Evaluation model construction method, battery health state evaluation method, device and equipment
The battery aging data and Gaussian process regression model are expanded through the TimeGAN model to optimize hyperparameters, solving the problem of accuracy and stability of the battery health status evaluation model under a small amount of data, and achieving efficient evaluation in battery health status evaluation such as lithium-ion batteries .
Patent Information
- Application Number
- CN202410999782.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-12-06
AI Technical Summary
The existing data-driven battery health status evaluation model cannot guarantee its accuracy and stability when there are only a small amount of battery aging experimental data, resulting in the inability to accurately evaluate battery health status.
The experimental data of battery aging is expanded through the TimeGAN model, health characteristic data related to health status is extracted, and hyperparameters are optimized using Gaussian process regression model and global optimization algorithm to build a health status evaluation model.
The accuracy and reliability of the battery health status evaluation model is significantly improved under a small amount of battery aging data, reducing the impact of data size on the model, reducing the risk of overfitting, and improving the stability of the model.
Smart Images

Figure CN119087221A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of battery life prediction, and in particular to a method for constructing a health status assessment model, a battery health status assessment method, a health status assessment model construction device, a battery health status assessment device, an electronic device and a computer-readable storage medium. Background Art
[0002] With the continuous promotion of electric vehicles, the continuous increase of electronic devices and the gradual maturity of new energy technologies, electrochemical energy storage represented by lithium-ion battery energy storage technology is entering a period of vigorous development. In order to ensure the safety and stability of lithium-ion batteries and reliable battery performance evaluation, accurate estimation of the health state (SOH) of lithium-ion batteries is essential.
[0003] Currently, commonly used battery health state estimation methods include direct measurement method, model-based method and data-driven method. Among them, the direct measurement method is limited by experimental requirements and cannot be used in actual operating environments. Model-based methods rely on precise initial conditions, and these condition parameters are often not convenient to obtain in practical applications. Therefore, model-based methods will be hindered in actual use. Although data-driven methods can achieve accurate estimation effects without studying the specific degradation mechanism of the battery, the long life of lithium-ion batteries and insufficient aging data limit the actual effect of data-driven modeling. Therefore, when using data-driven battery health state estimation methods, how to ensure that the model can maintain its accuracy and stability when there is only a small amount of battery aging experimental data is still a challenge.
[0004] In summary, existing battery health characteristic estimation methods usually consider how to increase the accuracy of the battery health characteristic evaluation model based on existing data. However, in actual use, the battery health characteristic evaluation model is often limited by the lack of power station operation data under actual conditions, resulting in an inability to accurately evaluate the battery health status. Summary of the invention
[0005] In view of the technical problem in the prior art that the data-driven battery health status assessment model cannot guarantee its accuracy and stability when there is only a small amount of battery aging test data, the present invention provides a health status assessment model construction method, which can be used to construct a model that supports efficient and accurate battery health status calculation when there is only a small amount of battery aging test data. In addition, the present invention also provides a battery health status assessment method, which can accurately estimate the health status of a target battery (e.g., a lithium-ion battery) when there is only a small amount of battery aging test data.
[0006] To achieve the above-mentioned purpose, the first aspect of the present invention provides a method for constructing a health status assessment model, which comprises the following steps: obtaining health feature data of a reference battery based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery; performing expansion processing on the health feature data of the reference battery through the TimeGAN model to generate expanded health feature data; and training a machine learning model using the expanded health feature data as a training sample and the health status as a label to obtain a health status assessment model.
[0007] In an exemplary embodiment of the present invention, the acquisition of health characteristic data of a reference battery based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery may include: determining a calculation result of a correlation coefficient between each test parameter and the health status in the aging experiment data of the reference battery; comparing the calculation result of the correlation coefficient between each test parameter and the health status with a correlation threshold; determining a test parameter corresponding to a correlation coefficient calculation result greater than the correlation threshold as a health factor; and extracting experimental data corresponding to the health factor from the aging experiment data of the reference battery to obtain the health characteristic data of the reference battery.
[0008] In an exemplary embodiment of the present invention, the correlation coefficient calculation result may include: a Pearson correlation coefficient and a Spearman correlation coefficient.
[0009] In an exemplary embodiment of the present invention, the expansion processing of the health feature data of the reference battery by the TimeGAN model to generate the extended health feature data may include: performing time series sampling on the health feature data of the reference battery to obtain the original time series data; training the TimeGAN model using the original time series data as training samples and the health status as a label; during the training process, optimizing the reconstruction loss function, the unsupervised loss function and the supervised loss function simultaneously by using the gradient descent method; and determining the new time series data generated by the trained TimeGAN model as the extended health feature data.
[0010] In an exemplary embodiment of the present invention, the machine learning model may be a Gaussian process regression model or an LSTM model.
[0011] In an exemplary embodiment of the present invention, when the machine learning model is a Gaussian process regression model, during the training of the Gaussian process regression model, the hyperparameters of the Gaussian process regression model can be optimized by a global optimization algorithm to obtain the globally optimal hyperparameters.
[0012] In an exemplary embodiment of the present invention, the global optimization algorithm may be an ant colony algorithm or a particle swarm algorithm.
[0013] In an exemplary embodiment of the present invention, when the global optimization algorithm is an ant colony algorithm, in the process of optimizing the hyperparameters of the Gaussian process regression model by the ant colony algorithm, if the pheromone concentration of the initially selected node reaches a specified level, the simulated annealing algorithm can be used to optimize the initially selected node.
[0014] In an exemplary embodiment of the present invention, the kernel function of the Gaussian process regression model may be a combination of an RBF kernel function and a Matérn kernel function.
[0015] In an exemplary embodiment of the present invention, the hyperparameters of the Gaussian process regression model may include: the weight of the RBF kernel function, the weight of the Matérn kernel function, the Parameters, Matérn kernel function Parameters and Matérn kernel function Nu parameter.
