A method and device for estimating the state of charge of a soft-pack lithium ion battery and a storage medium

Abstract

The present application relates to a kind of soft package lithium ion battery state of charge estimation method, device and storage medium based on dynamic stress and deep learning, wherein the method comprises: the battery charging and discharging data set containing dynamic stress signal is established, and the battery charging and discharging data set includes current, terminal voltage, dynamic stress data and reference state of charge;Offline training soft package lithium ion battery state of charge estimation model based on dynamic stress and deep learning: according to the input form required by deep learning network, the time series features of current, terminal voltage and dynamic stress in battery charging and discharging data set are reorganized into three-dimensional tensor sample set, the deep learning network structure of battery state of charge estimation model is built, and the battery state of charge estimation model is trained offline based on reference state of charge and sample set;Online battery state of charge estimation.Compared with prior art, the present application has the advantages of strong adaptability to different working conditions, good estimation performance when data length is limited, strong robustness to measurement noise interference and the like.

Description

Technical Field

[0001] This invention relates to the field of energy storage batteries for electric vehicles, and in particular to a method, apparatus, and storage medium for estimating the state of charge of a pouch lithium-ion battery based on dynamic stress and deep learning. Background Technology

[0002] Currently, the low-carbon vision is driving the development of new energy vehicles, with pure electric vehicles being a key route. The power battery is responsible for energy storage and power output. Lithium-ion batteries, due to their high energy / power density, long lifespan, and environmental friendliness, have become a research hotspot for vehicle manufacturers, universities, and research institutions. The state of charge (SOC) of a battery is defined as the ratio of remaining charge to maximum usable capacity. It indicates the remaining charge of the battery and is a crucial state variable for determining remaining driving range, charging / discharging power, and safe operating range.

[0003] Conventional methods for estimating the state of charge (SOC) include the ampere-hour integration method, model-based methods, and data-driven methods. The ampere-hour integration method, based on the definition of SOC, is an open-loop estimation method, highly susceptible to initial errors and measurement noise. Model-based methods estimate SOC by establishing a battery model and constructing adaptive filters; however, their accuracy, adaptability, and robustness to measurement noise are significantly influenced by the model and the observer. Data-driven methods estimate SOC based on the relationship between measurable battery information and its internal state.

[0004] The general idea behind data-driven methods is to measure battery current, voltage, temperature, etc., during charging and discharging, using this data directly as input to build and train a data-driven model to obtain suitable model parameters. The trained model is then used to estimate the battery's state of charge. Therefore, improving the estimation performance of data-driven methods mainly involves two issues: first, how to select and build a highly adaptive data-driven model; and second, how to acquire high-quality battery data.

[0005] Currently, data-driven models commonly used for state of charge estimation are mainly divided into traditional machine learning models and neural network models. Traditional machine learning methods include Gaussian process regression and support vector machines; neural network models include convolutional neural networks and recurrent neural networks. Compared with traditional machine learning techniques, deep learning techniques can handle relatively larger sliding window lengths, and deep learning models for processing time series include long short-term memory neural networks and gated recurrent neural networks.

[0006] Battery sensing technology and multidimensional data are key to data-driven battery state estimation. Conventional state-of-charge (POC) estimation methods typically use electrical and thermal signals for estimation. Current commonly used data-driven POC estimation methods usually use measurements of battery current, voltage, and temperature within a sliding window as model inputs. With the development of advanced battery sensing technology, it is hoped that more sensor signals can be utilized for battery state estimation, such as battery stress signals. During battery charging and discharging, the insertion and extraction of lithium ions from the electrode active material leads to structural changes in the electrode material, which in turn causes changes in battery volume and stress.

