High-precision battery residual capacity detection method, detection system and detection device

Abstract

The invention discloses a high-precision battery residual capacity detection method, system and device, and the method comprises the steps: S1, testing the AC impedance data of a plurality of batteries, and drawing an AC impedance spectrum; s2, obtaining the real residual capacity of the plurality of batteries; s3, utilizing the composite neural network prediction model to construct an evaluation model of the alternating current impedance presumption residual capacity; and S4, obtaining the residual capacity of the to-be-tested battery according to the evaluation model. The detection system is used for implementing the method, and the detection device is used for obtaining the alternating-current impedance data in the S1 and the S4. According to the method, rapid and accurate capacity presumption of the retired power battery is realized, the problems of high deviation, high energy consumption and long time of battery residual capacity estimation by adopting a traditional detection method are avoided, and high accuracy and reproducibility of detection data are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of battery detection, and in particular to a high-precision battery remaining capacity detection method, detection system and detection device. Background Art

[0002] At present, with the large-scale popularization of new energy vehicles and the large-scale construction of energy storage power stations, a large number of lithium batteries will be retired in the next 5 to 10 years.

[0003] Poor battery consistency during the energy storage cascade utilization process can lead to serious safety issues such as rapid capacity decay, thermal runaway and battery explosion. Therefore, retired batteries must be tested before cascade utilization to improve the overall performance of the battery pack. Research on retired battery detection and sorting technology is of great significance to further improve energy utilization and safety. At present, traditional capacity detection of retired batteries requires a large amount of data accumulation, including battery voltage, current, temperature and other information. However, different battery manufacturers and different application objects make these data independent, resulting in the problem of data islands. This greatly limits the effect of battery capacity detection and cannot fully maximize the utilization of data resources. Summary of the invention

[0004] The present invention provides a high-precision battery remaining capacity detection method for estimating the battery remaining capacity by measuring AC impedance, as well as a detection system and a detection device based on the detection method, which can realize rapid estimation of the remaining capacity of retired batteries, and is used to solve the technical problem of poor battery capacity detection effect mentioned in the background technology.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is: A high-precision method for detecting the remaining capacity of a battery comprises the following steps: S1. Testing the AC impedance data of several batteries at different given frequencies, drawing an AC impedance spectrum with the real part of the AC impedance of a single battery at different given frequencies as the horizontal axis and the imaginary part as the vertical axis, to obtain the AC impedance spectrum of each battery; S2. Obtaining the actual remaining capacity of the plurality of batteries; S3, constructing an evaluation model for estimating the remaining capacity by using AC impedance according to the AC impedance spectrum set and the actual remaining capacity of the plurality of batteries using a composite neural network prediction model; S4. Obtaining the remaining capacity of the battery to be tested according to the AC impedance spectrum of the battery to be tested and the evaluation model for estimating the remaining capacity by the AC impedance.

[0006] As a further optimization of the above technical solution, in step S3, the evaluation model of estimating the remaining capacity by using the composite neural network prediction model is constructed, including: first using the convolutional neural network to output the image feature matrix, and then inputting the image feature matrix into the fully connected neural network to obtain the composite neural network's predicted value of the battery capacity, and training the model based on the predicted value to obtain the evaluation model of estimating the remaining capacity by using the AC impedance.

[0007] As a further preferred embodiment of the above technical solution, in step S3, the specific steps of constructing an evaluation model for estimating the remaining capacity by AC impedance include: S31, converting the AC impedance spectrum of each battery into a gray value matrix; S32, using the convolution kernel of the convolution neural network to perform weighted summation on the grayscale values ​​within the convolution kernel window from a certain point of the image, and then outputting the weighted summation result to the corresponding position of the output matrix; the convolution kernel with a set moving step size is continuously moved to completely scan the grayscale value matrix, and when the elements at each position of the output matrix are filled, the output matrix is ​​output; S33, inputting the output matrix into a linear rectification unit to increase the nonlinearity of the mapping process, inputting the output matrix after the nonlinearity is increased into a pooling layer for a pooling operation, and outputting an image feature matrix; S34, input the image feature matrix output by the convolutional neural network into a fully connected neural network, the fully connected neural network includes an input layer, an output layer and a hidden layer, each neuron in each layer performs a weighted summation of the input vector of the layer and adds a bias value, and inputs the result into a set linear rectifier unit activation function, after the linear rectifier unit activation function is calculated, the calculation result is used as the output value of the neuron, the output values ​​of the neurons in the same layer are combined into a vector in a fixed order and input into the next layer, and the above steps are repeated in the next layer, the last layer has only one neuron, the weighted summation of the input vector of the neuron itself and the result of adding the bias is used as the predicted value of the battery capacity by the composite neural network; S35. Perform model training based on the predicted value and the actual remaining capacity of the plurality of batteries to construct an evaluation model for estimating the remaining capacity by AC impedance.

