CN115986229A

CN115986229A - Method for judging health degree of battery based on charging segment data

Info

Publication number: CN115986229A
Application number: CN202211441810.1A
Authority: CN
Inventors: 常伟; 肖伟
Original assignee: Shanghai Decepticon Electric Appliance Co ltd
Current assignee: Shanghai Decepticon Electric Appliance Co ltd
Priority date: 2022-11-17
Filing date: 2022-11-17
Publication date: 2023-04-18

Abstract

A method for judging the health degree of a battery based on charging segment data comprises the following steps: s001, acquiring data; step S002 data preprocessing; step S003 of characteristic engineering; step S004 model construction; and step S005, testing and online of the model. Compared with the prior art, the invention has the advantages that: compared with the laboratory environment, the method disclosed by the invention is more suitable for the practical situation, and has higher theoretical significance and application value. The model can be used for optimizing and maintaining the battery of the two-wheeled electric vehicle, and is favorable for timely replacing the battery with abnormal attenuation, thereby providing guarantee for the driving safety of the two-wheeled electric vehicle.

Description

Method for judging health degree of battery based on charging segment data

Technical Field

The invention relates to the field of battery health degree assessment and prediction, in particular to a method for judging the battery health degree through charging segment data of a two-wheeled electric vehicle.

Background

Since the industrial revolution, with the continuous thinking and improvement of people on the environmental destruction and resource exhaustion brought by the modern industrial system based on fossil energy, along with the continuous innovation of scientific technology and the shortage of fossil energy resources, the nation is strongly and actively advocating the development of novel clean energy to replace the traditional fossil energy, thereby relieving the pressure caused by energy crisis and environmental pollution. The electric bicycle is more and more the best choice for people to travel for a short distance in modern big cities with large population and rural areas with remote and laggard sizes. The electric bicycle rapidly occupies the market with the outstanding advantages of economy and convenience, and plays an important role in the field of the current electric bicycle. The core of electric bicycles, power batteries, is receiving more and more attention. The development of power batteries is also in the recent years, and the requirements of power batteries are increasing, especially in terms of power, energy, safety, stability, and service life.

The power battery is used as a power source of the electric bicycle, is not a single individual, but is formed by combining a plurality of individual batteries in a series-parallel connection mode, and further meets various voltage grades, output currents and maximum output powers required by power systems of different two-wheeled electric vehicles. The power battery is one of three electricity supporting electric automobiles as the heart of a new energy automobile, and is also more and more concerned by more scholars. Generally, the performance indexes of the power battery mainly include energy, power density, high-temperature performance, low-temperature performance and energy storage performance. However, the problem of performance attenuation of the power battery in the using process is always puzzling the further development of the two-wheeled electric vehicle industry. Generally, the capacity of the power battery is reduced to below 80%, and the power battery is not suitable for being used as the power battery. The service life of the power battery before retirement is determined by the attenuation speed of the capacity of the power battery, so the evaluation and prediction of the residual life and the state of health (SOH) of the battery become important research subjects in the field of power batteries.

In a great amount of existing research, many scholars establish a capacity fading model based on single factors or multiple factors, and mainly attribute the influence factors of the residual service life SOH of the power battery to temperature, depth of discharge and discharge rate. Among these influencing factors, temperature becomes the most dominant influencing parameter of the performance of the power battery in the vehicle stage compared with the discharge rate and the discharge depth. The design capacity of the power battery for the vehicle is often too large, and the situation of discharging near the maximum discharge rate rarely occurs in the using process of the electric vehicle. If the discharge rate of the battery does not exceed the maximum discharge rate allowed by the battery, the change of the discharge rate does not have additional influence on the capacity degradation. In the research of thermal runaway faults, electric heating characteristics, battery safety and residual life of a power battery of a two-wheeled electric vehicle, aiming at the aspect of temperature conduction, a method of experiment and mathematical modeling is adopted to perform experiment and modeling on the anisotropic heat conduction process of lithium ion Chi Gexiang, perform parameter estimation and experiment simulation on the heat transfer characteristics of the battery, determine parameters for battery thermal characteristic modeling according to a graphical method, design various acupuncture thermal runaway experiments and build a single acupuncture model aiming at the thermal runaway phenomenon of a ternary lithium ion power battery, and perform three-dimensional modeling by coupling three factors of thermal runaway side reaction, joule heat and heat transfer; the experimental research is carried out aiming at the thermal runaway and the spreading characteristic, and laboratory data such as the internal and external temperature difference of the battery, the thermal runaway spreading time and the like are found when the thermal runaway occurs.

The invention adopts a method for judging the health degree of a battery based on charging segment data, adopts a machine learning method to evaluate and predict the SOH of the battery, and solves the actual SOH attenuation prediction problem from the external temperature acquisition perspective by acquiring the external temperature data of the battery of the two-wheeled electric vehicle and exploring the data to dig out the prediction index related to the SOH attenuation.

Disclosure of Invention

The invention provides a method for judging the health degree of a battery through charging segment data of a two-wheeled electric vehicle, aiming at overcoming the defects in the existing research.

