A rapid detection method for health state of lithium iron phosphate battery pack for vehicle

Abstract

The present application relates to a kind of quick detection method for the health state of lithium iron phosphate battery pack for vehicle, belong to electric vehicle detection technical field;Method includes: based on the cycle aging data of lithium iron phosphate battery of pre-set, constructs SOH evaluation model;Collect the voltage data of the charge-discharge stage of battery pack of the vehicle to be detected;The voltage data of battery pack of the vehicle to be detected is input into the SOH evaluation model, and the health state of battery pack is quickly detected.The present application utilizes the method of deep learning, by mining the relationship between the charging capacity and battery health degree in differential voltage interval, constructs the effective evaluation model of lithium iron phosphate battery pack.Effectively expand the application of battery SOH detection method based on data driving in the field of vehicle detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric vehicle detection, in particular to a rapid detection method for the health state of a vehicle lithium iron phosphate battery pack. BACKGROUND

[0002] New energy vehicles have become an important direction of the development of the automobile industry due to their low pollution and high energy efficiency. As a core component of new energy vehicles, the performance and life of the battery pack directly affect the reliability, economy and user experience of the vehicle. The health assessment (SOH) of the battery pack is a key link to ensure the safety of the battery, improve the service life, optimize energy management and reduce operation and maintenance costs.

[0003] However, there is currently a lack of direct detection means for the health state of vehicle battery packs, especially for lithium iron phosphate battery packs. The main reasons are as follows: first, lithium-ion batteries convert chemical energy into electrical energy through a series of electrochemical coupling reactions. Without disassembling the pack, the battery pack is equivalent to a "black box" to the outside world, and it is difficult to detect the micro-chemical reactions related to aging occurring inside. Therefore, in actual measurement, only physical quantities such as voltage and current during the charging and discharging process can be used for indirect estimation, which poses a challenge to detection; second, the charging and discharging curve of ternary lithium changes relatively linearly with SOH, and the subtle changes in voltage can reflect the changes in SOC or capacity, which is conducive to the evaluation of SOH, while the charging and discharging curve of lithium iron phosphate battery has a clear "platform region", i.e. within most of the SOC range, the voltage change is very flat, resulting in the failure of most traditional detection methods; third, the capacity degradation of ternary lithium batteries is accompanied by a significant increase in internal resistance and lithium metal deposition, and the degradation process is relatively concentrated and easy to detect, while the degradation mechanism of lithium iron phosphate batteries includes gradual failure of active materials, increased polarization, changes in SEI film, etc., which do not always significantly affect the internal resistance or other characterization parameters, so it is difficult to evaluate by simple electrochemical methods.

[0004] In summary, in the field of vehicle detection, the SOH evaluation model and algorithm for vehicles equipped with lithium iron phosphate battery packs are relatively lagging behind. Many existing SOH evaluation methods are more suitable for ternary lithium batteries, and algorithms for lithium iron phosphate batteries still need more research and optimization.

[0005] The continuous progress of big data science and the wide application of deep learning technology in the engineering field provide an advanced approach to SOH evaluation of vehicle battery packs driven by operational data, meeting the new requirements for evaluation methods.

[0006] The existing latest technical solution is based on the acquisition of characteristic data such as current, voltage, temperature and the like (in most cases, voltage) during the battery charging process, and the health state of the battery is evaluated by analyzing and modeling these data, which is called a data-driven method. The general process is divided into: 1. Data acquisition. Collect voltage data from the battery system during battery charging; 2. Data preprocessing. Clean, correct and preprocess the collected data to ensure data consistency and accuracy; 3. Feature extraction. Extract features related to the health state of the battery from the processed data based on deep learning or machine learning; 4. Modeling. A mathematical model that relates the SOH of the battery to the features is established using the extracted features. 5. Calculate the result. Calculate the SOH of the battery through the model. Figure 1 A flowchart of the data-driven method.

