Lithium ion battery remaining capacity estimation method, device, equipment and storage medium

Abstract

The application relates to a lithium ion battery residual capacity estimation method, device, equipment and storage medium in the technical field of lithium ion batteries. The method comprises the following steps: obtaining experimental measurement values of a plurality of lithium ion batteries, extracting a capacity increment wave crest from a result obtained by fitting the measurement values in an SVM mode, respectively adopting SVM, an LSTM network and GRP for preliminary capacity prediction, taking the preliminary capacity prediction as input, and utilizing a random forest algorithm to fuse and obtain a capacity output. The method combines the advantages of SVM in nonlinear and high-dimensional space fitting problems, GRP in uncertainty prediction problems and the LSTM network in time series prediction problems, effectively solves the interference of voltage collection noise on the capacity increment curve, and solves the difficulty of effectively extracting the capacity increment curve features. Meanwhile, the random forest algorithm is adopted to estimate the battery capacity, and the shortcoming that a single machine learning algorithm is prone to local optimization is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium ion batteries, in particular to a lithium ion battery residual capacity estimation method, device, equipment and storage medium. BACKGROUND

[0002] Lithium ion batteries are a clean and efficient energy storage system and have been widely promoted in electric vehicles. However, due to the material properties of lithium ion batteries, their capacity will inevitably deteriorate gradually. Therefore, accurate monitoring of the residual capacity of lithium ion batteries not only helps to estimate the state of health, state of charge and improvement, but also plays a crucial role in ensuring the reliable operation of electric vehicles.

[0003] With the development of computer science, machine learning methods have been gradually introduced into capacity estimation problems. In machine learning methods, capacity is estimated by mapping the potential nonlinear relationship between health features and residual capacity. The incremental capacity (IC) curve is an important health feature that has been proven to have a strong correlation with battery capacity, which can reconstruct the voltage platform caused by internal electrochemical reaction equilibrium into an intuitive and identifiable peak or trough. However, the IC curve is sensitive to noise, and when the measured voltage has noise, the monotonicity of the voltage cannot be guaranteed, resulting in a zero or even negative denominator for IC calculation, which poses a great challenge to the calculation of the IC curve.

[0004] In existing machine learning methods, long short-term memory recurrent neural network is a typical neural network algorithm that captures long-term dependency information through a specially designed gate structure. Compared with traditional neural networks, long short-term memory recurrent neural network has higher nonlinear modeling capability and more accurate estimation performance when processing time series data. Gaussian process regression estimates the prediction distribution of the parameters at a specific time point rather than a single deterministic point estimate, which is beneficial for uncertainty quantification estimation. However, the estimation accuracy of this method will decrease when encountering high-dimensional space. Support vector machine algorithm is a famous classification and regression method that has shown high efficiency in solving nonlinear and high-dimensional model fitting problems. Although these methods can achieve capacity estimation, they are all based on a single learner, which can easily make the estimation result fall into local optimum, thereby limiting the fidelity of capacity estimation. SUMMARY

[0005] Therefore, it is necessary to provide a lithium ion battery residual capacity estimation method, device, equipment and storage medium to solve the above technical problems.

[0006] A lithium ion battery residual capacity estimation method, the method comprising:

[0007] Obtaining a plurality of sets of measurement values of a target lithium-ion battery in an aging experiment, the measurement values including voltage, current and capacity of the target lithium-ion battery.

[0008] According to the measurement values, a fitting method of support vector machine is adopted, and according to the voltage-capacity relationship obtained by fitting, capacity increment data is obtained, and a capacity increment peak of the capacity increment data is extracted.

[0009] A plurality of capacity increment peaks are taken as training samples.

[0010] The training samples are respectively input into a trained SVM, an LSTM network and a Gaussian process regression operator to obtain a first predicted battery capacity, a second predicted battery capacity and a third predicted battery capacity.

[0011] The first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity are input into a trained random forest operator to obtain a target lithium-ion battery capacity.