The technical scheme adopted by the invention for solving the technical problems is as follows:

a method for judging the health degree of a battery based on charging segment data comprises the following steps: step S001, data acquisition; step S002 data preprocessing; s003, characteristic engineering; step S004 model construction; and step S005, testing and online of the model.

Compared with the prior art, the invention has the advantages that:

the method has the advantages that: the method comprises the steps of carrying out multipoint temperature measurement on the surface of each battery cell, taking acquired data as a data source of a prediction model, acquiring battery pack data of two-wheeled electric vehicles in two years, carrying out temperature measurement from the outside of the battery, acquiring temperature information to further generate a temperature gradient curve, and estimating the temperature inside the battery through the generated temperature gradient curve to further judge the health state of the battery; compared with the laboratory environment, the method is more suitable for the actual situation, and has higher theoretical significance and application value.

The method has the advantages that (2): the SOH attenuation prediction method comprises the steps of respectively counting temperature data of optical fiber temperature measuring points and temperature data acquired by temperature sensors in different SOC intervals and different charging time intervals, predicting the SOH attenuation state of a battery by researching the temperature gradient change of the battery and an LSTM neural network method, and constructing an SOH attenuation prediction model; the model can be used for optimizing and maintaining the battery of the two-wheeled electric vehicle, and is favorable for timely replacing the battery with abnormal attenuation, thereby providing guarantee for the driving safety of the two-wheeled electric vehicle.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 is a schematic representation of the overall flow of the study protocol of the present invention;

FIG. 2 is a diagram of an LSTM network architecture according to the present invention;

FIG. 3 is a graphical representation of the present invention and the source of raw data;

FIG. 4 is a graphical representation of the comparison of the predicted value to true value gap (data in a partial test set) in accordance with the present invention;

FIG. 5 is a graph showing the importance of the characteristics of 16 temperature measuring points according to the present invention;

fig. 6 is a flowchart illustrating a method for determining battery health based on charging segment data according to the present invention.

Detailed Description

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings, in order that the present disclosure may be more fully understood and fully conveyed to those skilled in the art. While the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the invention is not limited to the embodiments set forth herein.

The principle of the method for judging the health degree of the battery based on the charging segment data is as follows:

the experimental battery of the research level is extended to the real vehicle battery of the practical use level. The battery temperature is measured by building an external temperature measuring point, so that the fault judgment rate of thermal runaway can be obviously improved. For the power battery in the vehicle stage, under the condition of lacking battery production experience, the method for directly acquiring the internal temperature not only can cause uncertain damage to the battery, but also can influence the prediction result.

In view of the above, in the aspect of measuring the battery temperature, the present invention intends to adopt a method of performing multi-point temperature measurement on the surface of each electrical core, and to use the acquired data as a data source of the prediction model. Compared with the data volume obtained by adopting a single internal temperature measurement method, the data sample volume based on multipoint temperature measurement is richer, and the advantages of multipoint temperature measurement are also reflected. In the aspect of temperature characteristic sequence extraction, the invention extracts characteristic variables for predicting SOH from mass data. And simultaneously, extracting the multipoint temperature measurement data, and constructing characteristic variables as data bases.

Firstly, battery pack data of two-wheeled electric vehicles in two years are collected, temperature measurement is carried out from the outside of a battery, temperature information is obtained, a temperature gradient curve is generated, the temperature inside the battery is estimated through the generated temperature gradient curve, and the health state of the battery is judged. The SOH attenuation state of the battery is predicted by researching the temperature gradient change of the battery and an LSTM neural network method, and an SOH attenuation prediction model is constructed. The model can be used for optimizing and maintaining the battery of the two-wheeled electric vehicle, and is favorable for timely replacing the battery with abnormal attenuation, thereby providing guarantee for the driving safety of the two-wheeled electric vehicle. Compared with the laboratory environment, the method is more suitable for the actual situation, and has higher theoretical significance and application value.

The method for judging the health degree of the battery based on the charging fragment data comprises the following steps:

step S001, data acquisition;

acquiring temperature data before power conversion station data light, performing data acquisition on a plurality of temperature measurement points corresponding to each battery, wherein the used data is power conversion station optical fiber temperature measurement data, and the method comprises a temperature data acquisition stage, wherein the optical fiber temperature measurement points are arranged on a battery bin for power conversion: each battery compartment is provided with a plurality of optical fiber temperature measuring points which are respectively arranged at the center and two sides of the battery compartment, the central optical fiber temperature points are more, and the non-central optical fiber temperature points are more (boundaries with different degrees); therefore, the battery can acquire the multipoint temperature outside the battery in the whole charging process in the charging process of the battery bin;

step S002 data preprocessing;

the temperature data is subjected to preliminary processing: and carrying out data deduplication, null value processing and abnormal value processing on the temperature data acquired by the temperature measuring point, and carrying out temperature data conversion.

