;
[0110] Step S2.4.2: Reconstruct phase space: define the phase space dimension p (p≤N-2),
[0111] x(v)=[T(v),T(v+1),...,T(v+p-1)]-T0(v),v=1,2,...L-p+1
[0112] in, is the mean of p consecutive T(i).
[0113] Step S2.4.3: Introduce fuzzy membership function:
[0114]
[0115] Where r is the similarity tolerance. Generally, r = 0.15*sd, and sd is the standard deviation of the analyzed temperature series [T(1), T(2), …, T(L)].
[0116] Step S2.4.4: Calculate the maximum absolute distance between window vectors x(v) and x(j)
[0117]
[0118] Among them, x(v) and x(j) are the maximum values of the differences with the corresponding endpoints.
[0119] Step S2.4.5: For v = 1, 2, ..., L-p+1, calculate
[0120] And j≠i
[0121] Step S2.4.6: For each v, find the average value to get
[0122]
[0123] Step S2.4.7: Definition
[0124] Step S2.4.8: Therefore, the fuzzy entropy of the original temperature sequence [T(1), T(2), …, T(L)] is:
[0125]
[0126] Finally, the temperature fuzzy entropy CCFE in the constant current charging stage is obtained c,i , Temperature fuzzy entropy CVFE in constant voltage charging stage c,i , Temperature fuzzy entropy DCFE in the discharge pulse stage c,i , collectively referred to as FE c,i .
[0127] S3: Based on the main aging indicators, auxiliary aging indicators and health status data of lithium-ion batteries in the dataset, a CMMOG health status estimation model based on convolutional neural network and hybrid multi-gated recurrent unit is established.
[0128] like Fig.15 and 16 As shown, the CMMOG health status estimation model includes a CNN layer, a GRU expert network layer and a subtask-specific network layer.
[0129] The CNN layer is used to fuse the input aging features of the lithium-ion battery into a high-dimensional battery health feature sequence; the aging features are the main aging indicators and auxiliary aging indicators of the lithium-ion battery.
[0130] Aging characteristics c,i ,CVArea c,i ,ED c,i ,FE c,i The input CNN network is fused into a high-dimensional battery health feature sequence through 5×5 convolution, maximum pooling, 3×3 convolution, and average pooling operations to form a network parameter sharing layer. The calculation method of convolution and pooling operations is as follows:
[0131]
[0132]
[0133] in, The numerical matrix representing the g-th channel of the l-th layer; Represents the g-th convolution channel of the l-1th layer, Represents the bias of the g-th channel in the l-th layer; represents the i-th convolution kernel of the l-1th layer, Cov2d(·) is a two-dimensional convolution operation, that is, a matrix multiplication operation; is the value of the Ωth point in the kth channel of the lth layer, Represents the numerical matrix of the Ωth region of the gth channel in layer l-1; Pooling(·) is a pooling operation, which includes maximum pooling and average pooling. Max(·) represents the maximum value of the matrix region, and Avg(·) represents the mean value of the matrix region.
[0134] There are at least two GRU expert network models in the GRU expert network layer, and each of the GRU expert network models is used to process the input high-dimensional battery health features and output an output result corresponding to the subtask.
[0135] Use GRU to establish the mapping relationship between the health feature sequence and the health status value; the specific formula is:
[0136] r d =σ(W r *[h d-1 ,x d ]+b r )
[0137] j d =σ(W d *[h d-1 ,x d ]+b j )
[0138]
[0139]
[0140] Among them, r d and j d represents the reset gate and update gate, and h d Respectively represent hidden state information and output gate; W r , W d , and b r , b j , Respectively represent the weight and bias of the corresponding gate; x d represents input information, σ(·) and tanh(·) represent the sigmoid activation function and the hyperbolic tangent activation function respectively; '*' represents matrix multiplication, and '·' represents matrix dot multiplication.
