CN111680848A - Battery life prediction method based on prediction model fusion and storage medium - Google Patents

Battery life prediction method based on prediction model fusion and storage medium Download PDF

Info

Publication number
CN111680848A
CN111680848A CN202010733589.1A CN202010733589A CN111680848A CN 111680848 A CN111680848 A CN 111680848A CN 202010733589 A CN202010733589 A CN 202010733589A CN 111680848 A CN111680848 A CN 111680848A
Authority
CN
China
Prior art keywords
battery
model
capacity
long
memory network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010733589.1A
Other languages
Chinese (zh)
Inventor
于天剑
甘沁洁
成庶
伍珣
代毅
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.)
Central South University
Original Assignee
Central South University
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 Central South University filed Critical Central South University
Priority to CN202010733589.1A priority Critical patent/CN111680848A/en
Publication of CN111680848A publication Critical patent/CN111680848A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Secondary Cells (AREA)

Abstract

The invention provides a battery life prediction method based on prediction model fusion and a storage medium, wherein a long and short memory network model is nested in a particle filter model, the fusion model is simple in structure, the long and short memory network model is trained by using the existing historical data to obtain a state transfer equation of a degradation trend equation to determine the particle filter model, the problem that the particle filter model depends too much on an empirical model is solved, the particle filter model can obtain uncertain expression of the residual life by using the weighted sum of particles and the predicted value of approximate capacity, in addition, a new sample obtained on line is added to the original training sample to intensively retrain the model, so that the model parameters are updated timely, better adaptability is realized, and the residual cycle life prediction of a cadmium-nickel storage battery can be realized.

Description

Battery life prediction method based on prediction model fusion and storage medium
Technical Field
The invention belongs to the technical field of battery life prediction, and particularly relates to a battery life prediction method based on prediction model fusion and a storage medium.
Background
No matter in electric locomotive or diesel locomotive, the storage battery and the charger are connected in parallel to form an energy source of a locomotive control circuit, once the storage battery breaks down, normal use of lighting, wireless communication devices and emergency devices in the locomotive cannot be maintained, and great threat is brought to life and property safety of passengers. Research has revealed that most of the batteries for high-speed railway vehicles are alkaline cadmium-nickel batteries, and in actual use, the batteries are generally replaced according to the number of kilometers of operation or the service life. At the moment, the service life of the battery is always provided with a large margin, and the replacement in advance undoubtedly improves the application cost of the motor train unit. Therefore, the research of accurate and reliable life prediction models is not slow. At present, methods for predicting the life of a battery are roughly classified into two types: model driven and data driven.
The model driving method is based on the internal structure principle and the degradation mechanism of the storage battery to establish a service life prediction model. The model driving method is, as disclosed in the prior art, applied to a battery tomography measurement technology and an electrochemical performance measurement technology, a power battery cycle life prediction model is constructed according to the internal structure of a lithium battery, but is influenced by factors such as the type and the model of the battery, and the method is difficult to be applied to practice. The model-driven method disclosed in the second prior art is a degradation model that uses extended kalman filtering to estimate the health and remaining life of a fuel cell (PEMFC) on-line, and is robust to operating conditions. The model driving method provided by the prior art is a model based on a new Particle Filter (PF) framework that uses the current measurement value to resample state particles, can prevent degeneracy of the particles, and can also adaptively adjust the number of particles, suitable for online applications. The experimental results show that compared with other standard models, the model can obtain more accurate prediction results in shorter time. Although the prediction performance of the model driving method is improved more and more, the model driving method is too dependent on a failure mechanism, the prediction accuracy is greatly dependent on a used state model, the working environment factors of the storage battery are complex and changeable, and the establishment of an accurate degradation model is difficult.
The data driving method obtains a battery performance degradation rule by mining and analyzing failure data, and then predicts the service life of the battery. The data driving method comprises the following steps: in the fourth prior art, a real-time remaining service life RUL estimation method based on a Support Vector Machine (SVM) is provided, cyclic data of a lithium battery under different working conditions are analyzed, key features are extracted from voltage and temperature curves, and a model is trained by using the features, so that the purpose of predicting the RUL of the lithium battery is achieved; in the prior art, equivalent circuit model parameters and aging process data are combined, and a prognosis framework of the PF is improved by using a Relevance Vector Machine (RVM), so that the prediction accuracy is further improved, and the prediction uncertainty is reduced; in the prior art, the service life prediction is carried out by adaptively optimizing a long-short-term memory network (LSTM) by using an elastic mean square back propagation method, and a more accurate prediction result can be obtained by using the method compared with a support vector machine and a standard recurrent neural network. The data-driven model based on the neural network is relatively good in performance in the prior art, but the neural network has good learning capacity on historical data, is difficult to determine the network structure, has high requirements on the sample size and quality of the data, and does not have output uncertainty expression.
In addition, the battery life research in the prior art mainly aims at lithium batteries and fuel cells, and cadmium-nickel storage batteries have no related life research at present due to long service life test time and harsh test conditions. The cycle life of the storage battery used in the existing related research is below 1000 times, while the life cycle of the cadmium-nickel storage battery used for a certain type of motor train unit is up to above 2000 times, and the battery capacity is reduced below the standard. With the increase of the periodicity, the offline method cannot update the model, errors are accumulated, and the model is difficult to have better accuracy, while the online prediction model can update the model along with the update of data, and the prediction accuracy of the model is higher. In addition, the cadmium-nickel battery has the characteristic of memory effect, and a good prediction result is difficult to obtain by a general prediction method.
Disclosure of Invention
In view of the above, the present invention provides a battery life prediction method and a storage medium based on prediction model fusion, so as to solve the problem that it is difficult to establish an accurate degradation model due to over dependence on a failure mechanism when the battery life is predicted based on a single particle filter model in the prior art, solve the problem of uncertainty of output based on a single neural network prediction model, and solve the technical problem that the model cannot be updated online in the prior art.
A battery life prediction method based on prediction model fusion, wherein the prediction model comprises a particle filter model and a long and short memory network model, and the battery life prediction method comprises the following steps:
step 1: the historical data of the battery is acquired,
step 2: predicting the capacity state of the battery according to the historical data through the particle filter model to obtain a predicted capacity value of the battery,
and step 3: comparing the predicted value of the capacity with a preset capacity failure threshold value of the battery, judging that the battery is failed when the predicted value of the capacity reaches the capacity failure threshold value, finishing the iteration of the particle filter model, and obtaining the residual life of the battery according to the iteration times,
the long and short memory network model is embedded in the particle filter model, and the long and short memory network model is trained and learned by taking battery capacity state value data in the historical data as training sample numbers, so that the long and short memory network model predicts the capacity state of the battery according to the battery capacity historical data, constructs a state transition equation of the particle filter model according to the trained output value of the long and short memory network model, and obtains a priori predicted value of the capacity state of the battery.
Preferably, the history data includes data on the number of charge and discharge times of the battery and data on a capacity state value of the battery corresponding to each time the charge and discharge is performed,
the battery life prediction method further includes:
before training and learning a long and short memory network model through the battery capacity data, preprocessing the historical data, wherein the preprocessing comprises removing useless data in the historical data and normalizing the historical data.
Preferably, the step 2 includes:
step 21: a randomly generated set of capacity values of the battery at a predicted starting time as initial particle groups of the particle filter model, and assigning initial weight coefficients to the respective initial particle groups,
step 22: training and learning the long and short memory network model by adopting the training sample data, so that the long and short memory network model predicts the capacity state of the battery according to the battery capacity historical data,
step 23: constructing the state equation according to the output of the long and short memory network model and obtaining the prior predicted value at the current moment,
step 24: generating a new set of particles based on the a priori value at the current time and the set of particles at the previous time,
step 25: the weights of the new set of particles are updated,
step 26: and correcting the prior predicted value at the current moment according to the new particle group and the weight of the new particle group to obtain the posterior predicted value at the current moment.
Preferably, the method for predicting battery life further includes step 4, step 5 and step 6, if the determination result in step 3 is that the battery is not out of service, the step 4 and the step 5 are sequentially executed,
the step 4 is to obtain the measured value of the battery capacity online,
step 5 is to determine whether the difference between the predicted value and the new measured value at the current time is greater than a set error value, if so, step 6 is executed, otherwise, the iteration number of the particle filter model is increased by 1, and the step 23 is returned,
and 6, adding the measured values obtained in the step 4 into the training sample to update the training sample, and turning to the step 22 after the training sample is updated so as to retrain the long and short memory network model through the updated training sample.
Preferably, a dropout module is arranged in the long and short memory network model to prevent overfitting of the long and short memory network model.
Preferably, the capacity status value data in the history data is compared with the capacity status value data
And establishing a time sequence, and training the long and short memory network model by using the time sequence.
Preferably, the step of determining the state equation and the a priori predicted value according to the trained long and short memory network includes:
the trained long and short memory network model predicts the state value of the battery capacity at the current time according to the capacity state value data at m times before the current time,
superposing the output value of the long and short memory network model at the current moment and the process noise of the battery capacity degradation at the previous moment at the current moment as the state prediction value of the current moment in the state equation to construct the state equation,
and calculating to obtain the prior predicted value at the current moment according to the state equation and the posterior predicted value at the previous moment of the current moment.
Preferably, the weight of the new particle group is updated using the importance samples such that the closer to the predicted value of the battery capacity state, the larger the weight coefficient corresponding to the particle is.
Preferably, the battery is a cadmium-nickel storage battery, and the long and short memory network model comprises an input and output layer, a long and short memory network layer, a dropout layer and a full connection layer.
A storage medium, wherein the storage medium is a readable storage medium of a computer, and wherein a computer program stored on the readable storage medium is executed by a processor to implement the battery life prediction method according to any one of the above.
The invention has the following beneficial effects: the method for predicting the remaining life of the battery on line based on the fusion of the particle filter model and the long and short memory network model can realize the prediction of the remaining cycle life of the cadmium-nickel storage battery, the long and short memory network model is nested in the particle filter model, the fusion model is simple in structure, the long and short memory network model is trained by using the existing historical data to obtain a degradation trend equation to determine a state transfer equation of the particle filter model, the problem that the particle filter model depends on an empirical model too much is solved, the particle filter model can obtain uncertain expression of the remaining life by using the weighted sum of particles and the predicted value of approximate capacity, in addition, a new sample obtained on line is added to an original training sample to intensively retrain the model, so that the model parameters are updated timely, and the method has better adaptability.
Drawings
Fig. 1 is a schematic flow chart of a method for predicting battery life based on fusion of prediction models according to an embodiment of the present invention;
fig. 2 is a comparison diagram of the prediction effect of predicting the battery life according to the fusion prediction model LSTM-P provided by the present invention under the setting conditions that the prediction starting point is T1100 cycles and the cycle number is RUL 1742 cycles;
fig. 3 is a comparison graph of the prediction effect of the battery life by using a standard PF model under the setting conditions that the prediction starting point is T1100 cycles and the cycle number is RUL 1742 cycles;
fig. 4 is a comparison graph of the prediction effect of battery life prediction by using a standard LSTM model under the setting conditions that the prediction starting point is T1100 cycles and the cycle number is RUL 1742 cycles;
fig. 5 is a comparison diagram of the prediction effect of predicting the battery life according to the fusion prediction model LSTM-P provided by the present invention under the setting conditions that the prediction starting point is T-200 cycles and the cycle number is RUL-842 cycles;
fig. 6 is a comparison graph of the prediction effect of the battery life by using a standard PF model under the setting conditions that the prediction starting point is T200 cycles and the cycle number is RUL 842 cycles;
fig. 7 is a comparison graph of the prediction effect of battery life prediction using a standard LSTM model under the setting conditions of a prediction starting point of T200 cycles and a cycle number of RUL 842 cycles.
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 any creative effort, shall fall within the protection scope of the present invention. It should be noted that "…" in this description of the preferred embodiment is only for technical attributes or features of the present invention.
In order to solve the above problems in the prior art, the present invention provides a battery life prediction method based on prediction model fusion, which mainly fuses two prediction models, namely a Particle Filter (PF) model and a long and short memory network (LSTM) model, that is, fuses the two prediction models in a manner of embedding the long and short memory network model in the particle filter model, thereby forming a fusion prediction model for predicting the battery life. In the fusion prediction model, the long and short memory network model takes the battery capacity state value data in the historical data as training sample numbers to train and learn, so that the long and short memory network model predicts the capacity state of the battery according to the battery capacity historical data, constructs a state transition equation of the particle filter model according to the trained output value of the long and short memory network model, and obtains a priori prediction value of the capacity state of the battery.
Specifically, the implementation steps of the battery life prediction method based on the fusion prediction model provided by the invention comprise:
step 1: historical data of the battery is acquired. The history data includes data of the number of times of charge and discharge of the battery and data of a capacity state value of the battery corresponding to each time the charge and discharge is performed.
Step 2: and predicting the capacity state of the battery according to the historical data through the particle filter model so as to obtain a predicted value of the capacity of the battery.
The step of obtaining the predicted value of capacitance comprises:
step 21: and randomly generating a group of capacity values of the battery at the predicted starting time to serve as initial particle groups of the particle filter model, and allocating initial weight coefficients to the initial particle groups.
Step 22: and training and learning the long and short memory network model by adopting the training sample data, so that the long and short memory network model predicts the capacity state of the battery according to the battery capacity historical data.
Step 23: and constructing the state equation according to the output of the long and short memory network model and obtaining the prior predicted value at the current moment.
Step 24: generating a new particle group according to the prior value at the current moment and the particle group at the last moment.
Step 25: updating the weights of the new set of particles.
Step 26: and correcting the prior predicted value at the current moment according to the new particle group and the weight of the new particle group to obtain the posterior predicted value at the current moment.
And step 3: and comparing the capacity predicted value with a set capacity failure threshold value of the battery, judging that the battery fails when the capacity predicted value reaches the capacity failure threshold value, finishing the iteration of the particle filter model, and obtaining the residual life of the battery according to the iteration times.
The battery in the method for predicting the service life of the battery is mainly a cadmium-nickel storage battery, and the battery is likely to fail only when the charge-discharge cycle number of the battery reaches more than 2000 times, which is far higher than the charge-discharge cycle number of the common lithium battery. Therefore, the prediction model for the battery life of a cadmium-nickel storage battery needs to have new data for obtaining the updated fusion prediction model online to update the fusion prediction model adaptively so that the prediction accuracy of the fusion prediction model will be higher.
Therefore, according to the method for predicting the battery life provided by the present invention, the method further includes step 4, step 5 and step 6, if the determination result in the step 3 is that the battery does not fail, the step 4 and the step 5 are sequentially executed,
step 4 is to obtain a new state value and a new measurement value of the battery capacity online, and specifically, the new state value and the new measurement value (observed value) generated in the online life test experiment may be obtained and added to the original time series established by providing the historical battery data. And 5, judging whether the difference value between the predicted value and the new measured value at the current moment is larger than a set error value, if so, executing step 6, otherwise, adding 1 to the iteration number of the particle filter model, and returning to the step 23. And step 6 is to add the new state value obtained in step 4 to the training sample to update the training sample, and after the training sample is updated, step 22 is performed to retrain the long and short memory network model through the updated training sample.
In addition, a dropout module is arranged in the long and short memory network model to prevent the long and short memory network model from being over-fitted. Therefore, the long and short memory network model provided by the embodiment of the invention comprises an input and output layer, a long and short memory network layer, a dropout layer and a full connection layer.
Before further elaborating the present invention by means of specific embodiments, we first describe the relevant principles of the particle filter model and the long-short memory network model.
The particle filtering algorithm is an algorithm which introduces Monte Carlo sampling to obtain the posterior probability and the estimated value of a random sample on the basis of Bayesian filtering. Assume a system (e.g., a battery system of the present invention) whose state equation and observation equation are shown in equation (1) and equation (2):
xk＝fk(xk-1，vk-1) (1)
Yk＝hk(xk，nk) (2)
wherein xk，YkSystem state and observed value, v, at time k respectivelyk-1Process noise (dynamic noise) at time k-1 of the system, nkObserved noise at time k, xk-1The system state at time k-1. In the battery life prediction application, the formula (1) is generally an empirical degradation equation, and an accurate degradation equation in actual engineering is difficult to obtain. In order to obtain the optimal estimation of the target state, the particle filter obtains the posterior probability density p (x) of the k-time system through two processes of prediction and updatingk|Yk). The prediction phase uses the probability density p (x) at time k-1k-1|Yk-1) Obtaining a prior probability p (x)k|Yk-1) Is shown in equation (3):
p(xk|Yk-1)＝∫p(xk|xk-1)p(xk-1|Yk-1)dxk-1(3)
in the updating stage, an importance probability density function q (x) is introduced by using an importance sampling methodk|Yk) From which sample particles are generated, approximating a posterior probability distribution p (x) using a weighted sum of the particlesk|Yk) Taking the obtained posterior probability calculation formula as a formula (4)
Figure BDA0002604143700000061
Wherein
Figure BDA0002604143700000062
The state of the ith particle at time k, with a weight of
Figure BDA0002604143700000063
The formula for distributing the weight is shown in formula (5):
Figure BDA0002604143700000071
the cyclic neural network (RNN) can process the non-linear time sequence by using its memory function, but when the sequence is long, the problem of gradient explosion and gradient disappearance is easy to exist, and the long-short term memory network (LSTM) is a special RNN designed to solve the problem. Compared with RNN, LSTM adds an information processing unit, namely a cell, which consists of a forgetting gate, an input gate, and an output gate.
The forgetting gate can discard the redundant information of the upper layer with a certain probability, and the calculation formula is shown as formula (6):
f(t)＝σ(Wfh(t-1)+Ufx(t)+bf) (6)
wherein h is(t-1)Hidden state of the previous layer, x(t)For current sequence position information, Wf、Uf、bfThe weight and the offset of the linear relation in the forgetting gate are shown, and sigma is a sigmoid activation function. The gate will output a value between 0 and 1, determining the degree of information loss, with 0 indicating "completely discarded" and 1 indicating "completely reserved".
The input gate can process the information of the current sequence position, and the calculation formula is shown as formula (7) and formula (8):
i(t)＝σ(Wih(t-1)+Uix(t)+bi) (7)
a(t)＝tanh(Wah(t-1)+Uax(t)+ba) (8)
wherein Wi、Ui、Wa、UaWeight of linear relation in input gate, bi、baIs an offset. The results of the forgetting gate and the input gate will be used for updating the cell state, the update equation of which is shown in equation (9):
C(t)＝C(t-1)⊙f(t)+i(t)⊙a(t)(9)
wherein C is(t)For the updated cell state, ⊙ is the Hadamard product.
The output gate can process the information of the current sequence, the cell state and the hidden state of the upper layer, and outputs a new hidden state to the next layer, and the calculation formula corresponding to the output gate is shown as the formula (10) and the formula (11):
o(t)＝σ(Woh(t-1)+Uox(t)+bo) (10)
h(t)＝o(t)⊙tanh(C(t)) (11)
wherein, Wo、Uo、boWeight and offset, h, which is a linear relationship in the output gate(t)And the hidden state of the current layer is used as the output of the current layer and is continuously transmitted into the next layer.
In view of the fact that the long and short memory networks (LSTM) have good learning ability, the Particle Filter (PF) can adapt to a nonlinear and non-gaussian system well, and can give uncertainty expression, the present document proposes a method flow diagram of a battery life prediction method based on the fusion prediction model, which is provided by combining two algorithms of LSTM and PF, and is shown in fig. 1, and further elaborates how to combine the particle filter model and the long and short memory network model to predict the life of the battery.
As shown in FIG. 1, when the cadmium-nickel storage battery is in use, the available capacity of the storage battery is reduced due to the deactivation of active matters, the reduction of electrolyte and the like, TB _ T3061-2016 specifies, the capacity value is used as a failure judgment basis, and when the capacity is reduced to 70% of the rated capacity, the storage battery is determined to be failed. Therefore, the battery capacity is generally used as a performance degradation factor, and the life is predicted according to the evolution law of the degradation factor. The service life of the storage battery is influenced by various factors such as temperature, charge and discharge multiplying power, working conditions and the like, and the failure process is nonlinear and non-Gaussian. The particle filter can be well suitable for a non-Gaussian nonlinear system, and can obtain inaccurate expression of a prediction result, but the standard particle filter needs a state transition equation shown in a formula (1), and the factors such as environment and the like change greatly in actual application, so that a more accurate state equation is difficult to obtain. The LSTM has a memory function, can learn a time sequence of a long time span, but cannot adapt to uncertain factors such as noise and the like appearing in a system and cannot give uncertain expressions, so that two prediction models are fused, and the service life prediction of the storage battery is better realized by combining respective advantages.
Selecting capacity as degradation hereinFactor, the existing battery capacity data in the early stage (such as after 1000 battery charge-discharge cycles) is established as a time series (x)1，x2，x3，...xn) Training and learning the LSTM model through the established time sequence so that the LSTM model can obtain a predicted value of k time based on the information of m times before k time (current time), wherein the formula of the method for predicting the capacitance capacity state of the k time by using the long-short memory network model is shown as a formula (12)
Figure BDA0002604143700000081
Figure BDA0002604143700000082
For the output of the LSTM at the k-th time, i.e. the capacity state of the battery at the k-th time predicted by the LSTM model, the state transition equation of the particle filter is determined according to a capacity degradation model equation (12) obtained by LSTM training:
Figure BDA0002604143700000083
ωk-1is process noise, xkIs the state prediction value in the particle filter model at the k-th moment. Randomly generated k time capacity value as initialization particle
Figure BDA0002604143700000084
The calculation formula for obtaining a new set of particles according to the prior probability of the state transition equation is shown in formula (14):
Figure BDA0002604143700000085
optimizing weights of new particles using importance sampling, the closer to the state prediction value xkThe weighted sum of the particles is used to approximate the predicted value of the capacity at the k-th time as the weight of the particles (2) is larger. The new time series are used to update the LSTM model parameters. The specific flow is shown in figure 1.
Step a: preprocessing the capacity data, removing unavailable data, and normalizing, wherein the step is equivalent to the step 1.
Step b: randomly generating N particles
Figure BDA0002604143700000086
As the initial particles, for example, a set of capacity values of the battery at the k-th time are randomly generated as the initial particles.
Step c: and taking the processed data as a training sample to be put into an LSTM model to perform learning training on the LSTM model, and adding a dropout module in the LSTM to prevent overfitting. In the present invention we select an LSTM model that includes an input-output layer, an LSTM layer and one of the dropout layers.
Step d: using the capacitance degradation equation determined by the output of the trained LSTM model as a state equation in the particle filter model, as shown in formula (13), and obtaining the prior predicted value of the observed value according to the state equation
Figure BDA0002604143700000091
The prior predicted value and the prior probability are obtained according to the prior probability.
Step e: generating a new particle group state transition equation according to the prior predicted value at the current moment and the particle group at the last moment to generate a new particle group
Figure BDA0002604143700000092
Step f: updating the particle weight to obtain the post-verification predicted value
Figure BDA0002604143700000093
The posterior predicted value is the posterior value of the particle filter model to the battery capacity.
Step g: and (4) comparing the predicted battery capacity with 70% of the rated capacity, if the predicted battery capacity is smaller than the rated capacity, judging that the battery is out of service, and finishing prediction to obtain the residual service life, otherwise, entering the step h.
Step h: online acquisitionAdding a time series (x)new，ynew) Arrival xnewFor the new state value of the battery currently obtained online, ynewJudging a posterior predicted value for the observed value of the current battery capacity
Figure BDA0002604143700000094
And ynewIf the difference exceeds the set error range PEB, the LSTM model does not need to be updated, and the sliding window with the width of m moves forward by one step to continue prediction; otherwise, adding the newly added time sequence into the training set, turning to the step c to retrain the LSTM model parameters, and performing subsequent prediction by using the retrained model.
In order to research the aging characteristics of the cadmium-nickel batteries, multiple groups of same-type exhaust cadmium-nickel batteries for motor train units are used, the nominal voltage of each battery is 1.2V, the rated capacity is 160 A.h, a high-low temperature test box is used for maintaining the test environment temperature, and a storage battery pack test system is used for monitoring parameters such as current and voltage.
The cycle life test is carried out under the environment of 25 +/-5 ℃ according to the specification of an iron standard TB _ T3061-2016, 50 cycles are taken as a group, and the first cycle in each group of cycles is 0.25ItCharging for 6h at 0.25ItDischarging for 2.5h, and circulating for 2-50 times with 0.2ItCharging for 7-8 h at 0.2ItDischarging to 1.0V/node until the discharge time of any 50 cycles is less than 3.5h, and discharging at 0.2ItAnd (4) performing one more group of cycles, and if the discharge time of the 50 th cycle of two continuous groups is less than 3.5h, indicating that the capacity is reduced to below 70% of the rated capacity, terminating the life test.
And calculating to obtain capacity according to an ampere-hour integral theorem, taking the capacity as the performance degradation characteristic of the battery, wherein the formula of the on-time integral theorem is shown as a formula (15):
Figure BDA0002604143700000095
Ckfor the capacity of the kth charge-discharge cycle, I is the discharge current, a time series of capacities is obtained, the data is normalized using a normalization function such asPreprocessing shown in equation (16):
Figure BDA0002604143700000101
the fitness evaluation function of the prediction model of the battery life is defined as shown in equation (17):
Figure BDA0002604143700000102
where n is the predicted total number of data points, CkAs a value of the actual capacity, the capacity value,
Figure BDA0002604143700000103
to predict capacity values.
To verify the predictive effect of the proposed method, experimental data were predicted using standard particle filtering for comparison. The state transition equation uses an exponential model, as shown in equation (18):
Ck＝ηcCk-1+β1exp(-β2/Δtk-1) (18)
η thereincFor coulombic efficiency, 0.998, Δ t is generally takenk-1＝tk-tk-1And the other parameters are obtained by fitting experimental data for the time interval of two adjacent periods.
The experiment shows that the cadmium-nickel storage battery presents a low-capacity phenomenon in the earlier stage due to the special memory effect, and after a plurality of thorough charge-discharge cycles, the capacity is recovered to a rated value and fails in the 2842 period. The capacity threshold of the battery failure is 112 A.h, T1100 cycle and T2000 cycle are respectively used as a prediction starting point, experimental data before the prediction starting point is used as a training set, and data after the prediction starting point is used as a test set. The LSTM model structure comprises an input-output layer, an LSTM layer, a dropout layer and a full connection layer, and the optimizer uses adam. The number of particles N is 300, and the observed noise covariance Q is 0.0001
Table 1T 1100 cycles, results of the experiment
Figure BDA0002604143700000104
Fig. 2 to fig. 4 are comparison graphs of the prediction effects of the fusion prediction model LSTM-PF, the standard PF prediction model and the LSTM prediction model provided by the present invention under the setting conditions of the prediction starting point T being 1100 cycles and the actual RUL being 1742 cycles,
table 1 shows the results of three models, including prediction, error and fitness. As can be seen from the fitness data, the fusion model error is smaller, and the prediction error is 27 cycles less than PF and 18 cycles less than LSTM.
TABLE 2 RUL prediction values at 2000 cycles
Figure BDA0002604143700000105
Fig. 5 to 7 are prediction comparison graphs of the fusion model LSTM-PF, the standard PF, and the LSTM model under the setting conditions of the prediction starting point T being 2000 cycles and the actual RUL being 842 cycles, respectively, and table 2 shows the result evaluation of the three models, where the prediction error of the fusion model is 9 cycles less than PF, 5 cycles less than LSTM, and the fusion model has a higher degree of fitting.
The experimental result shows that when the prediction is started from the same starting point, the fusion model has a more accurate prediction result than the standard PF and LSTM models, and for all three models, when T is 2000 cycles, the prediction effect is better than that when T is 1100 cycles, and the further back the starting point is, more data can be used for training the model, so that the model is more accurate. For the same model, the prediction model continuously learns the updated parameters along with the updating of the observation data, and the online prediction result is more accurate.
Therefore, the method for predicting the remaining life of the battery on line based on the fusion of the particle filter model and the long and short memory network model can realize the prediction of the remaining cycle life of the cadmium-nickel storage battery, the long and short memory network model is nested in the particle filter model, the fusion model is simple in structure, the long and short memory network model is trained by using the existing historical data to obtain a state transfer equation of a degradation trend equation for determining the particle filter model, the problem that the particle filter model depends too much on an empirical model is solved, the particle filter model can obtain uncertain expression of the remaining life by using the weighted sum of particles and the predicted value of the approximate capacity, and in addition, a new sample obtained on line is added to the original training sample to intensively retrain the model, so that the model parameters are updated timely and the method has better adaptability.
Furthermore, the present invention provides a storage medium, which is a readable storage medium of a computer, and a computer program stored on the readable storage medium is executed by a processor to implement the method for predicting battery life according to any embodiment of the present invention.
While embodiments in accordance with the invention have been described above, these embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments described. Many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention and various embodiments with various modifications as are suited to the particular use contemplated. The invention is limited only by the claims and their full scope and equivalents.

Claims (10)

1. A battery life prediction method based on prediction model fusion is characterized in that the prediction model comprises a particle filter model and a length memory network model, and the battery life prediction method comprises the following steps:
step 1: the historical data of the battery is acquired,
step 2: predicting the capacity state of the battery according to the historical data through the particle filter model to obtain a predicted capacity value of the battery,
and step 3: comparing the predicted value of the capacity with a preset capacity failure threshold value of the battery, judging that the battery is failed when the predicted value of the capacity reaches the capacity failure threshold value, finishing the iteration of the particle filter model, and obtaining the residual life of the battery according to the iteration times,
the long and short memory network model is embedded in the particle filter model, and the long and short memory network model is trained and learned by taking battery capacity state value data in the historical data as training sample numbers, so that the long and short memory network model predicts the capacity state of the battery according to the battery capacity historical data, constructs a state transition equation of the particle filter model according to the trained output value of the long and short memory network model, and obtains a priori predicted value of the capacity state of the battery.
2. The battery life prediction method according to claim 1, wherein the history data includes data on the number of times of charge and discharge of the battery and data on a capacity state value of the battery corresponding to each time the charge and discharge is performed,
the battery life prediction method further includes:
before training and learning a long and short memory network model through the battery capacity data, preprocessing the historical data, wherein the preprocessing comprises removing useless data in the historical data and normalizing the historical data.
3. The battery life prediction method of claim 1, wherein the step 2 comprises:
step 21: a randomly generated set of capacity values of the battery at a predicted starting time as initial particle groups of the particle filter model, and assigning initial weight coefficients to the respective initial particle groups,
step 22: training and learning the long and short memory network model by adopting the training sample data, so that the long and short memory network model predicts the capacity state of the battery according to the battery capacity historical data,
step 23: constructing the state equation according to the output of the long and short memory network model and obtaining the prior predicted value at the current moment,
step 24: generating a new set of particles based on the a priori value at the current time and the set of particles at the previous time,
step 25: the weights of the new set of particles are updated,
step 26: and correcting the prior predicted value at the current moment according to the new particle group and the weight of the new particle group to obtain the posterior predicted value at the current moment.
4. The method according to claim 3, further comprising a step 4, a step 5 and a step 6, wherein if the determination result in the step 3 is that the battery has not failed, the step 4 and the step 5 are sequentially executed,
said step 4 is to obtain a current measurement of said battery capacity online,
step 5 is to determine whether the difference between the predicted value and the new measured value at the current time is greater than a set error value, if so, step 6 is executed, otherwise, the iteration number of the particle filter model is increased by 1, and the step 23 is returned,
and 6, updating the training sample by adding the measured value obtained in the step 4 into the training sample, and turning to the step 22 after the training sample is updated so as to retrain the long and short memory network model through the updated training sample.
5. The battery life prediction method according to claim 1, wherein the over-fitting of the long and short memory network model is prevented by providing a dropout module in the long and short memory network model.
6. The battery life prediction method according to claim 2, wherein the capacity state value data in the history data is converted into the capacity state value data
And establishing a time sequence, and training the long and short memory network model by using the time sequence.
7. The method of claim 6, wherein the step of determining the state equation and the a priori predicted values according to the trained long-short memory network comprises:
the trained long and short memory network model predicts the state value of the battery capacity at the current time according to the capacity state value data at m times before the current time,
superposing the output value of the long and short memory network model at the current moment and the process noise of the battery capacity degradation at the previous moment at the current moment as the state prediction value of the current moment in the state equation to construct the state equation,
and calculating to obtain the prior predicted value at the current moment according to the state equation and the posterior predicted value at the previous moment of the current moment.
8. The battery life prediction method according to claim 3, wherein the weight of the new particle group is updated using the importance sample such that the weight coefficient corresponding to a particle closer to the predicted value of the battery capacity state is larger.
9. The method according to claim 1, wherein the battery is a cadmium-nickel battery, and the long and short memory network model includes an input and output layer, a long and short memory network layer, a dropout layer, and a full connection layer.
10. A storage medium, characterized in that the storage medium is a readable storage medium of a computer, and a computer program stored on the readable storage medium is executed by a processor to implement the battery life prediction method according to any one of claims 1 to 9.
CN202010733589.1A 2020-07-27 2020-07-27 Battery life prediction method based on prediction model fusion and storage medium Pending CN111680848A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010733589.1A CN111680848A (en) 2020-07-27 2020-07-27 Battery life prediction method based on prediction model fusion and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010733589.1A CN111680848A (en) 2020-07-27 2020-07-27 Battery life prediction method based on prediction model fusion and storage medium

Publications (1)

Publication Number Publication Date
CN111680848A true CN111680848A (en) 2020-09-18

Family

ID=72458018

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010733589.1A Pending CN111680848A (en) 2020-07-27 2020-07-27 Battery life prediction method based on prediction model fusion and storage medium

Country Status (1)

Country Link
CN (1) CN111680848A (en)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112307638A (en) * 2020-11-09 2021-02-02 中南大学 Capacitor life estimation method and device and electronic equipment
CN112327193A (en) * 2020-10-21 2021-02-05 北京航空航天大学 Lithium battery capacity diving early warning method
CN112526354A (en) * 2020-12-22 2021-03-19 南京工程学院 Lithium battery health state estimation method
CN112698207A (en) * 2020-12-03 2021-04-23 天津小鲨鱼智能科技有限公司 Battery capacity detection method and device
CN112763929A (en) * 2020-12-31 2021-05-07 华东理工大学 Method and device for predicting health of battery monomer of energy storage power station system
CN112765772A (en) * 2020-12-25 2021-05-07 武汉理工大学 Power battery residual life prediction method based on data driving
CN112782591A (en) * 2021-03-22 2021-05-11 浙江大学 Lithium battery SOH long-term prediction method based on multi-battery data fusion
CN113283632A (en) * 2021-04-13 2021-08-20 湖南大学 Early battery fault warning method, system, device and storage medium
CN113343560A (en) * 2021-05-25 2021-09-03 中车青岛四方车辆研究所有限公司 Capacitance value prediction method and device of direct current support capacitor
CN113420494A (en) * 2021-05-25 2021-09-21 四川轻化工大学 Super-capacitor Bayes probability fusion modeling method
CN113504483A (en) * 2021-07-09 2021-10-15 北京航空航天大学 Integrated prediction method for residual life of lithium ion battery considering uncertainty
CN114355184A (en) * 2022-01-05 2022-04-15 国网江苏省电力有限公司宿迁供电分公司 High-voltage circuit breaker state monitoring and early warning system and method based on online learning
CN114487848A (en) * 2022-01-17 2022-05-13 北京和利时系统集成有限公司 Method and device for calculating state of storage battery
CN116502544A (en) * 2023-06-26 2023-07-28 武汉新威奇科技有限公司 Electric screw press life prediction method and system based on data fusion
CN118362898A (en) * 2024-04-24 2024-07-19 苏州特瑞菲机械设备有限公司 New energy automobile battery performance detection system and method

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103335653A (en) * 2013-06-06 2013-10-02 北京航空航天大学 Adaptive incremental particle filtering method for Mars atmosphere entry section
CN105157704A (en) * 2015-06-03 2015-12-16 北京理工大学 Bayesian estimation-based particle filter gravity-assisted inertial navigation matching method
CN106931453A (en) * 2017-02-27 2017-07-07 浙江大学 The forecasting system and method for circulating fluid bed domestic garbage burning emission of NOx of boiler
CN107576963A (en) * 2017-09-11 2018-01-12 中国民航大学 The method of estimation of dual polarization radar difference travel phase shift based on particle filter
CN109917292A (en) * 2019-03-28 2019-06-21 首都师范大学 A kind of lithium ion battery life-span prediction method based on DAUPF
CN110174690A (en) * 2019-05-30 2019-08-27 杭州中科微电子有限公司 A kind of satellite positioning method based on shot and long term memory network auxiliary
CN110188920A (en) * 2019-04-26 2019-08-30 华中科技大学 A kind of lithium battery method for predicting residual useful life
CN110187290A (en) * 2019-06-27 2019-08-30 重庆大学 A kind of lithium ion battery residual life prediction technique based on pattern of fusion algorithm
CN110703120A (en) * 2019-09-29 2020-01-17 上海海事大学 Lithium ion battery service life prediction method based on particle filtering and long-and-short time memory network
CN110765897A (en) * 2019-10-08 2020-02-07 哈尔滨工程大学 Underwater target tracking method based on particle filtering
CN111044926A (en) * 2019-12-16 2020-04-21 北京航天智造科技发展有限公司 Method for predicting service life of proton exchange membrane fuel cell
CN111103544A (en) * 2019-12-26 2020-05-05 江苏大学 Lithium ion battery remaining service life prediction method based on long-time and short-time memory LSTM and particle filter PF

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103335653A (en) * 2013-06-06 2013-10-02 北京航空航天大学 Adaptive incremental particle filtering method for Mars atmosphere entry section
CN105157704A (en) * 2015-06-03 2015-12-16 北京理工大学 Bayesian estimation-based particle filter gravity-assisted inertial navigation matching method
CN106931453A (en) * 2017-02-27 2017-07-07 浙江大学 The forecasting system and method for circulating fluid bed domestic garbage burning emission of NOx of boiler
CN107576963A (en) * 2017-09-11 2018-01-12 中国民航大学 The method of estimation of dual polarization radar difference travel phase shift based on particle filter
CN109917292A (en) * 2019-03-28 2019-06-21 首都师范大学 A kind of lithium ion battery life-span prediction method based on DAUPF
CN110188920A (en) * 2019-04-26 2019-08-30 华中科技大学 A kind of lithium battery method for predicting residual useful life
CN110174690A (en) * 2019-05-30 2019-08-27 杭州中科微电子有限公司 A kind of satellite positioning method based on shot and long term memory network auxiliary
CN110187290A (en) * 2019-06-27 2019-08-30 重庆大学 A kind of lithium ion battery residual life prediction technique based on pattern of fusion algorithm
CN110703120A (en) * 2019-09-29 2020-01-17 上海海事大学 Lithium ion battery service life prediction method based on particle filtering and long-and-short time memory network
CN110765897A (en) * 2019-10-08 2020-02-07 哈尔滨工程大学 Underwater target tracking method based on particle filtering
CN111044926A (en) * 2019-12-16 2020-04-21 北京航天智造科技发展有限公司 Method for predicting service life of proton exchange membrane fuel cell
CN111103544A (en) * 2019-12-26 2020-05-05 江苏大学 Lithium ion battery remaining service life prediction method based on long-time and short-time memory LSTM and particle filter PF

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HERALDO ROZAS: "Comparison of different models of future operating condition in Particle-Filter-based Prognostic Algorithms", vol. 53, no. 53, pages 10336 - 10341 *
李校林，吴腾: "基于PF-LSTM网络的高效网络流量预测方法", vol. 36, no. 36, pages 3833 - 3836 *
肖仁鑫，宋新月，张梦帆，夏雪磊，肖佳鹏: "基于长短期记忆神经网络的健康状态估算", vol. 58, no. 58, pages 77 - 81 *

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112327193A (en) * 2020-10-21 2021-02-05 北京航空航天大学 Lithium battery capacity diving early warning method
CN112307638A (en) * 2020-11-09 2021-02-02 中南大学 Capacitor life estimation method and device and electronic equipment
CN112698207A (en) * 2020-12-03 2021-04-23 天津小鲨鱼智能科技有限公司 Battery capacity detection method and device
CN112526354A (en) * 2020-12-22 2021-03-19 南京工程学院 Lithium battery health state estimation method
CN112765772B (en) * 2020-12-25 2022-11-04 武汉理工大学 Power battery residual life prediction method based on data driving
CN112765772A (en) * 2020-12-25 2021-05-07 武汉理工大学 Power battery residual life prediction method based on data driving
CN112763929B (en) * 2020-12-31 2024-03-08 华东理工大学 Method and device for predicting health of battery monomer of energy storage power station system
CN112763929A (en) * 2020-12-31 2021-05-07 华东理工大学 Method and device for predicting health of battery monomer of energy storage power station system
CN112782591A (en) * 2021-03-22 2021-05-11 浙江大学 Lithium battery SOH long-term prediction method based on multi-battery data fusion
CN112782591B (en) * 2021-03-22 2022-07-22 浙江大学 Lithium battery SOH long-term prediction method based on multi-battery data fusion
CN113283632A (en) * 2021-04-13 2021-08-20 湖南大学 Early battery fault warning method, system, device and storage medium
CN113283632B (en) * 2021-04-13 2024-02-27 湖南大学 Early-stage fault early-warning method, system, device and storage medium for battery
CN113343560B (en) * 2021-05-25 2022-07-12 中车青岛四方车辆研究所有限公司 Capacitance value prediction method and device of direct current support capacitor
CN113420494A (en) * 2021-05-25 2021-09-21 四川轻化工大学 Super-capacitor Bayes probability fusion modeling method
CN113343560A (en) * 2021-05-25 2021-09-03 中车青岛四方车辆研究所有限公司 Capacitance value prediction method and device of direct current support capacitor
CN113504483A (en) * 2021-07-09 2021-10-15 北京航空航天大学 Integrated prediction method for residual life of lithium ion battery considering uncertainty
CN114355184A (en) * 2022-01-05 2022-04-15 国网江苏省电力有限公司宿迁供电分公司 High-voltage circuit breaker state monitoring and early warning system and method based on online learning
CN114355184B (en) * 2022-01-05 2023-09-26 国网江苏省电力有限公司宿迁供电分公司 Online learning-based high-voltage circuit breaker state monitoring and early warning method
CN114487848A (en) * 2022-01-17 2022-05-13 北京和利时系统集成有限公司 Method and device for calculating state of storage battery
CN114487848B (en) * 2022-01-17 2024-05-03 北京和利时系统集成有限公司 State calculation method and device for storage battery
CN116502544A (en) * 2023-06-26 2023-07-28 武汉新威奇科技有限公司 Electric screw press life prediction method and system based on data fusion
CN116502544B (en) * 2023-06-26 2023-09-12 武汉新威奇科技有限公司 Electric screw press life prediction method and system based on data fusion
CN118362898A (en) * 2024-04-24 2024-07-19 苏州特瑞菲机械设备有限公司 New energy automobile battery performance detection system and method

Similar Documents

Publication Publication Date Title
CN111680848A (en) Battery life prediction method based on prediction model fusion and storage medium
CN110221225B (en) Spacecraft lithium ion battery cycle life prediction method
Li et al. Big data driven lithium-ion battery modeling method based on SDAE-ELM algorithm and data pre-processing technology
CN109061504B (en) Method and system for predicting remaining service life of lithium ion batteries of same type and different types
CN112241608B (en) Lithium battery life prediction method based on LSTM network and transfer learning
Liu et al. Remaining useful life estimation for proton exchange membrane fuel cells using a hybrid method
Khalid et al. Prediction of li-ion battery state of charge using multilayer perceptron and long short-term memory models
KR102439041B1 (en) Method and apparatus for diagnosing defect of battery cell based on neural network
Vanem et al. Data-driven state of health modelling—A review of state of the art and reflections on applications for maritime battery systems
Kim et al. Estimation of Li-ion battery state of health based on multilayer perceptron: As an EV application
WO2022253038A1 (en) Method and system for predicting state of health of lithium battery on basis of elastic network, and device and medium
CN111736084B (en) Valve-regulated lead-acid storage battery health state prediction method based on improved LSTM neural network
CN112215434A (en) LSTM model generation method, charging duration prediction method and medium
CN104680024A (en) Method for predicting remaining useful life of lithium ion battery based on GA (Genetic Algorithms) and ARMA (Auto Regressive and Moving Average) models
CN113361692A (en) Lithium battery residual life combined prediction method
Ansari et al. Jellyfish optimized recurrent neural network for state of health estimation of lithium-ion batteries
CN117031310A (en) Method for predicting residual service life of power battery of electric automobile
Xu et al. Safety warning analysis for power battery packs in electric vehicles with running data
De Sousa et al. Comparison of different approaches for lead acid battery state of health estimation based on artificial neural networks algorithms
CN116298936A (en) Intelligent lithium ion battery health state prediction method in incomplete voltage range
CN116243194A (en) On-line prediction method for battery health state under mixed working condition
CN117420442A (en) MNQ-LSTM-based battery remaining life prediction method
Tiwari et al. Use of ML Techniques for Li-Ion Battery Remaining Useful Life Prediction-A Survey
Basia et al. Comparison of data driven algorithms for SoH estimation of Lithium-ion batteries
CN109977622B (en) Method for predicting residual life of power 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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20200918