Battery cell open-circuit voltage detection method based on gaussian mixture model

Abstract

The present application relates to the field of battery testing, and discloses a battery cell open-circuit voltage detection method based on a Gaussian mixture model. The method comprises: acquiring an open-circuit voltage dataset containing timestamps, battery cell numbers, and open-circuit voltage values; performing denoising and normalization preprocessing on the acquired open-circuit voltage dataset; by using the preprocessed open-circuit voltage dataset, constructing a Gaussian mixture model by means of an EM algorithm; and by using the constructed Gaussian mixture model, performing open-circuit voltage anomaly detection. To address the issue of low accuracy of open-circuit voltage detection in the prior art, the present application employs an EM algorithm to construct a Gaussian mixture model, and optimizes model parameters via iterative E-steps and M-steps. The model enables detection of anomalies in the open‑circuit voltage of battery cells, and supports dynamic adjustment of the number of model components and incremental learning to adapt to system changes. By calculating data responsibilities and setting a threshold, anomalies in the open-circuit voltage of battery cells can be efficiently and accurately determined, thereby improving the accuracy of open-circuit voltage detection.

Description

A method for detecting open-circuit voltage of battery cells based on Gaussian mixture model Technical Field

[0001] The present application relates to the field of battery detection, and more specifically, to a method for detecting the open-circuit voltage of a battery cell based on a Gaussian mixture model. Background Art

[0002] The open circuit voltage (OCV) of a battery cell is a key indicator for evaluating battery health and performance, and has a direct impact on the quality and safety of the cell. Existing cell OCV anomaly detection is primarily performed on a pallet basis, requiring centralized analysis and determination of the OCVs of multiple cells within each pallet. However, in practice, the number of cells in each pallet varies significantly, ranging from single digits to dozens, and the statistical distribution of the OCVs is highly complex. Traditional detection methods struggle to adapt to this complex distribution of varying cell counts, leading to false positives and missed detections.

[0003] Chinese patent application, application number CN202210904863.6, publication date July 29, 2022, discloses a method for detecting abnormal battery boxes based on deep autoencoder Gaussian mixture distribution, which belongs to the field of energy storage power station safety technology. The method includes: obtaining the feature data to be measured of the battery box, the feature data to be measured characterizes the current performance of the battery box; based on the pre-acquired deep autoencoder Gaussian mixture distribution model, determining the corresponding distribution probability according to the feature data to be measured, the deep autoencoder Gaussian mixture distribution model is trained based on the historical feature data of the battery box, and the historical feature data is of the same data type as the feature data to be measured; judging whether the battery box is an abnormal battery box according to the distribution probability. However, in this scheme, the Gaussian mixture distribution has a good fitting effect on single-mode data, but has a weak ability to describe complex multimodal distributions. Therefore, the detection accuracy in this scheme needs to be further improved. Summary of the Invention

[0004] 1. Technical problems to be solved

[0005] In response to the problem of low open-circuit voltage detection accuracy in the prior art, the present application provides a battery cell open-circuit voltage detection method based on a Gaussian mixture model, which improves the detection accuracy of the open-circuit voltage by constructing a Gaussian mixture model through the EM algorithm.

[0006] 2. Technical solution

[0007] The purpose of this application is achieved through the following technical solutions.

[0008] An embodiment of the present specification provides a method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model, comprising: collecting an open circuit voltage dataset containing a timestamp, a battery cell number, and an open circuit voltage value; preprocessing the collected open circuit voltage dataset by denoising and normalizing the dataset; constructing a Gaussian mixture model using the EM algorithm using the preprocessed open circuit voltage dataset; and performing open circuit voltage anomaly detection using the constructed Gaussian mixture model.

[0009] The open-circuit voltage acquisition equipment for the battery cells, such as acquisition cards and data cables, is set up. An acquisition program is written to scan and measure the open-circuit voltage of each battery cell at regular intervals. The acquisition program also uses the system time as a timestamp and records the unique cell number. Each scan records a timestamp, cell number, and corresponding open-circuit voltage value, ultimately forming an open-circuit voltage dataset. This dataset can be stored in a local database or uploaded to a cloud database. The open-circuit voltage sampling frequency can be set, such as every 10 minutes. It is also possible to collect instantaneous open-circuit voltage during the charge and discharge process. Other battery cell information, such as brand, model, and usage time, can also be recorded as auxiliary information for the dataset. During data acquisition, denoising can be performed to remove data with obvious errors. The collected dataset can be stored in standard formats such as CSV, XLSX, and JSON. This dataset provides input for subsequent open-circuit voltage analysis modeling and anomaly detection.

[0010] The collected open-circuit voltage dataset is denoised using algorithms such as median filtering and mean filtering to remove outliers. The dataset is preprocessed using normalization, such as min-max normalization, to map the data to the same order of magnitude for easier comparison. Using this preprocessed dataset, a Gaussian mixture model is constructed using the EM algorithm. The number of Gaussian components can be preset, and the model parameters are learned through iterative E- and M-steps. The E-step calculates the responsivity of each data point on each Gaussian component. The M-step updates the model parameters based on the responsivity. Termination conditions are set for the iterations to obtain the final model parameters, including the mean, variance matrix, and mixing coefficients. The constructed Gaussian mixture model is used to detect anomalies in new data, calculating the responsivity and comparing it to a threshold. When new data that does not conform to the existing pattern appears, the model parameters can be incrementally adjusted to continue learning. The model's expressive power can also be expanded by adding or removing Gaussian components. Finally, the model outputs the anomaly data results and locates the abnormal battery cells. A model performance evaluation process can be established to continuously optimize the model and parameters.

[0011] Furthermore, a Gaussian mixture model is constructed, including: presetting the component number H of the Gaussian mixture model and initializing the mean, variance and mixing coefficient of each Gaussian component i, where the component number H represents the number of Gaussian components contained in the Gaussian mixture model; iteratively performing E steps and M steps on the preprocessed open-circuit voltage data set; in the E step, calculating the responsiveness of each open-circuit voltage data on each Gaussian component i; in the M step, updating the parameters of each Gaussian component i according to the responsiveness obtained in the E step; setting a threshold value for the responsiveness change and a maximum number of iterations, and repeating the E and M steps until the responsiveness change is less than the preset threshold value or the set maximum number of iterations is reached.

[0012] Among them, Gaussian component refers to each Gaussian distribution component contained in the Gaussian mixture model. In this application, the Gaussian mixture model is composed of multiple Gaussian components, and each Gaussian component is used to describe a pattern / cluster in the data. The preset number of components H indicates that the model contains H Gaussian components. Each component represents a potential pattern in the data. Each Gaussian component has its own independent set of parameters, including mean, variance and mixing coefficient. These parameters need to be initialized separately for each Gaussian component during initialization. The mean represents the center of the pattern, the variance represents the distribution range, and the coefficient represents the weight of the pattern. In the E step, the responsiveness of the data point on each component is calculated. In the M step, only the corresponding component parameters are updated according to the responsiveness. Ultimately, each component describes the parameters of a major pattern in the data. The combination of components can adapt to the overall complex distribution. When a new pattern appears, components can be added to adapt.

[0013] Among them, the responsiveness change refers to the change in the responsiveness of the data points on each Gaussian component during the iteration process of the EM algorithm of the Gaussian mixture model. In this application, in the E step, the responsiveness of each data point on each Gaussian component is calculated. After the M step, the model parameters are updated and the responsiveness will also change. The absolute value or the sum of squares of the responsiveness change of each data point between two iterations is calculated. When the responsiveness change is less than the set threshold, it means that the model parameters tend to be stable and the iteration can be terminated. The maximum number of iterations can also be set to prevent too many invalid iterations. The responsiveness change reflects the parameter convergence state and can be used as one of the iteration termination conditions. Threshold setting requires comprehensive consideration of accuracy and convergence speed. Too large a change means that the model is still being adjusted, and a small change means that the model has converged. Finally, the converged model parameters are obtained for open circuit voltage anomaly detection.

[0014] Among them, the number of components H needs to be set in advance, which represents the number of Gaussian components contained in the mixture model. The setting of the H value will affect the fitting effect of the data set. A reasonable range of H can be given based on empirical values ​​or prior knowledge, such as 3-10. It is also possible to select H with the best fitting effect through multiple experiments. According to the set H value, the parameters of the corresponding number of Gaussian components need to be initialized. Each Gaussian component needs to initialize the mean, covariance matrix and mixing coefficient. The mean can be initialized randomly or the initial center can be determined based on statistical analysis. The covariance matrix can be initialized to the unit matrix or empirical value. The mixing coefficient can be initialized to a uniform distribution or distributed according to a certain proportion. Reasonable setting of the initialization parameters can accelerate model convergence. If the setting is inappropriate, it can also be automatically adjusted during the iteration process. You can also use the subnets method to first cluster the initial H class centers and set the corresponding Gaussian component parameters according to the class centers.

[0015] After preprocessing the open-circuit voltage data for denoising and normalization, cleaned input data is obtained. The EM algorithm is performed on the preprocessed data set to iteratively estimate the parameters of the Gaussian mixture model. E-step: Based on the current model parameters, the responsivity of each data point on each Gaussian component is calculated. The responsivity reflects the probability that the data point belongs to that Gaussian component. M-step: Based on the responsivity results calculated in the E-step, the parameters of each Gaussian component, including the mean, covariance matrix, and mixing coefficient, are updated. Through multiple iterations of the E-step and M-step, the model parameters gradually converge to the local optimal solution. The iterative termination conditions can be set, such as reaching the maximum number of iterations or the responsivity change being less than a threshold. Some methods can also be used to prevent model overfitting, such as the early stopping method. The computational complexity of each E-step and M-step is large, and vectorization can be used to accelerate the operation. The setting of the initialization parameters will also affect the convergence speed. Finally, the parameters of the trained Gaussian mixture model are obtained for subsequent open-circuit voltage anomaly detection.

[0016] Furthermore, the Gaussian mixture model is expressed as follows:

[0017] in, is the coefficient, which indicates the probability of selecting the kth Gaussian distribution, is a Gaussian distribution, represents the parameters of the k-th Gaussian distribution; As the kth Gaussian distribution model, the probability density distribution of the cell open circuit voltage data under the Gaussian mixture model is obtained.

[0018] Furthermore, in step E, based on the current model parameters, the sub-model k is calculated for the observed data y j Responsiveness

[0019] in, It is the probability that the jth observation data comes from the kth sub-model under the current model parameters, which is called the probability of sub-model k for observation data y j responsiveness.

[0020] Further, step M: calculate the model parameters for the new round of iteration:

[0021] in, is the mean of the kth sub-model in the new iteration, is the variance of the kth sub-model obtained in the new round, It is the proportion of the kth sub-model in the new round of iteration in the hybrid model.

[0022] Furthermore, the open circuit voltage detection includes: calculating the responsiveness of the open circuit voltage of each battery cell in the open circuit voltage data set on each Gaussian component through a Gaussian mixture model; presetting a threshold for abnormal responsiveness judgment; when the responsiveness of the open circuit voltage data on all Gaussian components is lower than the preset threshold, the corresponding open circuit voltage data is judged to be abnormal.

[0023] The newly collected open-circuit voltage sample data is tested using a trained Gaussian mixture model. For each sample data, its responsivity on each Gaussian component of the model is calculated. The responsivity reflects the degree of match between the data and the known pattern, with higher responsivity indicating a closer match. A responsivity threshold is pre-set as a criterion for determining whether there is an anomaly. The responsivity of each Gaussian component is compared with the threshold in turn. If the responsivity on all Gaussian components is lower than the threshold, it can be determined that the data does not match the known pattern and is abnormal data. Judging anomalies by threshold comparison is simple and intuitive, and statistical tests can also be used. Finally, the data judged to be abnormal is output, and the abnormal battery cell is located. The setting of the threshold needs to be determined based on the actual situation. The model can also output probabilities and perform probability accumulation to determine anomalies.

[0024] Furthermore, the number of components and corresponding Gaussian components in the Gaussian mixture model are adjusted through incremental learning. After model training is complete, the model is updated using incremental learning. When it is discovered that the original model cannot adequately describe newly acquired data, the incremental learning process is initiated. This new data is collected as an incremental dataset. A determination is made as to whether the new data belongs to an existing Gaussian component or requires a new component. If it belongs to an existing component, the parameters of the other components are fixed, and only the relevant component is adjusted to adapt to the new data. If a component needs to be added, the parameters of the new component are initialized, and the parameters of all components are adjusted simultaneously. The EM algorithm is used to iteratively train the new and original data, updating the model parameters. Based on the complexity of the new data, the number of components to be added is determined, and the number of components in the original model is adjusted. Incremental training is repeated until the model can adapt well to the new data. Alternatively, when anomalies occur, the corresponding component can be deleted and other components updated to maintain model performance. Incremental learning allows the model to dynamically evolve and adapt to changes in the data distribution. This avoids the need for complete model retraining each time, improving efficiency.

[0025] Furthermore, the number of components in the Gaussian mixture model can be adjusted, including adding or removing Gaussian components. After model training is complete, monitor model performance to determine whether the number of components needs to be adjusted. If the detection effect is poor and cannot adapt to the data, consider increasing the number of components. Determine the number of Gaussian components to be added based on the complexity of the emerging data pattern. Add a new Gaussian component and randomly initialize its mean, covariance, and coefficient. Fix the parameters of other existing components and only learn the parameters of the new component. Collect data for the new pattern and iterate using the EM algorithm to learn the parameters of the new component. You can also fine-tune the parameters of existing components simultaneously. If the abnormal pattern disappears, consider removing the corresponding Gaussian component. Set its responsiveness to 0 and iteratively learn and adjust the parameters of other components. After removing a component, you may need to renormalize the mixing coefficients. Check the performance of the model after removing the component and restore the original model if the performance is poor. Dynamically adding or removing components can make the model size reasonably adapt to the complexity of the data.

[0026] For further denoising, median filtering or mean filtering is used. The raw open-circuit voltage dataset collected may contain abnormal impulse noise. A sliding window is set, with the window size typically being an odd number such as 3, 5, or 7. Median filtering: The data within the window are sorted in ascending order and the median value is taken as the filtered output. This effectively suppresses impulse noise. Mean filtering: The arithmetic mean of the data within the window is calculated as the output, which can suppress random noise. Filtering is performed on the dataset using a sliding window to generate a denoised dataset. Median filtering is effective for removing impulse noise, but its detail preservation is slightly poor. Mean filtering retains good detail, but its effectiveness in suppressing impulse noise is slightly weaker. Choose between median filtering and mean filtering based on scenario requirements. Alternatively, median filtering can be used to remove impulse noise first, followed by mean filtering to preserve detail. The filter window size should be appropriately set, generally an odd number between 3 and 7.

[0027] Furthermore, normalization processing is performed, and a linear normalization function is used to map the open circuit voltage value to the range of 0 to 1. The original sampled values ​​of the open circuit voltage are collected, and there are different voltage ranges. Set the target interval of normalization to [0, 1]. Find the minimum value min and the maximum value max in all sampled data. Use the linear normalization function: y = (x-min) / (max-min), traverse all sampled data x, map it to the 0-1 interval, and calculate the corresponding y. Convert the original open circuit voltage data to a unified 0-1 range according to the above function. Linear normalization retains the characteristics of the original data distribution and relative size. It enables data in different voltage ranges to be directly compared, which is convenient for subsequent modeling. Other normalization functions can be used, such as Sigmoid, Tanh, etc. The linear function is simple to calculate and has a better normalization effect. Different target normalization intervals can also be set.

[0028] 3. Beneficial effects

[0029] Compared with the existing technology, the advantages of this application are:

[0030] By adopting a Gaussian mixture model, this method can more accurately model the distribution of open-circuit voltage data, thereby improving detection accuracy. The iterative optimization process of the EM algorithm helps to better fit complex data distributions, making the model more consistent with actual conditions.

[0031] The introduction of incremental learning and a strategy for dynamically adjusting the number of components enables the model to flexibly adapt to changes that may occur in the battery system during operation, thereby improving the dynamic adaptability of the system.

[0032] By calculating the responsivity of the open-circuit voltage data on each Gaussian component, the method can effectively detect anomalies. Setting a responsivity threshold helps accurately determine whether the cell open-circuit voltage is abnormal, improving the robustness of the system.

[0033] Introducing median filtering or mean filtering for denoising and linear normalization function for normalization can help eliminate interference and inconsistency in the data and improve the robustness of the model to various data conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] FIG1 is a schematic diagram of a method for detecting open-circuit voltage of a battery cell based on a Gaussian mixture model of the present application. DETAILED DESCRIPTION

[0035] The present application is described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] Figure 1 is a schematic diagram of a Gaussian mixture model-based open-circuit voltage detection method for battery cells in this application. This application proposes a statistically based Gaussian mixture model (GMM) method for detecting anomalies in the open-circuit voltage (OCV) of battery cells. This method can model complex OCV distributions, accurately distinguish between normal and abnormal samples, identify different operating modes and fault conditions, and reduce reliance on large amounts of labeled data. The open-circuit voltage of a battery cell is an important indicator of its health. However, OCV often exhibits complex statistical distributions, making anomaly detection difficult. Furthermore, different operating modes and fault conditions can cause variations in the OCV distribution. Existing methods based on thresholds and fixed models often fail to accurately describe the OCV distribution, resulting in high false positive rates. Battery OCV data under normal operating conditions is collected, and a Gaussian mixture model is constructed to describe the probability distribution of OCV. For newly measured OCV data, its probability score under the GMM model is calculated. Any score below a threshold is considered an anomaly. When an anomaly is detected, the clustering properties of the GMM are leveraged to identify the mode to which the OCV belongs and determine the fault type. The GMM parameters are updated online to adapt to changes in the OCV distribution. Compared to fixed models and threshold methods, GMM can more accurately describe the statistical distribution of battery OCV. Compared to purely data-driven methods, GMM requires only a small amount of labeled data. This method can detect anomalies and classify and identify different fault conditions, demonstrating its strong adaptability.

[0037] The open circuit voltage (OCV) of the battery cells, filter the pallet data. Filter all the open circuit voltage (OCV) data of the cells in a pallet during the capacity separation phase on a certain day. The data records the time, pallet number, and open circuit voltage (OCV) value. The cell open circuit voltage (OCV) value list [x1, x2, x3, ..., x M ] where x M Represents the open circuit voltage (OCV) value of the Mth battery cell.

[0038] Data preprocessing. First, remove the cells in each tray whose open circuit voltage (OCV) is higher than a fixed threshold by removing noise, and use the open circuit voltage (OCV) of all the remaining cells as the test data. The remaining cell open circuit voltage (OCV) value list is [x1, x2, x3, ..., x N ], where x N Represents the open circuit voltage (OCV) value of the Nth battery cell and N≤M.

[0039] Data normalization. Use linear function normalization to map the data to the interval [0, 1], effectively avoiding the impact of different channel feature weights caused by different dimensions. The specific formula is:

[0040] Among them, y j Indicates the normalized open circuit voltage (OCV) value of the battery cell, x j Represents a time series dataset [x1,x2,x3,......,x N ] any element, x max and x min Represents the minimum value in each characteristic sequence respectively. The normalized cell open circuit voltage (OCV) value list is [y1,y2,y3,......,y N ].

[0041] Battery cell open circuit voltage data often exhibits multi-modal mixed distribution characteristics, and a single probability distribution model is difficult to accurately describe its distribution. This solution uses a Gaussian mixture model to describe the distribution of battery cell open circuit voltage data, which can better reflect the multi-modal characteristics of the data. The Gaussian mixture model is a probabilistic generative model that assumes that all data points are mixed results from multiple different Gaussian distributions. Formally, the Gaussian mixture model can be expressed as:

[0042] in, is the coefficient, which indicates the probability of selecting the kth Gaussian distribution, is a Gaussian distribution, represents the parameters of the k-th Gaussian distribution, then: The probability density distribution of the cell open-circuit voltage data under the Gaussian mixture model is obtained as the kth Gaussian distribution model. The model is evaluated to ensure that it accurately describes the distribution characteristics of the cell open-circuit voltage. The multimodal characteristics of the cell open-circuit voltage distribution are analyzed based on the model. The Gaussian mixture model can flexibly describe the multimodal distribution characteristics of battery cell open-circuit voltage data. Compared with a single model, it can more accurately characterize different modes in the data, providing support for cell status analysis and quality prediction.

[0043] Algorithm flow: Initialize the parameters of Gaussian distribution: initial value of mixing ratio is a random variable from a uniform Dirichlet distribution with an initial mean value u k It is randomly generated by normal distribution, and the center is determined by the observed data; the initial value of the standard deviation σ k It is the reciprocal of the square root of a vector of random exponentially distributed values. The parameters of the exponential distribution are determined based on data analysis techniques.

[0044] Step E: Calculate the sub-model k for the observed data y based on the current model parameters. j Responsiveness

[0045] in, It is the probability that the jth (the range of j values) observation data comes from the kth sub-model under the current model parameters, which is called the probability that sub-model k has an effect on the observation data y j responsiveness.

[0046] Step M: Calculate the model parameters for the next iteration:

[0047] in, is the mean of the kth sub-model in the new iteration, is the variance of the kth sub-model obtained in the new round, It is the proportion of the kth sub-model in the new round of iteration in the hybrid model.

[0048] Repeat the E and M steps until convergence. Here, the EM iteration stopping threshold is set to 0.001 and the maximum number of iterations is set to 100.

[0049] Anomaly detection, for new observation data x, calculate its responsiveness on each Gaussian model k:

[0050] in, It is the probability that the jth (the range of j values) observation data comes from the kth sub-model under the current model parameters, which is called the probability that sub-model k has an effect on the observation data y j responsiveness.

[0051] If for all k=1,...,K, If the values ​​of are very small, x is judged to be abnormal data. In order to ensure the accuracy and recall rate of anomaly detection, the threshold is set to 0.3.

[0052] Adaptive parameter estimation, adjust Gaussian components, and add new Gaussian models. When new pattern data is observed, the model parameters need to be adjusted to adapt to the new pattern. The new pattern data is treated as a new Gaussian distribution, and the mean and variance are initialized to Reduce the new Gaussian model. If the response of a sub-model is continuously low, remove the sub-model. Establish a new Gaussian mixture Gaussian model. Add the new Gaussian distribution to the original mixture Gaussian model, where the model parameters are Adjust component parameters and repeat to continuously detect anomalies.

[0053] The above schematically describes the invention of the present application and its implementation methods. This description is not restrictive. Without departing from the spirit or basic features of the present application, the present application can be implemented in other specific forms. What is shown in the drawings is only one of the implementation methods of the invention of the present application. The actual structure is not limited to this. Any figure mark in the claims should not limit the claims involved. Therefore, if a person of ordinary skill in the art is inspired by it, without departing from the purpose of the invention, a structural method and embodiment similar to the technical solution without creativity should fall within the scope of protection of this patent. In addition, the word "including" does not exclude other elements or steps, and the word "one" before an element does not exclude the inclusion of "multiple" elements. The multiple elements stated in the product claim can also be implemented by one element through software or hardware. Words such as first and second are used to indicate names and do not indicate any specific order.

Claims

1. A method for detecting the open circuit voltage of a battery cell based on a Gaussian mixture model, comprising: Collect an open circuit voltage dataset containing timestamp, cell number, and open circuit voltage value; Perform denoising and normalization preprocessing on the collected open circuit voltage data set; The preprocessed open circuit voltage dataset is used to construct a Gaussian mixture model using the EM algorithm; The constructed Gaussian mixture model is used to detect anomalies in open circuit voltage.

2. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 1, wherein: Construct a Gaussian mixture model, including: Preset the component number H of the Gaussian mixture model and initialize the mean, variance and mixing coefficient of each Gaussian component i. The component number H represents the number of Gaussian components contained in the Gaussian mixture model; On the preprocessed open circuit voltage dataset, iteratively perform E-step and M-step; Step E: Calculate the responsivity of each open circuit voltage data on each Gaussian component i; In step M, the parameters of each Gaussian component i are updated according to the responsiveness obtained in step E; Set the threshold of the response degree change and the maximum number of iterations, and repeat the E-step and M-step iterations until the response degree change is less than the preset threshold or the maximum number of iterations is reached.

3. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 2, wherein: Gaussian mixture model, the expression is as follows: in, is the coefficient, which indicates the probability of selecting the kth Gaussian distribution, is a Gaussian distribution, represents the parameters of the k-th Gaussian distribution; As the kth Gaussian distribution model, the probability density distribution of the cell open circuit voltage data under the Gaussian mixture model is obtained.

4. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 3, wherein: Step E: Calculate the sub-model k for the observed data y based on the current model parameters. j Responsiveness in, It is the probability that the jth observation data comes from the kth sub-model under the current model parameters, which is called the probability of sub-model k for observation data y j responsiveness.

5. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 4, wherein: Step M: Calculate the model parameters for the next iteration: in, is the mean of the kth sub-model in the new iteration, is the variance of the kth sub-model in the new round, It is the proportion of the kth sub-model in the new round of iteration in the hybrid model.

6. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to any one of claims 2 to 5, wherein: Open circuit voltage detection, including: The open circuit voltage response of each cell in the open circuit voltage dataset on each Gaussian component is calculated using the Gaussian mixture model. Preset threshold for abnormal response determination; When the responsiveness of the open circuit voltage data on all Gaussian components is lower than a preset threshold, the corresponding open circuit voltage data is determined to be abnormal.

7. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 6, wherein: The number of components and the corresponding Gaussian components of the Gaussian mixture model are adjusted through incremental learning.

8. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 7, wherein: Adjust the number of components in the Gaussian mixture model, including adding or removing Gaussian components.

9. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 1, wherein: De-noising is done by using median filtering or mean filtering.

10. The method for detecting open circuit voltage of a battery cell based on a Gaussian mixture model according to claim 9, wherein: Normalization processing uses a linear normalization function to map the open circuit voltage value to the range of 0 to 1.

Images