[0007] Combining advanced sensing and machine learning technologies holds promise for achieving more refined battery state estimation. Currently, the main problems with this approach are: first, existing data-driven models do not consider the time-series characteristics of battery measurement data and cannot effectively utilize historical information; second, current state-of-charge (SOC) estimation methods fail to effectively utilize battery stress information, thus failing to achieve accurate, effective, and reliable SOC estimation. Summary of the Invention

[0008] The purpose of this invention is to provide a method, device, and storage medium for estimating the state of charge (SOC) of a pouch lithium-ion battery based on dynamic stress and deep learning. This method considers the time-series characteristics of battery measurement data, effectively utilizes the battery's stress information, and improves the reliability of SOC estimation.

[0009] The objective of this invention can be achieved through the following technical solutions:

[0010] A method for estimating the state of charge (SOC) of a pouch lithium-ion battery based on dynamic stress and deep learning includes the following steps:

[0011] Step 1) Establish a battery charge-discharge dataset containing dynamic stress signals, wherein the battery charge-discharge dataset includes current, terminal voltage, dynamic stress data and reference state of charge;

[0012] Step 2) Offline training of a state-of-charge estimation model for soft-pack lithium-ion batteries based on dynamic stress and deep learning;

[0013] Step 21) Reorganize the time series features of current, terminal voltage and dynamic stress in the battery charging and discharging dataset into a three-dimensional tensor sample set according to the input format required by the deep learning network;

[0014] Step 22) Build the deep learning network structure of the battery state of charge estimation model, and train the battery state of charge estimation model offline based on the reference state of charge and the sample set;

[0015] Step 3) Online battery state of charge estimation:

[0016] Step 31) Acquire battery current, terminal voltage, and dynamic stress signals online during the actual charging and discharging process of the battery;

[0017] Step 32) Based on the trained battery state of charge estimation model, perform online estimation of the state of charge of the power battery.

[0018] Step 1) includes the following steps:

[0019] Step 11) Construct an experimental device for measuring the stress of a soft-pack battery using constraint fixtures and strain sensors;

[0020] Step 12) Design a battery charge-discharge experiment based on the recommended ambient temperature and charge-discharge current of the battery, conduct the test according to the set operating conditions, record the battery current, terminal voltage and dynamic stress during the test, calculate the battery reference state of charge, and construct a battery charge-discharge dataset.

[0021] In step 11), plane stress is measured in a laboratory environment using a constraint fixture and a stress sensor. In vehicle applications, the stress sensor is placed between two pouch batteries to measure plane stress.

[0022] In step 12), the dynamic stress S recorded during the test is... d It is the stress related to the battery's operating state, and is the total stress S. t Subtract the initial static stress S s The value after:

[0023] S d =S t -S s

[0024] The total stress is the instantaneous stress measured directly on the battery surface by a pressure sensor during charging / discharging, while the static stress is the stress obtained after the battery has reached equilibrium with sufficient rest time.

[0025] The battery reference state of charge is calculated according to the definition:

[0026]

[0027] Among them, SOC k and I k These are the battery state of charge and current at time k, respectively, where Δt is the sampling time, and C is the current. m It refers to battery capacity.

[0028] In step 12), the set operating conditions include: long-term charge and discharge operating conditions, short-term charge and discharge operating conditions, pulse charge and discharge operating conditions, and dynamic driving operating conditions.

[0029] In step 21), the battery charge / discharge data at time k is x.k =[I k V k ,F k ], where I k It is the current at time k, V k F is the terminal voltage at time k. k It is the dynamic stress at time k;

[0030] Using a sliding window of length n, the input X used to train the deep learning network is obtained. k =[x k-n+1 ,…,x k-1 ,x k The data length is denoted as N. data The sample size obtained is:

[0031] N samples =N data -n+1

[0032] Reorganize the data into a three-dimensional tensor form: [N] samples ,n,N features ], where N features Indicates the number of features.

[0033] In step 22), the deep learning network includes one input layer, two LSTM hidden layers, one fully connected layer, and one output layer. The activation process of the LSTM network is as follows:

[0034]

[0035] Among them, f t Forgotten Gate, i t It's an input gate, o t It's an output gate. W is a candidate state in the update process. x It is the weight, b x σ is the bias, σ(·) is the sigmoid activation function, and tanh(·) is the hyperbolic tangent activation function.

