CN115389947B - Lithium battery health state prediction method and device, electronic equipment and storage medium - Google Patents
Lithium battery health state prediction method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN115389947B CN115389947B CN202211321901.1A CN202211321901A CN115389947B CN 115389947 B CN115389947 B CN 115389947B CN 202211321901 A CN202211321901 A CN 202211321901A CN 115389947 B CN115389947 B CN 115389947B
- Authority
- CN
- China
- Prior art keywords
- data
- charge
- lithium battery
- dtv
- layer
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60L—PROPULSION OF ELECTRICALLY-PROPELLED VEHICLES; SUPPLYING ELECTRIC POWER FOR AUXILIARY EQUIPMENT OF ELECTRICALLY-PROPELLED VEHICLES; ELECTRODYNAMIC BRAKE SYSTEMS FOR VEHICLES IN GENERAL; MAGNETIC SUSPENSION OR LEVITATION FOR VEHICLES; MONITORING OPERATING VARIABLES OF ELECTRICALLY-PROPELLED VEHICLES; ELECTRIC SAFETY DEVICES FOR ELECTRICALLY-PROPELLED VEHICLES
- B60L58/00—Methods or circuit arrangements for monitoring or controlling batteries or fuel cells, specially adapted for electric vehicles
- B60L58/10—Methods or circuit arrangements for monitoring or controlling batteries or fuel cells, specially adapted for electric vehicles for monitoring or controlling batteries
- B60L58/16—Methods or circuit arrangements for monitoring or controlling batteries or fuel cells, specially adapted for electric vehicles for monitoring or controlling batteries responding to battery ageing, e.g. to the number of charging cycles or the state of health [SoH]
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/392—Determining battery ageing or deterioration, e.g. state of health
-
- 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
Abstract
The embodiment of the specification relates to the technical field of lithium batteries, in particular to a method and a device for predicting the health state of a lithium battery, an electronic device and a storage medium. The lithium battery health state prediction method comprises the following steps: collecting cyclic aging data of a target lithium battery in a plurality of charging and discharging cyclic processes; the cyclic aging data comprises original voltage data and original temperature data; for each charge-discharge cycle, performing: cleaning the original voltage data and the original temperature data in the current charging and discharging cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data; performing feature extraction on the DTV data of each charge-discharge cycle to obtain time sequence data; inputting the time sequence data into a preset LSTM model for training to obtain a prediction model; and predicting the health state of the lithium battery to be tested by using the prediction model and quantifying the uncertain result.
Description
Technical Field
The embodiment of the specification relates to the technical field of lithium batteries, in particular to a method and a device for predicting the health state of a lithium battery, an electronic device and a storage medium.
Background
The electric automobile is widely popularized due to the advantages of low emission, environmental protection and low noise, and the lithium ion battery (hereinafter referred to as lithium battery) has the advantages of high voltage, long cycle life, small self-discharge and the like, so the lithium ion battery is widely applied to the electric automobile and is an important power source of the electric automobile. The health state of the lithium battery is closely related to safety, so that the timely and accurate prediction of the health state of the lithium battery is very important.
Disclosure of Invention
In order to be able to accurately predict the health state of a lithium battery in time, embodiments of the present specification provide a method and an apparatus for predicting the health state of a lithium battery, an electronic device, and a storage medium.
In a first aspect, an embodiment of the present specification provides a method for predicting a health state of a lithium battery, including:
collecting cyclic aging data of a target lithium battery in the process of multiple charging and discharging cycles; wherein the cyclic aging data comprises raw voltage data and raw temperature data;
for each of the charge and discharge cycles, performing: cleaning the original voltage data and the original temperature data in the current charging and discharging cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data;
performing feature extraction on the DTV data of each charge-discharge cycle to obtain time series data;
inputting the time sequence data into a preset LSTM model for training to obtain a prediction model;
and predicting the health state of the lithium battery to be tested by using the prediction model and quantifying an uncertain result.
In one possible design, the cleaning process includes at least one of: outliers are screened out, sampling intervals are fixed, and filtering is performed.
In one possible design, the extracting features of the DTV data for each charge and discharge cycle to obtain time series data includes:
for each of the charge and discharge cycles, performing: sequentially carrying out filtering processing and connection processing on the DTV data of the current charge-discharge cycle to obtain a DTV curve; extracting the size and the position of a peak and the size and the position of a trough in the DTV curve, and taking the sizes and the positions as initial characteristic data; performing correlation analysis on the battery capacity of the current charge-discharge cycle and the size and the position of the current peak or the current trough aiming at each peak or each trough to obtain target characteristic data with the correlation higher than a preset threshold value; wherein the battery capacity is derived from the cyclic aging data;
and arranging the target characteristic data of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
In one possible design, the performing feature extraction on the DTV data for each charge and discharge cycle to obtain time series data includes:
for each of the charge and discharge cycles, performing: constructing DTV data of the current charge-discharge cycle into a column vector; performing singular value decomposition on the column vector to obtain a singular value;
and arranging the singular values of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
In one possible design, the inputting the time-series data into a preset LSTM model for training to obtain a prediction model includes:
constructing at least two time series data to obtain a three-dimensional tensor;
and inputting the plurality of constructed three-dimensional tensors into a preset LSTM model for training to obtain a prediction model.