[0007] The above method has been successfully applied to the health state detection of ternary lithium battery packs, and ideal evaluation results have been achieved. However, when the evaluation object is changed to a lithium iron phosphate battery pack, the aging mechanism of lithium iron phosphate is more complex, and there is a significant "platform region" in the charging and discharging process, which reduces the evaluation accuracy or causes the model to fail. Therefore, it is difficult to extract the relationship between voltage and capacity change during the charging and discharging stage through traditional methods.

[0008] Therefore, how to improve the detection capability of the detection model for lithium iron phosphate battery packs, especially the detection capability of the detection model for the battery packs inside the vehicle without disassembling the battery pack, is a key problem for the practical application of the data-driven detection method in the health state evaluation of the battery packs in the vehicle. SUMMARY

[0009] The purpose of the present application is to provide a rapid detection method for the health state of a lithium iron phosphate battery pack for vehicles to solve the problems existing in the prior art. The evaluation model obtained based on part of the charging training data is successfully applied to the SOH detection of the battery pack inside the vehicle. The method of evaluating the health degree of the battery by detecting the change of the charging capacity in the differential voltage stage can sensitively capture the battery capacity in the voltage micro-change interval, thereby indirectly evaluating the degradation degree of the active material inside the battery, and improving the evaluation performance of the model for lithium iron phosphate batteries.

[0010] To achieve the above purpose, the present application provides the following scheme:

[0011] A rapid detection method for the health state of a lithium iron phosphate battery pack for vehicles, comprising:

[0012] Based on the preset cycle aging data of the lithium iron phosphate battery, an SOH evaluation model is constructed;

[0013] Collecting voltage data of the charging and discharging stage of the battery pack of the vehicle to be detected;

[0014] The voltage data of the vehicle battery pack to be detected is input into the SOH evaluation model to quickly detect the state of health of the battery pack.

[0015] Optionally, constructing the SOH evaluation model based on the preset cycle aging data of the lithium iron phosphate battery comprises:

[0016] Collecting voltage data during the cycle aging process of the preset lithium iron phosphate battery;

[0017] Standardizing all voltage data during the cycle aging process;

[0018] Based on the standardized voltage data, a △Q-△U curve is drawn; wherein the △Q-△U curve is a corresponding relationship curve between the voltage and the charging capacity △Q in each △U stage as the voltage rises from the starting voltage;

[0019] Constructing a deep learning model with a multi-head attention mechanism as the core;

[0020] Training the deep learning model using the △Q-△U curve to obtain an SOH evaluation model.

[0021] Optionally, collecting voltage data during the cycle aging process of the preset lithium iron phosphate battery comprises:

[0022] For the preset lithium iron phosphate battery, constant current charging and discharging is performed at a current of 1C rate until the maximum available capacity of the lithium iron phosphate battery is below a preset threshold value of the rated available capacity, and voltage data of all charging stages is collected.

[0023] Optionally, the expression of the standardization processing is:

[0024] U 包 ’=U 包 ×U 额，单 / U 额，包

[0025] Wherein, U 包 ’ represents the battery pack voltage data after standardization, U 额，包、 U 额，单 respectively represent the rated voltage of the battery pack, the single rated voltage, U 包 represents the voltage data during the cycle aging process.

[0026] Optionally, inputting the voltage data of the vehicle battery pack to be detected into the SOH evaluation model comprises:

[0027] Pretreating the voltage data of the vehicle battery pack to be detected to obtain a standardized time series;

[0028] The normalized time series is input into a multi-head attention mechanism, the correlation between time steps is calculated to capture global context dependency, and a hidden feature representation is obtained;

[0029] The hidden feature representation is input into a feedforward neural network, and a global average pooling layer is used to reduce the time step dimension of the feedforward neural network output to extract a global feature representation of the sequence.

[0030] The global feature representation is converted into a scalar output through a fully connected network, representing the predicted SOH value.