[0036] In step 32), the method for estimating the state of charge of the battery is as follows:

[0037] SOC * =LSTM(x * )

[0038] Wherein, LSTM is a trained battery state-of-charge estimation model, x * The input vector consists of the battery current, terminal voltage, and dynamic stress acquired online.

[0039] A plug-and-charge function testing device for a charging pile includes a memory, a processor, and a program stored in the memory. When the processor executes the program, it implements the method described above.

[0040] A storage medium having a program stored thereon, which, when executed, implements the method described above.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The battery state of charge estimation method of the present invention considers dynamic stress signals in addition to current and voltage, which can reflect the volume changes caused by lithium ion insertion and extraction during battery charging and discharging. Combined with deep learning network, it realizes online estimation of battery state of charge, making the model highly adaptable to different operating conditions, with good estimation performance under limited data length, and strong robustness to battery measurement noise interference. Attached Figure Description

[0043] Figure 1 This is a flowchart of the method of the present invention;

[0044] Figure 2 This is a schematic diagram illustrating the relationship between battery stress and state of charge.

[0045] Figure 3 The results show the state-of-charge estimation of a pouch lithium-ion battery based on dynamic stress and deep learning.

[0046] Figure 4 This is a diagram showing the error in the state of charge estimation obtained based on the method of this invention. Detailed Implementation

[0047] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0048] This embodiment provides a method for estimating the state of charge of a soft-pack lithium-ion battery based on dynamic stress and deep learning, such as... Figure 1 As shown, it includes the following steps:

[0049] Step 1) Establish a battery charge-discharge dataset containing dynamic stress signals, wherein the battery charge-discharge dataset includes current, terminal voltage, dynamic stress data and reference state of charge.

[0050] Step 11) Construct an experimental device for measuring the stress of a soft-pack battery using constraint fixtures and strain sensors.

[0051] In addition to measuring voltage, current, and temperature, the planar stress of a battery needs to be measured during its operation. In a laboratory environment, planar stress is measured using a constraint fixture and a stress sensor. In automotive applications, the stress sensor is placed between two pouch cells to measure the planar stress.

[0052] The battery used in this implementation case is a pouch lithium-ion battery, with lithium manganese oxide as the positive electrode and graphite as the negative electrode. This battery has a nominal capacity of 8Ah, a nominal voltage of 3.7V, a charging cut-off voltage of 4.2V, and a discharging cut-off voltage of 2.8V.

[0053] Step 12) Design a battery charge-discharge experiment based on the recommended ambient temperature and charge-discharge current of the battery, conduct the test according to the set operating conditions, record the battery current, terminal voltage and dynamic stress during the test, calculate the battery reference state of charge, and construct a battery charge-discharge dataset.

[0054] During battery charging and discharging, the insertion and extraction of lithium ions between the two electrodes causes changes in battery volume, resulting in stress. The dynamic stress S recorded during the test... d It is the stress related to the battery's operating state (such as current, state of charge, etc.), and is the total stress S. t Subtract the initial static stress S s The value after:

[0055] S d =S t -S s

[0056] The total stress is the instantaneous stress measured directly on the battery surface by a pressure sensor during charging / discharging, while the static stress is the stress obtained after the battery has reached equilibrium during a sufficient rest period (usually 2 hours).

[0057] In this step, the set operating conditions should cover the actual working scenarios of the battery as much as possible. In this embodiment, these include long-term charge and discharge conditions, short-term charge and discharge conditions, pulse charge and discharge conditions, and dynamic driving conditions, in order to improve the adaptability of the estimation model.

[0058] This embodiment designs four typical operating conditions to simulate the actual charging and discharging of the battery.

[0059] Operating Condition 1: Constant Current Condition, simulating long-term charge and discharge of the battery. The battery is charged and discharged at a constant current rate of 1C, denoted as Cy1 and Cy2.

[0060] Operating Condition 2: Short-term operating condition, simulating short-term charge and discharge of the battery. The battery is charged to a set state of charge (SOC) at different current rates (e.g., 0.25C, 0.5C, 1.5C, and 2C), and then discharged to 0% SOC at 1C, denoted as Cy3-Cy6. The set SOCs are 20%, 40%, 60%, and 80%, respectively.

[0061] Operating Condition 3: Pulse Operating Condition, simulating the power characteristics of the battery. This condition includes two sub-conditions: Sub-condition 1 uses different current rates (0.25C, 0.5C, 1.5C, and 2C) to charge the battery to the set state of charge, and then lets the battery rest for 2 hours, denoted as Cy7-Cy10; Sub-condition 2 first charges the battery to the set state of charge with a constant current of 1C, and then applies a 60-second discharge-charge pulse with different current rates (0.25C, 0.5C, 1.5C, and 2C), and lets it rest for 1 hour between two adjacent state of charge points, denoted as Cy11-Cy14 respectively.

[0062] Operating Condition 4: Dynamic Operating Conditions, simulating the actual driving of electric vehicles. Two commonly used driving conditions are adopted—the New European Driving Cycle (NEDC) and the Urban Road Cycle (UDDS), denoted as Cy15 and Cy16 respectively.

[0063] The measured battery strain is the strain value after subtracting the initial pressure. During the experiment, the fixture and test battery were placed in a constant temperature chamber to ensure environmental consistency, with the ambient temperature set at 25°C.

[0064] The battery reference state of charge (SOC) is calculated by current integration according to the definition, and serves as the output label for offline training. The battery SOC is defined as the ratio of the remaining usable charge to the battery's current capacity, which can be expressed using the previous time step's SOC. k-1 Add the current change in charge I k Δt (discharge time I) k When I is negative, it indicates that I is charging. k (positive) and capacity C m The ratio is calculated as follows: in this embodiment, the sampling time Δt = 1s, and the battery capacity C... m =8Ah. Therefore, the battery reference state of charge is calculated as follows:

[0065]

[0066] like Figure 2As shown, the relationship between battery stress and state of charge (SCC) is relatively unaffected by the battery's operating current. The relationship between battery stress / strain and SCC is quantitatively evaluated using Pearson and Spearman correlation coefficients. The Pearson correlation coefficient indicates a linear correlation between dynamic stress and SCC, while the Spearman correlation coefficient indicates a monotonic relationship. The two correlation coefficients between battery dynamic stress and SCC exceed 0.97 and 0.985, respectively, indicating a strong correlation between the battery dynamic stress signal and SCC, making them suitable as inputs for SCC estimation models.

[0067] Step 2) Offline training of a state-of-charge estimation model for soft-pack lithium-ion batteries based on dynamic stress and deep learning.

[0068] Step 21) Reorganize the time series features of current, terminal voltage and dynamic stress in the battery charging and discharging dataset into a three-dimensional tensor sample set according to the input format required by the deep learning network.

[0069] The battery charge / discharge data at time k is x k =[I k V k ,F k ], where I k It is the current at time k, V k F is the terminal voltage at time k. k It is the dynamic stress at time k.

[0070] A sliding window of length n is used, meaning that battery sensor data from time k-n+1 to k is used at time k to obtain the input X for training the deep learning network. k =[x k-n+1 ,…,x k-1 ,x k The data length is denoted as N. data The sample size obtained is:

[0071] N samples =N data -n+1

[0072] Reorganize the data into a three-dimensional tensor form: [N] samples ,n,N features ], where N features The number of features is 3 in this embodiment, which are current, terminal voltage and dynamic stress.