Step S003 feature engineering

Constructing difference sequence characteristics of the same temperature points, constructing difference sequence characteristics of different temperature points, and constructing a temperature increment sequence; the feature extraction mainly comprises two aspects of temperature feature sequence extraction and SOH sequence acquisition: in the aspect of temperature characteristic sequence extraction, the temperature change generated in the charging process of the real vehicle is selected in the charging process, and external multipoint temperature data are recorded, wherein the external multipoint temperature data mainly comprise the following three temperature characteristic sequences:

step S004 model construction;

selecting an algorithm and constructing a model according to various characteristics such as the difference sequence characteristics of the same temperature points constructed in the step S003; in the aspect of selection of a learning algorithm, fitting the extracted related feature sequences with a target SOH by a deep learning LSTM method or an integrated learning XGboost algorithm;

the method for selecting the long-short term memory neural network (LSTM) has excellent prediction accuracy on one hand, and can solve the problem of RNN gradient disappearance on the other hand, so that an innovative internal gate structure can be used, useful historical information is reserved, useless historical information is deleted, and the accuracy is greatly improved. Meanwhile, the method mainly adopts a mechanism of controlling the gate, and consists of a memory cell, an input gate, an output gate and a forgetting gate.

The method for predicting the SOH of the battery by adopting the LSTM neural network is characterized in that vehicle-mounted battery charging data of a new energy lithium battery automobile which is actually acquired for nearly two years are serialized, input features of the LSTM are extracted, part of data in the battery data are used as a verification set, indexes (such as maximum available capacity) which embody the SOH attenuation characteristics are used as output features of the LSTM, and the existing SOH condition of the battery is estimated by using a neural network method after training and verification. Therefore, a battery SOH attenuation prediction model which accords with the actual vehicle condition is constructed, the method can be suitable for estimating the SOH condition of the battery of the two-wheeled electric vehicle in the actual scene, the key problem that the estimation method is difficult to adapt to the complex working environment in the laboratory scene can be solved, and the method has important theoretical significance and application value.

Step S005, model testing and online;

and testing on the test set by using the optimal model, counting the test effect, and performing online on the model according to the test effect. The charging data of a plurality of batteries are calculated in the experiment, and the charging data are respectively calculated according to the training set: and (4) verification set: test set =7:2:1, the algorithm model obtains very good effects on a training set and a verification set, the effects on the test set are slightly lower than those of the training set and the verification set, but the difference is not large, and the effects are still good. But even if the effect on the test set is good, the final effect verification needs to be carried out by using a real vehicle. The charging data of a plurality of batteries and the temperature data collected by corresponding optical fibers, wherein the battery type is mainly lithium iron phosphate.

The method includes the steps of truly verifying real data of a plurality of real vehicles, and counting charging time span and charging times of batteries in detail for a plurality of vehicle numbers 1,2.. 5,n. According to the statistical real data of a plurality of real vehicles, the online actual capacity and the SOH of 6 vehicles are predicted, the absolute error of the rated capacity is analyzed, and the reliability of the model is further verified. In order to verify the effectiveness of the SOH prediction model established in the text, x new vehicles are selected for real vehicle testing. The temperature data of the battery is recorded by simulating the charge and discharge behaviors of an actual user, and the SOH of the battery is predicted by utilizing the model. And finally, comparing the actually measured SOC value of the battery with the SOH value predicted by the algorithm to obtain a relative error.

Example 1

Fig. 1 is a flowchart illustrating a method for determining battery health based on charging segment data according to an embodiment of the present invention. Wherein:

and S001, acquiring data, wherein the data is before the whole scheme, and the data acquisition stage mainly comprises a temperature data acquisition stage, and aims to acquire temperature data before data light of a power station is changed and acquire data of 16 temperature measurement points corresponding to each battery. The data used in the experiment of the invention is the optical fiber temperature measurement data of the power conversion station.

In the temperature data acquisition stage, optical fiber temperature measuring points are arranged on a battery chamber for battery replacement: each battery compartment has 16 optical fiber temperature measuring points which are respectively arranged at the center and two sides of the battery compartment, the central optical fiber temperature point is 4, and the non-central optical fiber temperature point is 12 (boundaries of different degrees). Therefore, the battery can acquire the multipoint temperature outside the battery in the whole charging process in the charging process of the battery bin. Because the optical fiber is adopted for temperature measurement, the temperature precision is higher (the precision reaches two digits after decimal point).

And step S002, preprocessing data, wherein the directly acquired temperature data cannot be directly used for training the model due to the problems of repetition, null value, abnormality and the like. Therefore, the following data preliminary processing procedure is carried out on the actual temperature data: and carrying out data duplicate removal, null value processing and abnormal value processing on the temperature data acquired by the temperature measuring points, and carrying out temperature data conversion.

Data deduplication: because of some instability in data upload, a small amount of duplicate data may result. For duplicate data, only one of them is retained.

Null value processing: in all data, half of the data are valid temperature values. But some nulls may be generated due to data upload or fiber optic problems. The filling is preferentially performed for these null values, that is, the filling is performed using the temperature value at the previous time or the temperature values at the other temperature measurement points at this time. If the temperature loss rate is large, the charging data is generally discarded.