[0141] The number of subtasks in this embodiment is 4, and the number of GRU expert network models is 3, that is, the subtasks in this embodiment are to learn four batteries with different charge and discharge rates.
[0142] The subtask-specific network layer is used to use the gating network corresponding to the subtask to allocate the mapping weights of all expert network models in each subtask; and determine the final output result of each subtask based on the mapping weights of all expert network models in each subtask and the output result corresponding to the subtask output by each expert network model.
[0143] The number of subtasks is consistent with the number of gating networks and corresponds one to one. During the training process of the CMM0G health status estimation model, the final output result of the subtask is the health status value of the lithium-ion battery corresponding to the subtask. During the testing process of the CMM0G health status estimation model, based on the final output result of each subtask, the subtask module length is mapped through a deep neural network to complete the health status estimation of the unknown lithium-ion battery.
[0144] The final output result y of the mth subtask m for:
[0145] y m =Liner m (f m (x))
[0146]
[0147] g m (x) = softmax(W gm x)
[0148] Among them, Liner m (·) is the linear mapping network output of the mth subtask, f m (x) is the weighted output of all GRU expert networks for the mth subtask, GRU(·) o is the oth GRU computing unit, G is the total number of GRU computing units, 1≤m≤N p ,1≤o≤G,N p is the number of multi-task tasks, that is, the number of batteries for a single training, g m (x) is the mapping weight of all expert network models in the mth subtask, W gm is the weight of the gating network of the mth subtask, softmax(·) is the normalized exponential function, x is the high-dimensional battery health feature sequence output by the CNN layer, and N in this example p =4, G=3.
[0149] The data set is divided into a training set and a test set. When establishing the CMOG health state estimation model, the training set is used for training, and the test set is used to test the trained CMOG health state estimation model.
[0150] Specifically, during training, the first 50% of the data of the aging feature sequence and the health state sequence are taken to construct the training set: and The weight information of the sample charge and discharge interval estimation model is obtained by training the CMM0G model.
[0151] Optimization objective function:
[0152]
[0153] Among them, Loss(Sample c-train ,label,σ) is the homoscedasticity loss between the battery health state value and the health state estimation value, N C is the total number of samples, label is the health status value of the cth battery sample in the training set, CMMOG (Sample c-train ) is the health status estimation value of the c-th battery sample input to the CMMOG health status estimation model, σ task is the noise regularization term of the task.
[0154] Through back propagation, the target optimization function reaches the minimum value of training.
[0155] The last 50% of the data will be aged Bring in the trained model to get the predicted value
[0156] Among them, the lithium battery health status estimation interval is the training sample Sample c , the health status of all lithium batteries of the same specification within the charge and discharge rate range of 1≤c≤N is estimated.
[0157] Specifically, during the test,
[0158] The aging features of the test set Bring it into the training network to get the health status estimation result Among them, Test e is the e-th test sample, and Len is the sequence length of the test sample.
[0159] The present invention overcomes the shortcoming of traditional methods that repeated training is required when conducting health status estimation for unknown batteries. It can conduct accurate health status estimation for unknown batteries, and achieve the effect of estimating the health status of all battery samples with different charge and discharge rates within this interval by training a group of data on batteries with different charge and discharge rates.
[0160] S4: Input the main aging index and auxiliary aging index of the unknown lithium-ion battery into the CMM0G health state estimation model to obtain the health state estimation value of the unknown lithium-ion battery.
[0161] When using the CMMOG health state estimation model to estimate the health state of an unknown lithium-ion battery, the subtask module lengths are mapped through a deep neural network based on the final output results of each subtask output by the subtask-specific network layer (specifically, the mapping relationship can be set according to expert experience, averaged or weighted) to complete the health state estimation of the unknown lithium-ion battery.
[0162] In order to demonstrate the feasibility and accuracy of the unknown lithium-ion battery health status estimation method based on the CMMOG model of the present invention, the present invention uses the following examples to illustrate, and compares the errors of various algorithms to illustrate the accuracy of the present invention.
