Lithium ion battery multi-scene health state detection method and system

Abstract

The invention relates to the technical field of lithium ion battery health detection, in particular to a lithium ion battery multi-scene health state detection method and system, and the method comprises the steps: S1, collecting data; collecting discharge data of the to-be-tested battery; the battery discharge data is discharge data of a battery monomer in a laboratory scene, a battery pack in a road scene or a detection line scene; s2, calculating fused net discharge energy and accumulated discharge energy according to the discharge data collected in the step S1; s3, model detection; inputting the fused net discharge energy and the accumulated discharge energy into a pre-trained detection model to obtain an SOH value of the to-be-detected battery; wherein the detection model takes fused net discharge energy and accumulated discharge energy of a battery monomer in a laboratory scene or a battery pack in a road scene as input, and takes a battery SOH value as output; and the value range of the SOH of the battery is 0%-100%. And the SOH detection can be quickly and accurately carried out in a random SOC interval or under an unknown working condition.

Description

Technical Field

[0001] This disclosure relates to the field of lithium-ion battery health testing technology, and in particular to a method and system for detecting the health status of lithium-ion batteries in multiple scenarios. Background Technology

[0002] As a key component of electric vehicles and energy storage systems, the State of Health (SOH) of lithium-ion batteries directly affects the safety and reliability of the entire vehicle operation.

[0003] In existing technologies, it is not possible to directly detect the state of harm (SOH) of a battery. Currently, when evaluating the SOH of a battery, it is usually necessary to statistically analyze a large amount of data from similar batteries and train a predictive model using historical data of lithium-ion batteries of the same type and under the same operating conditions. The predictive model is then used to predict the battery SOH. For example, patent CN114563711A uses this method to predict the battery SOH. This method relies on a large amount of historical data from similar batteries and complete discharge data, which is inefficient. In addition, existing technologies based on the rapid detection method of constant voltage charging segment use current interval as a feature, which cannot be applied to random SOC intervals, such as patent CN106885994A.

[0004] Existing methods typically rely on complete charge-discharge cycle data or large data volume characteristics, making it difficult to quickly and accurately detect SOH in random SOC ranges or under unknown operating conditions. Summary of the Invention

[0005] This disclosure is made in view of the above-mentioned problems. This disclosure provides a method and system for detecting the health status of lithium-ion batteries in multiple scenarios.

[0006] According to one aspect of this disclosure, a method for detecting the health status of a lithium-ion battery under multiple scenarios is provided, comprising the following steps: S1, Data Acquisition; Acquire discharge data of the battery under test; The battery discharge data is discharge data of a single battery cell in a laboratory setting, a battery pack in a road setting, or a test line setting. S2, calculate the fused net discharge energy and cumulative discharge energy based on the discharge data collected in step S1; S3, Model Detection; The fused net discharge energy and cumulative discharge energy are input into a pre-trained detection model to obtain the SOH value of the battery under test; The detection model takes the fused net discharge energy and cumulative discharge energy of a single battery cell in a laboratory scenario or a battery pack in a road scenario as input, and the battery SOH value as output; The battery SOH value ranges from 0% to 100%.

[0007] In the lithium-ion battery multi-scenario health status detection method described above, optionally, the detection model in step S3 is trained through the following steps: S31, acquire training data, the training data including charge and discharge training data in laboratory scenarios and charge and discharge training data in road scenarios; S32, preprocess the charge / discharge training data to obtain time-series training data; S33, extract feature parameters within a preset voltage range or a corresponding SOC range, the feature parameters including fused net discharge energy and cumulative discharge energy, and establish a multivariate regression model with the number of cycles; S34, using a multiple regression model, extracts effective feature parameters according to the selected voltage range and the corresponding SOC range; S35 uses the effective feature parameters as input to the detection model and the battery SOH value as output to train the detection model.

[0008] In the multi-scenario health status detection method for lithium-ion batteries described above, it is optional to integrate the net discharge energy with the product of the collected voltage and the battery capacity. The multivariate regression model takes the number of discharge cycles as input and the combined net discharge energy and cumulative discharge energy as output. Effective feature parameters are selected using PCC and linear correlation coefficient indices.