In one possible design, the LSTM model comprises an input layer, a first LSTM layer, a first dropout layer, a second LSTM layer, a second dropout layer, a full connection layer and an output layer which are connected in sequence, wherein an activation function in the first LSTM layer and an activation function in the second LSTM layer are tanh functions, and an activation function in the full connection layer is a sigmoid function; and/or the presence of a gas in the atmosphere,
the gradient descent algorithm of the LSTM model is an RMSprop algorithm; and/or the presence of a gas in the atmosphere,
and the LSTM model adopts a Bayesian algorithm to carry out parameter optimization.
In a possible design, the predicting the state of health of the lithium battery to be tested and quantifying the uncertainty result by using the prediction model includes:
the health state of the lithium battery to be tested is predicted for multiple times by using the prediction model, and multiple prediction results are obtained;
and carrying out uncertainty quantification on the plurality of prediction results by adopting a Monte Carlo method to obtain uncertainty results.
In a second aspect, an embodiment of the present specification further provides a device for predicting a health state of a lithium battery, including:
the acquisition module is used for acquiring cyclic aging data of the target lithium battery in the process of multiple charging and discharging cycles; wherein the cyclic aging data comprises raw voltage data and raw temperature data;
and the execution module is used for executing the following steps aiming at each charge and discharge cycle: cleaning the original voltage data and the original temperature data in the current charging and discharging cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data;
the characteristic extraction module is used for extracting the characteristics of the DTV data of each charge-discharge cycle to obtain time sequence data;
the training module is used for inputting the time series data into a preset LSTM model for training to obtain a prediction model;
and the prediction module is used for predicting the health state of the lithium battery to be tested by using the prediction model and quantifying an uncertain result.
In a third aspect, an embodiment of the present specification further provides an electronic device, which includes a memory and a processor, where the memory stores a computer program, and the processor executes the computer program to implement the method according to any embodiment of the present specification.
In a fourth aspect, the embodiments of the present specification further provide a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed in a computer, the computer program causes the computer to execute the method according to any one of the embodiments of the present specification.
The embodiment of the specification provides a method and a device for predicting the health state of a lithium battery, electronic equipment and a storage medium, wherein DTV data of each charge and discharge cycle is subjected to feature extraction to obtain time sequence data, and then the time sequence data is input into a preset LSTM model to be trained to obtain a prediction model, so that the effect of predicting the health state of the lithium battery by fusing a data driving method and a signal analysis method is achieved, and the defects of a single prediction means are overcome, so that the health state of the lithium battery can be timely and accurately predicted.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the description below are some embodiments of the present specification, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flowchart of a method for predicting a health status of a lithium battery according to an embodiment of the present disclosure;
fig. 2 is a hardware architecture diagram of an electronic device provided in an embodiment of the present specification;
fig. 3 is a structural diagram of a lithium battery health status prediction apparatus according to an embodiment of the present disclosure;
fig. 4 is a graph of DTV data obtained based on the voltage and temperature changes of a lithium battery during one charge-discharge cycle, provided by an embodiment of the present disclosure.
Detailed Description
In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure more clear, the technical solution in the embodiments of the present disclosure will be described clearly and completely with reference to the drawings in the embodiments of the present disclosure, and it is obvious that the described embodiments are a part of the embodiments of the present disclosure, but not all embodiments, and all other embodiments obtained by a person of ordinary skill in the art without creative efforts based on the embodiments of the present disclosure belong to the protection scope of the present disclosure.
The aging of the lithium battery is a long-term gradual change process, and the health state of the battery is influenced by various factors such as temperature, current multiplying power, cut-off voltage and the like. Currently, the lithium battery health status prediction is mainly classified into a model-based method, a data-driven method, a signal analysis-based method, and an experience-based method.
The model-based method is mainly used for predicting by constructing a mechanism model or an equivalent circuit model, has strong interpretability, and is commonly used in the fields of battery thermal management, battery design and the like in practical application; the method based on data driving is used for predicting through historical observation data directly, the data driving method does not need much knowledge about the working principle Of a battery generally, has higher fitting capacity and wider application field on nonlinear data, and is often used for cloud estimation Of the State Of Health (SOH) Of the lithium battery based on big data; the characteristic-based signal analysis method is based on an electrochemical analysis technology and comprises methods such as Incremental Capacity Analysis (ICA), differential Voltage Analysis (DVA) and Differential Thermal Voltammetry (DTV) for extracting appropriate and reliable battery degradation characteristics, the signal analysis method can macroscopically react the microscopic reaction in the lithium battery by obtaining and analyzing characteristic data, and is generally used as an auxiliary technical means in practical application; the experience-based method mainly carries out prediction by building an experience model or curve fitting and other methods, has the characteristics of low calculation cost, high real-time performance and the like, and is often used for predicting the health state of the lithium battery by a vehicle-end rapid model in practical application.