[0012] In one of the embodiments, a plurality of sets of measurement values of a target lithium-ion battery in an aging experiment are obtained, including:

[0013] A plurality of target lithium-ion batteries are subjected to experiments under a predetermined test environment, and are subjected to aging experiments by repeated charging and discharging.

[0014] In the cycle life test, after a predetermined aging period, a predetermined rated capacity value is charged and discharged, the current capacity is calibrated, data acquisition is performed at a predetermined sampling frequency, and the voltage, current and capacity of the target lithium-ion battery are obtained.

[0015] In one of the embodiments, according to the measurement values, a fitting method of support vector machine is adopted, and according to the voltage-capacity relationship obtained by fitting, capacity increment data is obtained, and a capacity increment peak of the capacity increment data is extracted, including:

[0016] The voltage of the measurement values is taken as input, and the capacity is taken as output, a fitting method of support vector machine is adopted, and a voltage-capacity relationship is obtained.

[0017] The voltage-capacity relationship is calculated by a capacity increment data calculation formula to obtain capacity increment data; the capacity increment data calculation formula is:

[0018]

[0019] Wherein: represents the differential of capacity with respect to voltage, represents the ratio of capacity increment to voltage increment, Q t , Q t-1 respectively represent the capacity values at time t and t-1, Vt , V t-1 respectively represent the voltage value of t and t-1.

[0020] extracting a capacity increment peak in the capacity increment data, and taking the capacity increment peak as a health feature parameter.

[0021] In one embodiment, the training sample is respectively input into the trained SVM, LSTM network and Gaussian process regression operator, and first, second and third predicted battery capacities are obtained, including:

[0022] The training sample is input into the trained SVM, the training sample is characterized into high-dimensional space features by using a kernel function, and the high-dimensional space features are obtained by using a nonlinear SVM to obtain the first predicted battery capacity.

[0023] The training sample is input into the trained LSTM network, and the long-term dependence relationship is captured through the gate structure to obtain the second battery capacity.

[0024] The training sample is input into the trained Gaussian process regression operator to obtain the third battery capacity.

[0025] In one embodiment, the first, second and third predicted battery capacities are input into the trained random forest operator to obtain the target lithium-ion battery capacity, including:

[0026] The first, second and third predicted battery capacities are input into the trained random forest operator to obtain the output of each tree in the random forest; the output expression of the jth tree in the forest of the random forest operator is:

[0027]

[0028] wherein, c j represents the output of the jth tree in the forest of the random forest operator, X i represents the ith point, represents the training sample set, Θ j represents a random variable with the same distribution, represents all cells containing x, represents the number of points falling into

[0029] The target lithium-ion battery capacity is obtained by using the bagging strategy according to the output of each tree in the random forest; the expression of the bagging strategy of the random forest operator is:

[0030] ​

[0031] wherein, represents the estimation result of the random forest, represents the estimation result of the jth tree in the random forest, and M represents the number of trees.

[0032] A lithium ion battery residual capacity estimation device, the device comprising:

[0033] An experimental data acquisition module is configured to acquire a plurality of groups of measurement values of a plurality of target lithium ion batteries in an aging experiment, the measurement values including voltages, currents, and capacities of the target lithium ion batteries.

[0034] A training sample determination module is configured to use a fitting method of a support vector machine according to the measurement values, obtain capacity increment data according to a voltage-capacity relationship obtained by fitting, and extract capacity increment peaks of the capacity increment data; and use the plurality of capacity increment peaks as training samples.

[0035] A target lithium ion battery capacity determination module is configured to input the training samples into a trained SVM, LSTM network, and Gaussian process regression operator respectively, to obtain a first predicted battery capacity, a second predicted battery capacity, and a third predicted battery capacity respectively; and input the first predicted battery capacity, the second predicted battery capacity, and the third predicted battery capacity into a trained random forest operator to obtain a target lithium ion battery capacity.

[0036] In one of the embodiments, the experimental data acquisition module is further configured to perform experiments on the plurality of target lithium ion batteries under a predetermined test environment, and perform aging experiments by repeatedly charging and discharging; in a cycle life test, after a predetermined aging period, perform a predetermined rated capacity value charging and discharging cycle, calibrate the current capacity, and use a preset sampling frequency to perform data acquisition to obtain the voltages, currents, and capacities of the target lithium ion batteries.

[0037] In one of the embodiments, the training sample determination module is further configured to use a fitting method of a support vector machine to obtain a voltage-capacity relationship by taking the voltages of the measurement values as inputs and the capacities as outputs; and use a capacity increment data calculation formula to calculate the capacity increment data from the voltage-capacity relationship; the capacity increment data calculation formula is:

[0038]

[0039] wherein, represents a differential of the capacity with respect to the voltage, represents a ratio of the capacity increment to the voltage increment, Q t , Q t-1 respectively represent the capacity values at time t and t-1, V t, V t-1 respectively represent the voltage value of t and t-1;

[0040] extracting a capacity increment wave crest in the capacity increment data, and taking the capacity increment wave crest as a health feature parameter.

[0041] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of any of the above methods when executing the computer program.

[0042] A computer readable storage medium, which stores a computer program, and the computer program implements the steps of any of the above methods when executed by a processor.

[0043] The above lithium ion battery remaining capacity estimation method, device, equipment and storage medium, the method comprises obtaining experimental measurement values of a plurality of lithium ion batteries, and fitting by using a support vector machine, extracting a capacity increment wave crest from the fitting result, and performing preliminary capacity prediction by using a support vector machine, a long short-term memory recurrent neural network and a Gaussian process regression, then taking the preliminary capacity prediction as input, and fusing the outputs of multiple machine learning machines by using a random forest algorithm. This method combines the advantages of support vector machine in nonlinear and high-dimensional space fitting problem, Gaussian process regression in uncertainty prediction problem and long short-term memory recurrent neural network in time series prediction problem, effectively solves the interference of voltage collection noise on the capacity increment curve, and solves the difficulty of effectively extracting the capacity increment curve feature. At the same time, the random forest algorithm is used to fuse the support vector machine, long short-term memory network and Gaussian process regression algorithm to estimate the battery capacity, effectively solving the shortcoming that a single machine learning algorithm is easy to fall into local optimum. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 It is a flowchart of the lithium ion battery remaining capacity estimation method in one embodiment;

[0045] Figure 2 It is a relationship curve between charging voltage and charging capacity in one embodiment;

[0046] Figure 3 It is a relationship curve between capacity increment wave crest and battery capacity in one embodiment;

[0047] Figure 4 It is a battery capacity estimation result in one embodiment;

[0048] Figure 5 It is a structural block diagram of the lithium ion battery remaining capacity estimation device in one embodiment;

[0049] Figure 6 It is an internal structure diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0050] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0051] Support Vector Machine, abbreviated as: SVM.

[0052] Gaussian Process Regressor, abbreviated as: GPR.

[0053] Long short-term memory recurrent neural network, abbreviated as: LSTM RNN.

[0054] In one embodiment, as shown in Figure 1 a lithium ion battery remaining capacity estimation method is provided, which includes the following steps:

[0055] Step 100: Obtain a plurality of target lithium ion batteries in the aging experiment of a plurality of groups of measurement values, including the voltage, current and capacity of the target lithium ion battery.

[0056] Specifically, the charging voltage, current and capacity data are collected from the experimental test.

[0057] Step 102: According to the measurement value, the fitting method of support vector machine is adopted, and according to the voltage capacity relationship obtained by fitting, the capacity increment data is obtained, and the capacity increment wave peak of the capacity increment data is extracted; the plurality of capacity increment wave peaks are taken as training samples.

[0058] Specifically, the charging voltage data is taken as the input, the charging capacity data is taken as the output, the fitting method of support vector machine is adopted to charge the voltage and charging capacity relationship, and the noise interference problem is solved.

[0059] The capacity increment wave peak in the capacity increment data is extracted as a health feature parameter, the relationship between the capacity increment wave peak and the battery capacity is obtained, and the peak value of the capacity increment curve continuously decreases with the decrease of the battery capacity

[0060] Step 104: The training samples are respectively input into the trained SVM, LSTM network and Gaussian process regression operator, and the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity are obtained.

[0061] Specifically, 1) compared with traditional neural networks, SVM converges to the optimal value much faster, and can meet the fitting of nonlinear and high-dimensional data. For nonlinear cases, kernel functions are used in SVM to feature the original data into high-dimensional space. The kernel functions used in SVM can be linear kernel functions, Gaussian radial basis kernel functions or polynomial kernel functions.

[0062] 2) LSTM network is an extended form of traditional neural network, which can capture long-term dependencies through gate structure and has higher nonlinear establishment potential. It has more accurate performance in processing time series prediction. LSTM is composed of an input gate, a forget gate, an output gate and different input-output connections controlled by these gates.

[0063] 3) GPR mainly describes the regression function based on probability distribution, reflects the empirical risk through the likelihood function, and then obtains the posterior probability distribution through the Bayesian theory.

[0064] The fusion framework proposed in the method combines the advantages of SVM in nonlinear and high-dimensional space fitting problems, GPR in uncertainty prediction problems and LSTM network in time series prediction problems.

[0065] Step 106: input the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity into the trained random forest operator to obtain the target lithium ion battery capacity.

[0066] Specifically, the estimation results of SVM, LSTM network and Gaussian process regression operator: the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity are used as the input of the fusion method, and the outputs of multiple machine learning machines are fused by using random forest (RF) algorithm. RF is a typical ensemble learning method, which can integrate single learners to obtain the optimal estimation result.

[0067] The above lithium ion battery residual capacity estimation method comprises obtaining experimental measurement values of a plurality of lithium ion batteries, and fitting by using a support vector machine mode, extracting a capacity increment wave crest from a fitting result, performing preliminary capacity prediction by using a support vector machine, a long short-term memory recurrent neural network and a Gaussian process regression respectively, and then inputting the preliminary capacity prediction into a random forest algorithm to fuse outputs of the multiple machine learning machines. The method combines the advantages of the support vector machine in nonlinear and high-dimensional space fitting problems, the Gaussian process regression in uncertainty prediction problems and the long short-term memory recurrent neural network in time series prediction problems, effectively solves the interference of voltage collection noise on the capacity increment curve, and solves the difficulty of effectively extracting the capacity increment curve features. Meanwhile, the random forest algorithm is used to fuse the support vector machine, the long short-term memory network and the Gaussian process regression algorithm to estimate the battery capacity, effectively solving the shortcoming that a single machine learning algorithm is easy to fall into local optimization.

[0068] In one of the embodiments, the step 100 comprises: performing experiments on a plurality of target lithium ion batteries under a predetermined test environment, and repeatedly charging and discharging to perform an aging experiment; in the cycle life test, after a predetermined aging period, a predetermined rated capacity value is charged and discharged to calibrate the current capacity, and a predetermined sampling frequency is used to collect data to obtain the voltage, current and capacity of the target lithium ion battery.

[0069] Specifically, all the batteries are tested in a temperature control box under a test condition of 40℃, and repeatedly charged and discharged to show their aging characteristics. During the charging process, a constant current charging scheme of 2C (C is the rated capacity value here) is used until the terminal voltage reaches 4.2V; then a constant voltage charging strategy is used to charge the battery until the current drops to a predefined cutoff threshold. In the discharging phase, the Artemis-based load working condition is used to simulate the aging process in actual application. In the cycle life test, 1C charging and discharging is performed every 100 aging periods to calibrate the current capacity, and the voltage, current and capacity of the target battery are recorded, and the sampling frequency is 1Hz.

[0070] In one of the embodiments, the step 102 comprises: taking the voltage of the measurement value as input and the capacity as output, using the fitting mode of the support vector machine to obtain the voltage-capacity relationship; calculating the capacity increment data by using a capacity increment data calculation formula on the voltage-capacity relationship; and the capacity increment data calculation formula is:

[0071]

[0072] Wherein: represents the differential of the capacity with respect to the voltage, represents the ratio of the capacity increment to the voltage increment, Q t , Q t-1 respectively represent the capacity values at time t and t-1, and Vt , V t-1 respectively represent the voltage value of t and t-1 time;

[0073] Extract the capacity increment wave crest in the capacity increment data, and take the capacity increment wave crest as a health characteristic parameter.

[0074] Specifically, taking the charging voltage data as input and the charging capacity data as output, the SVM algorithm is used to fit the relationship between the charging voltage and the charging capacity, and the fitting result is as follows Figure 2 As shown in the figure, Figure 2 It can be seen that the curve fitted by SVM eliminates the influence of noise and well fits the relationship between voltage and capacity.

[0075] The fitting data is processed by formula (1) to obtain the capacity increment data. Extract the capacity increment wave crest in the capacity increment data as a health characteristic parameter to obtain the relationship between the capacity increment wave crest and the battery capacity as shown in the following Figure 3 As shown in the figure, Figure 3 It can be seen that the peak value of the capacity increment curve continuously decreases with the decrease of the battery capacity. Therefore, this parameter is used as a health characteristic to be applied to the estimation of the battery capacity.

[0076] In one embodiment, step 104 includes: inputting the training sample into the trained SVM, using the kernel function to feature the training sample into high-dimensional space features, and using the high-dimensional space features to obtain the first predicted battery capacity by the nonlinear SVM; inputting the training sample into the trained LSTM network, capturing the long-term dependence relationship through the gate structure, and obtaining the second battery capacity; inputting the training sample into the trained Gaussian process regression operator to obtain the third battery capacity.

[0077] Specifically, 1) Compared with the traditional neural network, SVM converges to the optimal value much faster, and can meet the fitting of nonlinear and high-dimensional data. For nonlinear cases, the kernel function is used in SVM to feature the original data into high-dimensional space. The nonlinear problem of SVM is derived as:

[0078]

[0079] In the formula, ε represents the tolerance deviation, ω and b represent the weight and deviation, which can be solved by introducing Lagrange multipliers α and α * , that is:

[0080]

[0081] 2) LSTM network is an extended form of traditional neural network, which can capture long-term dependencies through gate structure, has higher nonlinear establishment potential, and has more accurate performance in processing time series prediction. LSTM network is composed of an input gate, a forget gate, an output gate and different input-output connections controlled by these gates. The calculation process of LSTM network is summarized as follows:

[0082]

[0083] where f k , i k , g k , O and c k represent the forget gate, the input gate, the input node, the output gate and the storage unit, respectively; b f , b i , b g represent the bias of the forget gate, the input gate and the input node, respectively; OP k represents the output at step k, OW f , OW i , OW g , OW O are the output weights of the forget gate, the input gate, the input node and the output gate, respectively; IW f , IW i , IW g , IW O are the input weights of the forget gate, the input gate, the input node and the output gate, respectively; IP k and IP k-1 correspond to the inputs at steps k and k-1; p k represents the internal variable of the LSTM storage unit; σ is the activation function, and as a preferred, the activation function adopts the sigmoid function; Tanh is defined as the hyperbolic function.

[0084] 3) GPR mainly describes the regression function based on probability distribution, reflects the empirical risk through the likelihood function, and then obtains the posterior probability distribution through the Bayesian theory. Therefore, the GPR problem can be simplified as:

[0085]

[0086] where f(x) is the output function, () Gaussian probability distribution function, x and y represent input variables and observed variables, m(x) and k f (x,x′) are the mean and covariance functions, ζ represents added noise, y, y * and represent the prior value, the predicted value, the mean of the predicted value, respectively, x and x *Represents the high-dimensional n input vectors and test input vector Let p(y) represent the noise covariance matrix. * |x,y,x * ) represents the prior distribution, K f (x,x) represents an n-dimensional symmetric positive definite matrix.

[0087] The training processes for SVM, LSTM networks, Gaussian process regression, and random forests are similar. The peak of the 70% capacity increment curve in the test data is arbitrarily selected as the input, and the corresponding 70% capacity value is used as the output. Then, the corresponding calculation formula is used for calculation.

[0088] In one embodiment, step 106 includes: inputting the first predicted battery capacity, the second predicted battery capacity, and the third predicted battery capacity into the trained random forest operator to obtain the output of each tree in the random forest; the output expression of the j-th tree in the forest of the random forest operator is:

[0089]

[0090] Among them, c j Let X represent the output of the j-th tree in the forest of the random forest operator. i This represents the i-th point. Represents the training sample set, Θ j Represent random variables with the same distribution. This represents all cells containing the letter 'x'. Indicates all that fall into The number of points.

[0091] The target lithium-ion battery capacity is obtained by using a bagging strategy based on the output of each tree in the random forest; the expression for the bagging strategy of the random forest operator is:

[0092]

[0093] in, This represents the estimation result of the random forest. Let M represent the estimated value of the j-th tree in the random forest, where M represents the number of trees.

[0094] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1At least one of the steps in the method can include a plurality of sub-steps or a plurality of stages, which are not necessarily performed at the same time, but can be performed at different times, and the order of the execution of the sub-steps or stages is not necessarily sequential, but can be performed alternately or alternately with at least one of the other steps or sub-steps or stages of the other steps.

[0095] In one specific embodiment, the lithium-ion battery capacity estimation results are as follows Figure 4 The estimated value is very similar to the measured value of all the study cells throughout the life cycle. The root mean square error (RMSE) of the estimated capacity is 3.91 mAh (about 0.53% of the nominal capacity), and the maximum absolute error (MAE) and the average absolute error (AAE) are 11.93 mAh and 3.12 mAh, respectively. Compared with traditional correlation methods, the advantages of the present application are that it can effectively solve the interference of voltage acquisition noise on the capacity increment curve, and solve the difficulty of effectively extracting the characteristics of the capacity increment curve. At the same time, in order to integrate the advantages of various machine learning, the random forest algorithm is used to fuse the currently popular single learning machine: support vector machine, long short-term memory network and Gaussian process regression algorithm to estimate the battery capacity, effectively solving the shortcoming of single machine learning algorithm that is easy to fall into local optimum.

[0096] In one embodiment, as shown in Figure 5 a lithium-ion battery remaining capacity estimation device is provided, comprising: an experimental data acquisition module, a training sample determination module and a target lithium-ion battery capacity determination module, wherein:

[0097] The experimental data acquisition module is configured to acquire a plurality of groups of measurement values of a plurality of target lithium-ion batteries in an aging experiment, the measurement values including the voltage, current and capacity of the target lithium-ion batteries.

[0098] The training sample determination module is configured to use a fitting method of a support vector machine according to the measurement values, obtain capacity increment data according to the voltage-capacity relationship obtained by fitting, and extract capacity increment wave peaks of the capacity increment data; and take the plurality of capacity increment wave peaks as training samples.

[0099] The target lithium-ion battery capacity determination module is configured to input the training samples into a trained SVM, an LSTM network and a Gaussian process regression operator respectively, to obtain a first predicted battery capacity, a second predicted battery capacity and a third predicted battery capacity respectively; and input the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity into a trained random forest operator to obtain the target lithium-ion battery capacity.

[0100] In one of the embodiments, the experimental data acquisition module is further configured to perform experiments on the plurality of target lithium-ion batteries under a predetermined test environment, and perform aging experiments through repeated charging and discharging; in the cycle life test, after a preset aging period, a predetermined rated capacity value is used for charging and discharging cycles, the current capacity is calibrated, data acquisition is performed at a preset sampling frequency, and the voltage, current and capacity of the target lithium-ion battery are obtained.

[0101] In one of the embodiments, the training sample determination module is further configured to use a support vector machine fitting method to obtain a voltage-capacity relationship by taking the voltage of the measurement value as input and the capacity as output; calculate the capacity increment data of the voltage-capacity relationship by using a capacity increment data calculation formula, and the capacity increment data calculation formula is shown in formula (1). Extract the capacity increment peak in the capacity increment data, and take the capacity increment peak as the health feature parameter.

[0102] In one of the embodiments, the target lithium-ion battery capacity determination module is further configured to input the training sample into the trained SVM, use a kernel function to feature the training sample into a high-dimensional space feature, and use a nonlinear SVM to obtain a first predicted battery capacity; input the training sample into the trained LSTM network, capture long-term dependencies through a gate structure, and obtain a second battery capacity; and input the training sample into the trained Gaussian process regression operator to obtain a third battery capacity.

[0103] In one of the embodiments, the target lithium-ion battery capacity determination module is further configured to input the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity into the trained random forest operator to obtain the output of each tree in the random forest; the output expression of the jth tree in the forest of the random forest operator is shown in formula (6); and obtain the target lithium-ion battery capacity by using a bagging strategy according to the output of each tree in the random forest; and the expression of the bagging strategy of the random forest operator is shown in formula (7).

[0104] The specific limitations of the lithium-ion battery remaining capacity estimation device can be referred to the limitations of the lithium-ion battery remaining capacity estimation method in the foregoing, which will not be repeated here. Each module in the above lithium-ion battery remaining capacity estimation device can be realized by software, hardware and combinations thereof, in whole or in part. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each of the above modules.

[0105] In one embodiment, a computer device is provided, which can be a terminal, and the internal structure diagram thereof can be as shown in Figure 6As shown in the figure. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement a lithium ion battery remaining capacity estimation method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0106] Those skilled in the art can understand that, Figure 6 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0107] In one embodiment, a computer device is provided, including a memory and a processor, the memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0108] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0109] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, storage, databases, or other media in the embodiments provided by the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0110] The technical features of the above embodiments can be combined in any way. In order to make the description simple, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present application.

[0111] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, some modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for estimating the remaining capacity of a lithium-ion battery, characterized in that, The method includes: Multiple sets of measurements were obtained for several target lithium-ion batteries during aging experiments; the measurements included the voltage, current, and capacity of the target lithium-ion batteries. Based on the measured values, a support vector machine is used to fit the voltage-capacity nonlinear relationship. With voltage as input and capacity as output, the data is characterized into a high-dimensional space through a kernel function to construct the voltage-capacity nonlinear relationship, eliminating the interference of voltage acquisition noise on the capacity increment curve, and obtaining a smoothed voltage-capacity relationship. The kernel function is one of a linear kernel function, a Gaussian radial basis kernel function, or a polynomial kernel function. Capacity increment data is obtained based on the smoothed voltage-capacity relationship, and the capacity increment peak of the capacity increment data is extracted. Use multiple of the aforementioned capacity increment peaks as training samples; The training samples are input into the trained SVM, LSTM networks and Gaussian process regression operators respectively to obtain the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity respectively. The first predicted battery capacity, the second predicted battery capacity, and the third predicted battery capacity are input into a trained random forest operator, and the output is fused through a Bagging strategy to obtain the target lithium-ion battery capacity.

2. The method according to claim 1, characterized in that, Obtain multiple sets of measurements of the target lithium-ion battery in the aging experiment, including: Multiple target lithium-ion batteries were tested under a predetermined test environment, and aging tests were conducted by repeated charge and discharge cycles. In the cycle life test, after a preset aging period, a predetermined multiple of the rated capacity value is charged and discharged to calibrate the current capacity. Data is collected using a preset sampling frequency to obtain the voltage, current and capacity of the target lithium-ion battery.

3. The method according to claim 1, characterized in that, Capacity increment data is obtained based on the smoothed voltage-capacity relationship, and the capacity increment peak of the capacity increment data is extracted, including: The smoothed voltage-capacity relationship is calculated using the capacity increment data calculation formula to obtain the capacity increment data; the capacity increment data calculation formula is: in: This represents the derivative of capacitance with respect to voltage. It represents the ratio of the increase in capacity to the increase in voltage. They represent t Time and t -1 capacity value, They represent t Time and t -1 is the voltage value; Extract the capacity increment peak from the capacity increment data and use the capacity increment peak as a health feature parameter.

4. The method according to claim 1, characterized in that, The training samples are input into the trained SVM, LSTM networks, and Gaussian process regression operator, respectively, to obtain the first predicted battery capacity, the second predicted battery capacity, and the third predicted battery capacity, including: The training samples are input into the trained SVM, and the kernel function is used to feature the training samples into high-dimensional space features. The high-dimensional space features are then used with a nonlinear SVM to obtain the first predicted battery capacity. The training samples are input into a trained LSTM network, and long-term dependencies are captured through a gate structure to obtain the second battery capacity. The training samples are input into the trained Gaussian process regression operator to obtain the third battery capacity.

5. The method according to claim 1, characterized in that, The first predicted battery capacity, the second predicted battery capacity, and the third predicted battery capacity are input into a trained random forest operator, and the output is fused using a Bagging strategy to obtain the target lithium-ion battery capacity, including: The first predicted battery capacity, the second predicted battery capacity, and the third predicted battery capacity are input into the trained random forest operator to obtain the output of each tree in the random forest; the output expression of the j-th tree in the forest of the random forest operator is: in, This represents the output of the j-th tree in the forest of the random forest operator. This represents the i-th point. Represents the training sample set, Represent random variables with the same distribution. Indicates all containing The cell, Indicates all that fall into Points; The target lithium-ion battery capacity is obtained by using a bagging strategy based on the output of each tree in the random forest. The expression for the bagging strategy of the random forest operator is: in, This represents the estimation result of the random forest. Represents the first in a random forest j The estimated number of trees, Indicates the number of trees.

6. A device for estimating the remaining capacity of a lithium-ion battery, characterized in that, The device includes: The experimental data acquisition module is used to acquire multiple sets of measurement values ​​of multiple target lithium-ion batteries in aging experiments, including the voltage, current and capacity of the target lithium-ion batteries; The training sample determination module is used to fit the voltage-capacity nonlinear relationship using a support vector machine based on the measured values. With voltage as input and capacity as output, the module uses a kernel function to feature the data into a high-dimensional space to construct the voltage-capacity nonlinear relationship, eliminating the interference of voltage acquisition noise on the capacity increment curve, and obtaining a smoothed voltage-capacity relationship. The kernel function is one of a linear kernel function, a Gaussian radial basis function kernel function, or a polynomial kernel function. Based on the smoothed voltage-capacity relationship, capacity increment data is obtained, and the capacity increment peaks of the capacity increment data are extracted. Multiple capacity increment peaks are used as training samples. The target lithium-ion battery capacity determination module is used to input the training samples into the trained SVM, LSTM network and Gaussian process regression operator respectively to obtain the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity respectively; and input the first predicted battery capacity, the second predicted battery capacity and the third predicted battery capacity into the trained random forest operator to obtain the target lithium-ion battery capacity.

7. The apparatus according to claim 6, characterized in that, The experimental data acquisition module is also used to conduct experiments on multiple target lithium-ion batteries under a predetermined test environment and to perform aging experiments by repeated charge and discharge. In the cycle life test, after a preset aging cycle, a predetermined multiple of the rated capacity value is charged and discharged to calibrate the current capacity. Data is collected using a preset sampling frequency to obtain the voltage, current and capacity of the target lithium-ion battery.

8. The apparatus according to claim 6, characterized in that, The training sample determination module is further used to take the voltage of the measured values ​​as input and the capacity as output, and use a support vector machine fitting method to obtain the voltage-capacity relationship; the voltage-capacity relationship is then calculated using the capacity increment data calculation formula to obtain the capacity increment data; the capacity increment data calculation formula is: in: This represents the derivative of capacitance with respect to voltage. It represents the ratio of the increase in capacity to the increase in voltage. They represent t Time and t -1 capacity value, They represent t Time and t -1 is the voltage value; Extract the capacity increment peak from the capacity increment data and use the capacity increment peak as a health feature parameter.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images