[0009] In the multi-scenario health status detection method for lithium-ion batteries described above, optionally, the formula for the multiple regression model is as follows: ; Where x is the number of loops N, p0, p1, p2…p n It is a constant. The output is the combined net discharge energy and the cumulative discharge energy.

[0010] In the lithium-ion battery multi-scenario health status detection method described above, optionally, the preset voltage range is a series of continuous intervals between the upper cutoff voltage and the lower cutoff voltage or a series of continuous intervals between the lower cutoff voltage and the upper cutoff voltage. The preset SOC range consists of multiple consecutive ranges corresponding to the preset voltage range.

[0011] In the lithium-ion battery multi-scenario health status detection method described above, optionally, in step S32, the collected data is subjected to Gaussian filtering for noise reduction, and the data is completed using polynomial interpolation.

[0012] In the lithium-ion battery multi-scenario health status detection method described above, optionally, the number of preset voltage ranges is not less than 5.

[0013] In the lithium-ion battery multi-scenario health status detection method described above, optionally, the detection model is a BiGRU-Attention model based on a bidirectional gated recurrent unit and an attention mechanism.

[0014] This invention also proposes a multi-scenario health status detection system for lithium-ion batteries, comprising: The data acquisition module is used to collect the discharge data of the battery under test. The preprocessing module, electrically connected to the data acquisition module, is used to filter and interpolate the charging and discharging data acquired by the data acquisition module. The feature extraction module, electrically connected to the preprocessing module, is used to extract effective feature parameters within a preset voltage range or SOC range; the effective feature parameters include fused net discharge energy and cumulative discharge energy. The detection module is electrically connected to the feature extraction module; it is used to output the SOH value of the battery under test based on the effective feature parameters. The battery discharge data refers to the voltage of a single battery cell in a laboratory setting or the SOC value of a battery pack in a road setting; the detection module is equipped with a pre-trained detection model as described in any of the above.

[0015] In the lithium-ion battery multi-scenario health status detection system described above, the feature extraction module is used to extract effective feature parameters based on a pre-established multivariate regression model with respect to the number of cycles.

[0016] As will be described in detail below, the lithium-ion battery multi-scenario health status detection method and system according to embodiments of this disclosure collects data from individual battery cells and the entire vehicle battery pack, using net discharge energy and cumulative discharge energy as inputs and the battery SOH value as output to construct a detection model. Furthermore, for the same type of battery, the model is trained using individual battery cell discharge data in a laboratory setting and battery discharge data in a road setting, achieving cross-scenario universality and adaptability. During model training, by fusing voltage data and battery pack SOC data and extracting features according to a preset voltage range or preset SOC range, the amount of data can be significantly reduced. This facilitates accurate acquisition of SOH values ​​within random SOC ranges.

[0017] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description

[0018] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0019] Figure 1 This is a flowchart of the steps of the method proposed in this invention.

[0020] Figure 2 This is a comparison diagram of the preset voltage range and SOC range of the ternary lithium battery proposed in this invention.

[0021] Figure 3 This is a graph showing the relationship between discharge energy and cycle number in a certain example of the present invention.

[0022] Figure 4 This is a schematic diagram illustrating the error in the prediction result of a laboratory scene in an example of this invention.

[0023] Figure 5 This is a schematic diagram of the error in the road scene prediction result in a certain example of the present invention.

[0024] Figure 6 This is a schematic diagram of the feature parameters extracted from the actual vehicle inspection line after following the operating conditions in a certain example.

[0025] Figure 7 This is a schematic diagram of the error in the actual vehicle test results in a certain example of the present invention.

[0026] Figure 8 This is a structural block diagram of Embodiment 3 of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this disclosure more apparent, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure. It should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0028] Example 1 To address the problems mentioned in the background section, this embodiment proposes a multi-scenario health status detection method for lithium-ion batteries. Please refer to [link / reference]. Figure 1 The detection method includes the following steps: S1, Data Acquisition; Acquire discharge data of the battery under test; the battery discharge data can be individual battery cells in a laboratory setting, battery packs in a road setting, or discharge data from a testing line setting. In specific implementations, the data collected varies depending on the testing scenario. In a laboratory setting, individual battery cells can be easily tested; therefore, only the discharge data of individual battery cells needs to be collected. Constant current discharge is performed at a set rate, and during this process, the individual battery cell voltage, battery capacity, and cumulative discharge amount are collected. When performing health status testing on a battery pack, the battery pack voltage, battery pack capacity, and cumulative discharge amount during the testing process are collected. Battery pack testing can be performed in a road setting or on a testing line. In practical applications, testing on a testing line is one of the main application scenarios of this testing method. When testing on a testing line, the battery pack voltage, battery pack capacity, and cumulative discharge energy are obtained through a BMS.