[0031] Optionally, the multi-head attention mechanism includes several independent attention heads, each of which extracts dependency characteristics in different feature subspaces, and finally splices the output results of all attention heads to form a hidden feature representation of a predetermined dimension.

[0032] Optionally, the feedforward neural network includes a two-layer structure; the first layer includes a plurality of neurons and uses a ReLU activation function to increase the non-linear representation capability, and the second layer compresses the features to a dimension matching the attention output to ensure the consistency of the feature space.

[0033] Optionally, constructing the SOH evaluation model further includes:

[0034] The mean square error of the predicted SOH and the actual SOH of the predetermined lithium iron phosphate battery is calculated, and the model parameters are updated with an initial learning rate combined with an Adam optimizer.

[0035] The beneficial effects of the present application are:

[0036] The present application uses a deep learning method to construct an effective evaluation model for lithium iron phosphate battery packs by mining the relationship between the charge capacity and the battery health degree in the differential voltage interval. The application effectively expands the application of data-driven battery SOH detection methods in vehicle detection.

[0037] Compared with existing methods, the advantages of the present application mainly lie in the following three aspects: first, the present application constructs an SOH evaluation model by calculating the charge capacity in the differential voltage interval, effectively solving the inaccurate measurement problem caused by the "platform area" of the lithium iron phosphate battery charging stage. Second, the present application can effectively detect the SOH of electric vehicles equipped with lithium iron phosphate battery packs without disassembly, filling the gap in the detection method of lithium iron phosphate battery packs in the vehicle detection field. Third, based on the artificial intelligence large model, the battery SOH detection model is constructed to extract the conversion relationship between the battery data and the battery SOH, which has higher accuracy than traditional models. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only illustrate some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0039] Figure 1 A schematic diagram of a data-driven method is shown.

[0040] Figure 2 A flowchart of a rapid detection method for the health state of a lithium iron phosphate battery pack for vehicles according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0043] The present embodiment proposes a rapid detection method for the health state of a lithium iron phosphate battery pack for vehicles, which comprises:

[0044] Based on the preset cycle aging data of the lithium iron phosphate battery, an SOH evaluation model is constructed.

[0045] Voltage data of the charging and discharging stage of the battery pack of the vehicle to be detected is collected.

[0046] The voltage data of the battery pack of the vehicle to be detected is input into the SOH evaluation model for rapid detection of the health state of the battery pack.

[0047] Specifically, the method of the present embodiment is divided into a model construction stage and a battery pack detection stage. In the model construction stage, based on the cycle aging data of the lithium iron phosphate battery prepared in advance, an SOH evaluation model is constructed. In the detection stage, the model is used to evaluate the battery pack of the vehicle to be detected, and the SOH evaluation result of the battery pack is output.

[0048] Further, based on the preset cycle aging data of the lithium iron phosphate battery, the SOH evaluation model comprises:

[0049] Voltage data in the cycle aging process of the preset lithium iron phosphate battery is collected.

[0050] standardizing voltage data in all cycle aging processes;

[0051] drawing a delta Q-delta U curve based on the standardized voltage data, wherein the delta Q-delta U curve is a curve of the corresponding relationship between the voltage and the charging capacity in each delta U stage starting from the initial voltage;

[0052] constructing a deep learning model with a multi-head attention mechanism as the core;

[0053] training the deep learning model using the delta Q-delta U curve to obtain an SOH evaluation model.

[0054] Further, collecting the voltage data of the preset lithium iron phosphate battery in the cycle aging process includes:

[0055] The preset lithium iron phosphate battery is subjected to constant current charging and discharging at a current of 1C rate until the maximum available capacity of the lithium iron phosphate battery is below a preset threshold value of the rated available capacity, and the voltage data of all charging stages is collected.

[0056] Further, constructing the SOH evaluation model further includes:

[0057] calculating the mean square error of the predicted SOH and the actual SOH of the preset lithium iron phosphate battery, and updating the model parameters with an initial learning rate combined with an Adam optimizer.

[0058] Specifically, in the model construction stage of the present embodiment, the lithium iron phosphate battery is subjected to constant current charging at a current of 1C rate until the maximum charging voltage of the battery. The voltage curve of this stage is extracted. Then, starting from the initial voltage, the charging capacity delta Q in each delta U stage where the voltage rises is calculated, and a delta Q-delta U curve is drawn. Based on the curves obtained from the charging voltage under different health states, an evaluation model of curve features to health state is constructed, and the model parameters are updated using the battery charging data.

[0059] Further, inputting the voltage data of the battery pack of the vehicle to be detected into the SOH evaluation model includes:

[0060] preprocessing the voltage data of the battery pack of the vehicle to be detected to obtain a standardized time series;

[0061] inputting the standardized time series into the multi-head attention mechanism, capturing global context dependency by calculating the correlation between time steps, and obtaining hidden feature representation;

[0062] inputting the hidden feature representation into the feedforward neural network, and using the global average pooling layer to reduce the dimension of the time step dimension output by the feedforward neural network to extract the global feature representation of the sequence;

[0063] The global feature representation is converted into a scalar output representing a predicted SOH value by a fully connected network.

[0064] Specifically, in the battery pack detection stage in the embodiment, for the vehicle battery pack to be detected, the rated capacity and the maximum charging voltage of the vehicle battery pack are obtained by communicating with the vehicle once before evaluation, and then the vehicle to be tested is charged at a 1C rate current until the maximum charging voltage Umax of the battery pack. The voltage data in the charging stage is extracted, and the AQ-AU curve is drawn according to the same operation as in the model construction stage. The SOH evaluation model obtained in the model construction stage is used to evaluate the AQ-AU curve characteristics of the battery pack, and finally the SOH of the vehicle battery pack to be tested is obtained.

[0065] Next, taking the application in the EV battery pack detection station as an example, as shown in Figure 2 , the detailed implementation process of the technical scheme in the embodiment is as follows:

[0066] Environment: EV power battery detection station;

[0067] Equipment: EV battery pack data reading equipment, which is internally equipped with a high-voltage charging gun, a current sensor, a voltage sensor, a memory, and a processor.

[0068] Steps:

[0069] 1. Data acquisition: Prepare a lithium iron phosphate single cell in advance, and perform constant current cycle charging and discharging on it at a 1C rate current until its maximum available capacity is below 80% of the rated available capacity. Collect all voltage data in the charging stage to form a training data set. Read the information of the vehicle battery pack to be tested by the EV battery pack data reading equipment, and obtain its rated capacity and maximum charging voltage. Perform a charging and discharging operation on the vehicle to be tested at a 1C rate current, and collect the voltage data in the charging stage to form a test data.

[0070] 2. Data standardization: Since the voltage data relationship of the battery pack and the cell satisfies: U 额，包 / U 额，单 =U 包 / U 单 , where U 额，包、 , U 额，单 , U 额，包 , U 额，单 , and U 包 are the rated voltage of the battery pack, the rated voltage of the single cell, the maximum charging voltage of the battery pack, the maximum charging voltage of the single cell, and the voltage of the battery pack and the voltage of the single cell collected in step 1, respectively. U 单 is a fixed value of 3.2V. Therefore, based on the formula U 包 ’=U 包 ×U 额，单 / U 额，包， Among them U 包 'This is the standardized battery pack voltage data, which is on the same order of magnitude as the voltage data of a single battery cell.

[0071] 3. Starting from the initial voltage U0 = 2.7V, calculate the amount of charge ΔQ of the battery for every ΔU = 60mV increase in voltage, and plot the ΔQ-ΔU curve.