The lithium battery health state prediction by using a single method cannot give consideration to high precision, practicability, stability and universality, and each method has own defects. The method based on the model has the disadvantages that the construction of an accurate model is very difficult, a large amount of physical and chemical knowledge is needed, various internal reaction mechanisms of the battery are involved, the accuracy of the model is easily influenced by variable current and temperature, and an accurate mechanism model is difficult to obtain under the influence of different external conditions; the data-driven-based method has the disadvantages that the interpretability is weak, the prediction result is sensitive to the data quality, the stability is poor, the required characteristics can be obtained only by a complex data processing process, and the calculated amount of the vehicle-mounted BMS is increased; the disadvantage of the signal analysis based method is that it is highly dependent on the sampling conditions and the data processing flow is complex; the method based on the empirical model has the defects of poor precision and poor anti-interference capability under complex working conditions.
The inventor finds out in the development process that: the method can be used for predicting the health state of the lithium battery by combining a data driving method and a signal analysis method, so that the defects of single prediction means can be avoided, and the health state of the lithium battery can be predicted timely and accurately.
The inventive concept of the illustrative embodiment is described below.
Referring to fig. 1, an embodiment of the present disclosure provides a method for predicting a health state of a lithium battery, including:
step 100: collecting cyclic aging data of a target lithium battery in a plurality of charging and discharging cyclic processes; the cyclic aging data comprises original voltage data and original temperature data;
step 102: for each charge-discharge cycle, performing: cleaning the original voltage data and the original temperature data in the current charging and discharging cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data;
step 104: performing feature extraction on the DTV data of each charge-discharge cycle to obtain time sequence data;
step 106: inputting the time sequence data into a preset LSTM model for training to obtain a prediction model;
step 108: and predicting the health state of the lithium battery to be tested by using the prediction model and quantifying the uncertain result.
In the embodiment of the specification, the DTV data of each charge-discharge cycle is subjected to feature extraction to obtain time sequence data, and then the time sequence data is input into a preset LSTM model to be trained to obtain a prediction model, so that the effect of predicting the health state of the lithium battery by fusing data driving and signal analysis methods is realized, the defects of a single prediction means are avoided, and the health state of the lithium battery can be predicted timely and accurately.
The technical scheme can be used for estimating the SOH by a vehicle-mounted Battery Management System (BMS), and has diversified application prospects in electric automobiles.
The manner in which the various steps shown in fig. 1 are performed is described below.
With respect to step 100:
in some embodiments, the cyclic aging data of the target lithium battery in the multiple charge-discharge cycle process can be acquired based on multiple ways such as a vehicle-mounted battery management system, a cloud battery monitoring platform, a factory-provided and public data set, and the acquisition path and the acquisition mode of the cyclic aging data are not limited.
With respect to step 102:
in one embodiment of the present description, the cleaning process includes at least one of: outliers are screened out, sampling intervals are fixed, and filtering is performed.
In the embodiment, the original voltage data and the original temperature data are cleaned, so that abnormal values and noise caused by sampling can be effectively removed, and the problems of reduction in prediction accuracy, poor stability and the like caused by abnormal and fluctuating data acquired by the vehicle-mounted battery management system in practical application are solved.
In one embodiment of the present description, the DTV data may be obtained by processing the target voltage data and the target temperature data based on differential thermal voltammetry. Specifically, DTV data is obtained by differentiating the temperature of the battery surface against the terminal voltage during charging and discharging, as shown in the following equation:
in the embodiment, the features are extracted through a differential thermal voltammetry method, and only voltage data and temperature data need to be acquired in practical application, so that the problems that the requirements on sampling conditions and equipment are too harsh and complex in the practical application process and the practicability is very high are solved, and meanwhile, the data volume in each cycle is greatly reduced through the feature extraction, and the problem that resources are consumed in subsequent LSTM model training is overlarge are solved.
It should be noted that the prediction result of the data driving method highly depends on the data quality and the selected features, and a plurality of features can be derived from the operation data of the battery, and the correlations between the features and the capacity are not necessarily high, and the complicated features may cause problems of increased computational burden, low model training efficiency, low model precision, and the like, and the same feature may be greatly different under different battery systems, different charging and discharging conditions, or different sampling conditions, and individual features may lack universality. The use of features that are highly correlated with cell aging and are compact and versatile is therefore of crucial importance to improve model efficiency and accuracy. The DTV method can highly reflect microscopic changes in the battery aging process on a macroscopic scale, and has strong correlation with the battery aging.
It should be further explained that differential thermal voltammetry is a fast electrochemical characterization method, and is often used as an important method for tracking the state of health of a battery in practical application. During the charging and discharging process of the lithium battery, along with the insertion and the separation of lithium in an electrode material, phase change occurs between the two to cause the change of entropy, and further cause the change of temperature, and DTV data is obtained by monitoring the change of voltage and temperature during the charging and discharging process in the practical application process. With reference to fig. 4, during the charging process, the aging of the battery is accompanied by the decrease of the peak value and the movement of the peak position to the high potential, the valley gradually becomes smaller and moves to the low potential, so as to reflect the changes such as the resistance increase and the electrode performance unevenness during the capacity fading process of the battery, and further effectively reflect the microscopic changes of the battery during the aging process in the macro scale.