[0029] It should be noted that when conducting tests on the testing line, the bench testing process disclosed in CN117268793A can be used. The similarities will not be repeated here, and those skilled in the art can implement it. The only difference is the data collected during the testing process.

[0030] S2, calculate the fused net discharge energy and cumulative discharge energy based on the discharge data collected in step S1. In practice, since the data used for training the detection model is the fused net discharge energy and cumulative discharge energy, this step requires calculation based on the data collected in step S1. Specifically, the formula for calculating the fused net discharge energy is as follows: ; in, The voltage collected. For collection The corresponding battery capacity at that time.

[0031] In this application, the fused net discharge energy is introduced as an input, wherein the fused net discharge energy is related to the collected voltage and battery capacity.

[0032] S3, Model Detection; The fused net discharge energy and cumulative discharge energy are input into a pre-trained detection model to obtain the SOH value of the battery under test; The detection model takes the fused net discharge energy and cumulative discharge energy of the battery cell in the laboratory scenario and the battery pack in the road scenario as input, and the battery SOH value as output; The battery SOH value ranges from 0% to 100%.

[0033] In this way, by using the combined net discharge energy and cumulative discharge energy as inputs, the SOH of the battery can be quickly detected, which is beneficial to improving the battery's cross-scenario application capability and enhancing its versatility and adaptability in different scenarios.

[0034] Example 2 The accuracy and cross-scenario application capability of the detection method disclosed in Embodiment 1 depend on the trained detection model. In practical applications, training the detection model is also a key factor. This embodiment discloses a specific training method for the detection model.

[0035] In this invention, the detection model is constructed as follows: This invention constructs a health detection module based on a bidirectional gated recurrent unit and an attention mechanism. This module runs a BiGRU-Attention algorithm within the system to perform regression calculations on low-data-volume features, thereby outputting the battery health status. In a laboratory scenario, low-data-volume feature parameters within multiple voltage ranges serve as the input to the detection model, with the battery health status detection result as the output. In a road scenario, low-data-volume feature parameters within multiple SOC ranges serve as the input to the detection model, with the battery health status result as the output. It should be noted that the constructed formula is implemented by the detection module running in the system and does not rely on manual calculation. In specific implementation, the number of voltage ranges is equal to and corresponds one-to-one with the number of SOC ranges; for example, the number of voltage ranges can be set to 7.

[0036] First, construct the GRU model, including update gates and reset gates. Update gate Z t Responsible for updating the hidden state. Input to (x) t and hidden state h t−1 The update gate is obtained through the sigmoid activation function. Specifically, the number of hidden units is h, the input is xt given as small sample data at time step t, the size of the dataset sample is n, and the number of input feature vectors is D and h from the previous time step. t−1 Its expression is: ; In the formula, σ represents the sigmoid activation function, which outputs a value between 0 and 1, and W... z It updates the weight parameters of the gate, b. z This updates the gate's bias parameters.

[0037] Reset door r t The calculation method for determining how much historical information should be ignored is as follows: ; In the formula, W r This is the weight matrix of the reset gate, b r It is the bias matrix of the reset gate.

[0038] Reset the door's output multiplied by its hidden state. The input vector is then added to it. The result is then passed to the `tanh` function to obtain the new memory content. When the element value in the reset gate is close to 0, it indicates that the hidden state from the previous time step will be discarded; when the element value is close to 1, it indicates that the hidden state from the previous time step will be retained. The calculation equation is as follows: ; In the formula, W h and b h These are the weight matrix and bias matrix of the new storage unit state, and tanh is used to transform the values ​​of all elements to the range [-1 1].