[0008] As a further optimization of the above technical solution, in step S33, the pooling operation refers to the use of the maximum pooling method to perform pooling processing on the output matrix after nonlinearity, and each element of the output matrix after pooling is the maximum element in the spatial window range corresponding to the input matrix of the pooling layer. After the pooling layer, only the only important point is retained in the corresponding spatial window range, which can reduce the dimension of the grayscale value of the input matrix, simplify the image matrix, and highlight the key information.

[0009] As a further preferred embodiment of the above technical solution, in step S35, the model training is carried out by establishing an error function, taking the square of the difference between the predicted value of the battery capacity of a single battery and the actual remaining capacity of the battery as the prediction error, and taking the sum of the prediction errors of each battery as the error function of the composite neural network prediction model, and using the error function to perform iterative training, calculate the corresponding error and adjustment value of each parameter of each layer of neurons in the current situation, and adjust the parameters of the corresponding neurons, and when the set number of iterations is reached or the loss value of the error function after the set number of iterations meets the set threshold, stop the training, and complete the construction of the evaluation model for estimating the remaining capacity by AC impedance.

[0010] As a further optimization of the above technical solution, in step S35, the back propagation algorithm and the Adam optimizer are used to calculate the corresponding error and adjustment value of each parameter of each layer of neurons in the current situation, and adjust the parameters of the corresponding neurons. During the model training process, the back propagation algorithm is responsible for calculating the gradient, and the Adam optimizer uses this gradient information to dynamically adjust the learning rate to optimize the model parameters. The combination of the back propagation algorithm and the Adam optimizer can improve the training efficiency and model performance, converge faster during the training process, and maintain high stability and generalization ability.

[0011] Based on the same technical concept, the present invention also provides a high-precision battery remaining capacity detection system, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the detection method described in the above technical solution when executing the computer program.

[0012] Based on the same technical concept, the present invention also provides a detection system including a fixing table, a detection instrument, a probe and the above technical solution, wherein the fixing table is provided with a clamping mechanism for fixing the battery, the probe contacts the battery and transmits the electrical signal to the detection instrument, and the detection instrument sends the data to the detection system; the battery fixing table and the clamping mechanism are used to position the battery to be tested; the detection instrument is used to perform AC impedance data detection on the battery to be tested; the detection system is used to execute a program to calculate the remaining capacity of the battery to be tested based on the AC impedance data. The battery fixing table is mainly used to position the battery to be tested, and can be designed to be mobile to achieve the longitudinal and lateral displacement of the battery to be tested in the horizontal plane; the clamping mechanism configured on the fixing table can realize the fixation of the battery, and a certain pre-tightening force is applied to the battery to be tested through the clamping mechanism to facilitate the detection and auxiliary devices to perform AC impedance (AC-IR) data detection on the battery. The detection instrument is mainly used to perform AC impedance (AC-IR) data detection function on the battery to be tested.

[0013] As a further preferred embodiment of the above technical solution, four probes are provided. The four-probe contact method can ensure that the fitting rate of multiple test results of the same battery reaches more than 99.9%, thereby ensuring the accuracy of the data.

[0014] As a further preferred embodiment of the above technical solution, the detection instrument is an impedance tester.

[0015] The present invention has the following beneficial effects: The present invention constructs an evaluation model and detection method for battery performance, especially the remaining battery capacity, by collecting data such as battery AC impedance. Based on AC impedance image modal data and (convolutional) neural network algorithm, an AC impedance and capacity estimation algorithm is developed, and an online detection tool is established. Through deep learning, the nonlinear fitting accuracy of battery performance parameters is continuously corrected to achieve fast and accurate capacity estimation of retired power batteries, avoiding the problems of high deviation, high energy consumption, and long time for estimating the remaining battery capacity using traditional detection methods. At the same time, a detection device is designed accordingly based on the above detection method, and the high accuracy and reproducibility of the detection data are guaranteed by the detection fixture of retired power batteries with precise positioning control.

[0016] The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 This is a schematic structural diagram of a high-precision battery remaining capacity detection device according to Example 1; Figure 2 This is a comparison chart of the battery remaining capacity result measured by the high-precision battery remaining capacity detection method of Example 1 and the actual battery remaining capacity.

