CN114880939A - Intelligent prediction method and device for service life of power battery - Google Patents

Intelligent prediction method and device for service life of power battery Download PDF

Info

Publication number
CN114880939A
CN114880939A CN202210559862.2A CN202210559862A CN114880939A CN 114880939 A CN114880939 A CN 114880939A CN 202210559862 A CN202210559862 A CN 202210559862A CN 114880939 A CN114880939 A CN 114880939A
Authority
CN
China
Prior art keywords
data
battery
fitness
value
service life
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202210559862.2A
Other languages
Chinese (zh)
Inventor
贾小娥
李佳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Telephase Technology Development Beijing Co ltd
Original Assignee
Telephase Technology Development Beijing Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Telephase Technology Development Beijing Co ltd filed Critical Telephase Technology Development Beijing Co ltd
Priority to CN202210559862.2A priority Critical patent/CN114880939A/en
Publication of CN114880939A publication Critical patent/CN114880939A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/086—Learning methods using evolutionary algorithms, e.g. genetic algorithms or genetic programming
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
    • G06F2119/04—Ageing analysis or optimisation against ageing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/60—Other road transportation technologies with climate change mitigation effect
    • Y02T10/70—Energy storage systems for electromobility, e.g. batteries

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Medical Informatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Physiology (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides an intelligent prediction method and device for the service life of a power battery. The method comprises the following steps: acquiring vehicle state data and vehicle-mounted battery pack data from an electric vehicle operation monitoring center; integrating the data to obtain various characteristic data influencing the service life of the battery; screening the characteristic data based on the correlation to obtain several characteristic data which have the most obvious influence on the service life of the battery; and (4) constructing a neural network prediction model by taking the screened characteristic data as input and the battery life as output, and predicting the battery life by using the trained prediction model. According to the invention, the acquired vehicle state data and the vehicle-mounted battery pack data are integrated to obtain various characteristic data which affect the service life of the battery, and the characteristic data are screened based on the correlation to determine the input variable of the prediction model, so that the prediction precision of the prediction model on the service life of the battery is improved.

Description

Intelligent prediction method and device for service life of power battery
Technical Field
The invention belongs to the technical field of power battery service life prediction, and particularly relates to an intelligent power battery service life prediction method and device.
Background
The lithium battery has the best performance in the power battery. With the development of energy industries of all countries in the world, the utilization rate of lithium batteries is also rapidly increased, and the prediction of Remaining Life (RUL) of lithium batteries has attracted extensive attention of scholars at home and abroad. The time that a lithium battery spends from a currently set moment to its failure moment is called the remaining service life of the lithium battery, which is usually characterized by capacity, internal resistance or other state quantities. The power battery of the electric vehicle is a typical example, when the capacity of the power battery of the electric vehicle is attenuated to be below 70-80% of the rated capacity, the power battery is already failed, and the capacity value at the current moment is the failure threshold value of the power battery.
At present, the remaining life prediction methods of lithium batteries mainly include two methods: model-based methods and data-driven based methods. The model-based residual life prediction method is that a physical model is established according to the performance, the working characteristics and the working process of equipment, sample data is input into the model, and prediction data is obtained through a conversion algorithm of an input-output relation. Generally, the prediction method mainly uses a failure mechanism model to predict the RUL of the lithium ion battery. Based on data driving, namely in the prediction process, only historical data of equipment operation needs to be known, the data are researched and analyzed by using an algorithm, useful information is provided and used as input, and model prediction based on the data is realized. Currently, the following methods are generally used in the prediction method of the lithium battery RUL: support vector machine method, particle filter method, autoregressive model method, artificial neural network method. In recent years, the artificial neural network method has been more and more widely used. The artificial neural network method is mainly characterized in that historical data are used for training a constructed network model, and specific data of a battery to be predicted are input into the trained model, so that the residual life of the battery can be obtained. However, because of many factors affecting the battery life, it is difficult to select the factor having the most significant effect on the battery life from the many factors as the input variable of the model, so that the prediction accuracy of the model cannot be ensured, and the prediction result is not ideal.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an intelligent power battery life prediction method and device.
In order to achieve the above object, the present invention adopts the following technical solutions.
In a first aspect, the invention provides an intelligent prediction method for the service life of a power battery, which comprises the following steps:
acquiring vehicle state data and vehicle-mounted battery pack data from an electric vehicle operation monitoring center;
integrating the data to obtain various characteristic data influencing the service life of the battery;
screening the characteristic data based on the correlation to obtain several characteristic data which have the most obvious influence on the service life of the battery;
and (4) constructing a neural network prediction model by taking the screened characteristic data as input and the battery life as output, and predicting the battery life by using the trained prediction model.
Further, the vehicle state data and the on-vehicle battery pack data include: the system comprises a vehicle number, a total voltage and a total current of a battery pack, an SOC, a highest voltage and a lowest voltage of a single battery, a forward accumulated electric quantity, a reverse accumulated electric quantity, a total mileage, an operation time and a vehicle speed.
Further, the method further comprises a data preprocessing step of: and eliminating abnormal data, supplementing missing data and distinguishing charge and discharge current data.
Further, the method for performing integrated processing on the data comprises the following steps: and taking different statistics of the same data as new feature data, and taking any combination of the different statistics of the same data as the new feature data, wherein the statistics comprise a maximum value, a minimum value, a median value, a mean value, a variance and a mean square error.
Further, the method for screening the feature data based on the correlation comprises the following steps:
calculating a correlation coefficient of any one feature data and the service life of the battery, and sequencing the feature data according to the sequence of the correlation coefficients from large to small;
deleting the characteristic data with the correlation coefficient smaller than a first threshold value;
and calculating a correlation coefficient between any two feature data in the remaining feature data, and deleting the feature data which is ranked later for the two feature data with the correlation coefficient being greater than a second threshold value, so that the correlation coefficient between any two feature data is not greater than the second threshold value.
Further, a radial basis function neural network is adopted to construct a battery life prediction model.
Further, an optimization algorithm is adopted to optimize the parameters of the radial basis function neural network.
Furthermore, the parameters of the radial basis function neural network are optimized by adopting a genetic algorithm, and the method comprises the following steps:
data normalization: determining a training sample and carrying out normalization processing on sample data;
chromosomal coding: data center c for neural network of radial basis function using real number coding i Spreading constant delta i And a weight w i Uniformly coding to form a chromosome string;
population initialization: the value range of the optimal initial population scale is 20-70;
constructing a fitness function: the fitness function is expressed as:
Figure BDA0003656106530000031
in the formula, F i The value of the fitness function is used as the fitness function value,
Figure BDA0003656106530000032
to predict value, y i Is the actual value, n is the total number of sample inputs;
selecting and operating: selecting individuals with high fitness in the population by adopting a roulette method, distributing the fitness of each individual in the population to a roulette plate, wherein the fitness value is proportional to the area occupied in the roulette plate, and selecting the individuals by rotating a pointer of the roulette plate; the probability of the ith individual being selected is:
Figure BDA0003656106530000033
in the formula, P i Probability of being selected for the ith individual, F i The fitness of the ith individual is shown, and N is the number of individuals in the population;
and (3) cross operation: performing cross operation by using an arithmetic cross method to obtain an individual g i 、g f The chromosomes of the cell are combined and crossed to form a new individual
Figure BDA0003656106530000034
The gene combination cross calculation formula is as follows:
Figure BDA0003656106530000035
Figure BDA0003656106530000036
in the formula, alpha is a scale factor, and alpha belongs to (0, 1);
cross probability P c Self-adaptive adjustment, the calculation formula is:
Figure BDA0003656106530000041
in the formula (f) max Is the maximum value of fitness in the population, f avg Is the average value of fitness in the population, f' is the fitness value of individuals in the cross, P C1 ＝0.9，P C2 ＝0.6；