[0016] A second aspect of the present invention provides a battery health status assessment method, which includes the following steps: obtaining health characteristic data of a target battery; inputting the health characteristic data of the target battery into a health status assessment model to obtain a health status assessment result of the target battery, wherein the health status assessment model is obtained by the health status assessment model construction method as described above.
[0017] The third aspect of the present invention provides a device for constructing a health status assessment model, the construction device comprising: a first data acquisition module, used to obtain health feature data of a reference battery based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery; a data expansion module, used to expand the health feature data of the reference battery through a TimeGAN model to generate extended health feature data; a model construction module, used to train a machine learning model using the extended health feature data as a training sample and the health status as a label to obtain a health status assessment model.
[0018] A fourth aspect of the present invention provides a battery health status assessment device, which includes: a second data acquisition module, used to obtain health characteristic data of a target battery; an assessment module, used to input the health characteristic data of the target battery into a health status assessment model to obtain a health status assessment result of the target battery, and the health status assessment model is obtained by the health status assessment model construction method as described above.
[0019] A fifth aspect of the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by one or more processors so that the processor executes the health status assessment model construction method as described above, or executes the battery health status assessment method as described above.
[0020] A sixth aspect of the present invention provides a computer-readable storage medium, which stores at least one program code, and the program code is loaded and executed by a processor to enable a computer to execute the health status assessment model construction method as described above, or to execute the battery health status assessment method as described above.
[0021] Through the technical solution provided by the present invention, the present invention has at least the following technical effects: (1) The method for constructing a health status assessment model of the present invention expands the battery aging test data through the TimeGAN model, and the expanded battery aging test data follows the same potential distribution law as the original battery aging test data, which solves the problem that the current data-driven battery health status estimation model cannot ensure the accuracy and stability of the model when there is only a small amount of battery aging test data, and reduces the impact of data scale on model feasibility to a certain extent; (2) When the health status assessment model is established by using the Gaussian process regression model, the method for constructing the health status assessment model of the present invention proposes to use the RBF kernel function, the Matérn kernel function and the assessment kernel function as the combined kernel function of the Gaussian process regression model. Compared with a single kernel function, the combined kernel function set by the present invention can not only balance the flexibility and fitting ability of the RBF kernel function and the Matérn kernel function to a certain extent, but also improve the stability of the Gaussian process regression model and reduce the overfitting risk that may be caused by a single kernel function. (3) When the health status assessment model is established by using the Gaussian process regression model, the health status assessment model construction method of the present invention proposes to introduce a global optimization algorithm (such as an ant colony algorithm, a particle swarm algorithm, etc.) to optimize the hyperparameters of the Gaussian process regression model, thereby improving the accuracy of the health status assessment model; (4) In the process of optimizing the hyperparameters by using the ant colony algorithm, the method for constructing the health status assessment model of the present invention also proposes to optimize and improve the ant colony algorithm by using a simulated annealing algorithm, thereby improving the optimization effect and reliability of the hyperparameters; (5) The battery health status assessment method of the present invention estimates the health status of the target battery by applying the above-mentioned health status assessment model, and can accurately estimate the health status of the target battery (for example, a lithium-ion battery) with only a small amount of battery aging experimental data.
[0022] Other features and advantages of the present invention will be described in detail in the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings: Figure 1 A flowchart of a method for constructing a health status assessment model provided by an embodiment of the present invention; Figure 2 A flowchart of another method for constructing a health status assessment model provided by yet another embodiment of the present invention; Figure 3 A flowchart of another method for constructing a health status assessment model provided by yet another embodiment of the present invention; Figure 4 A schematic diagram of a simulated annealing-ant colony algorithm process according to an embodiment of the present invention; Figure 5 A flowchart of another method for constructing a health status assessment model provided by yet another embodiment of the present invention; Figure 6 A schematic diagram of battery capacity change provided by an embodiment of the present invention; Figure 7 A schematic diagram of comparison results of a TimeGAN-generated dataset of a lithium-ion battery numbered B0006 provided in accordance with an embodiment of the present invention; Figure 8 A schematic diagram of health characteristic evaluation results of a lithium-ion battery numbered B0005 provided in accordance with an embodiment of the present invention; Fig. 9 A schematic diagram of health characteristic evaluation results of a lithium-ion battery numbered B0006 provided in accordance with an embodiment of the present invention; Fig.10 A schematic diagram of health characteristic evaluation results of a lithium-ion battery numbered B0007 provided in accordance with an embodiment of the present invention; Fig.11 A schematic diagram of health characteristic evaluation results of a lithium-ion battery numbered B0018 provided in accordance with an embodiment of the present invention; Fig.12 A structural block diagram of a device for constructing a health status assessment model provided by an embodiment of the present invention; Fig.13 A structural block diagram of a battery health status assessment device provided by an embodiment of the present invention; Fig.14A structural block diagram of an electronic device provided for one embodiment of the present invention.
[0024] Reference numerals 101 - first data acquisition module, 102 - data expansion module, 103 - model construction module, 201 - second data acquisition module, 202 - evaluation module, 301 - processor, 302 - memory. DETAILED DESCRIPTION
[0025] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.
[0026] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.
[0027] In the present invention, unless otherwise specified, “first”, “second”, etc. are merely used to facilitate description and distinction and cannot be understood as indicating or implying relative importance.
[0028] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "installation", "connection", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integrated connection; it can be a direct connection, or an indirect connection; it can be a wired connection, or a wireless connection. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.
[0029] The service life and health of lithium-ion batteries are closely related to the battery's state of health (SOH). However, due to the long service life of lithium-ion batteries, it is impossible to obtain more battery degradation process data during experiments and actual operation. Existing data-driven health status assessment models often rely on a large amount of battery aging experimental data. When the amount of data is reduced, the stability and accuracy of the health status assessment model will be greatly affected. Therefore, how to ensure that the health status assessment model can maintain its accuracy and stability when there is only a small amount of battery aging data is still a challenge.