[0018] Legend: 1. Fixed table; 2. Probe; 3. Clamping mechanism; 4. Battery. DETAILED DESCRIPTION

[0019] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0020] Embodiment 1: The high-precision battery remaining capacity detection method of this embodiment includes the following steps: (1) AC impedance test at different frequencies: Test the battery AC impedance at different given frequencies, and draw the AC impedance spectrum with the real part (R) of different test frequencies as the horizontal axis and the imaginary part (X) as the vertical axis. The AC impedance spectrum of each battery is formed by testing the AC impedance of a large number of batteries at different frequencies.

[0021] In this embodiment, 56 batteries of the same brand are tested for battery AC impedance at different frequencies (1000, 800, 630, 500, 320, 250, 160, 100, 80, 63, 50, 32, 25, 16, 10, 8, 6.3, 5, 3.2, 2.5, 1.6, 1, 0.8, 0.6, 0.4, 0.2, 0.12, 0.1 Hz) at 20% state of charge. The frequency and AC impedance data of one of the batteries (named as 1# battery) are shown in Table 1 below.

[0022] Table 1. AC impedance data of batteries at different frequencies

[0023] (2) Battery actual capacity test: A charge and discharge tester is used to determine the actual remaining capacity of the battery. Under a certain ambient temperature, by setting the battery charge and discharge cut-off voltage, rest time, charge and discharge current and other parameters, the remaining capacity of the battery can be accurately obtained through testing. The test steps are shown in Table 2.

[0024] Table 2. Steps for determining the actual remaining capacity of a battery

[0025] In this embodiment, 56 batteries of the same brand are tested for their actual remaining capacity, wherein the actual capacity of the above-mentioned 1# battery is 119.92 Ah.

[0026] (3) Construction of an evaluation model for estimating remaining capacity using AC-IR: The composite neural network (CNN) prediction model is used to construct the AC-IR (Alternating Current Impedance) remaining capacity estimation algorithm and model, where the composite neural network prediction model includes a convolutional neural network and a fully connected neural network.

[0027] The battery AC impedance image drawn above is converted into a gray value matrix. The convolution kernel (filter) is used to perform weighted summation of the gray values ​​within the convolution kernel window range starting from a certain point in the image, and then the weighted summation result is output to the corresponding position of the output matrix. The convolution kernel with a set moving step size is continuously moved to completely scan the gray value matrix until the elements at each position of the output matrix are filled, and then the output matrix is ​​output. The output matrix is ​​input into the Rectified Linear Unit (ReLU) to add nonlinearity to the mapping process. The matrix after adding nonlinearity is input into the pooling layer, and the maximum pooling method (max pooling) is used, that is, each element of the output matrix after pooling is the maximum element in the corresponding spatial window range of the input matrix of the pooling layer. In this way, after the pooling layer, only the only important point is retained in the corresponding spatial window range, reducing the dimension of the gray value of the input matrix, simplifying the image matrix, and highlighting key information.

[0028] The image feature matrix output by the convolutional neural network is input into the fully connected neural network. In the fully connected neural network of the composite neural network prediction model, there are three layers, namely the input layer, the output layer and the hidden layer. In the prediction process, each neuron in each layer will add a bias value to the weighted sum of the input vector of this layer, and then input it into the set linear rectifier unit (Rectified Linear Unit, ReLU) activation function. After the ReLU activation function is calculated, the calculation result is used as the output value of the neuron. The output values ​​of each neuron in the same layer will be combined into a vector in a fixed order and input into the next layer, and then the next layer will repeat the above steps. The last layer has only one neuron, which takes the weighted sum of its own input vector and adds the bias result as the composite neural network's prediction of the battery capacity. The prediction mechanism of the fully connected neural network is to iteratively optimize and update the composite neural network prediction model by establishing an error function and training the error through the back propagation function. The model training is stopped by setting the number of iterations or setting the loss value threshold of the error function to obtain the composite neural network prediction model. During the prediction model training process, the square of the difference between the predicted value of the battery capacity data of a certain battery and the actual data value of the battery capacity of the battery is used as the prediction error of this battery capacity prediction. The sum of the prediction errors of each n batteries is used as the total prediction error of the composite neural network prediction model, that is, the error function. The error function is used for iterative training, and the back propagation algorithm and the Adam optimizer are used to calculate the corresponding error and adjustment value of each parameter of each layer of neurons in the current situation, and the parameters of the corresponding neurons are adjusted. When the set number of iterations is reached or the loss value of the error function after the set number of iterations meets the set threshold, the training is stopped to obtain the evaluation model for estimating the remaining capacity by AC impedance. The comparison result of the remaining capacity of the battery estimated by the above-mentioned evaluation model for estimating the remaining capacity by AC impedance in this embodiment and the actual remaining capacity of the battery is as shown in the figure. Figure 2 As shown ( Figure 2 In the figure, the predicted capacity is the remaining capacity of the battery obtained by the method of this embodiment, and the actual capacity is the actual remaining capacity of the battery). Experiments on 56 real retired automotive power battery cells have proved that the average prediction accuracy of the method proposed in the present invention exceeds 95% (average prediction error <2.5Ah), and the closer the point is to the diagonal line, the more accurate the prediction.