[0039] In obtaining After that, time step Hidden state It is through the update gate at the current time step. Hidden state of the previous time step Candidate hidden states at the current time step When the combination operation is performed =0, When the value is 1, the GRU network will degenerate into a simple recurrent neural network; when... =0, When it is 0, the current state Only with the current input Related to, and in relation to historical state Irrelevant. When When it is 1, the current state This is equivalent to the hidden state of the previous moment. .

[0040] ; Once the memory content is acquired, the output of the current neuron can be derived. A Bidirectional GRU (BiGRU) model is then constructed, consisting of forward GRU layers and backward GRU layers. The forward GRU processes the input sequence from beginning to end, while the backward GRU processes it in the opposite direction, enabling the network to capture features from both the past and future, increasing the diversity of feature information and solving the problem of information propagation direction. The unique network structure of BiGRU allows each hidden layer to capture information from the past (forward) and future (backward) at a specific time step, enabling the extraction of more comprehensive degradation features in lithium-ion batteries. This overcomes the regeneration phenomenon in real-world battery applications and enhances the network's predictive performance. Dropout technology is employed to prevent overfitting by temporarily removing unnecessary neurons from the hidden layers to reduce model complexity.

[0041] Preferably, through the forward hidden layer state and reverse hidden layer state The hidden layer state H of BiGRU at time t can be obtained. t Forward hidden layer state From the current input X t and time X t-1 Forward hidden layer state at time Confirmed. Backward hidden layer state. From the current input and time Backward hidden layer state Sure.

[0042] Following the constructed BiGRU model, a transposed convolution is introduced to build a transposed self-attention mechanism layer. Even with limited training samples, this facilitates the rapid and accurate propagation of feature information within the network, resulting in improved regression performance. The added transposed layer reduces the computational burden of feature learning, enabling comprehensive feature extraction and fusion, thereby promoting the generation of high-quality fused images. The input matrix x, based on the feature parameters, is (x1, x2, x3, … x…). n After passing through a normalization layer, a matrix Y is generated. Then, a 1×1 convolutional layer is used to gather the information of Y, and a 3×3 convolutional layer is used to generate Q, K, and V. The specific expressions are as follows: ; ; ; In the formula, It is a 1×1 convolution kernel. It is a 3×3 depthwise convolution.

[0043] Then, reshape Q and K to generate a transposed attention map, specifically expressed as: ; In the formula, It is a 1×1 convolution kernel. This is a transposed self-attention mechanism.

[0044] Preferably, the BiGRU-Attention model is further optimized by adding a 1D-CNN layer after the Attention input layer, taking advantage of the low computational complexity of convolutional neural networks. Simultaneously, to reduce the loss of feature parameter information, a transposed convolutional layer is added to accelerate information propagation.

[0045] Specifically, the input data for the 1D-CNN model consists of two feature parts, extracted from the capacity, voltage, and energy curves of different discharge voltage ranges within a single discharge cycle. To improve the stability of the 1D-CNN model and accelerate the training process, all input data are normalized. Finally, the capacity and energy of the voltage segments in different ranges are filled into a matrix X of size m×2.

[0046] ; By applying averaging or max pooling techniques to reduce the dimensionality of the feature parameter map, the output is sent to a fully connected layer for detection after flattening. To measure the impact of each feature on the detection model and further improve accuracy, a product form is chosen to calculate the attention weights of the current hidden layer state, as shown in the following equation.

[0047] ; Secondly, after calculating the attention weights of the hidden layer states of the model, the attention weights F are obtained using the Softmax function. t The normalized results are all scaled to [0, 1] as shown in the following formula.

[0048] ; Attention scores are obtained at each time step; a higher attention score indicates a greater contribution of the feature parameters at that time step to the model's regression. The attention layer then performs a weighted summation of each attention score with the corresponding hidden layer output from the BiGRU layer. Finally, the attention mechanism outputs the battery health status detection result Y. t The calculation formula is shown below.

[0049] ; In the formula, For attention weights, F t Here, q represents the normalized weight values, W and U are the query vectors, and W and U are the training parameters.

[0050] Furthermore, to determine the effectiveness of the BiGRU-Attention model, it was evaluated by comparing it with an independent BiGRU-Attention neural network model, where all weights and biases were initially randomized. The entire process used the Adam gradient descent algorithm to iterate through the neural network's hyperparameters each time. The Adam gradient descent algorithm is specifically as follows: ; In the formula, and It is the first and the The hyperparameter vector of the next iteration. This is a hyperparameter, set to 0.01. As an additional parameter, set it to 10. -5 .

[0051] and Update in each iteration using the following method: In the formula, and These are the moving average coefficients for calculating the gradient and its square, respectively, defined as 0.9 and 0.999 in this paper. and They are and Deviation correction value, Represents the loss function The gradient.

[0052] For the above model, in this embodiment, it is trained by the following method, that is, in embodiment one, the detection model in step S3 is trained by the following steps: S31, acquire training data, which includes charge-discharge training data in laboratory scenarios and charge-discharge training data in road scenarios.

[0053] In laboratory scenario experiments, design a For 0.3C constant current charging excitation, the battery is charged to the upper cutoff voltage. After charging, let it stand for 1 hour. Then, Discharge to the lower cutoff voltage under constant current conditions at C-rate. After resting for 1 hour, the battery is charged and discharged. This charge-discharge cycle is repeated until the battery's state of harmonics (SOH) drops to 80% or below, at which point the test ends. To ensure the accuracy of the test data, a capacity test is performed every 25 battery aging cycles. In practical applications, test data is directly collected using a battery charge-discharge test system with an accuracy of 0.02% of full scale and a sampling frequency of [missing information]. The collected data (Hz) is transmitted to a host computer for storage via electrical signals. In a laboratory setting, the system's state of charge (SOC), voltage, capacity, cumulative discharge energy, and cycle count of individual battery cells are collected in real time. The state of equilibrium (SOH) of each battery cell is calculated, specifically by dividing the capacity by the factory-set capacity.

[0054] In a road scenario, the vehicle's power battery SOC is maintained between 20% and 80%. After a 5-second pause, the battery is tested following the bench test conditions described in patent number CN202311391324.8. After completing one test, the battery is paused for 5 seconds before the bench test is repeated, and this cycle continues until the battery's SOC drops to a certain level. To ensure accuracy and reliability, the percentage of errors is below 10%. The specific steps for bench testing are clearly described in existing technology (patent number ZL202311391324.8) and will not be repeated here. During repeated bench testing cycles, battery pack SOC, voltage, capacity, cumulative discharge energy, and the number of cycle times are collected. On-board diagnostics (OBD) is used to collect electric vehicle bench test data, with an OBD sampling frequency of [missing information]. Hz.

[0055] It should be noted that in this invention, the training data and the detection object used for the same detection model should be of the same type of battery. For example, for a detection model for ternary lithium batteries, the data used for training and detection should both come from ternary lithium batteries, but it is not limited to whether they come from individual ternary lithium battery cells or ternary lithium battery packs. Of course, this invention can also be applied to lithium iron phosphate batteries. When applied to lithium iron phosphate batteries, the training and detection objects are both lithium iron phosphate batteries.

[0056] S32 preprocesses the charge / discharge training data to obtain time-series training data. In practice, the purpose of preprocessing is to remove noise and complete the data. Gaussian filtering can be used to denoise the acquired data, as shown in the following formula. The formula is implemented by the module running in the system and does not rely on manual calculation.

[0057] ; Where: µ is the mean; σ is the standard deviation.

[0058] In practice, polynomial interpolation was used to fill in missing values, and the original data was processed. Data interpolation with intervals of s. The mathematical function for the polynomial interpolation is as follows, and the formula is implemented by the module running in the system without relying on manual calculation.

[0059] In the formula: Let be the degree of the polynomial.

[0060] In this step, Gaussian filtering is first used to denoise the original data, and then interpolation is performed on the filtered data. Direct interpolation of the original data would amplify the noise; filtered data is smoother, more representative of the true trend, and the interpolation effect is more stable. In this step, interpolation is used both for completion and to increase the data sampling density. In some cases, the time interval for the original data acquisition is 1 second, and the time interval after interpolation is 10 ms. This step eliminates outlier data and increases the data volume, which is beneficial for improving detection accuracy.

[0061] S33. Extract feature parameters within a preset voltage range or a corresponding SOC range. The feature parameters include fused net discharge energy and cumulative discharge energy, and establish a multivariate regression model with the number of cycles.