With respect to step 104:
in step 104, the present specification provides two technical solutions to implement further feature extraction on DTV data, namely peak-valley-based extraction and singular value-based extraction.
< extraction based on peaks and valleys >
In an embodiment of the present specification, step 104 may specifically include:
for each charge-discharge cycle, performing: sequentially carrying out filtering processing and connection processing on the DTV data of the current charge-discharge cycle to obtain a DTV curve; extracting the size and the position of a peak and the size and the position of a trough in the DTV curve, and taking the sizes and the positions as initial characteristic data; for each peak or each trough, performing correlation analysis on the battery capacity of the current charge-discharge cycle and the size and position of the current peak or the current trough to obtain target characteristic data with the correlation higher than a preset threshold; wherein the battery capacity is derived from cyclic aging data;
and arranging the target characteristic data of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
In this embodiment, the DTV data needs to be sorted into a DTV curve first, and then the size and position of a peak and the size and position of a valley in the DTV curve are extracted and used as initial feature data; in order to improve the prediction accuracy and the training efficiency of the model, the battery capacity of the current charge-discharge cycle and the size and the position of the current peak or the current trough are subjected to correlation analysis, so that the characteristics with high correlation are selected for use, the characteristic quantity is reduced, and the characteristics with poor correlation are screened out.
Since the prediction model is constructed based on the LSTM model, it is necessary to convert the target feature data into time-series data. The prediction model is built by taking an LSTM neural network as a core, the aging of a lithium battery belongs to the problem of long-term dependence, the LSTM is a variant of a circulating neural network (RNN), and the prediction model is mainly characterized in that a main line penetrating through a unit records information, the LSTM introduces a unit state and uses an input gate, a forgetting gate and an output gate to keep and control the information, the problems of gradient disappearance and gradient explosion are effectively solved, and the problem of long-term dependence is solved.
In some embodiments, the correlation analysis may be a pearson correlation analysis method.
It should be noted that, the DTV curve may be different in different battery systems and different charging/discharging conditions, and different sampling conditions and sampling errors may also affect the DTV curve. However, the existence of the peak-valley of the DTV curve and the information contained therein are slightly affected and relatively stable, and the evolution of the peak-valley information is highly correlated with the battery aging, so that the peak-valley information is extracted as a feature, and the universality and the stability are increased. Meanwhile, the data volume of each cycle is reduced to one point through feature extraction, so that the calculation consumption is greatly reduced, and the efficiency is improved.
As mentioned above, the DTV curve is obtained by filtering the DTV data (for example, filtering the DTV data by an SG filter), and generally, filtering the DTV data with good data quality needs to be performed once, and filtering the DTV data with poor data quality may need to be performed twice. It should be noted that filtering may bring a problem that if the filtering parameter setting is inaccurate, some insignificant peaks and valleys may be directly filtered out, which reduces the amount of calculation, but the feature quantity is reduced, so that the high prediction accuracy of the model cannot be guaranteed. Moreover, the DTV curves may be different in different battery systems and different charging and discharging conditions, which also deteriorates the universality of the peak-valley based extraction method.
In order to solve the technical problem, the inventors creatively consider that the singular value decomposition technology is applied to DTV data, so that the data volume can be greatly reduced (namely, peak-valley-based extraction generates several characteristic data volumes in one cycle, and singular value-based extraction generates only one characteristic data volume in one cycle), and important information and characteristics contained in the DTV can be effectively contained. The following describes a singular value-based extraction method.
< extraction based on singular value >
In an embodiment of the present specification, step 104 may specifically include:
for each charge-discharge cycle, performing: constructing DTV data of the current charge-discharge cycle into a column vector; performing singular value decomposition on the column vector to obtain a singular value;
and arranging the singular values of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
In this embodiment, since a column of DTV data can be regarded as a matrix of m × 1, a matrix of 1 × 1 can be obtained by performing singular value decomposition on a column of DTV data, and this value is a singular value. The singular value not only can effectively contain important information and characteristics contained in the DTV, but also can greatly reduce the data volume.
In addition, when the elements in the matrix slightly change, the singular values of the matrix do not change greatly, which indicates that the singular values have good stability and strong anti-interference capability, and further reduces the complexity of data processing and errors possibly caused by data quality. Therefore, the singular value decomposition method is applied to the DTV data, the characteristic of strong correlation is extracted by using a simpler characteristic extraction method, the characteristic of peak-valley information selected by a DTV curve can be replaced by a group of new characteristics for input in the conventional general analysis, the size of an input matrix is reduced, the efficiency of a model is improved, meanwhile, the dependence on the data quality is small, and the difficulty and the complexity of data preprocessing are reduced.
The principle of singular value decomposition is described below.
The singular value decomposition method is a matrix decomposition method, obtains singular values containing main information in a matrix through matrix decomposition, and performs singular value decomposition on column vectors constructed by DTV data respectively to obtain singular values containing the main information, and the principle is as follows:
m is a matrix with M rows and n columns, and after SVD decomposition, the matrix is split into 3 sub-matrixes, wherein the matrix U is a square matrix with M rows and M columns, and the column U is MM T Orthogonal eigenvectors of the matrix, V being a square matrix of n rows and n columns, each column of n being a matrix M T M orthogonal eigenvectors, Σ, are matrices having values only in the principal diagonal, where the values are called singular values, and are M T The square root of the M eigenvalues.
It is noted that the singular values are extracted for each cycle of DTV data, the singular values concentrate important information in the matrix among the larger singular values, and the more important the information, the larger the singular values. The singular values are decreased from the upper left corner to the lower right corner from large to small at a very fast speed, so that important information is concentrated on the upper left part, and the data can be basically restored by taking the front several maximum singular values. Since a column of DTV data can be regarded as a matrix of m × 1, a matrix of 1 × 1 can be obtained by performing singular value decomposition on a column of DTV data, and this value is the largest singular value at the top left corner.
In summary, in general, the singular value-based extraction method does not need to filter DTV data, and even for DTV data with poor data quality, the singular value-based extraction method performs feature extraction after filtering at most once, which has the following advantages over the peak-valley-based extraction method: the size of an input matrix is reduced, the efficiency of the model is improved, meanwhile, the method has small dependence on data quality, and the difficulty and the complexity of data preprocessing are reduced.
For step 106:
in an embodiment of the present specification, step 106 may specifically include:
constructing at least two time sequence data to obtain a three-dimensional tensor;
and inputting the plurality of three-dimensional tensors obtained by construction into a preset LSTM model for training to obtain a prediction model.
In this embodiment, first, target feature data is constructed into a two-dimensional matrix according to a certain time sequence length, one two-dimensional matrix is taken as a time sequence, then, a plurality of time sequences are constructed into a plurality of three-dimensional tensors, one three-dimensional tensor is taken as a batch, and training is performed by taking the batch as a unit during training, so that the speed and efficiency of model training are effectively improved.
In one embodiment of the present specification, the LSTM model includes an input layer, a first LSTM layer, a first dropout layer, a second LSTM layer, a second dropout layer, a fully-connected layer, and an output layer, which are connected in sequence, where an activation function in the first LSTM layer and the second LSTM layer is a tanh function, and an activation function in the fully-connected layer is a sigmoid function.
In this embodiment, dropout technique is used to prevent overfitting during training (i.e. Dropout layer is connected to the next layer of LSTM layer), which will make the activation value of a certain neuron stop working with a certain probability, and the neuron together with all its input and output connections are temporarily removed from the network, and each neuron has a fixed probability, independent of other neurons; by adopting the technology, the sensitivity of the network to the specific weight of the neuron is reduced, so that the model generalization is stronger, and the excessive dependence on some local characteristics is avoided.
In one embodiment of the present description, the gradient descent algorithm of the LSTM model is the RMSprop algorithm.
In the embodiment, the gradient descent algorithm selects the RMSprop algorithm, which is an improvement based on other gradient descent algorithms, the training efficiency and the training effect are optimized by updating the learning rate, and the attenuation coefficient is introduced, so that the too fast decline of the learning rate along with the increase of the training period is avoided.
In one embodiment of the present description, the LSTM model is optimized for parameters using a bayesian algorithm.
In the embodiment, the current adjustment of the hyperparameters of the neural network mostly adopts a manual adjustment mode, the mode highly depends on parameter adjustment experience, the efficiency is extremely low, the consumption of calculation power is great for the application with large data volume and complex model structure, and the efficiency is influenced, so that the Bayesian optimization algorithm is used for automatic optimization.
For step 108:
in an embodiment of the present specification, step 108 may specifically include:
the health state of the lithium battery to be tested is predicted for multiple times by using the prediction model, and multiple prediction results are obtained;
and carrying out uncertainty quantification on the plurality of prediction results by adopting a Monte Carlo method to obtain uncertainty results.
In the embodiment, the prediction result quantifies an uncertain result through a monte carlo method, the monte carlo method realizes calculation of certain certainty problems based on statistical results of a large number of events, and the problem that an LSTM neural network can predict multi-step advanced data points but cannot quantify the prediction confidence coefficient is solved.
It should be noted that the LSTM neural network can learn long-term correlation of time series data and predict multi-step advanced data points, but cannot quantify prediction confidence, and for this purpose, a monte carlo method is introduced to quantify prediction uncertainty, and in the application process, the monte carlo method needs a numerical algorithm for describing a system, and a deep learning model is defined as a numerical algorithm for describing a battery degradation process, that is, the output of the LSTM neural network is used as the input of the monte carlo process.
As shown in fig. 2 and fig. 3, an embodiment of the present specification provides a lithium battery state of health prediction apparatus. The apparatus embodiments may be implemented by software, or by hardware, or by a combination of hardware and software. In terms of hardware, as shown in fig. 2, for a hardware architecture diagram of an electronic device where a lithium battery health status prediction apparatus provided in the embodiment of the present disclosure is located, in addition to the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 2, the electronic device where the apparatus is located in the embodiment may also include other hardware, such as a forwarding chip responsible for processing a message, and the like. Taking a software implementation as an example, as shown in fig. 3, as a logical device, a CPU of the electronic device reads a corresponding computer program in the non-volatile memory into the memory for running.
As shown in fig. 3, the present embodiment provides a lithium battery health status prediction apparatus, including:
the acquisition module 300 is used for acquiring cyclic aging data of the target lithium battery in a plurality of charging and discharging cyclic processes; wherein the cyclic aging data comprises raw voltage data and raw temperature data;
an executing module 302, configured to, for each charge and discharge cycle, execute: cleaning the original voltage data and the original temperature data in the current charging and discharging cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data;
a feature extraction module 304, configured to perform feature extraction on the DTV data of each charge and discharge cycle to obtain time series data;
a training module 306, configured to input the time series data into a preset LSTM model for training, so as to obtain a prediction model;
and the prediction module 308 is configured to predict the health state of the lithium battery to be tested by using the prediction model and quantify an uncertainty result.
In this embodiment, the acquisition module 300 may be configured to perform step 100 in the above method embodiment, the execution module 302 may be configured to perform step 102 in the above method embodiment, the feature extraction module 304 may be configured to perform step 104 in the above method embodiment, the training module 306 may be configured to perform step 106 in the above method embodiment, and the prediction module 308 may be configured to perform step 108 in the above method embodiment.
In one embodiment of the present description, the cleaning process includes at least one of: outliers are screened out, sampling intervals are fixed, and filtering is performed.
In an embodiment of the present specification, the feature extraction module 304 is configured to perform the following operations:
for each of the charge and discharge cycles, performing: sequentially carrying out filtering processing and connection processing on the DTV data of the current charge-discharge cycle to obtain a DTV curve; extracting the size and the position of a peak and the size and the position of a trough in the DTV curve, and taking the sizes and the positions as initial characteristic data; performing correlation analysis on the battery capacity of the current charge-discharge cycle and the size and the position of the current peak or the current trough aiming at each peak or each trough to obtain target characteristic data with the correlation higher than a preset threshold value; wherein the battery capacity is derived from the cyclic aging data;
and arranging the target characteristic data of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
In an embodiment of the present specification, the feature extraction module 304 is configured to perform the following operations:
for each of the charge and discharge cycles, performing: constructing DTV data of the current charge-discharge cycle into a column vector; performing singular value decomposition on the column vector to obtain a singular value;
and arranging the singular values of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
In one embodiment of the present description, the training module 306 is configured to perform the following operations:
constructing at least two time series data to obtain a three-dimensional tensor;
and inputting the plurality of constructed three-dimensional tensors into a preset LSTM model for training to obtain a prediction model.
In one embodiment of the present specification, the LSTM model includes an input layer, a first LSTM layer, a first dropout layer, a second LSTM layer, a second dropout layer, a fully-connected layer, and an output layer, which are connected in sequence, where an activation function in the first LSTM layer and the second LSTM layer is a tanh function, and an activation function in the fully-connected layer is a sigmoid function; and/or the presence of a gas in the atmosphere,
the gradient descent algorithm of the LSTM model is an RMSprop algorithm; and/or the presence of a gas in the gas,
and the LSTM model adopts a Bayesian algorithm to carry out parameter optimization.
In one embodiment of the present specification, the prediction module 308 is configured to perform the following operations:
the health state of the lithium battery to be tested is predicted for multiple times by using the prediction model, and multiple prediction results are obtained;
and carrying out uncertainty quantification on the plurality of prediction results by adopting a Monte Carlo method to obtain uncertainty results.
It is to be understood that the schematic structure of the embodiment in this specification does not constitute a specific limitation to the lithium battery state of health prediction device. In other embodiments of the present description, a lithium battery state of health prediction device may include more or fewer components than those shown, or some components may be combined, some components may be split, or a different arrangement of components. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.
For the information interaction, execution process and other contents between the modules in the above-mentioned apparatus, because the same concept is based on as the method embodiment of this specification, specific contents can refer to the description in the method embodiment of this specification, and are not described herein again.
The embodiment of the present specification further provides an electronic device, which includes a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the method for predicting the health state of a lithium battery in any embodiment of the present specification is implemented.
The embodiment of the present specification further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the computer program causes the processor to execute a method for predicting a health state of a lithium battery in any embodiment of the present specification.
Specifically, a system or an apparatus equipped with a storage medium on which software program codes that realize the functions of any of the above-described embodiments are stored may be provided, and a computer (or a CPU or MPU) of the system or the apparatus is caused to read out and execute the program codes stored in the storage medium.
In this case, the program code itself read from the storage medium can realize the functions of any of the embodiments described above, and thus the program code and the storage medium storing the program code constitute a part of this specification.
Examples of the storage medium for supplying the program code include a flexible disk, hard disk, magneto-optical disk, optical disk (e.g., CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD + RW), magnetic tape, nonvolatile memory card, and ROM. Alternatively, the program code may be downloaded from a server computer via a communications network.
Further, it should be clear that the functions of any one of the above-described embodiments may be implemented not only by executing the program code read out by the computer, but also by causing an operating system or the like operating on the computer to perform a part or all of the actual operations based on instructions of the program code.
Further, it is to be understood that the program code read out from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then causes a CPU or the like mounted on the expansion board or the expansion module to perform part or all of the actual operations based on instructions of the program code, thereby realizing the functions of any of the above-described embodiments.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one of 8230" does not exclude the presence of additional like elements in a process, method, article, or apparatus comprising the element.
Those of ordinary skill in the art will understand that: all or part of the steps for realizing the method embodiments can be completed by hardware related to program instructions, the program can be stored in a computer readable storage medium, and the program executes the steps comprising the method embodiments when executed; and the aforementioned storage medium includes: ROM, RAM, magnetic or optical disks, etc. that can store program codes.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solutions of the present specification, and not to limit them; although the present description has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present specification.
Claims (8)
1. A method for predicting the health state of a lithium battery is characterized by comprising the following steps:
collecting cyclic aging data of a target lithium battery in a plurality of charging and discharging cyclic processes; wherein the cyclic aging data comprises raw voltage data and raw temperature data;
for each of the charge and discharge cycles, performing: cleaning the original voltage data and the original temperature data in the current charging and discharging cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data;
performing feature extraction on the DTV data of each charge-discharge cycle to obtain time sequence data;
inputting the time sequence data into a preset LSTM model for training to obtain a prediction model;
predicting the health state of the lithium battery to be tested by using the prediction model and quantifying an uncertainty result;
inputting the time series data into a preset LSTM model for training to obtain a prediction model, wherein the method comprises the following steps:
constructing at least two time series data to obtain a three-dimensional tensor;
inputting the plurality of constructed three-dimensional tensors into a preset LSTM model for training to obtain a prediction model;
the LSTM model comprises an input layer, a first LSTM layer, a first dropout layer, a second LSTM layer, a second dropout layer, a full connection layer and an output layer which are sequentially connected, wherein an activation function in the first LSTM layer and an activation function in the second LSTM layer are tanh functions, and an activation function in the full connection layer is a sigmoid function;
the gradient descent algorithm of the LSTM model is an RMSprop algorithm;
and the LSTM model adopts a Bayesian algorithm to carry out parameter optimization.
2. The method of claim 1, wherein the cleaning process comprises at least one of: outliers are screened out, sampling intervals are fixed, and filtering is performed.
3. The method of claim 1, wherein the performing feature extraction on the DTV data for each charge and discharge cycle to obtain time series data comprises:
for each of the charge and discharge cycles, performing: sequentially carrying out filtering processing and connection processing on the DTV data of the current charge-discharge cycle to obtain a DTV curve; extracting the size and the position of a peak and the size and the position of a trough in the DTV curve, and taking the sizes and the positions as initial characteristic data; performing correlation analysis on the battery capacity of the current charge-discharge cycle and the size and the position of the current peak or the current trough aiming at each peak or each trough to obtain target characteristic data with the correlation higher than a preset threshold value; wherein the battery capacity is derived from the cyclic aging data;
and arranging the target characteristic data of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
4. The method of claim 1, wherein the performing feature extraction on the DTV data for each charge and discharge cycle to obtain time series data comprises:
for each of the charge and discharge cycles, performing: constructing DTV data of the current charge-discharge cycle into a column vector; performing singular value decomposition on the column vector to obtain a singular value;
and arranging the singular values of different charge-discharge cycles according to a time sequence to obtain time sequence data in a two-dimensional matrix.
5. The method according to any one of claims 1 to 4, wherein the predicting the state of health of the lithium battery to be tested and quantifying uncertainty results by using the prediction model comprises:
the health state of the lithium battery to be tested is predicted for multiple times by using the prediction model to obtain multiple prediction results;
and carrying out uncertainty quantification on the plurality of prediction results by adopting a Monte Carlo method to obtain uncertainty results.
6. A lithium battery state of health's prediction unit characterized in that includes:
the acquisition module is used for acquiring cyclic aging data of the target lithium battery in the process of multiple charging and discharging cycles; wherein the cyclic aging data comprises raw voltage data and raw temperature data;
and the execution module is used for executing the following steps aiming at each charge and discharge cycle: cleaning the original voltage data and the original temperature data in the current charge-discharge cycle to obtain target voltage data and target temperature data; obtaining DTV data of the current charge-discharge cycle based on the target voltage data and the target temperature data;
the characteristic extraction module is used for extracting the characteristics of the DTV data of each charge-discharge cycle to obtain time series data;
the training module is used for inputting the time series data into a preset LSTM model for training to obtain a prediction model;
the prediction module is used for predicting the health state of the lithium battery to be tested by using the prediction model and quantifying an uncertainty result;
the training module is used for executing the following operations:
constructing at least two time series data to obtain a three-dimensional tensor;
inputting the plurality of constructed three-dimensional tensors into a preset LSTM model for training to obtain a prediction model;
the LSTM model comprises an input layer, a first LSTM layer, a first dropout layer, a second LSTM layer, a second dropout layer, a full connection layer and an output layer which are sequentially connected, wherein an activation function in the first LSTM layer and an activation function in the second LSTM layer are tanh functions, and an activation function in the full connection layer is a sigmoid function;
the gradient descent algorithm of the LSTM model is an RMSprop algorithm;
and the LSTM model adopts a Bayesian algorithm to carry out parameter optimization.
7. An electronic device comprising a memory in which a computer program is stored and a processor which, when executing the computer program, carries out the method according to any one of claims 1-5.
8. A computer-readable storage medium, on which a computer program is stored which, when executed in a computer, causes the computer to carry out the method of any one of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211321901.1A CN115389947B (en) | 2022-10-27 | 2022-10-27 | Lithium battery health state prediction method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211321901.1A CN115389947B (en) | 2022-10-27 | 2022-10-27 | Lithium battery health state prediction method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115389947A CN115389947A (en) | 2022-11-25 |
| CN115389947B true CN115389947B (en) | 2023-01-03 |
Family
ID=84127675
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211321901.1A Active CN115389947B (en) | 2022-10-27 | 2022-10-27 | Lithium battery health state prediction method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115389947B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111308377A (en) * | 2020-03-10 | 2020-06-19 | 北京理工大学 | Battery health state detection method and system based on temperature voltage differentiation |
| CN113917337A (en) * | 2021-10-13 | 2022-01-11 | 国网福建省电力有限公司 | Battery health state estimation method based on charging data and LSTM neural network |
| CN114441984A (en) * | 2022-01-27 | 2022-05-06 | 泉州装备制造研究所 | Lithium battery health state estimation method |
| CN114580496A (en) * | 2022-01-24 | 2022-06-03 | 北京航空航天大学 | Self-detection method for single fault of lithium ion battery |
| CN114814618A (en) * | 2022-05-31 | 2022-07-29 | 上海芯钛信息科技有限公司 | Lithium ion battery residual capacity estimation method, device, equipment and storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220284747A1 (en) * | 2021-03-08 | 2022-09-08 | Toyota Research Institute, Inc. | System and method for forecasting battery state with imperfect data |
-
2022
- 2022-10-27 CN CN202211321901.1A patent/CN115389947B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111308377A (en) * | 2020-03-10 | 2020-06-19 | 北京理工大学 | Battery health state detection method and system based on temperature voltage differentiation |
| CN113917337A (en) * | 2021-10-13 | 2022-01-11 | 国网福建省电力有限公司 | Battery health state estimation method based on charging data and LSTM neural network |
| CN114580496A (en) * | 2022-01-24 | 2022-06-03 | 北京航空航天大学 | Self-detection method for single fault of lithium ion battery |
| CN114441984A (en) * | 2022-01-27 | 2022-05-06 | 泉州装备制造研究所 | Lithium battery health state estimation method |
| CN114814618A (en) * | 2022-05-31 | 2022-07-29 | 上海芯钛信息科技有限公司 | Lithium ion battery residual capacity estimation method, device, equipment and storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于Stacking模型集成的LSTM网络短期负荷预测研究;丁斌等;《中国测试》;20200731;第46卷(第07期);第40-45页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115389947A (en) | 2022-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Luo et al. | A review of deep learning approach to predicting the state of health and state of charge of lithium-ion batteries | |
| Park et al. | LSTM-based battery remaining useful life prediction with multi-channel charging profiles | |
| Wang et al. | A critical review of online battery remaining useful lifetime prediction methods | |
| CN108896914B (en) | Gradient lifting tree modeling and prediction method for health condition of lithium battery | |
| Vanem et al. | Data-driven state of health modelling—A review of state of the art and reflections on applications for maritime battery systems | |
| CN113219343A (en) | Lithium battery health state prediction method, system, equipment and medium based on elastic network | |
| CN111999648A (en) | Lithium battery residual life prediction method based on long-term and short-term memory network | |
| CN114047452B (en) | Method and device for determining cycle life of battery | |
| CN112798960A (en) | Battery pack residual life prediction method based on migration deep learning | |
| CN115407211B (en) | Online prediction method and system for health state of lithium battery of electric vehicle | |
| CN112611976A (en) | Power battery state of health estimation method based on double differential curves | |
| CN115994441A (en) | Big data cloud platform online battery life prediction method based on mechanism information | |
| WO2022229611A1 (en) | Health monitoring of electrochemical energy supply elements | |
| CN115308608A (en) | All-vanadium redox flow battery voltage prediction method, device and medium | |
| CN114579644A (en) | Method, apparatus and medium for battery efficient charging data identification based on deep learning | |
| Xu et al. | State-of-charge estimation and health prognosis for lithium-ion batteries based on temperature-compensated Bi-LSTM network and integrated attention mechanism | |
| CN116896139B (en) | Novel BMS system based on communication power supply system | |
| CN115389947B (en) | Lithium battery health state prediction method and device, electronic equipment and storage medium | |
| CN115693916B (en) | Intelligent online monitoring method and system for direct-current power supply of transformer substation | |
| CN116643190A (en) | Real-time monitoring method and system for lithium battery health state | |
| CN116125279A (en) | Method, device, equipment and storage medium for determining battery health state | |
| Wang et al. | A conditional random field based feature learning framework for battery capacity prediction | |
| Zraibi et al. | Comparing single and hybrid methods of deep learning for remaining useful life prediction of lithium-ion batteries | |
| Uddin et al. | State of health estimation of lithium-ion batteries in vehicle-to-grid applications using recurrent neural networks for learning the impact of degradation stress factors | |
| CN117630682B (en) | Random degradation process-based RUL prediction method for lithium ion battery |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |