CN115329277A - A SOH Prediction Method for Retired Ni-MH Batteries - Google Patents

Abstract

The invention relates to a method for predicting SOH of a movement-withdrawing nickel-metal hydride battery, belonging to the technical field of energy storage. Firstly, collecting data indexes related to battery capacity decline in a battery charging state, ensuring data validity through data cleaning, constructing characteristic values by taking battery internal resistance and battery maximum internal pressure as health factors, and forming a sample set by the health factors and battery capacitance together; optimizing a LightGBM algorithm through an improved adaptive loss function, and obtaining an optimal parameter combination by applying a Hyperopt hyper-parameter framework; loading the sample set into the optimized algorithm model for training; and (4) predicting the SOH by adopting the trained algorithm training model. By adopting the method, the SOH prediction efficiency and precision of the returned batteries can be improved, and the returned batteries can be efficiently and accurately classified by utilizing the echelon.

Description

A SOH prediction method for degraded power Ni-MH batteries

technical field

The invention belongs to the technical field of energy storage, and relates to an SOH prediction method of a degraded power nickel-hydrogen battery.

Background technique

SOH prediction is a key step in the cascade utilization of returned batteries. The current SOH of the returned batteries determines the applicable scenarios of the batteries, and the accuracy of the SOH prediction determines the rationality of the batteries passing through screening and reconfiguration. With the coming of the decommissioning period of electric vehicle power batteries in my country, the number of returned batteries for electric vehicles is huge, and the traditional SOH test method is time-consuming, labor-intensive, and consumable. Therefore, it is necessary to invent a rapid SOH prediction for large-scale returned batteries method to promote the development of the field of cascade utilization of returned batteries.

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 realize SOH prediction, they are faced with a huge number of returned batteries. , they cannot give accurate and fast prediction results. Therefore, for the rapid and accurate prediction of returned batteries, the key is to efficiently obtain the eigenvalues strongly correlated with SOH. Secondly, it is also necessary to select an appropriate prediction algorithm model to ensure the prediction accuracy and operation speed.

Contents of the invention

In view of this, the object of the present invention is to provide a SOH prediction method for deactivated nickel-metal hydride batteries, which solves the problems of complex prediction process and insufficient prediction accuracy of traditional SOH prediction methods, and overcomes the need for a large number of training sets and calculations for general machine learning. It improves the adaptability of the algorithm to avoid over-fitting of the training model.

To achieve the above object, the present invention provides the following technical solutions:

A method for predicting SOH of a retired power nickel-metal hydride battery, the method comprising the following steps:

S1: Collect the battery capacity of the power Ni-MH battery in the charging state, the battery internal resistance of the 20% to 80% power state, and the maximum internal pressure data of the battery;

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

S3: The internal resistance of the battery and the maximum internal pressure of the battery are used to construct the health factor, and the health factor is used as the characteristic value to establish a mapping relationship with the battery capacity that matches it to form a sample library;

S4: Use the improved LightGBM as the main algorithm for SOH prediction of the health state; improve the original loss function through the adaptive loss function to reduce the influence of data outliers;

S5: Use the hyperparameter Hyperopt framework to build model parameter space, LightGBM model factory, and score acquirer to achieve efficient model optimization tuning;

S6: Establish a model framework by steps S5 and S4, and import the sample library processed in step S2 into model training;

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

S8: Calculate the battery SOH to obtain the battery SOH, the calculation formula is:

Among them: C _p is the predicted battery capacity, C ₀ is the rated capacity of the battery;

Optionally, in the S1, the battery capacity, battery internal resistance, and maximum internal pressure data of the power Ni-MH battery in the charging state are collected, wherein the battery internal resistance collects the data of the battery 20%-80% SOC state to ensure the accuracy of the internal resistance Reliability; the maximum internal pressure data collection battery 80% -100% SOC state data, to ensure the authenticity of the internal pressure.

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

Its calculation formula is:

为归一化后的输入特征。所有的数据进行归一化处理，能够极大地降低数据结构的复杂性，同时对减少计算时间也起着至关重要的作用。Among them, f _i is the input feature,

is the normalized input feature. All the data are normalized, which can greatly reduce the complexity of the data structure, and also play a vital role in reducing the calculation time.