[0030] In view of the technical problem in the prior art that the data-driven battery health status assessment model cannot guarantee its accuracy and stability when there is only a small amount of battery aging test data, the present invention provides a method for constructing a health status assessment model. The method first extracts battery health feature data with a high correlation with the battery health status from the battery aging test data, then uses the TimeGAN method to train and generate a battery health feature data set to achieve the expansion of the battery health feature data, and finally uses the expanded battery health feature data to train the machine learning model to achieve the construction of the health status assessment model. Compared with the health status assessment model construction method in the prior art, the present invention can significantly improve the accuracy and reliability of the assessment model when there is only a small amount of battery aging test data. In specific implementation, the above method can be executed by an electronic device, which can be a server, terminal or other device with processing function.
[0031] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and in combination with embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0032] In order to reduce the impact of data size on the feasibility of the health status assessment model to a certain extent, an embodiment of the present invention provides a method for constructing a health status assessment model, such as Figure 1 As shown, the construction method includes the following steps: Step S101, based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery, obtain the health characteristic data of the reference battery.
[0033] Here, the aging test data of the reference battery mainly refers to the data collected during long-term use or accelerated aging test of the reference battery. These data are used to evaluate the performance changes and life of the reference battery during use, usually including parameters such as capacity attenuation, charging efficiency, temperature change, terminal voltage, current, cycle life, impedance, etc.
[0034] The state of health (SOH) of a battery characterizes the ability of a current battery to store electrical energy relative to a new battery. It expresses the state of the battery from the beginning to the end of its life in the form of a percentage, and is used to quantitatively describe the performance state of the current battery. For example, the ratio between the full charge capacity and the rated capacity of the battery in its current state can be used as the SOH of the reference battery.
[0035] When extracting the health feature data of the battery, the correlation between the health feature data and the battery SOH should be considered. This is because when using machine learning algorithms to evaluate health features, the health feature data is actually mapped to SOH through certain model parameters. This requires that before using health feature data to train the health status assessment model, the correlation between each test parameter in the aging test data and SOH should be judged, and the test parameters with stronger correlation should be selected as the battery health feature data, so as to achieve accurate estimation of SOH.
[0036] Step S102, performing expansion processing on the health feature data of the reference battery through the TimeGAN model to generate expanded health feature data.
[0037] Here, the TimeGAN model is a generative adversarial network specifically designed for generating time series data. It consists of four network components: embedding function, recovery function, sequence generator, and sequence discriminator. The key to these components is that they are jointly trained, allowing TimeGAN to simultaneously learn to encode features, generate representations, and iterate across time. The embedding network provides a latent space in which the adversarial network operates, and the latent dynamics of real data and synthetic data are synchronized through supervised losses. The embedding function and the recovery function together constitute the auto-encoding component, which is responsible for encoding the original time series data into a latent space vector and decoding the data from the vector, with the goal of restoring the features and structure of the original time series data as much as possible. The sequence generator is responsible for generating new time series data based on the latent space vector provided by the embedding network, and the sequence discriminator determines whether the input time series data is real or generated by the generator. In this way, the generator and the discriminator are continuously optimized in the adversarial process, and finally the generator is able to generate synthetic time series data that is sufficient to deceive the discriminator.
[0038] That is to say, by inputting a small amount of health characteristic data of reference batteries to train the TimeGAN model, new time series data similar to the health characteristic data of the reference batteries can be generated, thereby achieving the expansion of the health characteristic data.
[0039] Step S103, training the machine learning model using the extended health feature data as training samples and the health status as labels.
[0040] Here, the data-driven approach to SOH estimation generally only requires the establishment of a health status assessment model based on the historical aging data of the battery, without the need to analyze the complex internal mechanisms of the battery. The commonly used data-driven approach is to use a machine learning model to construct a health status assessment model. For example, the machine learning model can be set to a Gaussian Process Regression model (GPR), a Long Short-term Memory model (LSTM), an Artificial Neural Network model (ANN), a Recurrent Neural Network (RNN), a Support Vector Machine model (SVM), a Relevance Vector Machine model (RVM), etc.
[0041] In addition, it should be noted that training samples refer to the data sets used by machine learning models for learning and training, and labels refer to the output information associated with each training sample, which is used to indicate or identify the category to which the training sample belongs or the expected output.
[0042] Step S104, using the trained machine learning model as a health status assessment model.
[0043] In this way, the battery aging experimental data is expanded through the TimeGAN model, and the expanded health feature data is used to train the machine learning model, so as to construct a mapping relationship between health features and SOH, and obtain a health status assessment model.
[0044] Further, in a possible implementation, in step S101, based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery, the process of obtaining the health characteristic data of the reference battery may include but is not limited to the following sub-steps S1011 to S1014.
[0045] Sub-step S1011, determining the calculation result of the correlation coefficient between each test parameter and the health state in the aging experiment data of the reference battery.
[0046] Sub-step S1012, comparing the calculation result of the correlation coefficient between each test parameter and the health status with the correlation threshold.
[0047] For example, the correlation threshold can be set to 0.7~0.9. If the calculated result of the correlation coefficient between the currently compared test parameter and the health status is greater than the correlation threshold, it is considered that the correlation between the test parameter and the health status is strong, and the machine learning model is trained as the health factor of the battery; conversely, if the calculated result of the correlation coefficient between the currently compared test parameter and the health status is less than the correlation threshold, it is considered that the correlation between the test parameter and the health status is weak.
[0048] Sub-step S1013, determining the test parameter corresponding to the correlation coefficient calculation result that is greater than the correlation threshold as a health factor.
[0049] Sub-step S1014, extracting experimental data corresponding to the health factor from the aging experimental data of the reference battery to obtain health characteristic data of the reference battery.
