A battery health status assessment method based on convolutional neural network

Abstract

The present invention relates to the technical field of battery health status assessment, and discloses a battery health status assessment method based on a convolutional neural network, comprising the following steps: Step 1: extracting charging segment data of several fully charged states; Step 2: correcting the SOC value of the charging segment data; and calculating the SOH value corresponding to each charging segment data; Step 3: intercepting the single cell voltage data within the same voltage interval in each charging segment data as an input value, and inputting it into an initial assessment model; the initial assessment model is an end-to-end model; Step 4: using the input value in Step 3, calculating the loss function and performing backpropagation to update the weight parameters until the iteration is completed; after the iteration is completed, a standard assessment model is obtained; Step 5: inputting the charging segment data of the battery to be assessed into the standard assessment model, and outputting the SOH value of the battery to be assessed. The present invention can accurately assess the battery health status of a vehicle, with high assessment accuracy and efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of battery health status assessment, and in particular to a battery health status assessment method based on convolutional neural network. Background Art

[0002] With the gradual reform of the current energy structure, a low-carbon economy has become one of the main development directions in the future. In this context, new energy vehicles are gradually replacing fuel vehicles. At the same time, the demand for low-carbon travel has also promoted the advancement of power battery technology and the vigorous development of electric vehicles.

[0003] At the same time, the driving safety of electric vehicles has received widespread attention. As the core energy source of electric vehicles, the power battery system's operational stability, safety, and service life directly affect the safety and stability of the entire vehicle. During actual use, the capacity of the power battery will gradually decrease with continuous charge and discharge cycles until it reaches the end of its life. After the end of its life, the battery's power and capacity will show an unstable, cliff-like decline, directly endangering driving safety. To accurately control this phenomenon, the battery state of health (SOH) is often used to characterize the performance changes of batteries after aging, as described above. This is usually obtained by comparing the actual battery capacity with the rated value. Through battery health assessment, the battery status can be effectively controlled to prevent it from affecting driving safety.

[0004] However, in the actual operation of electric vehicles, the attenuation of the battery health status often shows a nonlinear law and is difficult to accurately confirm. In addition, the current actual capacity of the battery needs to be calculated under full charge and discharge conditions, which is difficult to obtain in actual applications. The SOH value is difficult to calculate and confirm, making it difficult to accurately assess the current battery health status. Summary of the Invention

[0005] The present invention aims to provide a battery health status assessment method based on convolutional neural network, which can accurately assess the battery health status of a vehicle with high assessment accuracy and efficiency.

[0006] To achieve the above objectives, the present invention provides a basic solution: a battery health status assessment method based on a convolutional neural network, comprising the following steps:

[0007] Step 1: Extract charging fragment data of several fully charged states;

[0008] Step 2: Based on the charging segment data, correct the SOC value of the charging segment data; and calculate the current battery capacity value corresponding to each charging segment data based on the corrected SOC value, and then calculate the SOH value corresponding to each charging segment data;

[0009] Step 3: Extracting cell voltage data within the same voltage range from each charging segment as input values ​​and inputting them into an initial evaluation model; wherein the SOH value is used as a label for each charging segment data; the initial evaluation model is an end-to-end model and includes a convolutional neural network;

[0010] Step 4: The initial evaluation model uses the input values ​​in step 3 to calculate the loss function and perform backpropagation to update the weight parameters until the iteration is completed; after the iteration is completed, the standard evaluation model is obtained;

[0011] Step 5: Input the charging segment data of the battery to be evaluated into the standard evaluation model, and output the SOH value of the battery to be evaluated.

[0012] Among them, SOC (State of Charge) refers to the state of charge of the battery, that is, the available state of the remaining charge in the battery, which is generally expressed as a percentage.

[0013] SOH (State Of Health) refers to the battery health status, generally expressed as a percentage.

[0014] The working principle and advantages of the present invention are as follows: First, charging segment data in the fully charged state is specifically extracted as basic data, thereby obtaining input values ​​for training and optimizing the initial evaluation model. The selected basic data is relatively stable, the model training optimization effect is good, and the subsequent evaluation effect is good. In addition, during the actual operation of the battery, due to the complex data operating conditions, its battery charging data is uncertain. In this case, compared with the charging status data extracted in other states, the charging segment data in the fully charged state specifically extracted by this solution has relatively more stable data operating conditions and more stable data performance, making it more reliable as basic data.