Optionally, in the S3, the dynamic internal resistance and the maximum internal pressure are used to jointly construct the health factor as the characteristic value of the prediction model; the acquisition of the dynamic internal resistance data is simple and efficient; the characteristics of the stable internal pressure of the nickel-metal hydride battery are related to the battery life. Strong correlation improves the accuracy of battery SOH prediction;

When a NiMH battery is charged, the reactions at the positive and negative electrodes are as follows:

Positive electrode: Ni(OH) ₂ +OH ^- →NiOOH+H ₂ O+e ^-

Negative electrode: H ₂ O+e ^- → OH ^- +1/2H ₂

There is a corresponding relationship between the hydrogen storage alloy of the negative electrode of the nickel-hydrogen battery and the internal pressure, and there is a corresponding relationship between the battery SOH and the hydrogen storage alloy of the battery negative electrode. There is also a strong correlation between the internal pressure of the battery and the SOH of the battery. The battery internal pressure is used as a health factor to predict the remaining battery life life.

Optionally, in the S3, the improved LightGBM is used as the main algorithm of target detection, and the original loss function is improved through an adaptive loss function to reduce the influence of outliers on the prediction accuracy;

The loss function calculation formula is:

Different hyperparameters α correspond to loss functions:

When the hyperparameter α takes different values, it is expressed as an appropriate loss function according to the data characteristics, reducing the impact of the discrete group on the prediction accuracy.

Optionally, in the S4, the Hyperopt hyperparameter optimization framework, the training method based on the multi-threaded parallel histogram and the GOSS processing method are used to preprocess the data and create a LightGBM model factory and score acquirer, and produce each parameter through the factory The model and score acquirer evaluate each model, and finally screen out the best model parameters to realize the efficient optimization of the improved LightGBM model;

Among them, based on the multi-thread parallel histogram training method: convert the continuous floating-point number of the feature value into K discrete values, and finally construct a histogram with a width of K, where K is the number of calculations.

GOSS processing method: keep samples with large gradients during data processing, pre-set the threshold, and randomly remove samples with small gradients. The sampling rate of large gradients in the data is a, and the sampling rate of small gradients is b; reduce data splitting points, by setting The iterative termination condition trains the learner to improve learning efficiency and prediction accuracy.

Among them, the calculation formula of the maximum information gain point is:

_xi is the data sample of the training set, and _xij 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 at each gradient iteration; O represents the training of a fixed node Set, d represents the segmentation point under the segmentation feature j;

n _O ＝∑I[x _i ∈O]

n _o represents the number of training set samples of a fixed node;

n ^j represents the number of samples whose value on the jth feature is less than or equal to d;

n ^j represents the number of samples whose value is greater than d on the jth 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 For the objective function, set the number S of models to be trained, and randomly use the hyperparameter S' to achieve optimal parameter tuning.

The beneficial effects of the present invention are:

1. The purpose of the present invention is to provide a kind of SOH quick prediction method of deactivation power Ni-MH battery, this method solves the problem that uses traditional SOH prediction method to predict process complex, prediction precision is not enough, also overcomes in machine learning method for The number of model parameters is large, and the calculation is difficult. At the same time, the algorithm is also improved, so that the adaptive nature of the algorithm can avoid over-fitting of the training model.

2. The present invention adopts internal resistance and maximum internal pressure during battery charging to construct a health factor, wherein the internal resistance data is relatively simple to obtain and the internal resistance value is relatively constant in the state of 20%-80% SOC of the battery, which can improve the data acquisition efficiency; the maximum There is a strong one-to-one correlation between internal pressure and battery SOH, which greatly improves the accuracy of SOH prediction.

3. The present invention adopts the 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 a suitable loss function according to the data characteristics, thereby reducing the impact of discrete groups on the prediction accuracy.

4. The present invention uses the Hyperopt hyperparameter optimization framework, based on the multi-threaded parallel histogram training method and the GOSS processing method, to create a LightGBM model factory and score acquirer after preprocessing the data, and produce each parameter model and score acquirer through the factory Evaluate each model, and finally select the best model parameters, so as to realize automatic optimization of parameters, improve model accuracy, and improve prediction efficiency.

Other advantages, objects and features of the present invention will be set forth in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the investigation and research below, or can be obtained from Taught in the practice of the present invention. The objects and other advantages of the invention may be realized and attained by the following specification.

Description of drawings

In order to make the purpose of the present invention, technical solutions and advantages clearer, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

Fig. 1 is a flowchart of the present invention;

Fig. 2 is the histogram training method;

Fig. 3 is a flow chart of GOSS processing;

Figure 4 is the Hyperopt work flow chart.