[0050] Furthermore, in a possible implementation manner, the correlation coefficient calculation result may be set as a Pearson correlation coefficient (or Pearson correlation coefficient) or a Spearman correlation coefficient (or Spearman correlation coefficient).
[0051] Among them, the Pearson correlation coefficient is mainly used to measure the degree of linear relationship between two variables, and its value is between -1 and 1.
[0052] The Spearman correlation coefficient is a non-parametric statistic used to measure the correlation between two variables without assuming that the relationship between the two variables is linear.
[0053] Preferably, the correlation coefficient calculation results may be set as the Pearson correlation coefficient and the Spearman correlation coefficient.
[0054] The reason for choosing to use both the Pearson correlation coefficient and the Spearman correlation coefficient as the correlation coefficient calculation results is that when verifying the correlation between health characteristic data and SOH, the relationship between the two data is relatively complex. If only one of the correlation coefficients is selected, it may be impossible to determine the relationship between some data. Using two correlation coefficients at the same time can more comprehensively evaluate the correlation between health characteristic data and SOH.
[0055] Further, in a possible implementation, in step S102, the health feature data of the reference battery is extended by the TimeGAN model, and the process of generating extended health feature data includes but is not limited to the following sub-steps S1021 to S1024.
[0056] Sub-step S1021 , performing time series sampling on the health characteristic data of the reference battery to obtain original time series data.
[0057] Sub-step S1022, using the original time series data as training samples and the health status as labels, to train the TimeGAN model.
[0058] Sub-step S1023, during the training process of the TimeGAN model, the reconstruction loss function is reconstructed using the gradient descent method. , unsupervised loss function And the supervised loss function Make optimizations.
[0059] Reconstruction loss function The function is expressed as: .
[0060] In the formula, Represents a random variable The probability distribution of time period T, subscript Represents the original data distribution; Represents the original static features; Represents the static features after passing through the autoencoder; Represents the original tense feature; Represents the temporal features generated by the autoencoder.
[0061] Unsupervised loss function The function is expressed as: .
[0062] In the formula, It represents the judgment result of the static characteristics obtained by the model; It represents the judgment result of the dynamic characteristics obtained by the model; Represents the judgment result of the original static feature; Represents the original static features and the judgment results of dynamic features.
[0063] Supervised Loss Function The function is expressed as: .
[0064] In the formula, Represented as current random temporal features; Represented as the current random static feature; Represented as the current random temporal feature vector; Represents the time series data produced by the generator.
[0065] Sub-step S1024, determining the new time series data generated by the trained TimeGAN model as extended health feature data.
[0066] When the machine learning model is determined to be a Gaussian process regression model, since the Gaussian process regression model generally uses a single kernel function, the single kernel function may limit the generalization ability of the Gaussian process regression model and affect its adaptability on different data sets. Therefore, in view of the defect of the Gaussian process regression model having low generalization ability, the kernel function of the Gaussian process regression model can be adjusted, and the mapping relationship between health characteristics and SOH can be better captured by setting a combined kernel function, thereby achieving a more accurate SOH estimation.
[0067] Based on this, another embodiment of the present invention provides another method for constructing a health status assessment model, such as Figure 2 As shown, the construction method includes the following steps: Step S201, based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery, obtain the health characteristic data of the reference battery.
[0068] Step S202, performing expansion processing on the health feature data of the reference battery through the TimeGAN model to generate expanded health feature data.
[0069] Step S203, constructing a kernel function of a Gaussian process regression model by combining kernel functions to construct an improved Gaussian process regression model.
[0070] For example, the kernel function of the improved Gaussian process regression model can be composed of a Gaussian radial basis kernel function (or RBF kernel function) and a Matérn kernel function. The combination of the RBF kernel function and the Matérn kernel function can be described as the following relationship: .
[0071] in, Represents the kernel function of the improved Gaussian process regression model; Represents the hyperparameter that controls the width of the RBF kernel function, which can also be called the length size parameter; Represents the hyperparameter that controls the width of the Matérn kernel function; represents the Matérn kernel function Nu Parameter, which controls the smoothness of the Matérn kernel function; Represents the weight of the RBF kernel function; Represents the weight of the Matérn kernel function.
[0072] It should be noted that the above relationship is only for the purpose of briefly describing the combination of the RBF kernel function and the Matérn kernel function, and is not the actual calculation formula of the kernel function of the improved Gaussian process regression model.
[0073] The RBF kernel function is expressed as follows: .
[0074] In the formula, Represents the Euclidean distance between samples; is a hyperparameter that controls the width of the function; In K It is the abbreviation of kernel, which refers to the kernel function. and are the two eigenvectors of the input.
[0075] The Matérn kernel function is expressed as follows: .
[0076] In the formula, is a small positive number used to avoid division by zero; is the smoothness parameter, which determines the shape of the function; is the gamma function; is the modified Bissel function.
[0077] By combining the RBF kernel function with the Matérn kernel function, the flexibility and fitting ability of the two can be balanced to a certain extent. At the same time, the combined kernel function can also improve the stability of the Gaussian process regression model and reduce the overfitting risk that may be caused by a single kernel function.
[0078] Step S204, using the extended health feature data as training samples and the health status as labels, to train the improved Gaussian process regression model.
[0079] Step S205: Using the trained improved Gaussian process regression model as a health status assessment model.
[0080] Of course, the present invention is not limited to this. Other combined kernel functions, such as a combination of a square exponential kernel function and a linear kernel function, are also suitable for constructing the health status assessment model of the present invention, as long as the stability of the Gaussian process regression model can be improved and the overfitting risk that may be caused by a single kernel function can be reduced.