[0062] The main purpose of this step is to construct characteristic parameters and low-data-volume voltage or SOC ranges; among which, the characteristic parameters include the fused net discharge energy C. wp Accumulated discharge energy C w The net discharge energy C is integrated. wp The collected voltage V and capacity C c The product relationship between them yielded the net discharge energy C. wp As shown in the following formula: .

[0063] A multiple regression model is then established to extract feature parameters, namely the mapping relationship between Cwp, Cw, and the number of discharge cycles N. Based on the principle of minimizing fusion error, the linear correlation coefficient R² is used as the evaluation index, and the fitting parameters are adjusted by setting an R² threshold range. The multiple regression model takes N as input and Cwp, Cw as output, with the mapping function denoted as f(x). The MR model is shown in the following equation. Where x is the number of cycles N, p0, p1, p2…pn are constants, and f(x) are the output feature parameters Cwp, Cw. Simulation experiments show that the cubic polynomial has the smallest error; therefore, the cubic polynomial is chosen as the regression function.

[0064] .

[0065] In this step, a multivariate regression model is constructed to extract the feature parameters used for training. Since both the fused net discharge and the cumulative discharge are related to the number of cycles, extracting the feature parameters in this way helps to retain accurate data and improve the model's accuracy and reliability.

[0066] S34, using a multiple regression model, extracts effective feature parameters according to the selected voltage range and the corresponding SOC range; In this step, by dividing the voltage range and the corresponding SOC range, feature parameters are extracted separately within the preset voltage range and SOC range. Using the corresponding voltage range or SOC range as the feature parameter extraction range facilitates feature parameter extraction with low data volume, making it easier to complete battery health status detection after training by charging and discharging only within certain ranges. For example, in one implementation, the preset voltage range is used as the feature parameter extraction range for laboratory scenarios, and the corresponding SOC range is used as the feature parameter extraction range for road scenarios.

[0067] For preset voltage ranges: Preset voltage ranges are multiple consecutive intervals between the upper and lower cutoff voltages, or multiple consecutive intervals between the lower and upper cutoff voltages; preset SOC ranges correspond to multiple consecutive intervals of the preset voltage ranges. The number of preset voltage ranges should be no less than 5. In an example of a ternary lithium battery, the nominal capacity is selected for laboratory scenario testing. It is 2.5Ah, and the nominal voltage is 3.7V, upper cutoff voltage The lower cutoff voltage is 4.2V. A 2.5V ternary lithium battery was used as the test subject. The nominal capacity for road scene testing was also selected. 150Ah, nominal voltage 350V, upper cutoff voltage 385V, lower cutoff voltage This is a 350V vehicle battery pack. Please refer to [link / reference]. Figure 2 The following voltage ranges were selected for feature parameter extraction in low-data-volume scenarios: 0.1V intervals were used as the starting point, with a stable voltage of 4.1V as the starting point, and the ranges were selected sequentially downwards. Seven voltage ranges were chosen as the feature parameter extraction ranges for the laboratory scene: [4.0-4.1V], [3.9-4.0V], [3.8-3.9V], [3.7-3.8V], [3.6-3.7V], [3.5-3.6V], and [2.5-3.5V]. Please refer to [the relevant documentation / reference]. Figure 3 The extracted feature parameters are denoted as FP1, FP2, FP3, FP4, FP5, FP6, and FP7, respectively. The SOC intervals corresponding to the seven voltage intervals extracted for the laboratory scene feature parameters are [98%-85%], [85%-74%], [74%-65%], [65%-52%], [52%-34%], [34%-19%], and [19%-0], which are used as the feature parameter extraction intervals for the road scene.

[0068] In a specific example, under laboratory conditions, the characteristic parameters of a single ternary lithium battery cell within each preset voltage range are shown in Table 1: Table 1. Characteristic parameters of each preset voltage range in the laboratory setting. It's important to note that interval partitioning essentially divides the overall curve from the upper cutoff voltage to the lower cutoff voltage into multiple continuous smaller intervals. Since the performance degradation sensitivity of a battery varies across different voltage ranges, certain intervals are more sensitive to health parameters such as capacity retention and internal resistance change rate. Interval partitioning amplifies these sensitive features locally, avoiding the loss of degradation information caused by overall averaging. Simultaneously, interval processing reduces the impact of transient noise and generates multi-dimensional feature inputs, improving the robustness of the model. Therefore, this interval partitioning significantly improves the accuracy of health state estimation.

[0069] S35 uses the effective feature parameters as input to the detection model and the battery SOH value as output to train the detection model.

[0070] During model training, both laboratory and road scenarios were trained offline using datasets to detect battery health and achieve capacity estimation with a low error rate. Generally, a detection accuracy of less than 5% is considered acceptable. This invention uses the root mean square error (RMSE) as a metric for prediction error. It is important to note that a smaller RMSE value indicates higher prediction accuracy and stronger detection capability.

[0071] ; In the formula, f i Let y be the regression value of sample i. i Let i be the true value of sample i.

[0072] Example 3 This embodiment uses the detection system to implement the method in Embodiment 1. The similarities will not be repeated here. Only the similarities will be explained below.

[0073] The detection system disclosed in this implementation includes a data acquisition module, a preprocessing module, a feature extraction module, and a detection module.

[0074] The data acquisition module is used to collect discharge data of the battery under test; specifically, the collected discharge data and the method for collecting discharge data can refer to the method described in Embodiment 1.

[0075] The preprocessing module is electrically connected to the data acquisition module and is used to filter and interpolate the charging and discharging data acquired by the data acquisition module. The preprocessing module is used to preprocess the data before performing step S2. The preprocessing method is the same as step S32 in Embodiment 2, and will not be described again here.

[0076] The feature extraction module is electrically connected to the preprocessing module and is used to extract effective feature parameters within a preset voltage range or SOC range. The effective feature parameters include the fused net discharge energy and the cumulative discharge energy. The feature extraction module is used to complete step S2 in Embodiment 1, that is, to calculate the fused net discharge energy and the cumulative discharge energy based on the data collected by the data acquisition module. For details, please refer to the calculation process of the fused net discharge energy and the cumulative discharge energy in Embodiment 2; the similarities will not be repeated here.

[0077] The detection module is electrically connected to the feature extraction module; it is used to output the SOH value of the battery under test based on the effective feature parameters. The detection module is used to implement step S3 in Embodiment 1, that is, to obtain the battery SOH value through the model based on the fused net discharge energy and cumulative discharge energy. The battery discharge data is the individual cell voltage under laboratory conditions or the SOC value of the battery pack under road conditions; the detection module is equipped with a detection model pre-trained according to Embodiment 2.

[0078] In practical implementation, the feature extraction module is used to extract effective feature parameters based on a pre-established multivariate regression model with a certain number of iterations. Specifically, the extraction process of the feature extraction module can be referred to in Embodiment 2, and will not be repeated here.

[0079] In practical implementation, the data acquisition module can obtain the vehicle battery pack's discharge data through the on-board diagnostic (OBD) system. For example, when conducting tests on a testing line, the testing process described in Example 1 can be followed, maintaining real-time communication between the data acquisition module and the OBD system during the testing process.

[0080] In a specific example, the prediction result for the laboratory scene is as follows: Figure 4 As shown, the detection error within the seven voltage ranges is within 2.79%. The road scene prediction results are as follows... Figure 5 As shown, the detection error within the seven voltage ranges is within 3.61%, meeting the detection requirements. It should be noted that... Figure 4 , Figure 5 and Figure 7 In the diagram, the numbers on the vertical axis represent percentages, such as... Figure 4 In the figure, 2.9 to 3.8 refers to 2.9% to 3.8%, and the percentage is not shown in the figure.

[0081] During real-vehicle multi-excitation testing, taking another pure electric vehicle as an example, the test line followed the bench test conditions. The maximum speed error between the test conditions and the following conditions was set to [-2km / h, 2km / h]. If the speed exceeded the maximum error range, the test was retested or the speed was considered invalid. Based on the set SOC range, the seven SOC discharge ranges verified in the real vehicle test were: [99%-88%], [88%-75%], [75%-60%], [60%-45%], [45%-30%], [30%-15%], and [15%-0.5%]. These ranges cover the battery SOC setting range for the road scenario described in S3. The feature parameters extracted after the real-vehicle test line followed the conditions are as follows: Figure 6 The actual vehicle inspection results are as follows: Figure 7 As shown in the figure. The results calculated by S5 are compared with those of the single cell, and the results show that they are consistent with the detection results of the single cell. The maximum estimation error of the detection in the 7 voltage ranges is 4.92%, which is within the error range.

[0082] Through the above embodiments, the present invention has achieved health detection with low error rate in laboratory, road and real vehicle random SOC conditions, with a maximum error of no more than 5%. Compared with the traditional method that relies on the complete charge and discharge curve, the present invention only requires seven interval features to achieve detection, reducing the amount of data required by more than 70%, which meets the needs of engineering applications.

[0083] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0084] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0085] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0086] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0087] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0088] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0089] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for detecting the health status of lithium-ion batteries in multiple scenarios, characterized in that, Includes the following steps: S1, Data Acquisition; Acquire discharge data of the battery under test; The battery discharge data is discharge data of a single battery cell in a laboratory setting, a battery pack in a road setting, or a test line setting. S2, calculate the fused net discharge energy and cumulative discharge energy based on the discharge data collected in step S1; S3, Model Detection; The fused net discharge energy and cumulative discharge energy are input into a pre-trained detection model to obtain the SOH value of the battery under test; The detection model takes the fused net discharge energy and cumulative discharge energy of a single battery cell in a laboratory scenario or a battery pack in a road scenario as input, and the battery SOH value as output; The battery SOH value ranges from 0% to 100%.

2. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to claim 1, characterized in that, The detection model in step S3 is trained through the following steps: S31, acquire training data, the training data including charge-discharge training data in laboratory scenarios and charge-discharge training data in road scenarios; S32, preprocess the charge / discharge training data to obtain time-series training data; S33, extract feature parameters within a preset voltage range or a corresponding SOC range, the feature parameters including fused net discharge energy and cumulative discharge energy, and establish a multivariate regression model with the number of cycles; S34, using a multiple regression model, extracts effective feature parameters according to the selected voltage range and the corresponding SOC range; S35 uses the effective feature parameters as input to the detection model and the battery SOH value as output to train the detection model.

3. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to claim 2, characterized in that, The net discharge energy is related to the product of the collected voltage and the battery capacity; The multivariate regression model takes the number of discharge cycles as input and the combined net discharge energy and cumulative discharge energy as output. Effective feature parameters are selected using PCC and linear correlation coefficient indices.

4. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to claim 3, characterized in that, The formula for the multiple regression model is: ； Where x is the number of loops N, p0, p1, p2…p n It is a constant. The output is the combined net discharge energy and the cumulative discharge energy.

5. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to claim 2, characterized in that, The preset voltage range is a series of consecutive intervals between the upper cutoff voltage and the lower cutoff voltage, or a series of consecutive intervals between the lower cutoff voltage and the upper cutoff voltage. The preset SOC range consists of multiple consecutive ranges corresponding to the preset voltage range.

6. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to claim 2, characterized in that, In step S32, the collected data is subjected to Gaussian filtering to remove noise, and the data is completed using polynomial interpolation.

7. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to claim 1, characterized in that, The number of preset voltage ranges is no less than 5.

8. The method for detecting the health status of lithium-ion batteries in multiple scenarios according to any one of claims 1-7, characterized in that, The detection model is a BiGRU-Attention model based on bidirectional gated recurrent units and attention mechanisms.

9. A multi-scenario health status detection system for lithium-ion batteries, characterized in that, include: The data acquisition module is used to collect the discharge data of the battery under test; The preprocessing module, electrically connected to the data acquisition module, is used to filter and interpolate the charging and discharging data acquired by the data acquisition module. The feature extraction module is electrically connected to the preprocessing module and is used to extract effective feature parameters within a preset voltage range or SOC range. Effective characteristic parameters include net discharge energy and cumulative discharge energy; The detection module is electrically connected to the feature extraction module; Used to output the SOH value of the battery under test based on the effective characteristic parameters; The battery discharge data refers to the voltage of a single battery cell in a laboratory setting or the SOC value of a battery pack in a road setting; the detection module is equipped with a pre-trained detection model as described in any one of claims 1-8.

10. The lithium-ion battery multi-scenario health status detection system according to claim 9, characterized in that, The feature extraction module is used to extract effective feature parameters based on a pre-established multivariate regression model with a certain number of iterations.

Images