Vehicle cloud fusion battery state-of-charge estimation method and system based on expansive force

Abstract

The invention provides a vehicle cloud fusion battery state-of-charge estimation method and system based on expansibility, and the method comprises the following steps: obtaining battery state data at a vehicle end, and executing an IAEKF algorithm to obtain a first estimation value of the battery state-of-charge, the battery state data comprising a battery momentum expansibility; acquiring battery state data at the cloud, and outputting a second estimated value of the battery state of charge based on the battery state data and the expansive force characteristic parameters by using a joint prediction model comprising a GA model, a CNN model and an LSTM model; and performing one-dimensional Kalman filtering on the first estimation value and the second estimation value of the state of charge of the battery to obtain a final estimation value of the state of charge of the battery. According to the estimation method provided by the invention, the expansive force characteristic parameters which are more sensitive to the change of the state of charge of the battery can be extracted, and the SOC of the battery is quickly and accurately estimated by integrating the estimation results of the vehicle end and the cloud end.

Description

Technical Field

[0001] The present application belongs to the field of vehicle-mounted battery management technology, and relates to a vehicle-mounted battery state of charge estimation technology. Specifically, a vehicle-cloud fusion battery state of charge estimation method and estimation system based on expansion force are provided. Background Art

[0002] As a key component of electric vehicles, power batteries have an important impact on the driving performance of electric vehicles. Lithium batteries are widely used due to their outstanding advantages in weight, safety, energy density and cycle life. Accurate and timely acquisition of the status of soft-pack lithium batteries, especially the state of charge (SOC), is crucial for the reliable and efficient operation of electric vehicles.

[0003] At present, the mainstream calculation methods of SOC can be divided into four categories: traditional methods, methods based on battery models, methods based on data-driven, and hybrid algorithms. Among them, the typical representative of traditional methods is the ampere-hour integration method; model-based methods mainly include Kalman filter algorithm (KF), and further developed on its basis Extended Kalman filter algorithm (EKF), unscented Kalman filter algorithm (UKF), iterative adaptive Kalman filter algorithm (IAEKF), etc.; data-driven algorithms include feedforward neural networks, classification and regression algorithms, probability algorithms, recurrent neural networks (RNN) and rule-based algorithms. Since battery SOC is a typical time series model, with the development of the computer industry, RNN and improved RNN applications are gradually increasing. Among them, long short-term memory neural network (LSTM) and gated recurrent neural network (GRU) have also been widely used in SOC estimation; at the same time, hybrid algorithms have also grown, including hybrid optimization and hybrid deep learning. Hybrid optimization refers to using optimization algorithms based on deep learning to find the best parameters of the network, and hybrid deep learning uses multiple network fusions.

[0004] With the upgrade of wireless networks from 4G to 5G and even 6G, Internet of Vehicles technology is developing rapidly. By integrating vehicle-side data collection with big data processing on cloud computing platforms through vehicle-cloud fusion architecture, the computing power and storage resources of cloud platforms can be fully utilized to easily realize centralized processing and analysis of vehicle-side collected data, thereby effectively releasing the computing power resources of the vehicle-side control platform so that it can focus on processing key vehicle data. However, there is currently no vehicle-cloud fusion architecture for automotive battery SOC estimation.

[0005] In addition, most of the current SOC estimation methods or models use current, voltage, temperature and their rate of change or time window as input. However, the above data cannot fully characterize the internal changes of the battery during actual use. Therefore, it is necessary to find characteristic quantities that are more sensitive to SOC changes based on the inherent characteristics of the battery to improve the accuracy of the SOC estimation method. Summary of the invention

[0006] In order to solve the problems existing in the above-mentioned prior art, the present application provides a method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force through an embodiment, and the method comprises the following steps:

[0007] Acquire battery status data at the vehicle end and execute the IAEKF algorithm to obtain a first estimated value of the battery state of charge, wherein the battery status data includes an excitation current, a momentum voltage, a momentum expansion force, and a momentum temperature of the battery;

[0008] Acquire battery status data in the cloud and use a joint prediction model including a GA model, a CNN model, and an LSTM model to obtain a second estimated value of the battery state of charge, wherein the GA model is used to optimize the hyperparameters of the LSTM model, the CNN model is used to generate an expansion force characteristic parameter of the battery that is fed back to the LSTM model, and the LSTM model outputs a second estimated value of the battery state of charge based on the battery status data and the expansion force characteristic parameter;

[0009] A one-dimensional Kalman filter is performed on the first estimated value and the second estimated value of the battery state of charge to obtain a final estimated value of the battery state of charge.

[0010] Furthermore, the parameters required to execute the IAEKF algorithm are identified by the following steps:

[0011] Establish a second-order RC model of the battery considering the expansion force and determine its state expression:

[0012]

[0013] Among them, s is a complex variable, U L , U OC ,I L are the terminal voltage, open circuit voltage and excitation current of the battery respectively, p is the expansion force variable of the battery, R0(p), R1(p), C1(p), R2(p) and C2(p) are the internal resistance, concentration polarization resistance, concentration polarization capacitance, electrochemical polarization resistance and electrochemical polarization capacitance of the battery respectively;

[0014] Discretize the battery second-order RC model considering the expansion force to obtain its least squares recursive expression;

[0015] HPPC experiments were conducted on the battery under different initial expansion forces, initial temperatures and test conditions;

[0016] The HPPC experimental data were recursively fitted to achieve parameter identification, where the parameter identification results were the R0(p), R1(p), C1(p), R2(p), C2(p) and SOC-OCV fitting curves corresponding to different expansion forces and temperatures.

