CN110888059B - Charge state estimation algorithm based on improved random forest combined volume Kalman - Google Patents

Charge state estimation algorithm based on improved random forest combined volume Kalman Download PDF

Info

Publication number
CN110888059B
CN110888059B CN201911219341.7A CN201911219341A CN110888059B CN 110888059 B CN110888059 B CN 110888059B CN 201911219341 A CN201911219341 A CN 201911219341A CN 110888059 B CN110888059 B CN 110888059B
Authority
CN
China
Prior art keywords
algorithm
estimation
charge
state
power battery
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
Application number
CN201911219341.7A
Other languages
Chinese (zh)
Other versions
CN110888059A (en
Inventor
寇发荣
王甜甜
张宏
王思俊
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.)
Xian University of Science and Technology
Original Assignee
Xian University of Science and Technology
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 Xian University of Science and Technology filed Critical Xian University of Science and Technology
Priority to CN201911219341.7A priority Critical patent/CN110888059B/en
Publication of CN110888059A publication Critical patent/CN110888059A/en
Application granted granted Critical
Publication of CN110888059B publication Critical patent/CN110888059B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • B60L58/12—Methods or circuit arrangements for monitoring or controlling batteries or fuel cells, specially adapted for electric vehicles for monitoring or controlling batteries responding to state of charge [SoC]
    • 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]
    • G01R31/382—Arrangements for monitoring battery or accumulator variables, e.g. SoC
    • G01R31/3842—Arrangements for monitoring battery or accumulator variables, e.g. SoC combining voltage and current measurements
    • 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

Abstract

The invention discloses an algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation, which is used for solving the problem of accurate estimation of the state of charge of a power battery in working. The method combines random forest regression and a cubature Kalman filtering algorithm to jointly estimate the state of charge of the power battery, and weights and optimizes parameters of the random forest by a search algorithm containing a taboo whale so as to optimize the pruning threshold, the number of pretest samples and the number of decision trees of the algorithm, so that the optimization algorithm can quickly find the optimal solution, and the algorithm efficiency is improved; the residual service life of the power battery is predicted by memorizing the artificial neural network in a bidirectional long-time and short-time manner, so that the aims of correcting the maximum available capacity of the battery and improving the estimation precision of the state of charge of the power battery under the full-time working condition are fulfilled; the combined estimation algorithm integrates two algorithms of random forest and cubature Kalman filtering, the advantages of the two algorithms are brought into play, the defects of the two algorithms are avoided, and the estimation precision of the power battery charge state is higher.