[0029] The image reduction ratio (nDIV), convolution kernel size (wH×wW), pooling layer settings (pooling), convolution operator moving step size, pooling operator moving step size, activation function, learning rate and other parameters are all manually set. The setting values ​​are shown in Table 3. The parameters of the composite neural network will be randomly initialized and the parameters will be continuously calibrated to assign different weight values ​​to neurons in each layer and different bias values ​​to neurons in each layer during weighted summation.

[0030] Table 3. Parameter setting values ​​during prediction model construction

[0031] The high-precision battery remaining capacity detection system of this embodiment includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above detection method are implemented.

[0032] like Figure 1 As shown, the high-precision battery remaining capacity detection device of this embodiment includes the detection system of this embodiment, a fixing table 1, a detection instrument (an impedance tester, not shown in the figure to avoid obstruction, the detection instrument is connected to the probe 2 through a data line), and the probe 2. The fixing table 1 is provided with a clamping mechanism 3 (clamp) for fixing the battery 4. The probe 2 contacts the battery 4 and transmits the electrical signal to the detection instrument. The detection instrument transmits the data to the detection system for obtaining the remaining capacity of the battery.

[0033] When using the high-precision battery remaining capacity detection device of this embodiment, the following steps are adopted: (1) Place the battery 4 to be tested on the fixing table 1.

[0034] (2) Position and tighten the battery 4 to be tested, install and tighten the probe 2, and connect the data cable of the testing instrument.

[0035] (3) According to the operation instruction, the battery 4 to be tested is measured, the AC impedance data of the battery 4 to be tested is obtained and transmitted to the detection system, the detection system stores the data in a database, and obtains the remaining capacity through the AC impedance data AC-IR (AC-IR) remaining capacity estimation algorithm.

[0036] Further, according to actual needs, the following software functions can be designed for the detection system part of the high-precision battery remaining capacity detection device of this embodiment: (1) One-click capacity test function: Connect the impedance tester communication interface to the detection system, which has built-in operating software. Click the software test button, enter the test battery number, and operate the system to communicate with the impedance tester to perform an impedance test. After the test is completed, the operating system will read the real part (R) and imaginary part (X) values ​​in the EXCEL table returned by the impedance tester, draw a grayscale value image, and input the remaining capacity estimation algorithm to estimate the remaining capacity. After the calculation is completed, the operating software displays the capacity prediction value.

[0037] (2) Storage of corresponding numbered capacity data: Click the software test button, enter the corresponding test battery number, and then perform an AC impedance test on the battery. After the test process is completed, the corresponding test data table for the battery is generated. The software automatically records the test number and test data table of the battery to the specified path on the computer.

[0038] (3) Manual / automatic calibration of the test algorithm: The operating system has a built-in automatic calibration function for the remaining capacity estimation algorithm. After the test equipment has been used for a period of time, the test algorithm is recalibrated to obtain a capacity estimation algorithm that is more suitable for the current environment. The remaining capacity estimation algorithm is formed by training the n underlying data that have been input using a neural network algorithm. Click the algorithm automatic calibration button, and the operating software will use the n underlying data that have been recorded to relearn the built-in remaining capacity estimation algorithm to obtain a new algorithm estimation formula. The average error of the relearned algorithm estimation formula can be greater than / less than the average error of the original estimation formula. Manual algorithm calibration can add new underlying data to the algorithm calibration compared to automatic calibration. In the capacity calibration mode, enter the battery numbers and corresponding actual capacities of m batteries in turn. After clicking capacity calibration, the system will use the n underlying data that have been recorded in the system and the newly added m new data to relearn the built-in remaining capacity algorithm estimation formula to obtain a new algorithm estimation formula with underlying data of n+m data. Similarly, the average error of the relearned algorithm estimation formula can be greater than / less than the average error of the original estimation formula and can be used in the new test environment.

[0039] (4) Historical data query: The operating system can record the test numbers and corresponding capacity data of the tested batteries. The software automatically records the test number and test data table of the battery to the specified path of the computer.

[0040] (5) Probe not clamped alarm function: The operating system has a probe not clamped alarm function. When the probe is not clamped, the real part (R) and imaginary part (X) values ​​in the EXCEL table returned by the impedance tester will have specific values. When the system collects impedance values, if it finds the specific values, it will automatically pop up a clamp not clamped alarm notification.