[0081] In addition, when the machine learning model is determined to be a Gaussian process regression model, considering that the Gaussian process regression model has the problem of low prediction accuracy of the exported model due to unreasonable hyperparameter settings, in the process of training the Gaussian process regression model using extended health feature data, the global optimization algorithm can be used to adjust the hyperparameters of the Gaussian process regression model, thereby achieving optimization and improvement of the Gaussian process regression model, and then the optimized Gaussian process regression model can be used as a health status assessment model.
[0082] Based on this, another embodiment of the present invention provides another method for constructing a health status assessment model, such as Figure 3 As shown, the construction method includes the following steps: Step S301, based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery, obtain the health characteristic data of the reference battery.
[0083] Step S302, performing expansion processing on the health feature data of the reference battery through the TimeGAN model to generate expanded health feature data.
[0084] Step S303, using the extended health feature data as training samples and the health status as labels, the Gaussian process regression model is trained, and during the training process, the hyperparameters of the Gaussian process regression model are optimized using a global optimization algorithm.
[0085] Here, it should be noted that the hyperparameters of the Gaussian process regression model are generally obtained by minimizing the negative log-likelihood function and the conjugate gradient method, but the conjugate gradient method is essentially a local optimization algorithm, and the calculation structure depends on the artificial setting of the initial value, which easily causes the solved hyperparameters to fall into the local optimum. In order to avoid the local optimum, the embodiment of the present invention uses a global optimization algorithm to optimize the hyperparameters of the Gaussian process regression model, and the global optimal hyperparameters can be obtained.
[0086] For example, the global optimization algorithm can be set to an ant colony algorithm or a particle swarm algorithm.
[0087] Step S304: Use the trained Gaussian process regression model as a health status assessment model.
[0088] Of course, the present invention is not limited to this, and other global optimization algorithms, such as genetic algorithms, artificial fish swarm algorithms, etc., are also applicable to constructing the health status assessment model of the present invention, as long as the hyperparameters to be solved can be prevented from falling into the local optimum.
[0089] Furthermore, in a possible implementation, when the global optimization algorithm is determined to be the ant colony algorithm, due to some limitations of the ant colony algorithm, it may fall into a local optimal solution during the optimization process. It is possible to consider using a simulated annealing algorithm to improve the ant colony algorithm, and convert the solution space of the ant colony algorithm from a discrete space to a quasi-continuous space, so as to achieve a more reliable model optimization effect.
[0090] In the optimization process, the ant colony algorithm determines the pheromone concentration and heuristic information of each feasible solution node in the solution space. is the search space of the ant colony, that is, the solution space of the algorithm. Specifically, the selection probability of the node should satisfy the following formula: , .
[0091] In the formula, yes pheromone concentration of the node; yes The heuristic function of the node; yes Heuristic information of nodes. When performing parameter optimization, the loss function is usually used as heuristic information; and Determines the proportion of pheromone and heuristic information during state transition.
[0092] In addition, since the ant colony algorithm was originally used to find the optimal path, there is no correlation between adjacent paths in the scenario it considers. However, when faced with a continuous solution space, the heuristic information between adjacent nodes often has a certain potential correlation. For example, the heuristic information of nodes near the node with the highest heuristic information is also relatively high, and the ant colony algorithm for discrete space often ignores this feature. Therefore, in addition to the basic ant colony algorithm parameter search method, the embodiment of the present invention uses a simulated annealing method to expand the influence of the pheromone concentration of the current node on the selection probability to a certain range around it, so as to realize the probabilistic selection characteristics of the continuous space in the case of a discrete solution space.
[0093] The specific method is as follows: In the process of optimizing the hyperparameters of the Gaussian process regression model by using the ant colony algorithm, when the pheromone concentration of the selected node reaches the specified level, the simulated annealing algorithm is used to optimize the selected node. is the currently selected node, is a randomly selected new node, is the heuristic information calculation function, Indicates the temperature of the current simulated annealing algorithm. When using the simulated annealing algorithm, accept The probability of the node, that is, the Metropolis criterion is as follows: .
[0094] Moreover, in the simulated annealing algorithm, the probability of selecting a new node should satisfy the current node as the mean and is a normal distribution with variance is a function that is positively correlated with the pheromone concentration of the node. Through this method, the selection probability of the initial node is diffused to other surrounding nodes in the form of normal distribution, achieving the association effect between discrete nodes.
[0095] like Figure 4As shown in the figure, the actual calculation process of the ant colony algorithm improved by the simulated annealing method (referred to as simulated annealing-ant colony algorithm) is as follows: (1) Initialization.
[0096] a) Initialization parameters: determine the number of ant colonies, pheromone volatility coefficient, pheromone importance, heuristic information importance, etc.
[0097] b) Initialize the solution space: expand the parameter range into a discrete solution space as needed.
[0098] c) Initialize the pheromone matrix: Initially set the initial value of the pheromone for all solution space items, generally set to a small positive number. The feasible range of the parameter is 0.000001 to 1.0, which can be expanded into a solution space of length 6, which may be {0.00001, 0.0001, 0.001, 0.01, 0.1, 1.0}, and the value contained in it is The parameters may choose values, and in actual operation, in order to provide more options, the range of the solution space will be larger.
[0099] d) Initialize the ant colony: Before the algorithm starts, randomly select an initial node for the ant in the solution space. In the embodiment of the present invention, the initial node is a hyperparameter of the Gaussian process regression model. For example, assuming that the kernel function of the Gaussian process regression model is a combination of the RBF kernel function and the Matérn kernel function, the initial node of an ant may be (0.01, 0.01, 0.01, 0.01, 0.01), and the five parameters in the node correspond to the weight parameter of the RBF kernel function, the weight parameter of the Matérn kernel function, the RBF kernel function, and the weight parameter of the Matérn kernel function. Parameters, Matérn kernel function Parameters and Matérn kernel function parameter.
[0100] (2) Node selection: Select the next node to be visited by the ant colony. At this time, the node selection method needs to be determined according to the defined ant colony algorithm state transition rules.