Description

Charge state estimation algorithm based on improved random forest combined volume Kalman
Technical Field
The invention relates to the technical field of battery state of charge estimation of power battery systems of electric vehicles, in particular to an algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation.
Background
The near-pure electric vehicle (BEV), as one of new energy vehicles, is an automobile that uses a power battery as an energy storage power source and provides electric energy to a motor through the power battery, so as to propel the automobile to travel. Compared with the traditional diesel locomotive, the diesel locomotive has the characteristics of no emission, low noise, high energy conversion efficiency and the like, and gradually replaces the traditional diesel locomotive to become the mainstream form of vehicles in the future.
In the field of electric automobiles, a power battery is an indispensable part in the three-electricity technology and provides motion energy of the whole vehicle system. The lithium ion power battery has the advantages of high energy density, high power density, long service life, high safety, high reliability, low self-discharge rate, light weight, no memory and the like. Because the lithium ion power Battery has the defects of irreversible overcharge and overdischarge, severe characteristic change along with temperature change and the like, a complete Battery Management System (BMS) needs to be equipped so as to feed back and control the real-time state of the Battery pack and ensure the safety and reliability of the power Battery pack.
The State of Charge (SOC) is the most important parameter in the power battery management system, and is also the most important part in the battery State detection function, and can only be estimated according to a model or a corresponding algorithm. However, due to the fact that the interior of the chemical battery is complex, the quantity of internal measurable parameters is very limited, characteristics of the chemical battery are mutually coupled, namely the parameters are attenuated immediately after use, strong time variation and high nonlinearity, and in addition, parameters such as current and temperature under the working condition of an actual vehicle are wide in variation range and high in variation rate, and the research of an estimation algorithm with high precision and high robustness is the key point of the estimation of the state of charge of the power battery.
The electrochemical reaction process inside the power lithium ion battery is complex, the actual working condition is complex and severe, and the estimation method of the charge state of the invisible state quantity can be roughly divided into four categories: an ampere-hour integration method, a characterization parameter method, a model-based method and a data-driven method. The estimation values based on a residual capacity method, an impedance spectroscopy method and an open-circuit voltage method in the characterization parameter method are very accurate, but all the estimation values need to be calibrated in a laboratory environment, otherwise, the precision cannot be guaranteed; the charge state estimation based on the ampere-hour integration method has the mutual influence of factors such as acquisition of an initial charge state, sensor error accumulation, capacity decline and the like, and the precision of the charge state estimation is often difficult to ensure after the battery is used for a long time; the model-based estimation method usually needs to establish a power battery equivalent circuit model and a state equation thereof, apply a filtering algorithm and an observer and establish a state of charge estimation algorithm, and the estimation accuracy of the method is determined by an estimation process and a correction process, so that the accuracy is high; based on a data driving method, a mapping network between power battery parameters and the state of charge is established and trained through massive offline data, and the method has good advantages for solving the problem of high nonlinearity, and is high in estimation accuracy and fitting performance.
At present, a plurality of methods for estimating the state of charge exist, but all the methods are respectively a single method and have the advantages and the defects. The method combines a random forest and neural network algorithm based on a data driving method and a cubature Kalman filtering algorithm based on a model method, and effectively combines the advantages of the two algorithms through innovation switching, so that the state of charge estimation is more accurate. Parameters of random forests are weighted and optimized through a search algorithm containing the taboo whale so as to achieve optimization processing of an algorithm pruning threshold, the number of pretest samples and the number of decision trees, the optimization algorithm can quickly find an optimal solution, and algorithm efficiency is improved; the residual service life of the power battery is predicted by memorizing the artificial neural network in a bidirectional long-time and short-time manner, so that the aims of correcting the maximum available capacity of the battery and improving the estimation precision of the state of charge of the power battery under the full-time working condition are fulfilled; the combined estimation algorithm integrates two algorithms of random forest and cubature Kalman filtering, and the advantages of the two algorithms are adopted, so that the defects of the two algorithms are avoided, and the estimation precision of the power battery charge state is higher. The algorithm overcomes the influence of the fluctuation of working conditions and environments on the estimation precision of the state of charge, and improves the generalization and the robustness of the state of charge estimation.
Disclosure of Invention
In order to solve the defects in the prior art, the invention provides an algorithm based on the improved random forest combined Kalman volume Kalman power battery state of charge estimation, and aims to solve the problem of accurate estimation of the state of charge of a power battery in working.
In order to achieve the purpose, the invention adopts the technical scheme that:
an algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation is characterized in that: the algorithm for improving the estimation of the state of charge of the random forest combined cubature Kalman power battery comprises the following steps:
step one, preparation work: performing data offline acquisition on the power lithium ion battery under a cycle test working condition, and training a random forest based on optimization of a taboo whale optimization algorithm by using offline data;
step two, preparation work: and performing data offline acquisition on the power lithium ion battery under the cyclic test working condition, and training a bidirectional long-time and short-time memory artificial neural network by using offline data to complete the construction of an offline partial model.
Thirdly, collecting parameters of the power battery on line under the working condition of the real vehicle, and realizing the prediction of the residual service life of the battery through a bidirectional long-time memory artificial neural network; and inputting parameters such as residual life, current, voltage, temperature and the like into a random forest model under the whale taboo optimization, and realizing the real-time estimation of the charge state of the power battery.
Fourthly, power battery parameters are acquired on line under the working condition of the real vehicle, the parameters are identified on line through a recursive least square method containing forgetting factors, and the posterior estimation of the power battery charge state by the volume Kalman filter is realized through the identified parameters;
and fifthly, predicting the residual life of the battery by using a bidirectional long-time memory artificial neural network, further obtaining the maximum available capacity through a corresponding formula, and correcting the prior estimation of the charge state of the volume Kalman filter.
And step six, using the innovation of the cubature Kalman filtering algorithm as a judgment standard, and fusing the two algorithms to realize more accurate state of charge estimation.
Further, the step one specifically includes the steps of: the method comprises the steps of collecting external characteristic data of the power lithium ion battery, establishing an SOC estimation random forest model, and collecting the historical capacity of the power lithium ion battery.
Further, the second step specifically includes the following steps: the method comprises the steps of conducting RUL prediction, establishing a bidirectional LSTM neural network RUL prediction model in the second step, and conducting a whale taboo search algorithm optimization algorithm in the third step.
Further, the fourth step specifically includes the following steps: the method comprises the steps of firstly establishing least square method model parameter online identification containing forgetting factors, secondly establishing a second-order equivalent model, thirdly performing model-based ampere-hour method SOC prior estimation and fourthly performing volume Kalman filtering algorithm SOC posterior estimation.
Further, the step five specifically comprises the following steps: the method comprises the steps of initializing a bidirectional LSTM neural network structure, training a bidirectional cyclic neural network, performing unidirectional prediction, performing bidirectional prediction in a fourth step, and performing RUL prediction in a fifth step.
Further, the sixth step specifically includes the following steps: the first step is to perform an innovation switching algorithm and the second step is to obtain an SOC estimation result.
The invention has the beneficial effects that: the method combines random forest regression and a cubature Kalman filtering algorithm to jointly estimate the state of charge of the power battery, and weights and optimizes parameters of the random forest by a search algorithm containing a taboo whale so as to optimize the pruning threshold, the number of pretest samples and the number of decision trees of the algorithm, so that the optimization algorithm can quickly find the optimal solution, and the algorithm efficiency is improved; the residual service life of the power battery is predicted by memorizing the artificial neural network in a bidirectional long-time and short-time manner, so that the aims of correcting the maximum available capacity of the battery and improving the estimation precision of the state of charge of the power battery under the full-time working condition are fulfilled; the combined estimation algorithm integrates two algorithms of random forest and cubature Kalman filtering, the advantages of the two algorithms are brought into play, the defects of the two algorithms are avoided, and the estimation precision of the power battery charge state is higher.
The foregoing description is only an overview of the technical solutions of the present invention, and in order to clearly understand the technical solutions of the present invention and to implement the technical solutions according to the contents of the description, the following detailed description is given with reference to the preferred embodiments of the present invention and the accompanying drawings.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a flow chart of algorithm for optimizing whale by the taboo search algorithm of the invention.
FIG. 2 is a flow chart of optimizing random forests by a whale taboo algorithm according to the invention.
FIG. 3 is a flow chart of the bidirectional long-time and short-time memory artificial neural network of the present invention.
FIG. 4 is a flow chart of estimating the state of charge of the power battery by the cubature Kalman filter.
FIG. 5 is a flow chart of an algorithm based on improved stochastic forest combined cubature Kalman power battery state of charge estimation.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. 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.
Example 1:
as shown in fig. 1 and fig. 2, an algorithm based on improved random forest association volumetric kalman power battery state of charge estimation is characterized in that: the algorithm for improving the estimation of the state of charge of the random forest combined cubature Kalman power battery comprises the following steps:
step one, off-line training based on a data driving method:
1. let there be different samples in the set { x1,x2,...,xnIf there is one sample taken from the set S at a time with a put back, a total of n times are taken, thereby forming a new set S1Then S is1The probability of not including a sample is:
Figure GDA0003506344230000061
when n → ∞ is reached,
Figure GDA0003506344230000062
though, the new set S1The number of samples is n, equal to the number of samples in the original set S, but the new set may contain repeated samples, and if the repeated samples are removed, the new set S1Only 1-0.368 × 100% in the original set S is included in the set S, which is 63.2%. The detailed process is as follows:
1) resampling by using a Bootstrap method, and randomly generating training sequences S1, S2, … and ST;
2) generating corresponding decision trees C1, C2, … and CT by utilizing each training set; before selecting attributes on internal nodes, randomly extracting attributes from the attributes as a splitting attribute set of a current node, and splitting the node in the best splitting mode in the attributes (generally speaking, the value is kept unchanged in the whole forest growing process);
3) each tree grows intact without pruning;
4) for the test set sample, each decision tree is used for testing to obtain corresponding categories C1(X), C2(X), …, CT (X);
5) and adopting a voting method to take the category with the most output in the T decision trees as the category to which the test set sample belongs.
2. For describing the behavior of the whale, which is to surround a prey when hunting, the following mathematical model is proposed: d ═ CX '(t) -X (t) |, X (t +1) ═ X' (t) -AD, where t denotes the current number of iterations; a and C represent coefficients; x' (t) represents the eyeThe best previous whale position vector; x (t) represents the position vector of the current whale; a2 a rand1-a，C＝2·rand2，
Figure GDA0003506344230000071
Wherein the value of a decreases linearly from 2 to 0; t represents the current number of iterations; t ismaxIs the maximum iteration number; according to the hunting behavior of whale, which is a spiral sport game towards a prey, the mathematical model of the hunting behavior is as follows:
Figure GDA0003506344230000072
wherein D isp| X' (t) -X (t) | denotes the distance between whale and prey; x' (t) represents the best position vector so far; notably, whales swim to the prey in a spiral shape while contracting the enclosure. Therefore, in this synchronous behavior model, the assumption of some probability selects the shrink wrap-around mechanism and 1-PiThe spiral model is selected to update the whale position, and the mathematical model is as follows:
Figure GDA0003506344230000073
when the prey is attacked, a value for decreasing a is set close to the prey on the mathematical model, so that the fluctuation range of A also decreases with a. When the value of a decreases from 2 to 0 during the iteration, A is at [ -a, a [ -a [ ]]When A is at [ -1,1 [ -1 [ ]]While the next position of the whale can be anywhere between its present position and the location of the prey, the algorithm sets that when a < 1, the whale makes an attack on the prey.
In practical use, the random forest algorithm adopted by the invention has a large number of parameters, and no definite parameter selection rule exists for different training sample sets. According to the method, a whale taboo weighting optimization random forest algorithm is adopted to achieve optimization processing of the algorithm pruning threshold, the number of pretest samples and the number of decision trees, the optimization algorithm can quickly find the optimal solution, random parameter selection is avoided, and algorithm efficiency is improved.
1) Initializing according to empirical parameters;
2) selecting a training set and a test set according to a resampling method;
3) after all decision trees are generated, testing the result of each tree and calculating the corresponding weight;
4) calculating a regression result under the initial parameter;