[0163] The present invention conducts a battery aging experiment on 18650 lithium-ion batteries by using a battery testing device (Xinwei BTS-4000) and a thermostat (Xinwei MHW-200). The battery is charged at a constant current of 1.2A (0.5C) until the terminal voltage reaches 4.2V, then the terminal voltage is maintained at 4.2V until the charging current gradually decreases to 48mA, and finally discharged at a constant current of 2.4A (1C) until the terminal voltage drops to 3V. In addition, the temperature of the entire charging and discharging test process is controlled at 25°C.
[0164] The experimental data uses the charge and discharge data of 0.2C charging and 1C discharging, 0.2C charging and 0.5C discharging, 0.5C charging and 1C discharging, and 0.5C charging and 0.5C discharging. After extracting the aging characteristics, use CCArea, CVArea, ED i and FE i These aging characteristics are used as the input of the CMOG model. The first 50% charging data is used to train the model, and the subsequent 50% charging data is used to verify the performance of the model.
[0165] like Fig.11 The figure shows the comparison results of 0.2C charging and 1C discharging. In the comparative experiment, the RMSE values of LSTM, RNN, GRU and SVR are 0.6265%, 0.6264%, 0.5949% and 0.6684% respectively, while the RMSE of CMMOG model is only 0.1559%, which is the lowest. In addition, for the other five candidate models, the R 2 It is 0.9877, the highest among the five models. Table 1 shows the performance comparison of each model. Fig.12 for Fig.11 A partial enlargement of the last 50% of the charge and discharge cycles.
[0166] Table 1 Performance comparison of various models at 0.2C charging and 1C discharging rates
[0167] Estimation Model <![CDATA[R 2 ]]> MAE(%) RMSE(%) MAPE(%) CMMOG 0.9933 2.4306 0.1559 1.4151 LSTM 0.8905 0.4562 0.6265 0.5085 RNN 0.8905 0.4541 0.6264 0.5074 GRU 0.9013 0.5256 0.5949 0.5934
[0168] In order to verify the ability of CMMOG to estimate the health status of unknown lithium-ion batteries, the weights of the 0.2C charging and 1C discharging samples are used as the basis for testing the LSTM, GRU, RNN and SVR models. The 0.3C charging and 1C discharging samples are tested and verified. The results are as follows: Fig.13 As shown, Fig.14 for Fig.13 The local enlarged image of the last 50% charge and discharge cycles. Table 2 is a comparison of the generalization performance indicators of each model. In the comparative experiment, the RMSE values of LSTM, RNN, GRU and SVR are 0.9890%, 0.4538%, 0.4291% and 1.1637% respectively, while the RMSE of the CMMOG model is only 0.2885%, which is the lowest. In addition, for the other five candidate models, the R 2 The value is 0.9539, which is the highest among the five models. Therefore, the generalization ability of the CMMOG model is better than that of other models.
[0169] Table 2 Comparison of generalization performance of each model at 0.3C charging and 1C discharging rate
[0170] Model <![CDATA[R 2 ]]> MAE(%) RMSE(%) MAPE(%) CMMOG 0.9539 0.1913 0.2885 0.0021 LSTM 0.4586 0.8430 0.9890 0.0094 GRU 0.8981 0.3628 0.4291 0.0039 RNN 0.8860 0.3657 0.4538 0.0041 SVR 0.2506 1.0401 1.1637 0.0115
[0171] In another embodiment, a system for estimating the health status of an unknown lithium-ion battery based on a CMMOG model is provided, comprising:
[0172] An aging feature acquisition module is used to obtain the main aging indicators and auxiliary aging indicators of unknown lithium-ion batteries;
[0173] An estimation module, used for inputting the main aging index and the auxiliary aging index of the unknown lithium-ion battery into the CMMOG health state estimation model output by the model building module to obtain the health state estimation value of the unknown lithium-ion battery;
[0174] The model building module includes an acquisition unit, a determination unit and a building unit;
[0175] An acquisition unit is used to perform multiple charge and discharge cycles on all lithium-ion batteries in the data set, and collect battery data and health status data of each lithium-ion battery in each charge and discharge cycle; the charge and discharge cycle includes a constant current charging stage, a constant voltage charging stage, and a pulse discharge stage;
[0176] A determination unit, used to determine the main aging index and the auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle;
[0177] A construction unit is used to establish a CMMOG health state estimation model based on a convolutional neural network and a hybrid multi-gated cyclic unit based on the main aging indicators, auxiliary aging indicators and health state data of lithium-ion batteries in the data set.