[0101] In the embodiment of the present invention, the state transition rule can be divided into the following two parts: a) Based on the pheromone concentration and heuristic information of each node, the initial node selection is performed according to the node selection probability formula in the above ant colony algorithm.
[0102] b) When the pheromone concentration of the initially selected node reaches a certain level, in order to achieve the effect of extending the influence of the pheromone concentration of the node on the selection probability to the surrounding nodes, the simulated annealing method is used to optimize the selected node, so that the ant colony has a certain probability of selecting the node with a slightly lower pheromone concentration around it.
[0103] (3) Update pheromones.
[0104] a) Pheromone increase: When the ant colony passes through a node, pheromone will be produced on the node. The size of the pheromone is affected by the size of the heuristic information of the node. That is, when the node selected by the ant can produce a better effect in the model, the more pheromone is produced, thereby increasing the probability of the node being selected in the subsequent process.
[0105] b) Pheromone evaporation: In order to prevent the ant colony optimization process from falling into the local optimal solution and increase the probability of selecting poor nodes, pheromones will be evaporated after each iteration. Generally, the pheromone matrix is multiplied by the volatility coefficient p to achieve pheromone volatilization.
[0106] (4) Perform multiple rounds of iterations. In each round, all ants will reselect nodes and update pheromones until the predetermined number of iterations is reached or the stopping criterion is met (such as the quality of the solution no longer improves significantly or reaches the expected accuracy).
[0107] (5) The optimal node output generated by iteration is used as the hyperparameter of the Gaussian process regression model. Preferably, in order to reduce the impact of data scale on the feasibility of the health status assessment model to a certain extent and achieve the global optimal effect of the health status assessment model, it is also possible to consider using TimeGAN to expand the existing battery aging test data, and then construct the kernel function of the Gaussian process regression model by combining kernel functions. At the same time, the simulated annealing-ant colony algorithm is used to optimize the hyperparameters of the improved Gaussian process regression model.
[0108] Based on this, another embodiment of the present invention provides another method for constructing a health status assessment model, such as Figure 5 As shown, the construction method includes the following steps: Step S401, based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery, obtain the health characteristic data of the reference battery.
[0109] Step S402, performing expansion processing on the health feature data of the reference battery through the TimeGAN model to generate expanded health feature data.
[0110] Step S403, constructing the kernel function of the Gaussian process regression model by combining kernel functions to establish an improved Gaussian process regression model.
[0111] Among them, the combined kernel function of RBF kernel function and Matérn kernel function is used as the kernel function of the improved Gaussian process regression model.
[0112] Step S404, using the extended health feature data as training samples and the health status as labels to train the improved Gaussian process regression model, and using the simulated annealing-ant colony algorithm to optimize the hyperparameters of the improved Gaussian process regression model during the training process.
[0113] The specific process of optimizing the hyperparameters of the improved Gaussian process regression model by using the simulated annealing-ant colony algorithm can be referred to in the above embodiment. Figure 4 The calculation process shown will not be repeated here.
[0114] Step S405: Using the trained improved Gaussian process regression model as a health status assessment model.
[0115] In addition, in order to accurately estimate the health status of a target battery (e.g., a lithium-ion battery) when only a small amount of battery aging experimental data is available, an embodiment of the present invention further provides a battery health status evaluation method, which includes the following steps: Step S501, obtaining health characteristic data of a target battery.
[0116] Step S502: input the health characteristic data of the target battery into a health status assessment model to obtain a health status assessment result of the target battery.
[0117] The health status assessment model may be constructed and obtained by using the health status assessment model construction method in any of the above embodiments.
[0118] In order to verify the effectiveness and practicality of the health status assessment model constructed by the present invention, taking lithium-ion batteries as an example, the health status assessment model construction method of the embodiment of the present invention is implemented for the lithium-ion battery to obtain a health status assessment model, and verify whether the assessment result obtained by the model is accurate.
[0119] The lithium-ion battery data set uses lithium battery aging experimental data provided by NASA's Ames Center of Excellence in Prediction. Four 18650 lithium-ion batteries were selected, numbered B0005, B0006, B0007, and B0018. During the charging process, the four selected batteries were charged at a constant temperature of 24°C in a constant current (CC) mode of 1.5A until the battery voltage reached 4.2V, and then continued to charge in a constant voltage (CV) mode until the charging current dropped to 20mA. During discharge, the battery was discharged at a constant current of 2A until the voltages of B0005, B0006, B0007, and B0018 dropped to 2.7V, 2.5V, 2.2V, and 2.5V, respectively. The capacity changes of the four batteries can be seen in Figure 6 .
[0120] Based on the above lithium-ion battery dataset, the main process of the health status assessment model construction method based on TimeGAN and improved Gaussian process regression is as follows: Step S601: data processing and feature extraction.
[0121] After analyzing the correlation between the parameters of the NASA data set and the battery health status using the Pearson correlation coefficient and the Spearman correlation coefficient, multiple health factors were extracted to form health feature data. The extracted health factors are charging voltage rise time, discharge voltage drop time, charging current gradient, constant current charging time, maximum temperature, maximum temperature time, final temperature, and number of cycles.
[0122] Table 1 shows the correlation calculation results of the extracted 8 health factors and health status, where HF1 represents the charging voltage rise time, HF2 represents the discharge voltage drop time, HF3 represents the charging current gradient, HF4 represents the constant current charging time, HF5 represents the maximum temperature, HF6 represents the maximum temperature time, HF7 represents the final temperature, and HF8 represents the number of cycles. From the verification results of Table 1, the Pearson correlation of most health factors is above 0.9. In addition, the Spearman correlation is also above 0.7. It can be seen that the above health factors are strongly correlated with the battery SOH, so they can be used as battery health feature data for training and estimation of the TimeGAN model.
[0123] Table 1 Calculation results of the correlation between different health factors and health status
[0124] Step S602, TimeGAN expands the training data set.
[0125] The health feature data extracted in step S601 are sorted in a time series, and the data set is shifted and sampled by a sliding time window to obtain a three-dimensional data set of the time series, which is used as the training data of the TimeGAN model in step S602. The TimeGAN model is used for model training and data expansion, and the expanded health feature data set is output.
[0126] By reducing the dimension of the expanded health feature dataset into two-dimensional data, its similarity with the original data can be analyzed. Figure 7 The PCA distribution and t-SNE distribution of the extended data set of lithium-ion batteries numbered B0006 are shown. It can be seen that the data set expanded by the TimeGAN model follows the same potential distribution law as the original data set, so the extended data set can be used as a training data set for the estimation model.
[0127] Step S603: construct an improved Gaussian process regression model, and use simulated annealing-ant colony algorithm to optimize the improved Gaussian process regression model.
[0128] Firstly, the combined kernel function of Gaussian radial basis kernel function (RBF) and Matérn kernel function is used as the kernel function of Gaussian process regression model to construct an improved Gaussian process regression model.
[0129] Then, using the extended data set obtained in step S602, the ant colony algorithm optimized by the simulated annealing algorithm is used to optimize the hyperparameters of the Gaussian process regression model established in step S603, thereby constructing a mapping relationship between health characteristics and SOH. The hyperparameters involved in the optimization include RBF kernel function parameters, Matérn kernel function parameters, and the weight parameters of RBF and Matérn, respectively.
[0130] Step S604: using the optimized improved Gaussian process regression model as a health status assessment model for the lithium-ion battery.
[0131] When using the above-mentioned health status assessment model for lithium-ion batteries, the health status assessment result of the lithium-ion batteries can be obtained by directly inputting the original power station operation data under actual conditions into the health status assessment model for lithium-ion batteries.
[0132] In order to verify the effectiveness of the health status assessment model constructed by the embodiment of the present invention, the root mean square error RMSE, mean absolute error MAE and determination coefficient R² are used as evaluation indicators of model estimation accuracy. The calculation formulas of RMSE, MAE and R² are as follows: , , .
[0133] During the verification process, the charge and discharge data of each battery is divided into a training set and a test set. In order to verify the feasibility of the model with a small amount of data, the embodiment of the present invention selects the first 20% and the first 30% of the battery data set as the training set, and uses the data after that as the test set. The test results are as follows: Figure 8~Figure 11 As shown, it can be seen that when the first 30% of the battery data set is used as the training set, the SOH estimation curve is basically the same as the actual curve. When the first 20% of the battery data set is used as the training set, although there is some deviation compared to 30%, it is still close to the actual curve. Table 2 shows the evaluation results of the battery data set under the two conditions.
[0134] Table 2 Model evaluation results
[0135] More intuitively, the maximum error between the estimated SOH value and the actual value of the battery obtained by the analysis test is shown in Table 3. When the training set is the first 20% of the battery data set, the maximum error of the SOH estimated value of the B0006 battery with the largest error is 0.02378; and when the training set is the first 30% of the battery data set, the maximum error of the four data sets is less than 0.003. The above data show that the health status assessment model proposed in the embodiment of the present invention is still at a high level of accuracy when there are only a small number of battery aging experimental data sets.
[0136] Table 3 Maximum model error
[0137] In addition, it should be noted that the implementation environment of this embodiment includes at least one terminal and a server, and the health status assessment model construction method or battery health status assessment method in the above embodiment can be executed on the terminal or server respectively. The terminal and the server can be connected in communication to realize the interactive transmission of information.
[0138] Among them, the terminal can be any electronic product that can interact with the user through one or more methods such as keyboard, touchpad, touch screen, voice interaction, etc., such as PC (Personal Computer), PPC (Pocket Personal Computer), tablet computer, etc.
[0139] The server can be a single server or a server cluster composed of multiple servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.
[0140] like Fig.12 As shown, an embodiment of the present invention further provides a health status assessment model construction device, which includes: a first data acquisition module 101, a data expansion module 102 and a model construction module 103.
[0141] The first data acquisition module 101 is used to acquire health characteristic data of the reference battery based on the correlation between each test parameter and the health state in the aging experiment data of the reference battery.
[0142] The data expansion module 102 is used to expand the health feature data of the reference battery through the TimeGAN model to generate expanded health feature data.
[0143] The model building module 103 is used to train the machine learning model using the extended health feature data as training samples and the health status as labels to obtain a health status assessment model.
[0144] It should be noted that the health status assessment model construction device provided above only uses the division of the above-mentioned functional modules as an example when realizing its functions. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the health status assessment model construction device provided in the above embodiment and the health status assessment model construction method provided in the above embodiment belong to the same concept. The specific implementation process is detailed in the embodiment of the health status assessment model construction method, which will not be repeated here.
[0145] like Fig.13 As shown, an embodiment of the present invention further provides a battery health status evaluation device, which includes: a second data acquisition module 201 and an evaluation module 202.
[0146] The second data acquisition module 201 is used to acquire health characteristic data of a target battery.
[0147] The evaluation module 202 is used to input the health characteristic data of the target battery into a health status evaluation model to obtain a health status evaluation result of the target battery. The health status evaluation model can be obtained by the health status evaluation model construction method in any of the above embodiments.
[0148] Similarly, the battery health status assessment device provided above only uses the division of the above functional modules as an example to illustrate when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the battery health status assessment device provided in the above embodiment and the battery health status assessment method provided in the above embodiment belong to the same concept. The specific implementation process is detailed in the embodiment of the battery health status assessment method, which will not be repeated here.
[0149] like Fig.14 As shown, an embodiment of the present invention further provides an electronic device, which includes a processor 301 and a memory 302, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by one or more processors to enable the processor to implement the health status assessment model construction method in the above embodiment, or implement the battery health status assessment method in the above embodiment.
[0150] Of course, the electronic device may also have components such as a wired or wireless network interface, a keyboard, and an input / output interface for input and output. The electronic device may also include other components for realizing various functions of the device, which will not be described in detail here.
[0151] An embodiment of the present invention also provides a computer-readable storage medium, which stores at least one program code. The program code is loaded and executed by a processor so that the computer implements the health status assessment model construction method in the above embodiment, or implements the battery health status assessment method in the above embodiment.
[0152] Optionally, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, and an optical disc data storage device. Those skilled in the art may understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including several instructions for making a single-chip microcomputer, a chip or a processor perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc.
[0153] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, a variety of simple modifications can be made to the technical solution of the present invention, and these simple modifications all belong to the protection scope of the present invention.
[0154] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
[0155] In addition, various embodiments of the present invention may be arbitrarily combined, and as long as they do not violate the concept of the present invention, they should also be regarded as the contents disclosed by the present invention.
Claims
1. A method for constructing a health status assessment model, characterized in that: The construction method comprises: Based on the correlation between each test parameter and the health state in the aging experiment data of the reference battery, the health characteristic data of the reference battery is obtained; The health feature data of the reference battery is extended and processed through the TimeGAN model to generate extended health feature data; The machine learning model is trained using the extended health feature data as training samples and the health status as labels to obtain a health status assessment model.
2. The method for constructing a health status assessment model according to claim 1, characterized in that: The step of obtaining the health characteristic data of the reference battery based on the correlation between each test parameter and the health status in the aging experiment data of the reference battery includes: Determine the calculation results of the correlation coefficients between each test parameter and the health status in the aging test data of the reference battery; Compare the calculated results of the correlation coefficients between each test parameter and the health status with the correlation threshold; Determine the test parameter corresponding to the correlation coefficient calculation result that is greater than the correlation threshold as the health factor; The experimental data corresponding to the health factor is extracted from the aging experimental data of the reference battery to obtain the health characteristic data of the reference battery.
3. The method for constructing a health status assessment model according to claim 2, characterized in that: The correlation coefficient calculation results include: Pearson correlation coefficient and Spearman correlation coefficient.
4. The method for constructing a health status assessment model according to claim 1, characterized in that: The method of performing an extended processing on the health feature data of the reference battery by using the TimeGAN model to generate extended health feature data includes: Perform time series sampling on the health characteristic data of the reference battery to obtain original time series data; The TimeGAN model is trained using the original time series data as training samples and the health status as labels; During the training process, the reconstruction loss function, unsupervised loss function and supervised loss function are optimized simultaneously using gradient descent. The new time series data generated by the trained TimeGAN model is determined as the extended health feature data.
5. The method for constructing a health status assessment model according to claim 1, characterized in that: The machine learning model is a Gaussian process regression model or an LSTM model.
6. The method for constructing a health status assessment model according to claim 5, characterized in that: When the machine learning model is a Gaussian process regression model, during the training process of the Gaussian process regression model, the hyperparameters of the Gaussian process regression model are optimized by a global optimization algorithm to obtain the global optimal hyperparameters.
7. The method for constructing a health status assessment model according to claim 6, characterized in that: The global optimization algorithm is an ant colony algorithm or a particle swarm algorithm.
8. The method for constructing a health status assessment model according to claim 7, characterized in that: When the global optimization algorithm is the ant colony algorithm, in the process of optimizing the hyperparameters of the Gaussian process regression model by the ant colony algorithm, if the pheromone concentration of the initially selected node reaches a specified level, the simulated annealing algorithm is used to optimize the initially selected node.
9. The method for constructing a health status assessment model according to claim 5, characterized in that: The kernel function of the Gaussian process regression model is a combination of the RBF kernel function and the Matérn kernel function.
10. The method for constructing a health status assessment model according to claim 9, characterized in that: The hyperparameters of the Gaussian process regression model include: the weight of the RBF kernel function, the weight of the Matérn kernel function, the Parameters, Matérn kernel function Parameters and Matérn kernel function Nu parameter.
11. A battery health status assessment method, characterized in that: The evaluation methods include: Obtain health characteristic data of the target battery; The health characteristic data of the target battery is input into a health status assessment model to obtain a health status assessment result of the target battery, wherein the health status assessment model is obtained by the health status assessment model construction method described in any one of claims 1 to 10.
12. A device for constructing a health status assessment model, characterized in that: The construction device comprises: A first data acquisition module, used to acquire health characteristic data of a reference battery based on the correlation between each test parameter and the health state in the aging experiment data of the reference battery; A data extension module, used to extend the health feature data of the reference battery through the TimeGAN model to generate extended health feature data; The model building module is used to train the machine learning model using the extended health feature data as training samples and the health status as labels to obtain a health status assessment model.
13. A battery health status assessment device, characterized in that: The evaluation device comprises: A second data acquisition module is used to acquire health characteristic data of the target battery; An evaluation module is used to input the health characteristic data of the target battery into a health status evaluation model to obtain a health status evaluation result of the target battery, wherein the health status evaluation model is obtained by the health status evaluation model construction method according to any one of claims 1 to 10.
14. An electronic device, characterized in that: The electronic device includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by one or more processors so that the processor executes the health status assessment model construction method described in any one of claims 1 to 10, or executes the battery health status assessment method described in claim 11.
15. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one program code, and the program code is loaded and executed by a processor so that the computer executes the health status assessment model construction method described in any one of claims 1 to 10, or executes the battery health status assessment method described in claim 11.
Citation Information
Cited By
New energy heavy truck battery replacement system and method based on intelligent management
CN119294640A
Lithium iron phosphate battery monitoring system
CN119355541A
Lithium battery health degree evaluation method and system and electronic equipment
CN120046465A
Battery health state prediction method and system
CN120195574A
Wide-temperature-range lithium battery health state estimation method and system based on MGPR and improved time sequence RBF
CN120214615A