And (3) mutation operation: randomly selecting individuals in a population, and selecting certain position genes of chromosomes of the individuals with a certain probability P m Self-adapting, thereby forming a new individual; probability P m The calculation formula of the self-adaptive adjustment is as follows:
Figure BDA0003656106530000042
wherein f is the fitness value of the variant individual, P m1 ＝0.1，P m2 ＝0.01；
If the preset convergence condition is met or the maximum iteration number is exceeded, stopping; otherwise, the transchromosome encoding step is repeated to perform an iterative process.
In a second aspect, the present invention provides an intelligent prediction apparatus for power battery life, including:
the data acquisition module is used for acquiring vehicle state data and vehicle-mounted battery pack data from the electric vehicle operation monitoring center;
the data integration module is used for integrating and processing the data to obtain various characteristic data influencing the service life of the battery;
the data screening module is used for screening the characteristic data based on the correlation to obtain several kinds of characteristic data which have the most obvious influence on the service life of the battery;
and the service life prediction module is used for constructing a neural network prediction model by taking the screened characteristic data as input and the battery service life as output, and predicting the battery service life by using the trained prediction model.
Further, the device also comprises a parameter optimization module for optimizing the parameters of the radial basis function neural network by adopting a genetic algorithm, wherein the method comprises the following steps:
data normalization: determining a training sample and carrying out normalization processing on sample data;
chromosomal coding: data center c for neural network of radial basis function using real number coding i Spreading constant delta i And a weight w i Uniformly coding to form a chromosome string;
population initialization: the value range of the optimal initial population scale is 20-70;
constructing a fitness function: the fitness function is expressed as:
Figure BDA0003656106530000051
in the formula, F i Is suitable forThe value of the function of the degree of response,
Figure BDA0003656106530000052
to predict value, y i Is the actual value, n is the total number of sample inputs;
selecting and operating: selecting individuals with high fitness in the population by adopting a roulette method, distributing the fitness of each individual in the population to a roulette plate, wherein the fitness value is proportional to the area occupied in the roulette plate, and selecting the individuals by rotating a pointer of the roulette plate; the probability of the ith individual being selected is:
Figure BDA0003656106530000053
in the formula, P i Probability of being selected for the ith individual, F i The fitness of the ith individual is shown, and N is the number of individuals in the population;
and (3) cross operation: performing cross operation by using an arithmetic cross method to obtain an individual g i 、g f The chromosomes of the cell are combined and crossed to form a new individual
Figure BDA0003656106530000054
The gene combination cross calculation formula is:
Figure BDA0003656106530000055
Figure BDA0003656106530000056
in the formula, alpha is a scale factor, and alpha belongs to (0, 1);
cross probability P c Self-adaptive adjustment, the calculation formula is:
Figure BDA0003656106530000057
in the formula (f) max Is the maximum fitness in the populationValue f avg Is the average value of fitness in the population, f' is the fitness value of individuals in the cross, P C1 ＝0.9，P C2 ＝0.6；
And (3) mutation operation: randomly selecting individuals in a population, and selecting certain position genes of chromosomes of the individuals with a certain probability P m Self-adapting, thereby forming a new individual; probability P m The calculation formula of the self-adaptive adjustment is as follows:
Figure BDA0003656106530000061
wherein f is the fitness value of the variant individual, P m1 ＝0.1，P m2 ＝0.01；
If the preset convergence condition is met or the maximum iteration number is exceeded, stopping; otherwise, the transchromosome encoding step is repeated to perform an iterative process.
Compared with the prior art, the invention has the following beneficial effects.
According to the invention, vehicle state data and vehicle-mounted battery pack data are obtained from an electric vehicle operation monitoring center, the data are integrated to obtain various characteristic data which affect the battery life, the characteristic data are screened based on the correlation to obtain several characteristic data which affect the battery life most obviously, a neural network prediction model is constructed by taking the screened characteristic data as input and the battery life as output, and the trained prediction model is used for predicting the battery life, so that the intelligent prediction of the battery life is realized. According to the invention, the acquired vehicle state data and the vehicle-mounted battery pack data are integrated to obtain various characteristic data which affect the service life of the battery, and the characteristic data are screened based on the correlation to determine the input variable of the prediction model, so that the prediction precision of the prediction model on the service life of the battery is improved.
Drawings
Fig. 1 is a flowchart of an intelligent method for predicting the lifetime of a power battery according to an embodiment of the present invention.
Fig. 2 is a schematic view of an electric vehicle operation monitoring center.
FIG. 3 is a flow chart of a genetic algorithm.
Fig. 4 is a block diagram of an intelligent power battery life prediction apparatus according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention clearer and more obvious, the present invention is further described below with reference to the accompanying drawings and the detailed description. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope 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.
Fig. 1 is a flowchart of an intelligent prediction method for the service life of a power battery according to an embodiment of the present invention, including the following steps:
step 101, obtaining vehicle state data and vehicle-mounted battery pack data from an electric vehicle operation monitoring center;
102, integrating the data to obtain various characteristic data influencing the service life of the battery;
103, screening the characteristic data based on the correlation to obtain several types of characteristic data which have the most obvious influence on the service life of the battery;
and 104, taking the screened characteristic data as input and the battery life as output, constructing a neural network prediction model, and predicting the battery life by using the trained prediction model.
In this embodiment, step 101 is mainly used for data acquisition. The embodiment acquires data from an electric vehicle operation monitoring center. The vehicle-mounted information acquisition terminal installed on the electric vehicle captures vehicle state data and vehicle-mounted battery pack information data in real time through CAN bus data on the electric vehicle, and sends the data to the monitoring center through a wireless transmission network, as shown in FIG. 2. The monitoring center realizes the unified management of basic information related to services such as vehicles, drivers, equipment and the like through the communication with the vehicle-mounted terminal; the vehicle-mounted terminal senses information such as the position, the vehicle condition and the battery pack of the vehicle and monitors the running safety of the vehicle in real time. Therefore, the monitoring center stores a large amount of power battery data generated when the electric automobile runs. The battery data includes a total battery voltage, a total battery current, a cell voltage, and the like. It should be noted that the parameters acquired in the present embodiment include not only the battery pack data but also the vehicle running state data because the battery life is related to not only the parameters of the battery pack itself but also the vehicle state data, such as the vehicle speed.
In this embodiment, step 102 is mainly used to integrate the data to obtain various feature data that affect the battery life. In the embodiment, the battery life prediction is realized by constructing a battery life prediction network, and the prediction accuracy of the prediction model is closely related to the selection of the input variables, so that candidate input variables as many as possible are determined first, and then the most effective input variables are screened from the candidate input variables. In the prior art, characteristic factors which have large influence on the service life of the battery are generally selected intuitively, or only quantities which can be directly measured or conveniently obtained, such as the voltage and the current of the battery, are noticed, and the quantities are directly used as input variables. Practice has shown that it is sometimes more efficient to use the statistics of a certain feature as an input variable than to use the feature itself directly as an input variable, such as a maximum, an average, etc. Further studies have shown that sometimes it is more efficient to use a combination of multiple statistics of the same feature as an input variable than to use a single statistic as an input variable, such as "maximum minus average", "minimum plus average plus median", etc. Therefore, in order to obtain the most effective input variables, the present embodiment first integrates the acquired feature data to obtain as many candidate variables as possible.
In this embodiment, step 103 is mainly used for feature data screening. Because many candidate input variables are obtained in the last step, all the candidate input variables cannot be used as the input variables of the model, if the feature data which has little influence on the battery life is used as the input variables, the model structure becomes very complicated, and the prediction precision is reduced, so that feature data screening is needed, a large amount of feature data which has little influence are deleted, and only a few feature data which has obvious influence are reserved. The present embodiment judges the degree of influence of various feature data based on the correlation between the various feature data and the battery life. The correlation magnitude can be expressed by a correlation coefficient, and the larger the absolute value of the correlation coefficient, the higher the degree of correlation. When the correlation coefficient is positive, the correlation is positive, that is, the larger the input is, the larger the output is; when the correlation coefficient is negative, the correlation is negative, i.e., the larger the input, the smaller the output. The absolute value of the correlation coefficient and the influence degree are in a relation of: 0.8-1.0 is extremely strong correlation, 0.6-0.8 is strong correlation, 0.4-0.6 is moderate correlation, 0.2-0.4 is weak correlation, and 0.0-0.2 is extremely weak correlation or no correlation. The following example will give a specific screening protocol.
In this embodiment, step 104 is mainly used to predict the battery life. In the embodiment, the service life of the battery is predicted by constructing the neural network prediction model and utilizing the trained prediction model. The input of the model is the input variable obtained after the last screening, and the output is the service life of the battery. The neural network structure that can be used as a prediction model is many, for example, the most common BP (Back Propagation) neural network is a Back Propagation neural network, and the weight and the threshold of the network are continuously corrected through error Back Propagation during model training, so that the error is reduced and the prediction accuracy is improved. The BP neural network has high nonlinear mapping capability and strong generalization capability, and is widely applied to the fields of intelligent control, data processing, pattern recognition, prediction estimation and the like.
As an alternative embodiment, the vehicle state data and the vehicle-mounted battery pack data include: the system comprises a vehicle number, a total voltage and a total current of a battery pack, an SOC, a highest voltage and a lowest voltage of a single battery, a forward accumulated electric quantity, a reverse accumulated electric quantity, a total mileage, an operation time and a vehicle speed.
This example presents several collected data. As mentioned above, the required acquisition data can be obtained from the monitoring center, and the data is generally data that can be obtained directly from the monitoring center or obtained through simple statistical calculation. The collected data of the embodiment mainly comprises the total voltage and the total current of the battery pack, the SOC, the highest voltage and the lowest voltage of the battery unit and the like, the vehicle number, the running time, the vehicle speed and the like.
As an optional embodiment, the method further comprises a data preprocessing step: and eliminating abnormal data, supplementing missing data and distinguishing charge and discharge current data.
In order to improve the prediction accuracy, data needs to be preprocessed. The data preprocessing step of the implementation comprises the steps of eliminating abnormal data, supplementing missing data and distinguishing charge and discharge current data. In the actual data acquisition process, the vehicle state information and the power battery information are influenced by weather, building density, stability of a data acquisition end and interference signals, and abnormal data can be generated in the signal transmission process, so that the abnormal data need to be identified and eliminated. Data loss is inevitably caused in the data acquisition, transmission and storage processes. Data missing can reduce the accuracy of a data mining model, rules implicit in data are difficult to mine, inaccurate output is caused, and meanwhile, data incompleteness can be caused when abnormal data are removed. Therefore, the missing data must be supplemented by interpolation processing so that the data of each time period is as complete as possible. The charge and discharge current of a battery is an important parameter that affects the life of the battery, and the charge current and the discharge current have different effects and therefore must be distinguished. The most obvious difference between the charging current and the discharging current is that the directions of the charging current and the discharging current are different relative to the positive (negative) pole of the battery, the charging current flows into the positive pole of the battery, and the discharging current flows out of the positive pole of the battery. Therefore, two different currents can be distinguished, and when data is stored, the charging current is generally set to be a negative value, and the discharging current is generally set to be a positive value.
As an optional embodiment, the method for performing integrated processing on the data includes: and taking different statistics of the same data as new feature data, and taking any combination of the different statistics of the same data as the new feature data, wherein the statistics comprise a maximum value, a minimum value, a median value, a mean value, a variance and a mean square error.
The embodiment provides a technical scheme for integrating and processing the collected data. As described above, the feature data is integrated to obtain the new feature data that has the most effective influence on the battery life, so that the most effective prediction model can be established. The integration method provided by the embodiment is performed based on the statistic of a feature data, such as the maximum value, the minimum value, the variance, and the like. In order to obtain as many and new feature data as possible, these statistics may be used as new feature data, or any combination thereof may be used as new feature data. The combination may be various arithmetic operations between any two or more statistics, or may be a concatenation, or the like.
As an optional embodiment, the method for screening the feature data based on the correlation includes:
calculating a correlation coefficient of any one feature data and the service life of the battery, and sequencing the feature data according to the sequence of the correlation coefficients from large to small;
deleting the characteristic data with the correlation coefficient smaller than a first threshold value;
and calculating a correlation coefficient between any two feature data in the remaining feature data, and deleting the feature data which is ranked later for the two feature data with the correlation coefficient being greater than a second threshold value, so that the correlation coefficient between any two feature data is not greater than the second threshold value.
This embodiment provides a technical solution for screening the feature data based on correlation. As previously mentioned, feature screening is performed to determine the most efficient input variables for the predictive model. The most efficient input variables would be the several characteristic data that are most strongly correlated with battery life. Therefore, the correlation coefficient of each feature data with the battery life is first calculated and sorted in order from large to small. Then, the characteristic data with the correlation coefficient smaller than the first threshold value is deleted, and the characteristic data with the strongest correlation (ranked in the top) with the service life of the battery is obtained. However, the number of feature data obtained after such screening may not be the minimum, and some of the feature data may have strong correlation, and the effect of these strongly correlated feature data on the battery life is approximately the same or repeated, and if they are all used as input variables, the effect (complexity and accuracy) of the model is also affected, only one of the strongly correlated features should be retained. Therefore, the correlation coefficient between any two of the feature data obtained after the first screening is calculated, and the feature data ranked next in the two features with the correlation coefficient larger than the second threshold value is deleted.
As an alternative embodiment, a radial basis function neural network is used to construct a battery life prediction model.
The embodiment provides a neural network structure of a battery life prediction model. The prediction model of this embodiment adopts a Radial Basis Function (RBF) structure. The RBF neural network is a feed-forward neural network proposed by J.Moody and C.Darken in the 80 th century, is a function approximation-based neural network, has the characteristics of high convergence speed, simple structure, strong nonlinear approximation capability and the like, can approximate any function with any precision, is widely concerned by researchers in various fields, and is widely applied to various research fields such as information prediction, automatic control, data classification, pattern recognition and the like. The RBF neural network includes an input layer, a hidden layer, and an output layer. The input layer is mainly used for transmitting the received signals to the hidden layer; the hidden layer is composed of a plurality of neurons, is the core of the whole RBF neural network, and the neurons select different radial basis functions according to practical problems, and are mainly used for mapping input vectors into a high-dimensional characteristic space from a low-dimensional space, and the vectors are changed from low-dimensional inseparability into linear divisibility in the high-dimensional space. The radial basis function is the kernel function of the hidden layer of the RBF neural network, and the function image is symmetrical about the radial direction. The selection of the radial basis function is a main factor influencing the prediction performance of the RBF, so that the corresponding radial basis function needs to be selected according to practical problems. The Gaussian function is the most commonly used basis function because of its advantages such as symmetry, smoothness and strong analytic power. The expression is as follows:
φ i (x)＝exp(-||x-c i || 2 /2δ i 2 ),i＝1,2,…,n
wherein x is an input vector, c i For the data center of the ith function, | | x-c i I is input data and data center c i Distance of δ i As a Gaussian functionThe width, referred to as the spreading constant.
Since the RBF hidden layer to output layer is a linear transformation, the kth neuron outputs y k Comprises the following steps:
Figure BDA0003656106530000111
in the formula, R i (x) Ith output of the hidden layer, w i Is R i (x) The weight of (c).
As an alternative embodiment, an optimization algorithm is used to optimize the parameters of the radial basis function neural network.
The embodiment provides a technical scheme for improving a prediction network model. Three parameters of RBF neural network (data center c) i Spreading constant delta i And a weight w i ) Has variability and certain relationships between them, so that it is very difficult to determine the parameters by using a fixed formula or model in the conventional sense. Therefore, the optimization algorithm is adopted to optimize the three parameters of the radial basis function neural network, so that the network iteration times are reduced, and the convergence speed and generalization performance of the RBF neural network are improved. There are many optimization algorithms such as gradient descent algorithm, newton's method, genetic algorithm, etc. The embodiment does not limit the specific optimization algorithm.
As an alternative embodiment, a genetic algorithm is used to optimize the parameters of the radial basis function neural network, and the method is as follows:
data normalization: determining a training sample and carrying out normalization processing on sample data;
chromosomal coding: data center c for neural network of radial basis function using real number coding i Spreading constant delta i And a weight w i Uniformly coding to form a chromosome string;
population initialization: the value range of the optimal initial population scale is 20-70;
constructing a fitness function: the fitness function is expressed as:
Figure BDA0003656106530000121
in the formula, F i The value of the fitness function is used as the fitness function value,
Figure BDA0003656106530000122
to predict value, y i Is the actual value, n is the total number of sample inputs;
selecting and operating: selecting individuals with high fitness in the population by adopting a roulette method, distributing the fitness of each individual in the population to a roulette plate, wherein the fitness value is proportional to the area occupied in the roulette plate, and selecting the individuals by rotating a pointer of the roulette plate; the probability of the ith individual being selected is:
Figure BDA0003656106530000123
in the formula, P i Probability of being selected for the ith individual, F i The fitness of the ith individual is shown, and N is the number of individuals in the population;
and (3) cross operation: performing cross operation by using an arithmetic cross method to obtain an individual g i 、g f The chromosomes of the cell are combined and crossed to form a new individual
Figure BDA0003656106530000124
The gene combination cross calculation formula is as follows:
Figure BDA0003656106530000125
Figure BDA0003656106530000126
in the formula, alpha is a scale factor, and alpha belongs to (0, 1);
cross probability P c Self-adaptive adjustment, the calculation formula is:
Figure BDA0003656106530000127
in the formula (f) max Is the maximum value of fitness in the population, f avg Is the average value of fitness in the population, f' is the fitness value of individuals in the cross, P C1 ＝0.9，P C2 ＝0.6；
And (3) mutation operation: randomly selecting individuals in a population, and selecting certain position genes of chromosomes of the individuals with a certain probability P m Self-adapting, thereby forming a new individual; probability P m The calculation formula of the self-adaptive adjustment is as follows:
Figure BDA0003656106530000131
wherein f is the fitness value of the variant individual, P m1 ＝0.1，P m2 ＝0.01；
If the preset convergence condition is met or the maximum iteration number is exceeded, stopping; otherwise, the transchromosome encoding step is repeated to perform an iterative process.
The embodiment provides a technical scheme for optimizing RBF neural network parameters by adopting a genetic algorithm. The genetic algorithm is the most commonly used optimization algorithm, and its flow chart is shown in fig. 3, and will not be described herein. Optimization of data center c using genetic algorithms is detailed above i Spreading constant delta i And a weight w i The specific steps for optimization are to be explained, that is, the crossover probability P in the genetic algorithm of this embodiment c And probability P m All the adjustment methods are self-adaptive, and the specific adjustment method is shown in the formula.
Fig. 4 is a schematic composition diagram of an intelligent power battery life prediction apparatus according to an embodiment of the present invention, where the apparatus includes:
the data acquisition module 11 is used for acquiring vehicle state data and vehicle-mounted battery pack data from an electric vehicle operation monitoring center;
the data integration module 12 is used for integrating the data to obtain various characteristic data which affect the service life of the battery;
the data screening module 13 is configured to screen the feature data based on the correlation, so as to obtain several feature data that most significantly affect the battery life;
and the service life prediction module 14 is used for constructing a neural network prediction model by taking the screened characteristic data as input and the battery service life as output, and predicting the battery service life by using the trained prediction model.
The apparatus of this embodiment may be used to implement the technical solution of the method embodiment shown in fig. 1, and the implementation principle and the technical effect are similar, which are not described herein again. The same applies to the following embodiments, which are not further described.
As an optional embodiment, the apparatus further comprises a parameter optimization module, configured to optimize parameters of the radial basis function neural network by using a genetic algorithm, where the method includes:
data normalization: determining a training sample and carrying out normalization processing on sample data;
chromosomal coding: data center c for neural network of radial basis function using real number coding i Spreading constant delta i And a weight w i Uniformly coding to form a chromosome string;
population initialization: the value range of the optimal initial population scale is 20-70;
constructing a fitness function: the fitness function is expressed as:
Figure BDA0003656106530000141
in the formula, F i The value of the fitness function is used as the fitness function value,
Figure BDA0003656106530000142
to predict value, y i Is the actual value, n is the total number of sample inputs;
selecting and operating: selecting individuals with high fitness in the population by adopting a roulette method, distributing the fitness of each individual in the population to a roulette plate, wherein the fitness value is proportional to the area occupied in the roulette plate, and selecting the individuals by rotating a pointer of the roulette plate; the probability of the ith individual being selected is:
Figure BDA0003656106530000143
in the formula, P i Probability of being selected for the ith individual, F i The fitness of the ith individual is shown, and N is the number of individuals in the population;
and (3) cross operation: performing cross operation by using an arithmetic cross method to obtain an individual g i 、g f The chromosomes of the cell are combined and crossed to form a new individual
Figure BDA0003656106530000144
The gene combination cross calculation formula is as follows:
Figure BDA0003656106530000145
Figure BDA0003656106530000146
in the formula, alpha is a scale factor, and alpha belongs to (0, 1);
cross probability P c Self-adaptive adjustment, the calculation formula is:
Figure BDA0003656106530000147
in the formula (f) max Is the maximum value of fitness in the population, f avg Is the average value of fitness in the population, f' is the fitness value of individuals in the cross, P C1 ＝0.9，P C2 ＝0.6；
And (3) mutation operation: randomly selecting individuals in a population, and selecting certain position genes of chromosomes of the individuals with a certain probability P m Self-adapting, thereby forming a new individual; probability P m Adaptive adaptationThe whole calculation formula is as follows:
Figure BDA0003656106530000151
wherein f is the fitness value of the variant individual, P m1 ＝0.1，P m2 ＝0.01；
If the preset convergence condition is met or the maximum iteration number is exceeded, stopping; otherwise, the transchromosome encoding step is repeated to perform an iterative process.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. An intelligent prediction method for the service life of a power battery is characterized by comprising the following steps:
acquiring vehicle state data and vehicle-mounted battery pack data from an electric vehicle operation monitoring center;
integrating the data to obtain various characteristic data influencing the service life of the battery;
screening the characteristic data based on the correlation to obtain several characteristic data which have the most obvious influence on the service life of the battery;
and (4) constructing a neural network prediction model by taking the screened characteristic data as input and the battery life as output, and predicting the battery life by using the trained prediction model.
2. The intelligent prediction method of power battery life as claimed in claim 1, wherein the vehicle state data and vehicle battery pack data comprise: the system comprises a vehicle number, a total voltage and a total current of a battery pack, an SOC, a highest voltage and a lowest voltage of a single battery, a forward accumulated electric quantity, a reverse accumulated electric quantity, a total mileage, an operation time and a vehicle speed.
3. The intelligent prediction method of power battery life as claimed in claim 1, characterized in that the method further comprises a data preprocessing step: and eliminating abnormal data, supplementing missing data and distinguishing charge and discharge current data.
4. The intelligent prediction method for the service life of the power battery as claimed in claim 1, wherein the method for integrating the data comprises: and taking different statistics of the same data as new feature data, and taking any combination of the different statistics of the same data as the new feature data, wherein the statistics comprise a maximum value, a minimum value, a median value, a mean value, a variance and a mean square error.
5. The intelligent prediction method for the service life of the power battery as claimed in claim 1, wherein the method for screening the feature data based on the correlation comprises:
calculating a correlation coefficient of any one feature data and the service life of the battery, and sequencing the feature data according to the sequence of the correlation coefficients from large to small;
deleting the characteristic data with the correlation coefficient smaller than a first threshold value;
and calculating a correlation coefficient between any two feature data in the remaining feature data, and deleting the feature data which is ranked later for the two feature data with the correlation coefficient being greater than a second threshold value, so that the correlation coefficient between any two feature data is not greater than the second threshold value.
6. The intelligent prediction method for the service life of the power battery as claimed in claim 1, characterized in that a radial basis function neural network is adopted to construct a battery service life prediction model.
7. The intelligent prediction method of power battery life according to claim 6, characterized in that an optimization algorithm is used to optimize the parameters of the radial basis function neural network.
8. The intelligent prediction method for the service life of the power battery according to claim 7, characterized in that the parameters of the radial basis function neural network are optimized by adopting a genetic algorithm, and the method comprises the following steps:
data normalization: determining a training sample and carrying out normalization processing on sample data;
chromosome coding: data center c for neural network of radial basis function using real number coding i Spreading constant delta i And a weight w i Uniformly coding to form a chromosome string;
population initialization: the value range of the optimal initial population scale is 20-70;
constructing a fitness function: the fitness function is expressed as:
Figure FDA0003656106520000021
in the formula, F i The value of the fitness function is used as the fitness function value,
Figure FDA0003656106520000022
to predict value, y i Is the actual value, n is the total number of sample inputs;
selecting and operating: selecting individuals with high fitness in the population by adopting a roulette method, distributing the fitness of each individual in the population to a roulette plate, wherein the fitness value is proportional to the area occupied in the roulette plate, and selecting the individuals by rotating a pointer of the roulette plate; the probability of the ith individual being selected is:
Figure FDA0003656106520000023
in the formula, P i Probability of being selected for the ith individual, F i The fitness of the ith individual is shown, and N is the number of individuals in the population;
and (3) cross operation: performing cross operation by using an arithmetic cross method to obtain an individual g i 、g f The chromosomes of the cell are combined and crossed to form a new individual
Figure FDA0003656106520000024
The gene combination cross calculation formula is as follows:
Figure FDA0003656106520000025
Figure FDA0003656106520000026
in the formula, alpha is a scale factor, and alpha belongs to (0, 1);
cross probability P c Self-adaptive adjustment, the calculation formula is:
Figure FDA0003656106520000031
in the formula (f) max Is the maximum value of fitness in the population, f avg Is the average value of fitness in the population, f' is the fitness value of individuals in the cross, P C1 ＝0.9，P C2 ＝0.6；
And (3) mutation operation: randomly selecting individuals in a population, and selecting certain position genes of chromosomes of the individuals with a certain probability P m Self-adapting, thereby forming a new individual; probability P m The calculation formula of the self-adaptive adjustment is as follows:
Figure FDA0003656106520000032
wherein f is the fitness value of the variant individual, P m1 ＝0.1，P m2 ＝0.01；
If the preset convergence condition is met or the maximum iteration number is exceeded, stopping; otherwise, the transchromosome encoding step is repeated to perform an iterative process.
9. An intelligent power battery life prediction device, comprising:
the data acquisition module is used for acquiring vehicle state data and vehicle-mounted battery pack data from the electric vehicle operation monitoring center;
the data integration module is used for integrating and processing the data to obtain various characteristic data which affect the service life of the battery;
the data screening module is used for screening the characteristic data based on the correlation to obtain several types of characteristic data which have the most obvious influence on the service life of the battery;
and the service life prediction module is used for constructing a neural network prediction model by taking the screened characteristic data as input and the battery service life as output, and predicting the battery service life by using the trained prediction model.
10. The intelligent prediction device of power battery life as claimed in claim 9, further comprising a parameter optimization module for optimizing the parameters of the radial basis function neural network by using a genetic algorithm, wherein the method comprises the following steps:
data normalization: determining a training sample and carrying out normalization processing on sample data;
chromosomal coding: data center c for neural network of radial basis function using real number coding i Spreading constant delta i And a weight w i Uniformly coding to form a chromosome string;
population initialization: the value range of the optimal initial population scale is 20-70;
constructing a fitness function: the fitness function is expressed as:
Figure FDA0003656106520000041
in the formula, F i The value of the fitness function is used as the fitness function value,
Figure FDA0003656106520000042
to predict value, y i Is the actual value, n is the total number of sample inputs;
selecting and operating: selecting individuals with high fitness in the population by adopting a roulette method, distributing the fitness of each individual in the population to a roulette plate, wherein the fitness value is proportional to the area occupied in the roulette plate, and selecting the individuals by rotating a pointer of the roulette plate; the probability of the ith individual being selected is:
Figure FDA0003656106520000043
in the formula, P i Probability of being selected for the ith individual, F i The fitness of the ith individual is shown, and N is the number of individuals in the population;
and (3) cross operation: performing cross operation by using an arithmetic cross method to obtain an individual g i 、g f The chromosomes of the cell are combined and crossed to form a new individual
Figure FDA0003656106520000044
The gene combination cross calculation formula is as follows:
Figure FDA0003656106520000045
Figure FDA0003656106520000046
in the formula, alpha is a scale factor, and alpha belongs to (0, 1);
cross probability P c Self-adaptive adjustment, the calculation formula is:
Figure FDA0003656106520000047
in the formula (f) max Is the maximum value of fitness in the population, f avg Is the average value of fitness in the population, f' is the fitness value of individuals in the cross, P C1 ＝0.9，P C2 ＝0.6；
And (3) mutation operation: randomly selecting individuals in a population, and selecting certain position genes of chromosomes of the individuals with a certain probability P m Self-adapting, thereby forming a new individual; probability P m The calculation formula of the self-adaptive adjustment is as follows:
Figure FDA0003656106520000051
wherein f is the fitness value of the variant individual, P m1 ＝0.1，P m2 ＝0.01；
If the preset convergence condition is met or the maximum iteration number is exceeded, stopping; otherwise, the transchromosome encoding step is repeated to perform an iterative process.
CN202210559862.2A 2022-05-23 2022-05-23 Intelligent prediction method and device for service life of power battery Pending CN114880939A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210559862.2A CN114880939A (en) 2022-05-23 2022-05-23 Intelligent prediction method and device for service life of power battery

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210559862.2A CN114880939A (en) 2022-05-23 2022-05-23 Intelligent prediction method and device for service life of power battery

Publications (1)

Publication Number Publication Date
CN114880939A true CN114880939A (en) 2022-08-09

Family

ID=82678509

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210559862.2A Pending CN114880939A (en) 2022-05-23 2022-05-23 Intelligent prediction method and device for service life of power battery

Country Status (1)

Country Link
CN (1) CN114880939A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116027204A (en) * 2023-02-20 2023-04-28 山东大学 Lithium battery residual service life prediction method and device based on data fusion
CN116910637A (en) * 2023-04-03 2023-10-20 山东科技大学 Improved IGA-RBF neural network-based short-term load prediction method

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116027204A (en) * 2023-02-20 2023-04-28 山东大学 Lithium battery residual service life prediction method and device based on data fusion
CN116027204B (en) * 2023-02-20 2023-06-20 山东大学 Lithium battery residual service life prediction method and device based on data fusion
CN116910637A (en) * 2023-04-03 2023-10-20 山东科技大学 Improved IGA-RBF neural network-based short-term load prediction method
CN116910637B (en) * 2023-04-03 2024-04-26 山东科技大学 Improved IGA-RBF neural network-based short-term load prediction method

Similar Documents

Publication Publication Date Title
CN114880939A (en) Intelligent prediction method and device for service life of power battery
CN108681973A (en) Sorting technique, device, computer equipment and the storage medium of power consumer
CN111476435B (en) Charging pile load prediction method based on density peak value
CN113459871B (en) Battery charging method and device of battery replacement station, storage medium and electronic equipment
CN112791997A (en) Method for cascade utilization and screening of retired battery
CN111525587B (en) Reactive load situation-based power grid reactive voltage control method and system
CN113191547B (en) New energy vehicle power battery charging optimization method and system
CN115204038A (en) Energy storage lithium battery life prediction method based on data decomposition and integration model
CN116643178B (en) SOC estimation method and related device of battery management system
CN115372828A (en) Battery cell consistency evaluation method based on charging segment data and unsupervised algorithm
CN112287980A (en) Power battery screening method based on typical feature vector
CN114285086B (en) Electrochemical energy storage power station energy scheduling method and system based on improved ant colony algorithm
CN117673580B (en) Temperature control equipment for new energy battery
CN117543791B (en) Power supply detection method, device, equipment and storage medium for power supply
CN114384423A (en) Fuel cell health state identification method based on typical operation parameters
CN117129899B (en) Battery health state prediction management system and method
CN117318249A (en) Battery charging cloud monitoring method and system
CN117034762A (en) Composite model lithium battery life prediction method based on multi-algorithm weighted sum
CN116488219A (en) Distributed energy storage acquisition strategy control method based on multi-level architecture
CN112305442B (en) Method for quickly estimating SOH of power battery based on kmeans clustering
CN115034608A (en) Distribution network risk assessment method based on distribution network element and neural network
CN113212244A (en) New energy vehicle power battery life prediction method and system
CN117634931B (en) Electric automobile adjustment capability prediction method and system considering charging behavior
US11733313B2 (en) Method and apparatus for operating a system for providing states of health of electrical energy stores for a multiplicity of devices with the aid of machine learning methods
CN116191423B (en) Power collection system and method of photovoltaic power generation system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination