A Method for Predicting the State of Health (SOH) of a Rechargeable Nickel-Metal Hydride Battery

Abstract

The present invention relates to a method for predicting the SOH of a retired power nickel-metal hydride battery, belonging to the technical field of energy storage. First, data indicators related to battery capacity decline under the battery charging state are collected, and data cleaning is carried out to ensure data validity. Taking battery internal resistance and battery maximum internal pressure as health factors, eigenvalue is constructed, and the health factors and battery capacitance together form a sample set; Then, the LightGBM algorithm is optimized through an improved adaptive loss function, and the Hyperopt hyperparameter framework is applied to obtain the optimal parameter combination; The sample set is loaded into the optimized algorithm model for training; The trained algorithm training model is used to predict the SOH. By using this method, the efficiency and accuracy of SOH prediction of retired batteries can be improved, ensuring that the hierarchical utilization of retired batteries can be classified efficiently and accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy storage, and relates to a method for predicting the SOH of retired power nickel-metal hydride batteries. Background Art

[0002] SOH prediction is a key step in the cascade utilization of retired batteries. The applicable scenarios of batteries are determined according to the current SOH of the retired batteries, and the accuracy of its SOH prediction determines the rationality of battery screening and reconfiguration. With the advent of the retirement period of electric vehicle power batteries in China, the number of retired electric vehicle batteries is huge. Traditional SOH testing methods are time-consuming, laborious, and consumptive of materials. Therefore, it is necessary to invent a method for rapid SOH prediction of large-scale retired batteries to promote the development of the field of cascade utilization of retired batteries.

[0003] Nowadays, parameters such as voltage and equal voltage drop time are used as health factors, and methods such as SVM algorithm, genetic algorithm, and LSTM neural network can be well applied. Although these algorithms can achieve SOH prediction, in the face of a huge number of retired batteries, they cannot give accurate and rapid prediction results. Therefore, for the rapid and accurate prediction of retired batteries, it is the key to efficiently obtain eigenvalue strongly related to SOH. Secondly, it is also necessary to select a suitable prediction algorithm model to ensure prediction accuracy and operation speed. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method for predicting the SOH of retired power nickel-metal hydride batteries, which solves the problems of complex prediction process and insufficient prediction accuracy of traditional SOH prediction methods, overcomes the problems of general machine learning that requires a large amount of training sets, large amount of calculation, long calculation time, and high requirements for computer hardware, and improves the self-adaptability of the algorithm to avoid overfitting of the training model.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] A method for predicting the SOH of retired power nickel-metal hydride batteries, the method comprising the following steps:

[0007] S1: Collect the battery capacity under the charging state of the power nickel-metal hydride battery, the internal resistance of the battery in the power state of 20% - 80% of the battery power, and the maximum internal pressure data of the battery;

[0008] S2: Clean the collected data by filling missing values, deleting outliers, and normalizing;

[0009] S3: Jointly construct a health factor from the internal resistance of the battery and the maximum internal pressure of the battery, establish a mapping relationship between the health factor as an eigenvalue and the battery capacity that matches it, and form a sample library;

[0010] S4: Use the improved LightGBM as the main algorithm for predicting the state of health (SOH); improve the original loss function through an adaptive loss function to reduce the impact of data outliers;

[0011] S5: Use the Hyperopt framework for hyperparameters to construct a model parameter space, a LightGBM model factory, and a score getter to achieve efficient model optimization and tuning of parameters;

[0012] S6: Establish a model framework from steps S5 and S4, and import the sample library processed in step S2 into the model for training;

[0013] S7: Import the maximum internal pressure and battery internal resistance data of the current returned battery under the charging state into the model to quickly predict the battery capacity.

[0014] S8: Calculate the battery SOH, and obtain the battery SOH. The calculation formula is:

[0015]

[0016] Where: C p is the predicted battery capacity, and C0 is the rated battery capacity;

[0017] Optionally, in S1, collect the battery capacity, battery internal resistance, and maximum internal pressure data of the power nickel-metal hydride battery under the charging state. Among them, the battery internal resistance collects data in the 20%-80% SOC state of the battery to ensure the accuracy of the internal resistance; the maximum internal pressure data collects data in the 80%-100% SOC state of the battery to ensure the authenticity of the internal pressure.

[0018] Optionally, in S2, the specific data normalization processing method is: the mean-variance normalization method, which is a method of mapping data to a data distribution with a mean of 0 and a variance of 1.

[0019] Its calculation formula is:

[0020]

[0021] Where, f i is the input feature, is the input feature after normalization. Normalizing all data can greatly reduce the complexity of the data structure and also play a crucial role in reducing the calculation time.