[0015] Second, this solution's evaluation model is end-to-end, utilizing only a single model and objective function to complete the battery health assessment. This significantly reduces the complexity of the overall evaluation method, enabling a single network to address all steps and resulting in high overall evaluation efficiency. Conventional evaluation methods, on the other hand, often do not employ end-to-end models, instead employing non-end-end models consisting of multiple independent learning tasks and modules. These non-end-end models, due to the independence of each learning task and the independent and separate training of each module, are prone to errors between components, which can amplify errors and lead to poor overall model evaluation accuracy. Furthermore, conventional evaluation methods often assume that a comprehensive battery health assessment requires comprehensive processing of various feature values. Therefore, they extract a variety of feature data from the underlying data as parameter inputs, forcing conventional methods to employ non-end-end models to process these feature data separately. Furthermore, employing an end-to-end model requires a large amount of data for each feature type to train the end-to-end model, which is difficult to collect and process.

[0016] In reality, most of the excessive feature data cannot accurately represent the battery health status, and will increase the difficulty of feature selection, and even generate noise data, affecting the accuracy and efficiency of the evaluation. However, this solution starts from the battery's own characteristics, specifically selecting the single cell voltage data in the same voltage range in each charging segment data as the input value, and uses the calculated SOH value as the label to build an end-to-end model. It can be seen from the analysis of the battery IC curve (see the attached figure) that Figure 1 As shown in the figure, a specific single cell voltage change can cause a huge change in the charged capacity. Based on this huge change, the change in SOC value can be known and then the SOH value can be obtained. Taking this as input can not only accurately characterize the health status of the battery, but also help optimize the model and improve the evaluation accuracy. It also simplifies the basic data categories. When the total amount of data is the same, the effective data volume of this solution is larger, and the data is more representative and easier to process. It can better train the end-to-end model and further improve the accuracy of model evaluation.

[0017] In addition, the evaluation model of this solution uses a convolutional neural network, which has translation invariance and locality, and has good representation learning capabilities. It can well capture the changes in voltage during battery aging, that is, it can accurately extract high-order features from the input information and accurately evaluate the health status of the battery.

[0018] Furthermore, in step 1, the charging segment data is charging segment data of the battery after being left at rest.

[0019] The battery activity is more stable after standing still, and accordingly, its charging segment data is also more stable. This setting of this solution ensures that the basic data is more stable and reliable, which helps improve the model training effect and thus improve the accuracy of health status assessment.

[0020] Furthermore, in step 1, the charging segment data is parsed and extracted from the battery's message log that complies with the GB32960 protocol.

[0021] With this setting, the charging segment data meets the standard requirements and has high reliability.

[0022] Furthermore, in step 2, when correcting the SOC value of the charging segment data, the low-end SOC value is calibrated using the OCV-SOC curve.

[0023] This setting can further improve the accuracy and reliability of charging segment data and help improve evaluation accuracy.

[0024] Further, in step 3, the same voltage range is 3250mv to 3450mv.

[0025] Specifically, single-cell voltage values ​​within the range of 3250mv to 3450mv are selected as input. Within this range, small voltage changes will cause large changes in the amount of charge. Therefore, the data within this range contains richer information and is more representative. It can fully characterize the changes in the battery's state of charge, and thus effectively characterize the changes in the battery's health status. Using such representative data as model training input can achieve better training results, obtain a better-performing evaluation model, and effectively improve evaluation accuracy. In addition, compared to conventional data selection methods that believe that the larger the range, the better, conventional methods have less detailed data selection, and data values ​​in larger ranges are more difficult to obtain. In addition, the data in the range is not very representative, which increases the difficulty of data processing and poor subsequent evaluation results. This solution does not have such problems.

[0026] Furthermore, in step 3, the initial evaluation model further includes a pooling layer; after the pooling layer downsamples the input value, the input value is input into the convolutional neural network.

[0027] Downsampling the input values ​​further simplifies the input information, retaining only valid information and reducing the feature dimensionality, effectively avoiding overfitting. Furthermore, downsampling reduces the size of the input matrix for the convolutional neural network, facilitating network processing by keeping only a single value per cell at each voltage point.

[0028] Furthermore, the convolutional neural network includes 7 convolutional layers and 3 fully connected layers.

[0029] With this setting, the overall network structure is simple and has strong adaptability.

[0030] Furthermore, in step 4, the loss function is calculated as:

[0031]

[0032] In this way, the loss function is set accurately, which helps to accurately update the model parameters and optimize the model performance.

[0033] Furthermore, the method further includes step 6: when the output SOH value of the battery to be evaluated is ≤80%, determining that the battery to be evaluated has a health problem; when the output SOH value of the battery to be evaluated is >80%, determining that the battery to be evaluated has no health problem.

[0034] This setting further determines whether there are health problems with the battery, and the judgment is reliable based on certain numerical standards.

[0035] Furthermore, in step 2, the SOH value is calculated according to the following formula:

[0036]

[0037] Among them, Q i Refers to the charge in a single charging segment; ΔSOC i Refers to the SOC value increased by a single charging segment; C N Refers to the rated capacity of the battery.

[0038] This setting ensures accurate SOH calculation and label value setting, enabling better model training and improving the generalization ability of large models. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a battery IC curve according to an embodiment of a battery health status assessment method based on a convolutional neural network of the present invention;

[0040] Figure 2 A schematic diagram of battery state of charge changes within a specific voltage range according to an embodiment of a battery health status assessment method based on a convolutional neural network of the present invention;

[0041] Figure 3 This is a schematic diagram of a method flow of an embodiment of a battery health status assessment method based on a convolutional neural network according to the present invention;

[0042] Figure 4 Schematic diagram of the overall performance of a standard assessment model of a battery health status assessment method based on a convolutional neural network according to an embodiment of the present invention on multiple vehicles;

[0043] Figure 5 This is a schematic diagram of the overall performance of a standard assessment model of an embodiment of a battery health status assessment method based on a convolutional neural network in the present invention on a bicycle. DETAILED DESCRIPTION

[0044] The following is a further detailed description through specific implementation methods:

[0045] The embodiment is basically as shown in the attached Figure 3 、 Figure 4 and Figure 5 A battery health status assessment method based on a convolutional neural network is shown, comprising the following steps:

[0046] Step 1: Extract charging segment data of several fully charged states.

[0047] The charging segment data is the charging segment data of the battery in the fully charged state after being stationary. In this embodiment, the historical battery operation data of multiple vehicles of the same model are used as the basis for extracting the charging segment data. This setting can eliminate the differences in charging segment data caused by different vehicles, and the data is more targeted and can better evaluate the health status of the battery of a single vehicle.

[0048] Specifically, the fully charged state refers to a fully charged state, i.e., a state where the battery is charged to 100%. In this state, battery activity is relatively stable, and data stability is good. Furthermore, the charging segment data is parsed and extracted from the battery's message logs that comply with the GB32960 protocol. This data complies with the standard and is highly reliable. The GB32960 protocol specifically refers to the communication protocol and data format specified in the "GB / T32960 Technical Specification for Electric Vehicle Remote Service and Management Systems."

[0049] Step 2: According to the charging segment data, the SOC value of the charging segment data is corrected; and the current battery capacity value corresponding to each charging segment data is calculated based on the corrected SOC value, and then the SOH value corresponding to each charging segment data is calculated.

[0050] Specifically, when correcting the SOC value of the charging segment data, the low-end SOC value is calibrated using the OCV-SOC curve.

[0051] The SOH value is calculated according to the following formula:

[0052]

[0053] Among them, Q i Refers to the charge in a single charging segment; ΔSOC i Refers to the SOC value increased by a single charging segment; C N Refers to the rated capacity of the battery.

[0054] Step 3: Intercept the single cell voltage data within the same voltage range in each charging segment data as input value and input it into the initial evaluation model; wherein, the SOH value is used as the label of each charging segment data. Specifically, the same voltage range is a specific voltage range, specifically 3250mv ~ 3450mv; as shown in the attached Figure 1 and attached Figure 2 As shown, this interval contains peak battery variation. Smaller voltage changes within this interval correspond to more significant changes in the charge level and state of charge. This makes the data in this interval highly representative, fully characterizing changes in the battery's state of charge and, by extension, its health. Using this representative data as model training input can achieve better training results, resulting in a higher-performing evaluation model and significantly improving evaluation accuracy.

[0055] The initial evaluation model is an end-to-end model that includes a convolutional neural network (CNN) consisting of seven convolutional layers and three fully connected layers. The initial evaluation model also includes a pooling layer that downsamples input values ​​before feeding them into the CNN.

[0056] Specifically, during the downsampling process, each voltage point in the voltage range is downsampled to a single value per cell, and this value is the average of the cell voltage data within the same voltage range in each charging segment. This setting determines the input matrix size of the convolutional neural network to be (200 × n), where n is the number of cells. This effectively reduces the input matrix size of the convolutional neural network and facilitates network processing.

[0057] Step 4: The initial evaluation model uses the input values ​​in step 3 to calculate the loss function and perform backpropagation to update the weight parameters until the iteration is completed; after the iteration is completed, the standard evaluation model is obtained.

[0058] The calculated loss function is:

[0059]

[0060] Step 5: Input the charging segment data of the battery to be evaluated into the standard evaluation model, and output the SOH value of the battery to be evaluated.

[0061] Step 6: When the output SOH value of the battery to be evaluated is ≤80%, it is determined that the battery to be evaluated has a health problem; when the output SOH value of the battery to be evaluated is >80%, it is determined that the battery to be evaluated has no health problem.

[0062] This embodiment provides a battery health status assessment method based on a convolutional neural network. Starting from the battery's own characteristics and according to the battery IC curve, it takes voltage data in a specific voltage range as input, and uses the calculated SOH value as a label to build an end-to-end model for assessment. This method can accurately assess the health status of a single vehicle battery with high assessment accuracy and efficiency.

[0063] The input data is highly representative and can fully characterize changes in the battery's state of charge, and thus effectively characterize changes in the battery's state of health, achieving good model training results. The end-to-end model structure is extremely simple, using only one model and one objective function to complete battery health assessment. A single network can address all steps, resulting in high overall evaluation efficiency. The convolutional neural network in the end-to-end model can accurately capture feature changes and achieve high evaluation accuracy. Furthermore, the standard evaluation model constructed by this solution has undergone sufficient iterative training and can be used to make relatively timely assessments of the current battery health status of the battery being evaluated, ensuring a certain degree of timeliness.

[0064] Furthermore, compared to schemes that use non-end-to-end models for evaluation, and conventional evaluation methods that use multiple types of feature values ​​as input, this scheme streamlines the types of input values ​​while preserving the amount of data. Given the same total amount of training data as conventional schemes, this scheme has a larger amount of effective data, allowing conventional end-to-end models that require larger amounts of training data to be fully trained and achieve better evaluation results. Furthermore, because the training data input by this scheme is all data within a specific voltage range, it is more capable of representing changes in the battery's state of charge than other data and is easier to collect. More representative values ​​make model training more effective, improve model performance, and increase evaluation accuracy, with evaluation accuracy increasing by at least 20% compared to conventional models.

[0065] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A battery health status assessment method based on convolutional neural network, characterized in that: The following steps are involved: Step 1: Extract charging fragment data of several fully charged states; Step 2: Correct the SOC value of the charging segment data according to the charging segment data; The current battery capacity value corresponding to each charging segment data is calculated based on the corrected SOC value, and then the SOH value corresponding to each charging segment data is calculated; When correcting the SOC value of the charging segment data, the OCV-SOC curve is used to calibrate the low-end SOC value; Step 3: intercept the single cell voltage data in the same voltage range in each charging segment data as input value, and input it into the initial evaluation model; wherein, the same voltage range is ; The SOH value is used as a label for each charging segment data; the initial evaluation model is an end-to-end model; and the initial evaluation model includes a convolutional neural network; the convolutional neural network includes 7 convolutional layers and 3 fully connected layers; the initial evaluation model also includes a pooling layer; the pooling layer downsamples the input value before inputting it into the convolutional neural network; Step 4: The initial evaluation model uses the input values ​​in step 3 to calculate the loss function and perform backpropagation to update the weight parameters until the iteration is completed; after the iteration is completed, the standard evaluation model is obtained; Step 5: Input the charging segment data of the battery to be evaluated into the standard evaluation model, and output the SOH value of the battery to be evaluated.

2. The battery health status assessment method based on convolutional neural network according to claim 1, characterized in that: In step 1, the charging segment data is charging segment data of the battery after being left at rest.

3. The battery health status assessment method based on convolutional neural network according to claim 1, characterized in that: In step 1, the charging segment data is parsed and extracted from the battery's message log that complies with the GB32960 protocol.

4. The battery health status assessment method based on convolutional neural network according to claim 1, characterized in that: In step 4, the calculated loss function is: 。 5. The battery health status assessment method based on convolutional neural network according to claim 1, characterized in that: Also includes step 6: when the SOH value of the battery to be evaluated is output When the SOH value of the battery to be evaluated is , it is determined that the battery to be evaluated has no health problems.

6. The battery health status assessment method based on convolutional neural network according to claim 1, characterized in that: In step 2, the SOH value is calculated according to the following formula: ； in, Refers to the amount of charge in a single charging segment; Refers to the SOC value increased by a single charging segment; Refers to the rated capacity of the battery.

Images