Abnormal temperature numerical value processing: based on the temperature condition of a large amount of data under the condition of a box diagram method and a business background, abnormal high-temperature or low-temperature data is deleted.

Temperature data conversion: because the temperature data acquired by the optical fiber are all temperature data of different surfaces of the battery, the temperature data are not real battery temperature data, and the difference between the surface temperature and the actual temperature is different at different stages of charging. In contrast, the actual experiment performed offline obtains the conversion process from the optical fiber surface collection temperature to the actual battery temperature. The method comprises the following specific steps:

the experiment was carried out with the batteries selected for different cycles, on which thermistors were mounted as temperature sensors.

The following experiments were conducted on the basis of manually controlling the ambient temperature in the battery compartment to have different temperature gradients, which are currently [0 ℃ to 10 ℃ ], [10 ℃ to 20 ℃ ], [20 ℃ to 30 ℃ ], [30 ℃ to 40 ℃).

These cells were discharged offline (uniformly discharged to 5% soc) and then charged in the battery compartment (tests were performed for each temperature gradient).

And respectively counting temperature data of the optical fiber temperature measuring point and temperature data acquired by the temperature sensor in different SOC intervals and different charging time intervals.

And establishing a relation table among the temperature, the SOC interval, the charging time interval, the environment temperature interval and the temperature collected by the temperature sensor.

And establishing a regression model for the relation table, and obtaining the corresponding relation between each characteristic and the real temperature of the battery by using logistic regression. Note: the SOC selected interval is: [0% -10% ], [10% -20% ].

The interval of charging time selection is: [0 min-10 min ], [10 min-20 min ]., [60+ min ].

In the modeling process, the interval is directly converted into numerical values, and the numerical values are sequentially started from 1, such as the environmental temperature of 0-10 DEG C]、[10℃-20℃]、[20℃-30℃]、[30℃-40℃]Converted to 1,2,3,4 the resulting conversion equation is: note: t represents the true temperature of the battery, T _m Represents the mean value of the core temperature (8 th and 9 th mean values), T, of the temperature measuring point of the optical fiber _e Represents the ambient temperature, s represents the SOC interval, and t represents the charging time interval.

In the real SOH result calculation process, an ampere-hour integration method is also needed for data calculation, and after temperature data with considerable quality is obtained, the following method is adopted to calculate the capacity of each charging rush-in:

the formula:

wherein n represents the number of the data pieces charged at this Time, I represents the current magnitude in the charging process, and Time represents the data Time. The calibration is performed. The calibration method is mainly to standardize the actual capacity of different SOC intervals of each type of battery obtained by experiments in advance. The relative capacities of the different SOC intervals are shown in table 1, where the relative capacity refers to the ratio of the capacity of each interval to one-tenth of the rated capacity.

TABLE 1 relative capacities in different SOC intervals

SOC interval	Relative capacity
		0-10	Q1
10-20	Q2
		20-30	Q3
30-40	Q4
		40-50	Q5
50-60	Q6
		60-70	Q7
70-80	Q8
		80＝90	Q9
90-100	Q10

The formula after normalization is:

where Δ Q is the charge capacity calculated by ampere-hour integration, n is the SOC ten-digit part where charging is started (n =2 if charging is started at 25), and j is the ones-digit part where charging is started (j =5 if charging is started at 25).

1) And processing abnormal values of Q. Although the charging data of the power swapping station is relatively pure, a plurality of data exist and the overall deviation is relatively large, so that the abnormal data are deleted.

2) And performing moving average on the Q to obtain a smoother Q sequence.

3) The SOH sequence is obtained by converting SOH and dividing each value of the Q sequence by the starting Q.

Training model establishment, process based on SOH real result training

And S003, performing characteristic engineering, namely constructing difference sequence characteristics of the same temperature points, constructing difference sequence characteristics of different temperature points and constructing a temperature increment sequence. The feature extraction mainly comprises two aspects of temperature feature sequence extraction and SOH sequence acquisition: in the aspect of temperature characteristic sequence extraction, the temperature change generated in the charging process of the real vehicle is selected in the charging process, and external multipoint temperature data are recorded, wherein the external multipoint temperature data mainly comprise the following three temperature characteristic sequences:

1. and (3) extracting the temperature of a plurality of temperature measuring points of each monomer in the charging process, wherein the temperature extracting mainly comprises a temperature difference sequence, a rising speed sequence and an entropy sequence.

2. During the charging process, a difference sequence, a rising speed sequence and an entropy sequence of the average temperature between different monomers are extracted.

3. In the charging process, temperature increment sequences of unit SOC are extracted at all temperature measuring points of the monomers and the average temperature of different monomers respectively.

In the aspect of acquiring the SOH sequence of the battery health, firstly, selecting a traditional ampere-hour integration method for measuring the capacity of each charge (selecting a fixed voltage interval); removing abnormal values according to the copper beam charged each time; and finally, processing the capacity data by adopting a moving average method to obtain an SOH sequence. Taking a part of the SOH sequence as a model training set, and fitting the temperature-related characteristic sequence with a target SOH sequence; another portion of the SOH sequence is used as a validation set for the fitting method to validate the robustness and accuracy of the model.

After the characteristics are extracted, multipoint temperature measurement is carried out on the surface of each electric core in a data acquisition stage; and evaluating and correcting the constructed data based on the time unit: screening data with errors, and subsequently correcting the error data, wherein the method comprises the steps of storing an average value of a missing value and an abnormal value of the missing value and the abnormal value and performing subsequent calculation; for the value with the wrong time period, the time period is definitely acquired, and the data is adjusted and re-run; for values where the specification is wrong to calculate, the caliber is explicitly adjusted and the data sample quality is re-improved.

The extracted temperature characteristic sequence and the target sequence of the SOH are matched in a fitting manner by a deep learning method, so that the direct or indirect relation contained in the sequence is found.

And S004, constructing a model, and selecting an algorithm and constructing the model according to various characteristics such as the difference sequence characteristics of the same temperature points constructed in the step S003. In the aspect of selection of a learning algorithm, the invention fits the extracted related characteristic sequence with a target SOH through an LSTM method of deep learning or an integrated learning XGboost algorithm.

The method for selecting the long-short term memory neural network (LSTM) has excellent prediction accuracy on one hand, and can solve the problem of RNN gradient disappearance on the other hand, so that an innovative internal gate structure can be used, useful historical information is reserved, useless historical information is deleted, and the accuracy is greatly improved. Meanwhile, the method mainly adopts a mechanism of controlling the gate and consists of a memory cell, an input gate, an output gate and a forgetting gate. Fig. 1 is a network architecture diagram of the LSTM.

The method adopts the method for predicting the SOH of the battery of the LSTM neural network, carries out serialization processing on vehicle-mounted battery charging data of a new energy lithium battery automobile which is actually acquired for nearly two years, extracts the input characteristics of the LSTM, uses part of data in the battery data as a verification set, uses indexes (such as maximum available capacity and the like) which embody the SOH attenuation characteristics as the output characteristics of the LSTM, and estimates the existing SOH condition of the battery by using the neural network method after training verification. Therefore, a battery SOH attenuation prediction model which accords with the actual vehicle condition is constructed, the method can be suitable for estimating the SOH condition of the battery of the two-wheeled electric vehicle in the actual scene, the key problem that the estimation method is difficult to adapt to the complex working environment in the laboratory scene can be solved, and the method has important theoretical significance and application value.

The method selects the battery surface temperature gradient sequence data in each charging process as training data, wherein the temperature gradient is data after the influence of the environmental temperature is removed.

Selecting a charging SOC interval: the amount of charging data per battery is different, but the same length of sequence data needs to be used, so temperature gradient data of the charging process with SOC between 35% and 100% is extracted each time. 35 is selected herein as the initial SOC value because, through statistics of the distribution of the initial SOC of each battery charge by the charging station, studies have found that the percentage of batteries charged before 35SOC reaches over 95%, while the percentage charged before 30SOC is over 87%. Therefore, selecting 35 as the initial SOC value can ensure that not only most of the charging conditions are covered, but also the charging SOC interval is sufficiently large. The end SOC =100 is because the battery of the charging station is charged to SOC =100 every time it is charged.

Calculating the temperature gradient: in the 35SOC-100SOC range, temperature data is recorded every other SOC from the initial temperature, so that the final temperature gradient data is 66. There are 16 temperature measurement points, so the temperature gradient data of the final charge is 66 × 16 matrix data.

Converting real temperature gradient data: the average of the temperature data at the 8 th and 9 th temperature monitoring points is used, plus the current ambient temperature, charge SOC, charge time. The temperature transformation was performed by equation (1) to obtain the true temperature gradient data (66 x 1 sequence data).

Data set: the data set is temperature gradient data for each charge 66 x 16 and corresponding calculated battery SOH data.

Constructing a model: because the data is temperature gradient data, the gradient is based on the change of a time sequence and is also a multi-dimensional characteristic (16 temperature points), an LSTM model is selected as a training model.

And constructing a battery temperature gradient sequence through the sequence data of the battery temperature, feeding back the attenuation state of the target SOH from the gradient information and the temperature correlation characteristic sequence, and mining the corresponding relation between the temperature and the SOH. After the algorithm training and verification are finished, the SOH attenuation condition can be predicted through the temperature sequence characteristics in each charging process. The general scheme flow of the modeling of the invention is shown in FIG. 2.

And S005, testing and online of the model, performing testing on the test set by using the optimal model, counting the test effect, and performing online of the model according to the test effect. The experiment of the invention calculates 82216 times charging data of 5600 batteries, and the charging data are respectively calculated according to a training set: and (3) verification set: test set =7:2:1, the algorithm model obtains very good effects on a training set and a verification set, the effects on the test set are slightly lower than those of the training set and the verification set, but the difference is not large, and the effects are still good. But even if the effect on the test set is good, the final effect verification needs to be carried out by using a real vehicle. The charging data of 5600 batteries and the corresponding optical fiber acquisition temperature data, wherein the battery type is mainly lithium iron phosphate. The temperature data collected by the fiber is shown in fig. 3.

The real data of 6 real vehicles are truly verified, and the statistical table is shown in table 2. The charging time span and the charging times of the battery are counted in detail for 6 vehicles with the number 1,2.. 5,6. According to the statistical real data of 6 real vehicles, the offline actual capacity and the SOH of the 6 vehicles are predicted, the absolute error of the rated capacity is analyzed, and the data verification result is shown in the table 3, so that the reliability of the model is further verified. To verify the validity of the SOH prediction model established herein, 3 new vehicles were selected for real vehicle testing, with data distribution as shown in table 4. The temperature data of the battery is recorded by simulating the charge and discharge behaviors of an actual user, and the SOH of the battery is predicted by utilizing the model. And finally, comparing the actually measured SOC value of the battery with the SOH value predicted by the algorithm to obtain a relative error. The results of the experiment are shown in table 5, note: the rated capacity of the test cell in table 5 was 70.5AH. The error represented on the three latest battery data of the algorithm model is an average error larger than the effect on the test set, and after all, the error is a new battery which may have various dimensions different from the previous battery, so that the effect predicted by the algorithm can be truly represented by the effect tested on the latest battery. From the prediction effect of the three batteries at present, the errors are all less than 3%, and the errors of two batteries are all less than 1%, which is enough to say that the current algorithm model has better generalization capability and higher prediction accuracy.

TABLE 2 basic cases of the validation data

Vehicle number	Type of battery	Time span	Region of origin	Number of charges
					01	Lithium iron phosphate	3 months in 2017-11 months in 2021	Shanghai province	1960
02	Lithium iron phosphate	3 months in 2017-11 months in 2021	Shanghai province	1525
					03	Lithium iron phosphate	3 months in 2017-11 months in 2021	Shanghai province	1667
04	Lithium iron phosphate	5 months in 2019 to 11 months in 2021	Shanghai province	986
					05	Lithium iron phosphate	5 months in 2019 to 11 months in 2021	Shanghai province	1139

Table 3 data verification results

TABLE 4 data set distribution

Data set	Data volume	Average relative error
			Training set	57551	0.176％
Verification set	16443	0.179％
			Test set	8222	0.223％

TABLE 5 real vehicle charging and discharging data

Vehicle number	Actual capacity under line	Algorithm prediction of SOH	Predicting SOH->Predicted capacity	Relative error
					01	65.6AH	93.30％	65.78AH	0.27％
02	58.9AH	85.64％	60.37AH	2.50％
					03	61.7AH	88.32％	62.27AH	0.92％

Since the actual training data volume is large and the average test index has been written in the above text, the comparison data of part of the training set and the test set is shown, as shown in fig. 4, the abscissa represents 20 sets of experimental data randomly selected in the test set, and the ordinate represents the comparison effect between the actual value and the predicted value of SOH.

As a plurality of external temperature measuring points are distributed on the surface of the battery, the temperature measuring points at the center of the battery are numbered as 8 and 9, the number of the temperature measuring points which extend from the center point to two sides to the edge area is 1-16, the ordinate in the figure 5 represents the number of the temperature measuring points, the ordinate represents the importance sequence of the characteristics of each temperature measuring point, the numerical value of the importance degree of the abscissa calculates the information gain value through information entropy modeling, and the quantitative value of the importance degree of the characteristics is extracted. The numerical value measured by the temperature measuring point closer to the center of the single battery is closer to the actual temperature of the center of the battery, so the overall characteristic ranking of the temperature measuring points is in a similar step-shaped distribution. Based on the data from the real vehicle testing (table 3), the following conclusions were made: (1) The difference between the actual battery capacity (SOH) of the three vehicles and the SOH value predicted by the algorithm is small, and the relative errors are within 3 percent (the prediction error of the vehicle No. 01 is even smaller than 0.3 percent). (2) As the battery capacity (SOH) decreases, the prediction error of the algorithm increases gradually, increasing from 0.27% to 2.5%.

In conclusion, the charging and discharging tests of the two-wheeled electric vehicle battery prove that the model can achieve high prediction precision, the error does not exceed 3%, and the effectiveness of judging the health degree of the battery through the charging fragment data of the two-wheeled electric vehicle is fully verified. In addition, it has also been found that the prediction accuracy of the model is gradually decreasing as the SOH of the battery decreases. But due to the limitation of experimental conditions, the algorithm verification at low SOH value is not performed (which is also the focus of the subsequent work).

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are clearly and completely described above with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are a part of the embodiments of the present invention, but not all of the embodiments. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.

Thus, the above detailed description of the embodiments of the invention presented in the drawings is not intended to limit the scope of the invention as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Claims

1. A method for judging the health degree of a battery based on charging segment data is characterized by comprising the following steps:

step S001, data acquisition;

step S002 data preprocessing;

the temperature data is subjected to preliminary processing: carrying out data duplicate removal, null value processing and abnormal value processing on temperature data acquired by a temperature measuring point, and carrying out temperature data conversion;

step S003 of characteristic engineering;

constructing difference sequence characteristics of the same temperature points, constructing difference sequence characteristics of different temperature points, and constructing a temperature increment sequence; the feature extraction mainly comprises two aspects of temperature feature sequence extraction and SOH sequence acquisition: in the aspect of temperature characteristic sequence extraction, the method comprises the following steps of selecting temperature change in the charging process of an actual vehicle, recording external multipoint temperature data, and mainly comprising the following three temperature characteristic sequences:

in the aspect of acquiring the SOH sequence of the battery health, firstly, selecting a traditional ampere-hour integration method for measuring the capacity of each charge (selecting a fixed voltage interval); then, removing abnormal values according to the copper beam charged each time; finally, processing the capacity data by adopting a moving average method to obtain an SOH sequence; taking a part of the SOH sequence as a model training set, and fitting the temperature-related characteristic sequence with a target SOH sequence; another part of the SOH sequence is used as a verification set of the fitting method, so that the robustness and the accuracy of the model are verified;

after the characteristics are extracted, multipoint temperature measurement is carried out on the surface of each electric core in a data acquisition stage; and the constructed data is evaluated and corrected based on the time unit: screening data with errors, and subsequently correcting the error data, wherein the method comprises the steps of storing an average value of a missing value and an abnormal value of the missing value and the abnormal value and performing subsequent calculation; for the value with the wrong time period, the time period is definitely acquired, and the data is adjusted and re-run; for the numerical value with the wrong calculation specification, the caliber is definitely adjusted and the quality of the data sample is improved again;

step S004 model construction;

selecting an algorithm and constructing a model according to various characteristics such as the difference sequence characteristics of the same temperature points constructed in the step S003; in the aspect of selection of a learning algorithm, the extracted relevant feature sequences are fitted with a target SOH through an LSTM method of deep learning or an integrated learning XGboost algorithm;

the Method (LSTM) for selecting the long-short term memory neural network has excellent prediction accuracy on one hand, and can solve the problem of RNN gradient disappearance on the other hand, so that an innovative internal gate structure can be used, useful historical information is reserved, useless historical information is deleted, and the accuracy is greatly improved; meanwhile, the method mainly adopts a mechanism of a control gate, and consists of a memory cell, an input gate, an output gate and a forgetting gate;

the method for predicting the SOH of the battery by adopting the LSTM neural network is characterized in that vehicle-mounted battery charging data of a new energy lithium battery automobile which is actually acquired for nearly two years are subjected to serialization processing, input characteristics of the LSTM are extracted, part of data in the battery data is used as a verification set, indexes (such as maximum available capacity and the like) which embody SOH attenuation characteristics are used as output characteristics of the LSTM, and the existing SOH condition of the battery is estimated by utilizing a neural network method after training verification, so that a battery SOH attenuation prediction model which accords with actual vehicle conditions is constructed;

step S005, testing and online of the model;

testing on the test set by using the optimal model, counting the test effect, and performing model online according to the test effect; the charging data of a plurality of batteries are calculated in the experiment, and the charging data are respectively calculated according to the training set: and (4) verification set: test set =7:2:1, the algorithm training and parameter adjustment are carried out, the algorithm model obtains very good effects on a training set and a verification set, the effect on a test set is slightly lower than that of the training set and the verification set, but the difference is not large, and the effect is still good; but even if the effect on the test set is good, the final effect verification needs to be carried out by using a real vehicle; charging data of a plurality of batteries and temperature data collected by corresponding optical fibers, wherein the type of the batteries is mainly lithium iron phosphate;

the method includes the steps that real data of a plurality of real vehicles are truly verified, and detailed statistics is conducted on charging time span and charging times of batteries for a plurality of vehicle numbers 1,2.. 5,n; predicting the offline actual capacity and SOH of 6 vehicles according to the statistical real data of a plurality of real vehicles, analyzing the absolute error of rated capacity, and further verifying the reliability of the model; selecting new x vehicles to carry out real vehicle test in order to verify the effectiveness of the SOH prediction model established in the text; recording temperature data of the battery by simulating the charge and discharge behaviors of an actual user, and predicting the SOH of the battery by using the model; and finally, comparing the actually measured SOC value of the battery with the SOH value predicted by the algorithm to obtain a relative error.

2. The method for determining the health of a battery based on charging segment data according to claim 1, wherein: step S002 data preprocessing:

data deduplication: for duplicate data, only one of the data is reserved;

null value processing: in all data, half of the data are valid temperature values; but some null values can be generated due to data uploading or problems of the optical fiber; for the null values, filling is preferentially carried out, namely the filling is carried out by using the temperature value at the previous moment or the temperature values of other temperature measuring points at the current moment; if the temperature loss rate is large, the charging data is generally discarded;

abnormal temperature numerical value processing: based on the temperature condition of a large amount of data under the condition of a box diagram method and a business background, deleting abnormal high-temperature or low-temperature data;

temperature data conversion: because the temperature data acquired by the optical fiber are temperature data of different surfaces of the battery, the temperature data are not real battery temperature data, and the difference between the surface temperature and the actual temperature is different at different stages of charging; in contrast, the actual experiment performed offline obtains the conversion process from the collection temperature on the surface of the optical fiber to the actual temperature of the battery.

3. The method for determining the health of a battery based on charging segment data according to claim 1, wherein: step S002 data preprocessing:

the conversion process is specifically as follows:

selecting batteries with different periods to carry out the experiment, and mounting a thermistor on the batteries as a temperature sensor;

manually controlling the environmental temperature in the battery compartment to be different temperature gradients, [0 ℃ -10 ℃), [10 ℃ -20 ℃), [20 ℃ -30 ℃), and [30 ℃ -40 ℃), and carrying out experiments on the basis:

offline uniform discharge to 5% SOC, and then in-bin charging, at each temperature gradient, test was performed;

respectively counting temperature data of an optical fiber temperature measuring point and temperature data acquired by a temperature sensor in different SOC intervals and different charging time intervals;

establishing a relation table among the temperature, the SOC interval, the charging time interval, the environment temperature interval and the temperature collected by the temperature sensor;

establishing a regression model for the relation table, and obtaining a corresponding relation between each characteristic and the real temperature of the battery by using logistic regression; note: the SOC selected interval is: 0% -10%, [10% -20% ]., [90% -100% ];

the interval of charging time selection is: 1., [60+ min ];

in the modeling process, the interval is directly converted into numerical values, and the numerical values are sequentially converted into 1,2,3,4 from 1, such as the ambient temperature [ 0-10 ℃) ], [ 10-20 ℃) ], [ 20-30 ℃) and [ 30-40 ℃; the final conversion formula is:

T＝T _m +0.173*T _e -0.036*s-0.021*t+0.83 (1)

the formula:

note: Δ Q is the charge capacity calculated by ampere-hour integration, T represents the true temperature of the battery, T _m Represents the mean value of the core temperature of the optical fiber temperature measuring point, T _e Represents an ambient temperature, s represents an SOC interval, and t represents a charging time interval; wherein n represents the number of the data in the current charging, I represents the current in the charging process, and Time represents the data Time; carrying out calibration; the calibration method is mainly characterized in that the actual capacity of different SOC intervals of each type of battery obtained through experiments in advance is used as a standardMelting;

the formula after normalization is:

wherein Δ Q is a charging capacity calculated by ampere-hour integration, n is a SOC ten-digit part at which charging is started, and j is a ones-digit part at which charging is started;

1) Abnormal value processing of Q; although the charging data of the battery replacement station is pure, a plurality of data exist and the integral deviation is large, so that the abnormal data are deleted;

2) Performing sliding average on Q to obtain a smoother Q sequence;

3) Converting SOH, and dividing each value of the Q sequence by the initial Q to obtain an SOH sequence;

and (4) establishing a training model, and training based on the SOH real result.

4. The method of determining the health of a battery based on charging segment data according to claim 1, wherein: step S003 is characterized by:

temperature characteristic sequence 1, temperature extraction is carried out on a plurality of temperature measuring points of each monomer in the charging process, and mainly a difference sequence, a rising speed sequence and an entropy sequence of the extracted temperature are obtained;

a temperature characteristic sequence 2, in the charging process, extracting a difference sequence, a rising speed sequence and an entropy sequence of average temperatures among different monomers;

and 3, extracting temperature increment sequences of unit SOC at all temperature measuring points of the monomers and the average temperature of different monomer individuals respectively in the charging process.

5. The method for determining the health of a battery based on charging segment data according to claim 1, wherein: step S004:

selecting a charging SOC interval: selecting temperature gradient data of a charging process with SOC between 35% and 100% each time;

calculating the temperature gradient: in a certain temperature interval, starting from the initial temperature, the temperature data is recorded every other SOC, so that the final temperature gradient data is n; a total of m temperature measuring points are arranged, so that the temperature gradient data of the final charging is matrix data of n x m;

converting real temperature gradient data: adding the current ambient temperature, the charging SOC and the charging time to the average value of the temperature data of the Xth temperature monitoring point and the Yth temperature monitoring point; carrying out temperature conversion through a formula (1) to obtain real temperature gradient data;

data set: the data set is temperature gradient data of n × m in each charging and corresponding calculated SOH data of the battery;

constructing a model: selecting an LSTM model as a training model;

constructing a battery temperature gradient sequence through the sequence data of the battery temperature, feeding back the attenuation state of the target SOH from the gradient information and the temperature correlation characteristic sequence, and mining the corresponding relation between the temperature and the SOH; after the algorithm training and verification are finished, the SOH attenuation condition can be predicted through the temperature sequence characteristics in each charging process.