[0022] Optionally, in S3, use the dynamic internal resistance and the maximum internal pressure to jointly construct a health factor as the eigenvalue of the prediction model; among them, the dynamic internal resistance data is simple and efficient to obtain; the stable internal pressure characteristic of the nickel-metal hydride battery has a strong correlation with the battery life, which improves the accuracy of battery SOH prediction;

[0023] During the charging of a nickel-metal hydride battery, the reactions occurring at the positive and negative electrodes are as follows:

[0024] Positive electrode: Ni(OH)2 + OH - →NiOOH + H2O + e -

[0025] Negative electrode: H2O + e - →OH - + 1 / 2H2

[0026] There is a corresponding relationship between the hydrogen storage alloy at the negative electrode of the nickel-metal hydride battery and the internal pressure. There is a corresponding relationship between the SOH of the battery and the hydrogen storage alloy at the negative electrode of the battery. There is also a strong correlation between the internal pressure of the battery and the SOH of the battery. The internal pressure of the battery is used as a health factor to predict the remaining service life of the battery.

[0027] Optionally, in S3, an improved LightGBM is used as the main algorithm for object detection. The original loss function is improved through an adaptive loss function to reduce the impact of outliers on the prediction accuracy;

[0028] The calculation formula of the loss function is:

[0029]

[0030] Loss functions corresponding to different hyperparameters α:

[0031]

[0032] When the hyperparameter α takes different values, a suitable loss function is represented according to the data characteristics to reduce the impact of discrete groups on the prediction accuracy.

[0033] Optionally, in S4, the Hyperopt hyperparameter optimization framework, the training method based on multi-threaded parallel histograms, and the GOSS processing method are used. After preprocessing the data, a LightGBM model factory and a score acquirer are created. Each parameter model is produced by the factory, and each model is evaluated by the score acquirer. Finally, the best model parameters are selected to achieve efficient optimization tuning of the improved LightGBM model;

[0034] Among them, the training method based on multi-threaded parallel histograms: The continuous floating-point numbers of the feature values are converted into K discrete values, and finally a histogram with a width of K is constructed, where K is the number of calculations.

[0035] GOSS processing method: Samples with large gradients are retained during data processing. A threshold is preset, and samples with small gradients are randomly removed. The large-gradient sampling rate in the data is a, and the small-gradient sampling rate is b; The data splitting points are reduced, and the learning machine is trained by setting the iteration termination conditions to improve the learning efficiency and prediction accuracy.

[0036] Among them, the calculation formula for the maximum information gain point is as follows:

[0037]

[0038] x i is the data sample of the training set, and x ij is the data sample of the training set under the segmentation feature j; g i represents the negative gradient direction of the loss function of the model data variable during each gradient iteration; O represents the training set of a certain fixed node, d represents the segmentation point under the segmentation feature j;

[0039] n O = ∑I[x i ∈ O]

[0040] n o represents the number of samples in the training set of a certain fixed node;

[0041]

[0042] n j represents the number of samples whose value on the j-th feature is less than or equal to d;

[0043]

[0044] n j represents the number of samples whose value on the j-th feature is greater than d;

[0045] Hyperopt hyperparameter optimization framework: Through the set objective function and parameter space, using the TPE (Tree-of-Parzen-Estimators, TPE) algorithm in the Hyperopt framework, first randomly sample some hyperparameters, and then use the sampled hyperparameters to evaluate the objective function. Set the number S of models to be trained, and randomly adopt the hyperparameters S', to achieve optimal tuning of parameters.

[0046] The beneficial effects of the present invention are as follows:

[0047] 1. The purpose of the present invention is to provide a method for quickly predicting the SOH of a retrograde motive nickel-metal hydride battery. This method solves the problems of complex prediction process and insufficient prediction accuracy using traditional SOH prediction methods, and also overcomes the problems of large model parameter quantity and difficult calculation in machine learning methods. At the same time, the algorithm is improved, so that the self-adaptability of the algorithm avoids overfitting of the training model.

[0048] 2. The present invention constructs a health factor using the internal resistance and the maximum internal pressure during the battery charging process. Among them, the acquisition of internal resistance data is relatively simple, and the internal resistance value is relatively constant in the 20%-80% SOC state of the battery, which can improve the data acquisition efficiency; there is a strong one-to-one correspondence between the maximum internal pressure and the battery SOH, which greatly improves the accuracy of SOH prediction.

[0049] 3. The present invention uses an improved LightGBM as the main prediction algorithm, introduces a loss function to improve the original loss function, and reduces the influence of outliers on the prediction accuracy. The hyperparameters are expressed as appropriate loss functions according to the data characteristics, thereby reducing the influence of discrete groups on the prediction accuracy.

[0050] 4. The present invention uses the Hyperopt hyperparameter optimization framework. Based on the training method of multi-threaded parallel histograms and the GOSS processing method, after preprocessing the data, a LightGBM model factory and a score acquirer are created. Each parameter model is produced by the factory, and each model is evaluated by the score acquirer. Finally, the best model parameters are selected, thereby realizing automatic optimal tuning of parameters, improving the model accuracy, and enhancing the prediction efficiency.

[0051] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0053] Figure 1 is the flowchart of the present invention;

[0054] Figure 2 is the histogram training method;

[0055] Figure 3 is the GOSS processing flowchart;

[0056] Figure 4 is the Hyperopt working flowchart. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0058] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0059] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0060] like Figure 1 As shown, the SOH rapid prediction method of the depowered nickel-hydrogen battery includes the following specific steps:

[0061] (1) Collect data on battery capacity, battery internal resistance (20%-80% SOC state), and battery maximum internal pressure of power nickel-hydrogen batteries under charging status;

[0062] (2) Clean the collected data, fill missing values with the mean value method, delete outliers, and normalize with the mean-variance method;

[0063] (3) The internal resistance and maximum internal pressure data of the battery after cleaning are used together to construct a health factor, and a mapping relationship is established between the health factor as a characteristic value and the battery capacity that matches it to form a sample library;

[0064] (4) The improved LightGBM is used as the main algorithm for SOH prediction, and the loss function is introduced to improve the original loss function to reduce the impact of data outliers;

[0065] (5) Use the Hyperopt framework for hyperparameters to construct the model parameter space, the LightGBM model factory, and the score getter to achieve efficient optimization of model tuning parameters;

[0066] (6) Establish a model framework from steps (4) and (5), and import the sample library processed in step (2) into model training;

[0067] (7) Import the maximum internal pressure and battery internal resistance data of the battery under the current returned battery charging state into the model to quickly calculate the battery capacity.

[0068] (8) Calculate the battery SOH to obtain the real-time SOH of the battery.

[0069] Based on the multi-threaded parallel histogram training method: convert the continuous floating-point numbers of feature values into K discrete values, and finally construct a histogram with a width of K, where K is the number of calculations. The histogram training method is as Figure 2 shown.

[0070] The GOSS processing method: retain the samples with large gradients (pre-set threshold) during data processing, randomly remove the samples with small gradients (the large-gradient sampling rate in the data is a, and the small-gradient sampling rate is b), reduce the data splitting points, and train the learner by setting the iteration termination condition to improve the learning efficiency and prediction accuracy. The GOSS processing flow is as Figure 3 shown.

[0071] Among them, the calculation formula for the maximum information gain point is:

[0072]

[0073] x i is the data sample of the training set, and x ij is the data sample of the training set under the splitting feature j; g i represents the negative gradient direction of the loss function of the model data variable during each gradient iteration; O represents the training set of a certain fixed node, and d represents the splitting point under the splitting feature j;

[0074] n O =ΣI[x i ∈O]

[0075] n o represents the number of samples in the training set of a certain fixed node;

[0076]

[0077] n j represents the number of samples whose value on the j-th feature is less than or equal to d;

[0078]

[0079] n j represents the number of samples with a value greater than d on the j-th feature;

[0080] Hyperopt hyperparameter optimization framework: Through the set objective function and parameter space, using the TPE (Tree-of-Parzen-Estimators) algorithm in the Hyperopt framework, first randomly sample some hyperparameters, and then use the sampled hyperparameters to evaluate the objective function (set the number of models S to be trained, randomly adopt hyperparameters S'), to achieve optimal tuning of parameters. The Hyperopt work process is as Figure 4 shown.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the SOH of a nickel-metal hydride battery for reverse driving force, characterized in that: The method includes the following steps: S1: Collect data on the battery capacity, the internal resistance of the battery in the 20% - 80% state of charge, and the maximum internal pressure of the battery when the power nickel-metal hydride battery is charging; S2: Clean the collected data by filling in missing values, deleting outliers, and normalizing; S3: jointly construct a health factor from the battery internal resistance and the maximum internal pressure of the battery, establish a mapping relationship between the health factor as a characteristic value and the battery capacity it matches, and form a sample library; Use the dynamic internal resistance and the maximum internal pressure to jointly construct a health factor as the characteristic value of the prediction model; among them, the acquisition of dynamic internal resistance data is simple and efficient; the stable internal pressure characteristic of nickel-metal hydride batteries has a strong correlation with the battery life, improving the accuracy of battery SOH prediction; When the nickel-metal hydride battery is charging, the reactions occurring at the positive and negative electrodes are as follows: Positive electrode: Ni(OH)2+OH - →NiOOH+H2O+e - Negative electrode: H2O + e - → OH - + 1 / 2H2 There is a corresponding relationship between the hydrogen storage alloy at the negative electrode of the nickel-metal hydride battery and the internal pressure, there is a corresponding relationship between the battery SOH and the hydrogen storage alloy at the negative electrode of the battery, and there is also a strong correlation between the battery internal pressure and the battery SOH. Use the battery internal pressure as a health factor to predict the remaining service life of the battery; S4: Use the improved LightGBM as the main algorithm for predicting the health state SOH; improve the original loss function through an adaptive loss function to reduce the influence of data outliers; S5: Use the Hyperopt framework for hyperparameters, construct a model parameter space, a LightGBM model factory, and a score acquirer to achieve efficient optimization of model hyperparameters; S6: Establish a model framework from S5 and S4, and import the sample library processed by S2 into model training; S7: Import the maximum internal pressure and internal resistance data of the current returned battery when it is charging into the model to quickly predict the battery capacity; S8: Calculate the battery SOH, and obtain the battery SOH. The calculation formula is: Where: C p is the predicted battery capacity, and C0 is the rated battery capacity.

2. The SOH prediction method for a nickel-metal hydride battery with reverse driving force according to claim 1, characterized in that: In S1, collect data on the battery capacity, internal resistance, and maximum internal pressure of the power nickel-metal hydride battery when it is charging. Among them, the internal resistance of the battery collects data in the 20% - 80% SOC state to ensure the accuracy of the internal resistance; the maximum internal pressure data collects data in the 80% - 100% SOC state of the battery to ensure the authenticity of the internal pressure.

3. The SOH prediction method for a nickel-metal hydride battery with reverse movement power according to claim 2, characterized in that: In S2, the specific method for data normalization is: the mean-variance normalization method, which is a method of mapping data to a data distribution with a mean of 0 and a variance of 1. The calculation formula is: where f i is the input feature, and is the normalized input feature; normalizing all data can greatly reduce the complexity of the data structure and also plays a crucial role in reducing the calculation time.

4. The SOH prediction method for a nickel-metal hydride battery with reverse driving force according to claim 1, wherein: In S3, use the improved LightGBM as the main algorithm for object detection, and improve the original loss function through an adaptive loss function to reduce the influence of outliers on the prediction accuracy; The calculation formula of the loss function is: Loss functions corresponding to different hyperparameters α: When the hyperparameter α takes different values, represent it as an appropriate loss function according to the data characteristics to reduce the influence of discrete groups on the prediction accuracy.

5. A method for predicting the SOH of a nickel-metal hydride battery for reverse movement power according to claim 4, characterized in that: In S4, use the Hyperopt hyperparameter optimization framework, the training method based on multi-threaded parallel histograms, and the GOSS processing method. After preprocessing the data, create a LightGBM model factory and a score acquirer. Through the factory, produce models with various parameters, and use the score acquirer to evaluate each model. Finally, select the best model parameters to achieve efficient optimization of the improved LightGBM model; Among them, based on the multi-threaded parallel histogram training method: convert the continuous floating-point numbers of feature values into K discrete values, and finally construct a histogram with a width of K, where K is the number of calculations; GOSS processing method: Retain samples with large gradients during data processing, preset a threshold, randomly remove samples with small gradients, the large-gradient sampling rate in the data is a, and the small-gradient sampling rate is b; reduce the data splitting point, train the learning machine by setting the iteration termination condition, and improve the learning efficiency and prediction accuracy; Among them, the calculation formula for the maximum information gain point is: x u is a data sample of the training set, x ij is a data sample of the training set under the segmentation feature j; g i represents the negative gradient direction of the loss function of the model data variable at each gradient iteration; O represents the training set of a certain fixed node, and d represents the segmentation point under the segmentation feature j; n O = ΣI[x i ∈O] n o represents the number of training set samples of a certain fixed node; n j represents the number of samples whose value on the j-th feature is less than or equal to d; n j represents the number of samples with a value greater than d on the j-th feature; Hyperopt hyperparameter optimization framework: Through the set objective function and parameter space, adopt the TPE (Tree-of-Parzen-Estimators, TPE) algorithm in the Hyperopt framework, first randomly sample some hyperparameters, and then use the sampled hyperparameters to evaluate the objective function, set the number of models S to be trained, and randomly adopt hyperparameters S', to achieve optimal tuning of parameters.

Images