Detailed ways

Embodiments of the present invention are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various modifications or changes can be made to the details in this specification based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the diagrams provided in the following embodiments are only schematically illustrating the basic concept of the present invention, and the following embodiments and the features in the embodiments can be combined with each other in the case of no conflict.

Wherein, the accompanying drawings are for illustrative purposes only, and represent only schematic diagrams, rather than physical drawings, and should not be construed as limiting the present invention; in order to better illustrate the embodiments of the present invention, some parts of the accompanying drawings may be omitted, Enlargement or reduction does not represent the size of the actual product; for those skilled in the art, it is understandable that certain known structures and their descriptions in the drawings may be omitted.

In the drawings of the embodiments of the present invention, the same or similar symbols correspond to the same or similar components; , "front", "rear" and other indicated orientations or positional relationships are based on the orientations or positional relationships shown in the drawings, which are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred devices or elements must It has a specific orientation, is constructed and operated in a specific orientation, so the terms describing the positional relationship in the drawings are for illustrative purposes only, and should not be construed as limiting the present invention. For those of ordinary skill in the art, the understanding of the specific meaning of the above terms.

As shown in Figure 1, the rapid SOH prediction method for degraded power Ni-MH batteries includes the following specific steps:

(1) Collect the battery capacity, battery internal resistance (20%-80% SOC state), and battery maximum internal pressure data under the charging state of the power Ni-MH battery;

(2) Clean the collected data, apply the mean value method to fill in the missing values, delete the observed abnormal values, and normalize the mean value and variance method;

(3) Combine the cleaned battery internal resistance and battery maximum internal pressure data together to construct a health factor, and use the health factor as a characteristic value to establish a mapping relationship with the battery capacity that matches it to form a sample library;

(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 and reduce the influence of data outliers;

(5) Use the hyperparameter Hyperopt framework to build model parameter space, LightGBM model factory, and score acquirer to achieve efficient model optimization tuning;

(6) Model framework is established by steps (4) and (5), and the sample library after step (2) is processed is imported into model training;

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

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

Based on the multi-thread parallel histogram training method: convert the continuous floating-point number of the feature value 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 shown in Figure 2.

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

Among them, the calculation formula of the maximum information gain point is:

_xi is the data sample of the training set, and _xij 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 at each gradient iteration; O represents the training of a fixed node Set, 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 fixed node;

n ^j represents the number of samples whose value on the jth feature is less than or equal to d;

n ^j represents the number of samples whose value is greater than d on the jth feature;

Hyperopt hyperparameter optimization framework: through the set objective function and parameter space, adopt 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 S of models that need to be trained, and randomly use the hyperparameter S') to achieve optimal parameter tuning. The Hyperopt workflow is shown in Figure 4.

Finally, it is noted that the above embodiments are only used to illustrate the technical solutions of the present invention without limitation. 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 carried out Modifications or equivalent replacements, without departing from the spirit and scope of the technical solution, should be included in the scope of the claims of the present invention.

Claims (6)

1. A method for predicting the SOH of a withdrawing force nickel-metal hydride battery is characterized by comprising the following steps: the method comprises the following steps:

s1: collecting battery capacity, battery internal resistance of 20-80% of electric quantity state and maximum internal pressure data of the power nickel-hydrogen battery in a charging state;

s2: cleaning the acquired data by filling missing values, deleting abnormal values and carrying out normalization processing;

s3: constructing a health factor by using the internal resistance of the battery and the maximum internal pressure of the battery together, and establishing a mapping relation by using the health factor as a characteristic value and the battery capacity matched with the characteristic value to form a sample library;

s4: adopting an improved LightGBM as a health state SOH prediction subject algorithm; the original loss function is improved through the self-adaptive loss function, and the influence of data outliers is reduced;

s5: constructing a model parameter space, a LightGBM model factory and a score acquirer by using a hyper-parameter Hyperopt frame, and realizing efficient model optimization parameter adjustment;

s6: establishing a model frame by the steps S5 and S4, and importing the sample library processed in the step S2 into a model for training;

s7: and importing the maximum internal pressure and internal resistance data of the battery in the current quit-transported battery charging state into a model, and rapidly predicting the battery capacity.

S8: calculating the SOH of the battery to obtain the SOH of the battery, wherein the calculation formula is as follows:

wherein: c _p To predict battery capacity, C ₀ The rated capacity of the battery.

2. The method for predicting the SOH of the deactivatable nickel-metal hydride battery according to claim 1, wherein: in the step S1, acquiring battery capacity, battery internal resistance and maximum internal pressure data of the power nickel-hydrogen battery in a charging state, wherein the battery internal resistance acquires the data of the battery in a 20-80% SOC state, and the accuracy of the internal resistance is ensured; maximum internal pressure data acquisition of 80% -100% data of SOC state, ensuring authenticity of internal pressure.

3. The method for predicting the SOH of the deactivatable nickel-metal hydride battery according to claim 2, wherein: in the S2, a specific data normalization processing method comprises the following steps: the mean variance normalization method is a method for mapping data to data distribution with a mean value of 0 and a variance of 1.

The calculation formula is as follows:

wherein f is _i As a characteristic of the input, the input is,

is the normalized input features. All data are normalized, so that the complexity of a data structure can be greatly reduced, and the method plays a crucial role in reducing the calculation time.

4. The method for predicting the SOH of a depravation force nickel-metal hydride battery according to claim 3, wherein: in the S3, a health factor is constructed by using the dynamic internal resistance and the maximum internal pressure together to serve as a characteristic value of a prediction model; the dynamic internal resistance data is simple and efficient to obtain; the characteristic of stable internal pressure of the nickel-metal hydride battery has strong correlation with the service life of the battery, and the accuracy of predicting the SOH of the battery is improved;

when the nickel-hydrogen battery is charged, the reactions of the positive electrode and the negative electrode are as follows:

and (3) positive electrode: ni (OH) ₂ +OH ^- →NiOOH+H ₂ O+e ^-

Negative electrode: h ₂ O+e ^- →OH ^- +1/2H ₂

The hydrogen storage alloy of the cathode of the nickel-metal hydride battery has a corresponding relation with the internal pressure, the hydrogen storage alloy of the cathode of the battery SOH has a corresponding relation with the hydrogen storage alloy of the cathode of the battery, the internal pressure of the battery has a strong correlation with the SOH of the battery, and the internal pressure of the battery is used as a health factor to predict the residual service life of the battery.

5. The method of claim 4 for predicting the SOH of the deactivatable nickel-metal hydride battery, wherein: in the S3, the improved LightGBM is adopted as a main algorithm of target detection, and the original loss function is improved through a self-adaptive loss function, so that the influence of outliers on the prediction precision is reduced;

the loss function is calculated as:

different superparameters α correspond to the loss functions:

when the hyper-parameter alpha takes different values, the hyper-parameter alpha is expressed as a proper loss function according to the data characteristics, and the influence of the discrete group on the prediction precision is reduced.

6. The method of claim 5 for predicting the SOH of the deactivatable nickel-metal hydride battery, wherein: in the S4, a Hyperopt hyper-parameter optimization frame, a training mode based on a multi-thread parallel histogram and a GOSS processing mode are adopted, a LightGBM model factory and a score acquirer are created after data are preprocessed, each parameter model and each score acquirer are produced in the factory, each model is evaluated, the optimal model parameters are finally screened out, and efficient optimization parameter adjustment of the improved LightGBM model is achieved;

the method is based on a multi-thread parallel histogram training mode: and converting continuous floating point numbers of the characteristic values into K discrete values, and finally constructing a histogram with the width of K, wherein K is the calculation times.

GOSS processing mode: reserving samples with large gradients in the data processing process, presetting a threshold value, and randomly removing samples with small gradients, wherein the sampling rate of the large gradients in the data is a, and the sampling rate of the small gradients in the data is b; data split points are reduced, and learning efficiency and prediction accuracy are improved by setting an iteration termination condition to train the learner.

Wherein, the maximum information gain point calculation formula is:

x _i for the data samples of the training set, x _ij Dividing training set data samples under the characteristic j; g _i Representing the direction of the negative gradient of the loss function of the model data variable at each gradient iteration; o represents a training set of a certain fixed node, d represents a segmentation point under a segmentation characteristic j;

n _O ＝∑I[x _i ∈O]

n _o representing the number of training set samples of a certain fixed node;

n ^j representing the number of samples with j-th characteristic upper value less than or equal to d;

n ^j representing the number of samples with the value larger than d on the jth characteristic;

hyperopt hyper-parameter optimization framework: according to the set target function and parameter space, a TPE (Tree-of-Parzen-Estimators) algorithm in a Hyperopt frame is adopted, some hyper-parameters are sampled randomly, then the sampled hyper-parameters are used for evaluating the target function, the number S of models needing to be trained is set, and the hyper-parameters S' are adopted randomly to realize optimized parameter adjustment.

Images