[0073] Table 1 shows the data partitioning matrix used for training the state of charge estimation model and for online testing in this embodiment:

[0074] Table 1 Dataset Partition Matrix

[0075]

[0076] Step 22) Build the deep learning network structure of the battery state of charge estimation model, and train the battery state of charge estimation model offline based on the reference state of charge and the sample set.

[0077] In this embodiment, the deep learning network includes a sequence input layer, two LSTM hidden layers with 0.05 Dropout, a fully connected layer, and a regression layer. Each LSTM layer selects 100 hidden units. The sequence input layer is responsible for inputting sequence data into the constructed network, the fully connected layer multiplies the LSTM output by the weight matrix and adds a bias, and the regression layer performs the regression task.

[0078] The hidden layer consists of multiple repeating LSTM units. Each LSTM unit transitions two states to the next unit, i.e., the unit state (c t ) and hidden state (h t ). x t and y t These are the corresponding model input and output sequences. The cell states contain information learned from the previous time step, and the hidden states are also called output states. At each time step, the LSTM layer adds valid information to or discards invalid information from previous cell states, including forgetting, updating, and outputting cell and hidden states. These operations are controlled by three different gates, including the input gate i... t Forgotten Gate t and output gate o t The forgetting process determines which information should be discarded or retained using forget gates. The update process controls the level of state updates and adds information to the cell state using input gates. Finally, the output gates and the cell state are used to determine the next hidden state. W represents the candidate state of the update process. x b represents the weight. x The bias is σ(·), which is the sigmoid activation function, and tanh(·), which is the hyperbolic tangent activation function. The key activation operations are as follows:

[0079]

[0080] The LSTM model uses a reference state of charge (SOC) and a three-dimensional tensor sample set as training data. The label at time k is SOC. k The input is X k .

[0081] The LSTM model was trained using the Adam optimizer, with an initial learning rate of 0.001. The learning rate was decreased in segments every 100 epochs with a decrease factor of 0.2; the gradient decay factor β1 was 0.9, and the squared gradient decay factor β2 was 0.999. Due to the inherent randomness of deep learning model training, each model was trained three times, and the final state-of-charge estimation result was the average of the three training iterations.

[0082] Step 3) Online battery state of charge estimation

[0083] Step 31) Acquire battery current, terminal voltage and dynamic stress signals online during the actual charging and discharging process of the battery.

[0084] In automotive applications, the battery management system collects the battery's current, terminal voltage, and dynamic stress in real time as model input. It's important to note that, following step 21), the online measured battery current, terminal voltage, and dynamic stress data must also be reconstructed into a three-dimensional tensor form.

[0085] Step 32) Based on the trained battery state-of-charge estimation model, perform online estimation of the power battery's state of charge:

[0086] SOC * =LSTM(x * )

[0087] Wherein, LSTM is a trained battery state-of-charge estimation model, x * The input vector consists of the battery current, terminal voltage, and dynamic stress acquired online.

[0088] like Figure 3 The results shown are the state-of-charge (SOC) estimation results for Cy14. It can be seen that both the mechanical signal-based and deep learning-based SOC estimation methods can accurately estimate the SOC of the power battery in all three scenarios. Figure 4 As shown, the overall RMSE and MAE of the state of charge estimation for each cycle are 1.88% and 1.35%, respectively. The maximum RMSE in these cycles is less than 5%, and the maximum MAE is less than 4%, demonstrating good estimation accuracy. As shown in Table 1, in this embodiment, operating condition 4 in scenarios 1 and 2 was not included in the training, but it obtained small RMSE and MAE during online estimation, indicating that the method is adaptable to different operating conditions.

[0089] In addition, due to the addition of stress data, the accuracy of battery state of charge estimation is improved by 0.24% compared to using only current and voltage information. Therefore, it is less sensitive to the length of the sliding window and can achieve good estimation performance with limited data length.

[0090] In step 3), after adding 0.5% measurement noise to the three measurements, the state of charge estimation error only increased by 5.09% compared to the noise-free data, demonstrating the good robustness of the present invention.

[0091] In summary, one embodiment of the present invention is feasible, and the estimation result has a small error with the actual state of charge data, is adaptable to different operating conditions, supports estimation with small data lengths, and has good robustness to measurement noise.

[0092] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0093] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for estimating the state of charge of a soft-pack lithium-ion battery based on dynamic stress and deep learning, characterized in that, Includes the following steps: Step 1) Establish a battery charge-discharge dataset containing dynamic stress signals, wherein the battery charge-discharge dataset includes current, terminal voltage, dynamic stress data, and reference state of charge; Step 11) Construct an experimental device for measuring the stress of a soft-pack battery using constraint fixtures and stress sensors; Step 12) Design a battery charge and discharge experiment based on the recommended ambient temperature and charge and discharge current of the battery, conduct the test according to the set working conditions, record the battery current, terminal voltage and dynamic stress during the test, calculate the battery reference state of charge, and construct a battery charge and discharge dataset. Among them, the dynamic stress recorded during the test It is the stress related to the battery's operating state, and is the total stress. Subtract initial static stress The value after: The total stress is the instantaneous stress measured directly on the battery surface by a pressure sensor during charging / discharging, while the static stress is the stress obtained after the battery has reached equilibrium with sufficient rest time. The battery reference state of charge is calculated according to the definition: in, and They are k The battery state of charge and current at any given time. It is the sampling time. It refers to battery capacity; Step 2) Offline training of a state-of-charge estimation model for soft-pack lithium-ion batteries based on dynamic stress and deep learning; Step 21) Reorganize the time series features of current, terminal voltage and dynamic stress in the battery charging and discharging dataset into a three-dimensional tensor sample set according to the input format required by the deep learning network; in, k Battery charge / discharge data at any given time ,in, yes k Current at any moment yes k Terminal voltage at time , yes k Dynamic stress at any given moment; Use a length of n The sliding window provides the input for training the deep learning network. The data length is denoted as The sample size obtained is: Reorganize the data into a three-dimensional tensor form: [ , n , ],in, Indicates the number of features; Step 22) Build the deep learning network structure of the battery state of charge estimation model, and train the battery state of charge estimation model offline based on the reference state of charge and the sample set; Step 3) Online battery state of charge estimation: Step 31) Acquire battery current, terminal voltage, and dynamic stress signals online during the actual charging and discharging process of the battery; Step 32) Based on the trained battery state of charge estimation model, perform online estimation of the state of charge of the power battery; In step 22), the deep learning network includes one input layer, two LSTM hidden layers, one fully connected layer, and one output layer. The activation process of the LSTM network is as follows: in, Forgotten Gate It's an input gate. It's an output gate. It is a candidate state in the update process. It's weight. It's a bias. It is the sigmoid activation function. It is the hyperbolic tangent activation function.

2. The method for estimating the state of charge of a soft-pack lithium-ion battery based on dynamic stress and deep learning according to claim 1, characterized in that, In step 11), plane stress is measured in a laboratory environment using a constraint fixture and a stress sensor. In vehicle applications, the stress sensor is placed between two pouch batteries to measure plane stress.

3. The method for estimating the state of charge of a soft-pack lithium-ion battery based on dynamic stress and deep learning according to claim 1, characterized in that, In step 12), the set operating conditions include: long-term charge and discharge operating conditions, short-term charge and discharge operating conditions, pulse charge and discharge operating conditions, and dynamic driving operating conditions.

4. The method for estimating the state of charge of a soft-pack lithium-ion battery based on dynamic stress and deep learning according to claim 1, characterized in that, In step 32), the method for estimating the state of charge of the battery is as follows: Here, LSTM is a trained battery state-of-charge estimation model. The input vector consists of the battery current, terminal voltage, and dynamic stress acquired online.

5. A testing device for the plug-and-charge function of a charging pile, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1-4.

6. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method as described in any one of claims 1-4.

Images