[0178] For more specific processes of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.
[0179] In another embodiment, the present invention provides a computer device, including a processor and a memory; wherein, when the processor executes a computer program stored in the memory, the steps of the above-mentioned method for estimating the health status of an unknown lithium-ion battery based on the CMMOG model are implemented.
[0180] For more specific processes of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.
[0181] In another embodiment, the present invention provides a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the steps of the above-mentioned method for estimating the health status of an unknown lithium-ion battery based on the CMMOG model are implemented.
[0182] For more specific processes of the above method, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.
[0183] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system, device, and storage medium disclosed in the embodiment, since they correspond to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0184] Those skilled in the art can clearly understand that the technology in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution in the embodiments of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention or some parts of the embodiments.
[0185] The above are only preferred embodiments 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 technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. Health status estimation method of unknown lithium-ion batteries based on CMM0G model, It is characterized in that include: All lithium-ion batteries in the data set are subjected to multiple charge and discharge cycles, and the battery data and health status data of each lithium-ion battery in each charge and discharge cycle are collected; the charge and discharge cycle includes a constant current charging stage, a constant voltage charging stage, and a pulse discharge stage; Determine the main aging index and auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle; Based on the main aging indicators, auxiliary aging indicators and health status data of lithium-ion batteries in the dataset, a CMMOG health status estimation model based on convolutional neural network and hybrid multi-gated recurrent unit was established; Input the main aging index and auxiliary aging index of the unknown lithium-ion battery into the CMM0G health state estimation model to obtain the health state estimation value of the unknown lithium-ion battery; In the step of determining the main aging index and auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle, The battery data includes battery voltage, battery current, battery capacity and battery surface temperature during the charging and discharging process; The main aging indicators include the IC curve area in the constant current charging stage and the constant voltage charging stage and the voltage capacity Euclidean distance in the pulse discharge stage; the auxiliary aging indicators include the fuzzy entropy value of the battery surface temperature sequence in the charge and discharge cycle; In the method, based on the main aging indicators, auxiliary aging indicators and health status data of lithium-ion batteries in the data set, a CMMOG health status estimation model based on a convolutional neural network and a hybrid multi-gated cyclic unit is established. The CMMOG health status estimation model includes a CNN layer, a GRU expert network layer and a subtask-specific network layer; The CNN layer is used to fuse the input aging features of the lithium-ion battery into a high-dimensional battery health feature sequence; the aging features are the main aging indicators and auxiliary aging indicators of the lithium-ion battery; The GRU expert network layer has at least two GRU expert network models, each of which is used to process the input high-dimensional battery health features and output an output result corresponding to the subtask; The subtask-specific network layer is used to allocate the mapping weights of all expert network models in each subtask using the gating network corresponding to the subtask; And according to the mapping weights of all expert network models in each subtask and the output results corresponding to the subtask output by each expert network model, the final output result of each subtask is determined; During the training process of the CMM0G health status estimation model, the final output result of the subtask is the health status value of the lithium-ion battery corresponding to the subtask. During the testing process of the CMM0G health status estimation model, based on the final output result of each subtask, the subtask module length is mapped through a deep neural network to complete the health status estimation of the unknown lithium-ion battery.
2. The method for estimating the health status of an unknown lithium-ion battery based on the CMM0G model according to claim 1, It is characterized in that All lithium-ion batteries in the data set are subjected to multiple charge and discharge cycles, and battery data and health status data in each charge and discharge cycle are collected. The battery health status H of the cth lithium-ion battery in the ith charge and discharge cycle c,i for: Among them, C c,i is the maximum discharge capacity of the lithium battery in the ith charge and discharge cycle of the cth battery, C c is the rated capacity of the cth lithium battery.
3. The method for estimating the health status of an unknown lithium-ion battery based on the CMM0G model according to claim 1, It is characterized in that IC curve area CCArea of the cth battery in the constant current charging stage of the i-th charge and discharge cycle c,i for: Among them, u is the number of the sampling point in the constant current charging stage, Q c,i,u+1 is the battery state of charge of the cth battery at the i-th charge-discharge cycle u+1 sampling point, Q c,i,u is the battery state of charge of the cth battery at the uth sampling point in the ith charge-discharge cycle; The IC curve area CVArea of the cth battery in the constant voltage charging stage of the i-th charge and discharge cycle c,i for: Among them, m is the number of the sampling point in the constant voltage charging stage, Q c,i,m+1 is the battery state of charge of the cth battery at the m+1th sampling point of the ith charge-discharge cycle, Q c,i,m is the battery state of charge of the cth battery at the mth sampling point of the i-th charge and discharge cycle; The voltage capacity Euclidean distance ED of the cth battery in the pulse discharge phase of the i-th charge and discharge cycle c,i for: Among them, q is the initial sequence number of the pulse discharge, t is the data sampling interval during charge and discharge, z is the number of sampling times, Q c,i,q×t is the initial battery state of charge of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle; U c,i,q×t is the initial voltage of the cth battery in the pulse discharge stage of the i-th charge and discharge cycle; Q c,i,(q+1)×t is the battery state of charge of the cth battery at the i-th charge-discharge cycle (q+1)×t sampling time, U c,i,(q+1)×t is the voltage of the cth battery at the i-th charge / discharge cycle (q+1)×t sampling time, Q c,i,z×t is the final battery state of charge of the cth battery at the pulse discharge stage of the i-th charge and discharge cycle, U c,i,z×t is the termination voltage of the cth battery in the pulse discharge stage of the ith charge and discharge cycle.
4. The method for estimating the health status of an unknown lithium-ion battery based on the CMM0G model according to claim 1, It is characterized in that The method is used to fuse the input aging characteristics of lithium-ion batteries into a high-dimensional battery health feature sequence, The aging features are fused into a high-dimensional battery health feature sequence through 5×5 convolution, maximum pooling, 3×3 convolution, and average pooling operations in the CNN layer; The final output result of each subtask is determined according to the mapping weights of all expert network models in each subtask and the output result corresponding to the subtask output by each expert network model. The final output result y of the mth subtask m for: y m =Liner m (f m (x)) g m (x)=softmax(W gm x) Among them, Liner m (·) is the linear mapping network output of the mth subtask, f m (x) is the weighted output of all GRU expert networks for the mth subtask, GRU(·) o is the oth GRU computing unit, G is the total number of GRU computing units, g m (x) is the mapping weight of all expert network models in the mth subtask, W gm is the weight of the gating network of the mth subtask, softmax(·) is the normalized exponential function, and x is the high-dimensional battery health feature sequence output by the CNN layer.
5. The method for estimating the health status of an unknown lithium-ion battery based on the CMM0G model according to claim 1, It is characterized in that In the method, based on the main aging indicators, auxiliary aging indicators and health status data of lithium-ion batteries in the data set, a CMMOG health status estimation model based on a convolutional neural network and a hybrid multi-gated cyclic unit is established. The data set is divided into a training set and a test set. When establishing the CMOG health state estimation model, the training set is used for training, and the test set is used to test the trained CMOG health state estimation model. The objective optimization function of the CMM0G health status estimation model during training is: Among them, Loss(Sample c-train ,label,σ) is the homoscedasticity loss between the battery health state value and the health state estimation value, N C is the total number of samples, label is the health status value of the cth battery sample in the training set, CMMOG (Sample c-train ) is the health status estimation value of the c-th battery sample input to the CMMOG health status estimation model, σ task is the noise regularization term of the task.
6. A health status estimation system for unknown lithium-ion batteries based on CMMOG model, It is characterized in that include: An aging feature acquisition module is used to obtain the main aging indicators and auxiliary aging indicators of unknown lithium-ion batteries; An estimation module, used for inputting the main aging index and the auxiliary aging index of the unknown lithium-ion battery into the CMMOG health state estimation model output by the model building module to obtain the health state estimation value of the unknown lithium-ion battery; The model building module includes an acquisition unit, a determination unit and a building unit; An acquisition unit is used to perform multiple charge and discharge cycles on all lithium-ion batteries in the data set, and collect battery data and health status data of each lithium-ion battery in each charge and discharge cycle; the charge and discharge cycle includes a constant current charging stage, a constant voltage charging stage, and a pulse discharge stage; A determination unit, used to determine the main aging index and the auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle; A construction unit is used to establish a CMMOG health state estimation model based on a convolutional neural network and a hybrid multi-gated cyclic unit based on the main aging indicators, auxiliary aging indicators and health state data of lithium-ion batteries in the data set; In the step of determining the main aging index and auxiliary aging index of the lithium-ion battery according to the battery data of each lithium-ion battery in each charge and discharge cycle, The battery data includes battery voltage, battery current, battery capacity and battery surface temperature during the charging and discharging process; The main aging indicators include the IC curve area in the constant current charging stage and the constant voltage charging stage and the voltage capacity Euclidean distance in the pulse discharge stage; the auxiliary aging indicators include the fuzzy entropy value of the battery surface temperature sequence in the charge and discharge cycle; In the method, based on the main aging indicators, auxiliary aging indicators and health status data of lithium-ion batteries in the data set, a CMMOG health status estimation model based on a convolutional neural network and a hybrid multi-gated cyclic unit is established. The CMMOG health status estimation model includes a CNN layer, a GRU expert network layer and a subtask-specific network layer; The CNN layer is used to fuse the input aging features of the lithium-ion battery into a high-dimensional battery health feature sequence; the aging features are the main aging indicators and auxiliary aging indicators of the lithium-ion battery; The GRU expert network layer has at least two GRU expert network models, each of which is used to process the input high-dimensional battery health features and output an output result corresponding to the subtask; The subtask-specific network layer is used to allocate the mapping weights of all expert network models in each subtask using the gating network corresponding to the subtask; And according to the mapping weights of all expert network models in each subtask and the output results corresponding to the subtask output by each expert network model, the final output result of each subtask is determined; During the training process of the CMM0G health status estimation model, the final output result of the subtask is the health status value of the lithium-ion battery corresponding to the subtask. During the testing process of the CMM0G health status estimation model, based on the final output result of each subtask, the subtask module length is mapped through a deep neural network to complete the health status estimation of the unknown lithium-ion battery.
7. A computer device, It is characterized in that It comprises a processor and a memory; wherein, when the processor executes the computer program stored in the memory, the steps of the method for estimating the health state of an unknown lithium-ion battery based on the CMMOG model as described in any one of claims 1 to 5 are implemented.
8. A computer-readable storage medium, It is characterized in that Used to store computer programs; when the computer programs are executed by the processor, the steps of the method for estimating the health status of an unknown lithium-ion battery based on the CMM0G model as described in any one of claims 1 to 5 are implemented.
Citation Information
Patent Citations
Lithium battery life prediction method based on CNN-GRU combined neural network
CN115166561A