[0072] Constructing a SOH Evaluation Model: A deep learning model based on a multi-head attention mechanism is constructed. First, the input data is preprocessed into a standardized time series, shaped as a three-dimensional tensor (number of samples, time step, feature dimension), where the time step length is set to 100 and the feature dimension is 1. This data is then fed into the multi-head attention mechanism, which captures global contextual dependencies by calculating the correlation between time steps. The multi-head attention mechanism contains eight independent attention heads, each extracting dependency features in a different feature subspace. Finally, the outputs of all heads are concatenated to form a 64-dimensional hidden feature representation. The extracted temporal features are passed to a two-layer feedforward neural network. The first layer contains 128 neurons and uses the ReLU activation function to increase non-linear expressive power. The second layer compresses the features to 64 dimensions to match the attention output, ensuring consistency in the feature space. Subsequently, a global average pooling layer is used to reduce the dimensionality of the time steps, extracting the global feature representation of the sequence and mapping the time series to a fixed-length vector. Finally, these global features are converted into a scalar output, representing the predicted SOH value, through a fully connected layer. The mean square error between the predicted SOH and the actual SOH of the experimental battery cell was calculated, and the model parameters were updated using the Adam optimizer with an initial learning rate of 0.001.

[0073] Application: In EV battery pack testing stations, this method can quickly and effectively assess the State of Health (SOH) of new energy vehicles equipped with lithium iron phosphate battery packs.

[0074] The following is a detailed implementation process of the technical solution in this embodiment, using an application at a vehicle inspection station as an example:

[0075] Environment: Vehicle inspection station

[0076] Equipment: Fixed vehicle testing equipment, internally equipped with a high-voltage charging gun, current sensor, voltage sensor, memory, and processor.

[0077] step:

[0078] Data Acquisition: At the vehicle inspection station, the battery health testing equipment is connected to the battery system of the vehicle to be inspected. A high-voltage charging gun is used to charge the vehicle at a high rate current (80A) to ensure that the battery quickly reaches a stable charging state. Data from the battery pack during the charging process, including current, voltage, and temperature information, is collected and saved as the dataset to be tested.

[0079] Model Construction: A deep learning-based electric vehicle power battery health status detection model with an attention mechanism at its core was built. The raw data was standardized to ensure it was suitable for model input. Then, the processed data was input into the network. The attention mechanism calculated the characteristics of the data in different feature subspaces. Following the attention mechanism were two fully connected neural networks with 128, 64, and 1 neurons per layer, respectively. Deep neuron features were extracted from the data using the fully connected neural network. Finally, ReLU was used as the activation function to output the final SOH prediction value.

[0080] Parameter determination: Prepare individual lithium iron phosphate battery cells in advance. Divide 80A by the rated capacity of the battery pack to obtain the charging rate 1C. Then, perform constant current charging and discharging on the individual battery cells at the 1C rate until the usable capacity of the battery is less than 80% of the rated capacity of the individual cell. Extract voltage data for all charging stages. Using 2.7V as the starting voltage, calculate the battery charging capacity for each 60mV increase in voltage and plot the ΔQ-ΔU curve. Based on the obtained ΔQ-ΔU curve and the backpropagation algorithm, adjust the model parameters to minimize the difference between the model output and the actual SOH. The model after parameter determination is deployed in the device's memory before the testing equipment leaves the factory.

[0081] Battery SOH detection: The features in the dataset to be tested are input into a model with determined parameters to obtain an estimate of the state of health (SOH) of the vehicle under test.

[0082] Applications: At vehicle inspection stations, this technology enables rapid and accurate assessment of battery health, ensuring that electric vehicle battery systems meet inspection standards. This helps improve the efficiency of inspection stations and ensures road safety. Simultaneously, it provides vehicle owners with a convenient way to understand the health status of their vehicle's battery and take necessary maintenance measures.

[0083] This embodiment proposes a novel State of Health (SOH) testing method for new energy vehicles equipped with lithium iron phosphate (LFP) battery packs. Utilizing deep learning, it constructs an effective evaluation model for LFP battery packs by exploring the relationship between charging capacity and battery health within the differential voltage range. This effectively expands the application of data-driven battery SOH testing methods in the field of vehicle testing.

[0084] Compared to existing methods, the advantages of this embodiment are mainly reflected in the following three aspects: First, by constructing a SOH evaluation model by calculating the charging capacity within the differential voltage range, the inaccuracy problem of existing methods caused by the "plateau region" during the charging stage of lithium iron phosphate batteries is effectively solved. Second, without disassembling the battery pack, the SOH of electric vehicles equipped with lithium iron phosphate battery packs can be effectively detected, filling the gap in lithium iron phosphate battery pack detection methods in the field of vehicle inspection. Third, a battery SOH detection model is constructed based on a large artificial intelligence model, extracting the conversion relationship between battery data and battery SOH, achieving higher accuracy compared to traditional models.

[0085] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A rapid detection method for the state of health of a lithium iron phosphate battery pack for vehicles, characterized by, The method comprises the following steps: Based on the preset cycle aging data of lithium iron phosphate battery, a SOH evaluation model is constructed; Collecting voltage data of the battery pack of the vehicle to be detected during the charging and discharging stage; Input the voltage data of the battery pack of the vehicle to be detected into the SOH evaluation model to quickly detect the health status of the battery pack; Based on the preset cycle aging data of lithium iron phosphate battery, a SOH evaluation model is constructed, which comprises: Collecting voltage data during the cycle aging process of the preset lithium iron phosphate battery; Standardizing the voltage data during all cycle aging processes; Based on the standardized voltage data, draw a △Q-△U curve; wherein the △Q-△U curve is a corresponding relationship curve between the voltage and the charging capacity △Q during each △U stage when the voltage rises from the starting voltage; Construct a deep learning model with a multi-head attention mechanism as the core; Train the deep learning model using the △Q-△U curve to obtain a SOH evaluation model; Collecting voltage data during the cycle aging process of the preset lithium iron phosphate battery comprises: Carry out constant current cycle charging and discharging on the preset lithium iron phosphate battery at a current of 1C rate until the maximum available capacity of the lithium iron phosphate battery is lower than the preset threshold of the rated available capacity, and collect the voltage data during all charging stages; Inputting the voltage data of the battery pack of the vehicle to be detected into the SOH evaluation model comprises: Pretreating the voltage data of the battery pack of the vehicle to be detected to obtain the standardized time series; Input the standardized time series into the multi-head attention mechanism, capture the global context dependency by calculating the correlation between time steps, and obtain the hidden feature representation; Input the hidden feature representation into the feedforward neural network, and use the global average pooling layer to reduce the dimension of the time step output by the feedforward neural network, and extract the global feature representation of the sequence; Convert the global feature representation into a scalar output through a layer of fully connected network to represent the predicted SOH value; The multi-head attention mechanism comprises a plurality of independent attention heads, each of which extracts dependency characteristics in different feature subspaces, and finally splices the output results of all attention heads to form a hidden feature representation with a preset dimension; The expression of the standardization processing is: U 包 ’=U 包 ×U 额，单 / U 额，包 where U 包 represents the battery pack voltage data after standardization, U 额，包、 represents the battery pack voltage data after standardization, U 额，单 represents the battery pack voltage data after standardization, U 包 represents the battery pack voltage data after standardization, U 2. The method for quick detection of state of health of lithium iron phosphate battery pack for vehicle of claim 1, wherein, The feedforward neural network comprises two layers; wherein the first layer comprises a plurality of neurons and adopts ReLU activation function to increase the nonlinear expression ability, and the second layer compresses the features to the dimension matching the attention output to ensure the consistency of the feature space.

3. The method for quick detection of state of health of lithium iron phosphate battery pack for vehicle of claim 1, wherein, Constructing the SOH evaluation model further comprises: Calculate the mean square error of the predicted SOH and the actual SOH of the preset lithium iron phosphate battery, and update the model parameters with the initial learning rate combined with the Adam optimizer.

Images