5) and taking the regression result as a fitness value, performing iterative optimization on the initial parameters by adopting a taboo whale optimization algorithm, and comparing the initial parameters with historical results to select optimal model parameters.
6) Initializing parameters: i.e. the size SN of the whale population and the maximum iteration number TmaxMaximum value w ', minimum value w' of inertial weight, logarithmic spiral shape constant b, random number rand1rand2rand3Initial iteration times;
7) calculating the corresponding fitness value of each whale, sorting according to the fitness value, and selecting SN as initial populations;
8) calculating the sizes of the SN individual fitness values, and finding out the individual position with the minimum fitness value as the optimal position;
9) when A is more than or equal to 1, updating the position of the next generation, and when A is less than 1, updating the position of the next generation by adopting the improved position vector;
10) entering a tabu search stage, judging whether the convergence criterion is met, and if so, outputting a result; if not, generating a candidate solution;
11) judging whether the candidate solution meets the scofflaw criterion, if so, taking the solution of the scofflaw criterion as the current solution, otherwise, taking the optimal solution of the non-taboo object as the current solution, and judging whether the scofflaw criterion is met;
12) if the terminal condition is reached, outputting the optimal individual, namely the optimal solution found by the algorithm, and outputting the position and the fitness value of the optimal individual;
13) and establishing an optimal random forest by using the optimal parameters, and predicting by using the model.
Example 2:
an algorithm based on the improved random forest combined cubature Kalman power battery state of charge estimation as shown in FIG. 2 is characterized in that: the method comprises the following steps:
step two, bidirectional long-time and short-time memory circulation artificial neural network
On a time scale, the aging of a power battery is a long process, and thousands of charge and discharge cycles are covered, and a capacity prediction method based on historical data is often caused by noise doped in the acquisition process of the historical data. In order to solve the above problems, the present patent proposes a method for predicting the RUL of a power battery based on a timing prediction idea, in which the mixing of abnormal data has a certain influence on the extraction of power battery capacity information: a bi-directional recurrent neural network. The forgetting gate can discard redundant information in data noise, abnormal values and adjacent cycle numbers, so that the prediction accuracy of the attenuation sequence is improved; the input gate needs to decide which information needs to be saved in the internal state; the output gate determines the problem of the current output information.
Wherein x istIs the neuron input at time t; h ist-1Is the hidden layer information at time t-1; scIs an internal energy state; forget the door:
Figure GDA0003506344230000091
is forgotten gate input; wfXAnd WfhAre respectively input xtAnd ht-1The forget gate weight value; bfIs the forgetting gate threshold. An input gate:
Figure GDA0003506344230000092
Figure GDA0003506344230000093
and
Figure GDA0003506344230000094
respectively corresponding to gate input information under sigma function transformation and tanh function transformation; wiXAnd WgXRespectively corresponding to x in the two-part transformationtInputting a gate weight value; wihAnd WghRespectively corresponding to h in the two-part transformationt-1Inputting a gate weight value; biAnd bgRespectively corresponding to the input gate threshold values in the two part transformation; inner partAnd (3) updating the state:
Figure GDA0003506344230000095
an output gate:
Figure GDA0003506344230000096
Figure GDA0003506344230000097
WoXand WohRespectively are the input output weight; boIs the output gate threshold. When the RUL predictor is started, the BMS needs to perform structure initialization on the LSTM RNN, specifically including setting parameters such as input, output, neuron number, hidden layer number, and activation function type of the deep neural network. The BMS then begins extracting historical capacity data for the power cell and constructing training samples for the LSTM RNN, each training sample having as an input a capacity value at a previous time and as an output a capacity value at a next time during each training. After the construction of the training samples is completed, the network needs to be trained in order to obtain a timing model. Because the traditional batch gradient descent and random gradient descent methods are difficult to adapt to deep learning environments, the convergence rate of the network weight coefficient is easy to reduce; moreover, in order to avoid the over-fitting problem, an optimization method of Root Mean Square backward propagation (Root Mean Square prop) is adopted to ensure the convergence rate of the network weight coefficient, and the calculation process of the neural network parameters is as follows:
Figure GDA0003506344230000101
Figure GDA0003506344230000102
wherein x and y are the input and output of the network, respectively; [ x ] ofi:i+N,yi:i+N]Representing the Nth sample in each small batch of samples; j [ theta ] thetat；xi:i+N,yi:i+N]Is the objective function for each small batch; gamma raylstmIs the coefficient that determines the mean of the squared gradient; etalstmIs the learning efficiency of the training algorithm; epsilonlstmIs to avoid a smoothing term with a divisor of zero; is composed ofThe overfitting is prevented, and the sensitivity of the neurons in the neural network to a certain specific weight is reduced by adopting the Dropout technology, so that the effect of preventing overfitting training is achieved. A part of neurons are randomly discarded in the training process to achieve a large-scale sparse network training process with a large amount of weight sharing.
In actual use, after completing network training of the LSTM RNN, the BMS needs to input historical capacity data into the network for multiple forward recursive predictions until the predicted capacity value is below a failure threshold. Then, counting the recursion steps experienced during the recursion prediction period and taking the counted recursion steps as the RUL of the power battery.
The probability density function calculation predicted by the method is completely consistent with the solving process of the time sequence prediction method. And randomly generating a plurality of capacity sequence samples according to the statistical characteristics of the historical capacity data adjacent to the prediction starting point, then respectively inputting each group of samples into the LSTM RNN based on the idea of the MC method, and performing forward multi-step prediction simulation to further obtain N RUL simulation prediction values.
Example 3:
an algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation is characterized in that: the method comprises the following steps:
step three: online estimation is driven based on data.
In actual use, the parameters of the power battery are collected on line under the working condition of an actual vehicle, and the residual service life of the battery is predicted by memorizing an artificial neural network in a bidirectional long-time and short-time manner; and inputting parameters such as residual life, current, voltage, temperature and the like into a random forest model under the whale taboo optimization, and realizing the real-time estimation of the charge state of the power battery.
Example 4:
as shown in fig. 4, an algorithm based on improved random forest association cubkalman power battery state of charge estimation is characterized in that: the method comprises the following steps:
step four: cubature kalman filter posteriori estimation
The non-linear filtering in the Gaussian domain solves the problem of how to calculate the productThe problem of points, the integral of which exists in the form of a non-linear function gaussian density, in particular, in a cartesian coordinate system, the following integral form is considered:
Figure GDA0003506344230000111
wherein I (f) is the integral sought; rnIs an n-dimensional integral domain; (x) is a non-linear function; x is a state vector; to compute the integral value of the expression, the following two steps are taken: 1) transforming the formula into a sphere-radial integral form; 2) a third order spherical radial criterion is proposed.
In the sphere-radial transformation, the key step is to determine the vector x ∈ R in the Cartesian coordinate systemnThe process of transformation into radius r and direction vector y is given by x ry, yTy 1, r ∈ [0, + ∞), thus xTx＝r2R ∈ [0, ∞). The sphere-radial coordinate system can be expressed as:
Figure GDA0003506344230000112
in the formula of Un＝{y∈Rn|yTy is 1, which is the surface of the sphere; delta is the integral domain UnThe infinitesimal of (1); form of radial integration:
Figure GDA0003506344230000113
in the formula, s (r) is defined by a spherical integral with a unit weight function w (y) of 1. Definition of spherical integral:
Figure GDA0003506344230000121
assuming that the radial integral is numerically calculated by the point gaussian integral criterion:
Figure GDA0003506344230000122
assuming that the spherical integral is numerically calculated by the point sphere criterion:
Figure GDA0003506344230000123
(mr×ms) The sphere-radial volume criterion for a point is:
Figure GDA0003506344230000124
for the third-order spherical radial criterion, mr＝1，ms2n, containing a total of 2n volume points.
Thus, the third-order spherical radial criterion is extended to calculate the standard gaussian weighted integral, as follows:
Figure GDA0003506344230000125
in the formula (I), the compound is shown in the specification,
Figure GDA0003506344230000126
is a volume point set;
Figure GDA0003506344230000127
a weight corresponding to each volume point; m is the number of volume points, and when the three-order spherical radial criterion is applied, the number of the volume points is 2 times of the state vector dimension n.
[1]iFor the ith volume point, as follows:
Figure GDA0003506344230000128
thus, for a non-linear system such as an observation equation and a state equation, the steps of the volume kalman filtering are as follows: 1) initialization: quantity of initialized state
Figure GDA0003506344230000129
Error covariance PKProcess and measurement noise Q, r.; 2) calculating a volume point:
Figure GDA00035063442300001210
ⅰ＝1，2,....,2n，εiwherein n is the dimension of the state quantity; epsiloniIs a volume point set; as follows:
Figure GDA00035063442300001211
wherein [1 ]]Representing an identity matrix; 3) propagation volume point:
Figure GDA00035063442300001212
4) form of calculationThe state quantity predicted value and the error covariance predicted value are as follows:
Figure GDA00035063442300001213
5) calculating a volume point:
Figure GDA00035063442300001214
6) propagation volume point:
Figure GDA00035063442300001215
7) calculating a measurement predicted value:
Figure GDA00035063442300001216
8) calculating the measurement error covariance and cross covariance:
Figure GDA0003506344230000131
9) calculating Kalman gain, updating state quantity and corresponding error covariance:
Figure GDA0003506344230000132
Figure GDA0003506344230000133
example 5:
an algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation is characterized in that: the method comprises the following steps:
and fifthly, predicting the residual life of the battery by using a bidirectional long-time memory artificial neural network, further obtaining the maximum available capacity through a corresponding formula, and correcting the prior estimation of the charge state of the volume Kalman filter.
In practical use, the ampere-hour integration method adopted by the patent is used as the prior of the whole cubature Kalman filtering algorithm to estimate the state of charge of the power battery, and has congenital defects. The storage capacity of the power battery is reduced along with the continuous aging of the battery, and the reduction of the storage capacity means the reduction of the maximum available capacity of the power battery. The ampere-hour integral estimation battery state-of-charge algorithm depends on the accuracy of the maximum available capacity. Therefore, the prediction of the residual service life of the battery is realized by using the bidirectional long-time memory artificial neural network, the maximum available capacity is further obtained through a corresponding formula, the prior estimation of the charge state of the volume Kalman filter is corrected, and the accuracy of the prior estimation can be improved.
Example 6:
as shown in fig. 5, an algorithm based on improved random forest association volumetric kalman power battery state of charge estimation is characterized in that: the method comprises the following steps:
and step six, using the innovation of the cubature Kalman filtering algorithm as a judgment standard, and fusing the two algorithms to realize more accurate state of charge estimation.
In practical use, measurement errors inevitably exist in the battery terminal voltage and the terminal current, and the established model cannot completely and accurately express a complex dynamic battery system, so that the estimation of the state of charge by adopting a volume Kalman filtering algorithm based on a model method has inherent defects. The estimation of the state of charge of the power battery by the data driving method can effectively compensate the defect time interval based on the model method, and the combined estimation of the two algorithms can realize more accurate state of charge estimation.
1) The innovation is calculated from a volumetric kalman filter prior estimate:
Figure GDA0003506344230000141
2) if the innovation is less than 0.001V, the posterior estimation of the state of charge is realized by continuously using a volume Kalman filtering algorithm;
3) and if the innovation is more than 0.001V, estimating the state of charge by adopting a random forest algorithm based on a data-driven model, namely, under the optimization of a taboo whale optimization algorithm.
In conclusion, the invention adopts the algorithm for improving the estimation of the state of charge of the random forest combined cubature Kalman power battery to solve the problem of accurate estimation of the state of charge of the power battery in working. The method combines random forest regression and a cubature Kalman filtering algorithm to jointly estimate the state of charge of the power battery, and weights and optimizes parameters of the random forest by a search algorithm containing a taboo whale so as to optimize the pruning threshold, the number of pretest samples and the number of decision trees of the algorithm, so that the optimization algorithm can quickly find the optimal solution, and the algorithm efficiency is improved; the residual service life of the power battery is predicted by memorizing the artificial neural network in a bidirectional long-time and short-time manner, so that the aims of correcting the maximum available capacity of the battery and improving the estimation precision of the state of charge of the power battery under the full-time working condition are fulfilled; the combined estimation algorithm integrates two algorithms of random forest and cubature Kalman filtering, and the advantages of the two algorithms are adopted, so that the defects of the two algorithms are avoided, and the estimation precision of the power battery charge state is higher.
It should be noted that all the directional indicators (such as up, down, left, right, front, and rear … …) in the embodiment of the present invention are only used to explain the relative position relationship between the components, the movement situation, etc. in a specific posture (as shown in the drawing), and if the specific posture is changed, the directional indicator is changed accordingly.
In addition, the descriptions related to "first", "second", etc. in the present invention are for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature.
Technical solutions between various embodiments may be combined with each other, but must be realized by those skilled in the art, and when the technical solutions are contradictory or cannot be realized, such a combination should not be considered to exist, and is not within the protection scope of the present invention.

Claims (5)

1. An algorithm based on improved random forest combined cubature Kalman power battery state of charge estimation is characterized in that: the algorithm based on the improved random forest combined cubature Kalman power battery state of charge estimation comprises the following steps:
step one, preparation work: performing data offline acquisition on the power lithium ion battery under a cycle test working condition, and training a random forest based on optimization of a taboo whale optimization algorithm by using offline data;
step two, preparation work: performing data offline acquisition on the power lithium ion battery under a cycle test working condition, and training a bidirectional long-time and short-time memory artificial neural network by using offline data to complete the construction of an offline partial model;
thirdly, collecting parameters of the power battery on line under the working condition of the real vehicle, and realizing the prediction of the residual service life of the battery through a bidirectional long-time memory artificial neural network; inputting the parameters of residual life, current, voltage and temperature into a random forest model under the whale taboo optimization to realize the real-time estimation of the state of charge of the power battery;
fourthly, power battery parameters are acquired on line under the working condition of the real vehicle, the parameters are identified on line through a recursive least square method containing forgetting factors, and the posterior estimation of the power battery charge state by the volume Kalman filter is realized through the identified parameters;
step five, predicting the residual life of the battery by using a bidirectional long-time memory artificial neural network to obtain the maximum available capacity, and correcting the prior estimation of the charge state of the volume Kalman filtering;
step six, using innovation of a cubature Kalman filtering algorithm as a judgment standard, fusing algorithms of random forests under optimization based on the cubature Kalman filtering algorithm and a taboo whale optimization algorithm, and realizing more accurate state of charge estimation;
the innovation is calculated from the volumetric kalman filter prior estimate as:
Figure FDA0003506344220000011
if the innovation is less than 0.001V, the posterior estimation of the state of charge is realized by continuously using a volume Kalman filtering algorithm; and if the innovation is more than 0.001V, estimating the state of charge by adopting a random forest algorithm based on a data-driven model, namely, under the optimization of a taboo whale optimization algorithm.
2. The algorithm for improving the estimation of the state of charge of the stochastic forest joint cubature kalman power battery according to claim 1, wherein the step one comprises the following steps: the method comprises the steps of collecting external characteristic data of the power lithium ion battery, establishing an SOC estimation random forest model, and performing a whale taboo search algorithm optimization algorithm.
3. The algorithm for improving the estimation of the state of charge of the stochastic forest joint cubkalman power battery according to claim 1, wherein the second step specifically comprises the following steps: the first step is to predict RUL, and the second step is to establish a RUL prediction model of the bidirectional LSTM neural network.
4. The algorithm for improving the estimation of the state of charge of the stochastic forest joint cubkalman power battery according to claim 1, wherein the fourth step specifically comprises the following steps: the method comprises the steps of firstly establishing least square method model parameter online identification containing forgetting factors, secondly establishing a second-order equivalent model, thirdly performing model-based ampere-hour method SOC prior estimation and fourthly performing volume Kalman filtering algorithm SOC posterior estimation.
5. The algorithm for improving the estimation of the state of charge of the stochastic forest joint cubkalman power battery according to claim 1, wherein the fifth step specifically comprises the following steps: the method comprises the steps of initializing a bidirectional LSTM neural network structure, training a bidirectional cyclic neural network, performing unidirectional prediction, performing bidirectional prediction in a fourth step, and performing RUL prediction in a fifth step.
CN201911219341.7A 2019-12-03 2019-12-03 Charge state estimation algorithm based on improved random forest combined volume Kalman Active CN110888059B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911219341.7A CN110888059B (en) 2019-12-03 2019-12-03 Charge state estimation algorithm based on improved random forest combined volume Kalman

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911219341.7A CN110888059B (en) 2019-12-03 2019-12-03 Charge state estimation algorithm based on improved random forest combined volume Kalman

Publications (2)

Publication Number Publication Date
CN110888059A CN110888059A (en) 2020-03-17
CN110888059B true CN110888059B (en) 2022-04-19

Family

ID=69750096

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911219341.7A Active CN110888059B (en) 2019-12-03 2019-12-03 Charge state estimation algorithm based on improved random forest combined volume Kalman

Country Status (1)

Country Link
CN (1) CN110888059B (en)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111976543B (en) * 2020-07-20 2023-01-31 南京工程学院 Charging and discharging control method for battery pack of electric automobile
CN112100902B (en) * 2020-08-10 2023-12-22 西安交通大学 Lithium ion battery life prediction method based on flow data
CN111965562B (en) * 2020-10-20 2020-12-29 江苏慧智能源工程技术创新研究院有限公司 Method for predicting residual cycle life of lithium battery based on random forest model
CN112327170B (en) * 2020-11-13 2023-04-28 中汽研（天津）汽车工程研究院有限公司 Power battery full-period residual life estimation method based on neural network
CN112763917B (en) * 2020-12-04 2022-07-19 国网浙江省电力有限公司电力科学研究院 Method and system for correcting SOC (state of charge) of battery pack of energy storage power station in real time
CN113129927B (en) * 2021-04-16 2023-04-07 平安科技（深圳）有限公司 Voice emotion recognition method, device, equipment and storage medium
CN113253050B (en) * 2021-05-08 2022-07-12 南昌大学 Traveling wave fault location method based on whale optimization Kalman filtering algorithm
CN114755586B (en) * 2022-04-27 2023-08-01 电子科技大学 Lithium ion battery residual life prediction method
CN116911574B (en) * 2023-09-12 2024-02-02 华侨大学 Three-level supply chain optimization method and device based on whale algorithm and random forest

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102881022A (en) * 2012-07-20 2013-01-16 西安电子科技大学 Concealed-target tracking method based on on-line learning
CN107167741A (en) * 2017-06-06 2017-09-15 浙江大学 A kind of lithium battery SOC observation procedures based on neutral net
CN108872866A (en) * 2018-06-04 2018-11-23 桂林电子科技大学 A kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method
WO2019043600A1 (en) * 2017-08-29 2019-03-07 Emerson Electric (Us) Holding Corporation (Chile) Limitada Remaining useful life estimator
CN109506647A (en) * 2018-12-24 2019-03-22 哈尔滨工程大学 A kind of INS neural network based and magnetometer combined positioning method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102881022A (en) * 2012-07-20 2013-01-16 西安电子科技大学 Concealed-target tracking method based on on-line learning
CN107167741A (en) * 2017-06-06 2017-09-15 浙江大学 A kind of lithium battery SOC observation procedures based on neutral net
WO2019043600A1 (en) * 2017-08-29 2019-03-07 Emerson Electric (Us) Holding Corporation (Chile) Limitada Remaining useful life estimator
CN108872866A (en) * 2018-06-04 2018-11-23 桂林电子科技大学 A kind of charge states of lithium ion battery dynamic evaluation and long-acting prediction fusion method
CN109506647A (en) * 2018-12-24 2019-03-22 哈尔滨工程大学 A kind of INS neural network based and magnetometer combined positioning method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Online Parameter Identification and State of Charge Estimation of Lithium-Ion Batteries Based on Forgetting Factor Recursive Least Squares and Nonlinear Kalman Filter;Bizhong Xia等;《energies》;20181231;第11卷(第3期);1-23 *
基于随机加权自适应容积卡尔曼的电池ＳＯＣ估计;马建等;《中国公路学报》;20191130;第32卷(第11期);234-244 *
基于随机森林的锂离子电池荷电状态估算;韦振汉等;《广西师范大学学报（自然科学版）》;20181031;第36卷(第4期);27-33 *

Also Published As

Publication number Publication date
CN110888059A (en) 2020-03-17

Similar Documents

Publication Publication Date Title
CN110888059B (en) Charge state estimation algorithm based on improved random forest combined volume Kalman
CN110596593B (en) Lithium ion battery SOC estimation method based on intelligent adaptive extended Kalman filtering
CN110888058B (en) Algorithm based on power battery SOC and SOH joint estimation
CN111856287B (en) Lithium battery health state detection method based on stacked residual causal convolutional neural network
CN113702843B (en) Lithium battery parameter identification and SOC estimation method based on suburb optimization algorithm
CN114325450A (en) Lithium ion battery health state prediction method based on CNN-BilSTM-AT hybrid model
CN113253116A (en) Lithium ion battery state of charge estimation method and storage medium
CN110659722A (en) AdaBoost-CBP neural network-based electric vehicle lithium ion battery health state estimation method
CN113109717B (en) Lithium battery state of charge estimation method based on characteristic curve optimization
CN115856678A (en) Lithium ion battery health state estimation method
CN114726045B (en) Lithium battery SOH estimation method based on IPEA-LSTM model
CN114970332B (en) Lithium battery model parameter identification method based on chaotic quantum sparrow search algorithm
CN112834927A (en) Lithium battery residual life prediction method, system, device and medium
CN115219906A (en) Multi-model fusion battery state of charge prediction method and system based on GA-PSO optimization
CN114167295B (en) Lithium ion battery SOC estimation method and system based on multi-algorithm fusion
CN113033104B (en) Lithium battery state of charge estimation method based on graph convolution
Mazzi et al. Lithium-ion battery state of health estimation using a hybrid model based on a convolutional neural network and bidirectional gated recurrent unit
CN116840724A (en) Lithium ion battery health condition estimation and residual service life prediction method
CN116774088A (en) Lithium ion battery health state estimation method based on multi-objective optimization
CN114325393A (en) Lithium ion battery pack SOH self-adaptive estimation method based on PF and GPR
CN114895206A (en) Lithium ion battery SOH estimation method based on RBF neural network of improved wolf optimization algorithm
CN114966436A (en) Lithium battery state of charge prediction method, device, equipment and readable storage medium
CN114896865A (en) Digital twin-oriented self-adaptive evolutionary neural network health state online prediction method
CN117630682B (en) Random degradation process-based RUL prediction method for lithium ion battery
CN117110877A (en) BO-LSTM-based lithium ion battery residual service life prediction method

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