[0041] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. For those skilled in the art, improvements and changes obtained without departing from the technical concept of the present invention should also be regarded as the protection scope of the present invention.

[0042] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A high-precision method for detecting the remaining capacity of a battery, characterized in that: The following steps are involved: S1. Testing the AC impedance data of several batteries at different given frequencies, drawing an AC impedance spectrum with the real part of the AC impedance of a single battery at different given frequencies as the horizontal axis and the imaginary part as the vertical axis, to obtain the AC impedance spectrum of each battery; S2. Obtaining the actual remaining capacity of the plurality of batteries; S3, constructing an evaluation model for estimating the remaining capacity by using AC impedance according to the AC impedance spectrum set and the actual remaining capacity of the plurality of batteries using a composite neural network prediction model; S4, obtaining the remaining capacity of the battery to be tested according to the AC impedance spectrum of the battery to be tested and the evaluation model for estimating the remaining capacity by the AC impedance; In step S3, the specific steps of constructing an evaluation model for estimating the remaining capacity by using AC impedance include: S31, converting the AC impedance spectrum of each battery into a gray value matrix; S32, using the convolution kernel of the convolution neural network to perform weighted summation on the grayscale values ​​within the convolution kernel window from a certain point of the image, and then outputting the weighted summation result to the corresponding position of the output matrix; the convolution kernel with a set moving step size is continuously moved to completely scan the grayscale value matrix, and when the elements at each position of the output matrix are filled, the output matrix is ​​output; S33, inputting the output matrix into a linear rectification unit to increase the nonlinearity of the mapping process, inputting the output matrix after the nonlinearity is increased into a pooling layer for a pooling operation, and outputting an image feature matrix; S34, input the image feature matrix output by the convolutional neural network into a fully connected neural network, the fully connected neural network includes an input layer, an output layer and a hidden layer, each neuron in each layer performs a weighted summation of the input vector of the layer and adds a bias value, and inputs the result into a set linear rectifier unit activation function, after the linear rectifier unit activation function is calculated, the calculation result is used as the output value of the neuron, the output values ​​of the neurons in the same layer are combined into a vector in a fixed order and input into the next layer, and the above steps are repeated in the next layer, the last layer has only one neuron, the weighted summation of the input vector of the neuron itself and the result of adding the bias is used as the predicted value of the battery capacity by the composite neural network; S35. Perform model training based on the predicted value and the actual remaining capacity of the plurality of batteries to construct an evaluation model for estimating the remaining capacity by AC impedance.

2. The high-precision battery remaining capacity detection method according to claim 1, characterized in that: In step S33, the pooling operation refers to performing pooling processing on the output matrix after nonlinearity by using the maximum pooling method, and each element of the output matrix after pooling is the maximum element in the spatial window range corresponding to the input matrix of the pooling layer.

3. The high-precision battery remaining capacity detection method according to claim 1, characterized in that: In step S35, the model training is carried out by establishing an error function, taking the square of the difference between the predicted value of the battery capacity of a single battery and the actual remaining capacity of the battery as the prediction error, and taking the sum of the prediction errors of each battery as the error function of the composite neural network prediction model. The error function is used to perform iterative training, calculate the corresponding error and adjustment value of each parameter of each layer of neurons in the current situation, and adjust the parameters of the corresponding neurons. When the set number of iterations is reached or the loss value of the error function after the set number of iterations meets the set threshold, the training is stopped to complete the construction of the evaluation model for estimating the remaining capacity by AC impedance.

4. The high-precision battery remaining capacity detection method according to claim 3, characterized in that: In step S35, the back propagation algorithm and the Adam optimizer are used to calculate the corresponding error and adjustment value of each parameter of each layer of neurons in the current situation, and the parameters of the corresponding neurons are adjusted.

5. A high-precision battery remaining capacity detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 4 are implemented.

6. A high-precision battery remaining capacity detection device, characterized in that: The invention comprises a fixing table, a detection instrument, a probe and the detection system according to claim 5, wherein the fixing table is provided with a clamping mechanism for fixing the battery, the probe contacts the battery and transmits an electrical signal to the detection instrument, and the detection instrument sends the data to the detection system; the battery fixing table and the clamping mechanism are used to position the battery to be tested; The detection instrument is used to detect the AC impedance data of the battery to be tested; the detection system is used to execute a program calculation based on the AC impedance data to obtain the remaining capacity of the battery to be tested.

7. The high-precision battery remaining capacity detection device according to claim 6, characterized in that: The number of probes is four.

Images