[0017] Preferably, the IAEKF algorithm comprises the following steps:

[0018] Initialize the algorithm;

[0019] The prior estimate of the battery state vector at time k is determined based on the following formula:

[0020]

[0021] in, is the a posteriori estimate of the battery state vector at time k-1, u k-1 is the battery excitation current at time k-1, Φ is the battery state transfer matrix;

[0022] The a priori error d of the battery terminal voltage at time k is determined based on the following formula: k :

[0023]

[0024] Among them, z k is the battery terminal voltage at time k, h() is the battery observation function;

[0025] Determine the a posteriori estimate of the battery state vector at time k by iterative Kalman filtering and from The first estimated value of the battery state of charge at time k is extracted from the iterative Kalman filter, wherein the iterative Kalman filter adaptively updates the measurement noise covariance R at time k. k , process noise covariance Q k , prior error covariance matrix and the posterior error covariance matrix

[0026] The residual ε of the battery terminal voltage at time k is determined based on the following formula: k :

[0027]

[0028] Update the k time and re-execute the IAEKF algorithm until the estimation of the battery state of charge is completed.

[0029] Preferably, the iterative Kalman filter determines the a posteriori estimate of the battery state vector at time k by the following formula:

[0030]

[0031] Where superscript T is the matrix transpose, superscript i and i+1 represent two consecutive iterations, i≥0, are the i-th and i+1-th iteration results at time k, respectively, and is the Kalman gain of the i-th iteration at time k, v k is the observation noise at time k, δ is the preset accuracy threshold;

[0032] And, the iterative Kalman filter determines the prior error covariance matrix at time k by the following formula: Posterior error covariance matrix

[0033]

[0034] Among them, Φ k-1 , are the process transfer matrix and the posterior error covariance matrix at time k-1 respectively, and I is the unit matrix.

[0035] Preferably, the measurement noise covariance R at time k is k Determined based on the following formula:

[0036]

[0037] Among them, α is the forgetting factor, E[] is the expected value function, R k-1 , ε k-1 are the measurement noise covariance and the residual of the battery terminal voltage at time k-1, is the observation matrix corresponding to the end of the iteration at time k;

[0038] The process noise covariance Q at time k k Determined based on the following formula:

[0039]

[0040] Among them, β is the forgetting factor, Q k-1 ,d k-1 are the process noise covariance at time k-1 and the prior error of the battery terminal voltage, is the Kalman gain corresponding to the end of the iteration at time k.

[0041] Preferably, the GA model uses a genetic algorithm to optimize the hyperparameters of the LSTM model, wherein the hyperparameters include the time window length, the number of hidden layer neurons, the learning rate and the training batch size of the LSTM model, and the fitness function L of the genetic algorithm is shown as follows:

[0042] L=||SOC real -SOC pre ||,

[0043] Among them, SOC real SOC is the estimated value of the battery state of charge obtained by the ampere-hour integration method. pre It is an estimated value of the battery state of charge output by the LSTM model, and the estimated value is used to train the GA model and the CNN model.

[0044] Preferably, the input data of the CNN model are the excitation current I, momentum voltage U, momentum expansion force p and momentum temperature T of the battery at each moment, and the output data are two expansion force characteristic parameters of the battery at each moment, wherein the two expansion force characteristic parameters are respectively used to characterize the [I, U, p] related characteristics and [U, p, T] related characteristics of the battery at each moment;

[0045] The convolution kernel size of the CNN model is 1*3, and the convolution layer size is M*2*N, where M is the time window length of the LSTM model, and N is the length of the time series for battery state of charge estimation;

[0046] The CNN model is based on the SOC output of the LSTM model pre Conduct training.

[0047] Preferably, the input data of the LSTM model are M groups of [I, U, p, T] data of the battery at each time, and two expansion force characteristic parameters output by the CNN model, and the output data are the estimated SOC of the battery state of charge at each time. pre and a second estimate of the battery state of charge.

[0048] The present application also provides a vehicle-cloud fusion battery state of charge estimation system based on expansion force through an embodiment, and the estimation system includes:

[0049] A vehicle-side estimation unit disposed inside the vehicle, used to obtain battery status data and execute an IAEKF algorithm to obtain a first estimated value of the battery state of charge, wherein the battery status data includes an excitation current, a momentum voltage, a momentum expansion force, and a momentum temperature of the battery;

[0050] A cloud estimation unit provided in a cloud server, for acquiring battery status data and using a joint prediction model including a GA model, a CNN model and an LSTM model to obtain a second estimated value of the battery state of charge, wherein the GA model is used to optimize the hyperparameters of the LSTM model, the CNN model is used to generate an expansion force characteristic parameter of the battery which is fed back to the LSTM model, and the LSTM model predicts and outputs a second estimated value of the battery state of charge based on the battery status data and the expansion force characteristic parameter;

[0051] The fusion estimation unit is used to perform a one-dimensional Kalman filter on the first estimation value and the second estimation value of the battery state of charge to obtain a final estimation value of the battery state of charge.

[0052] The vehicle-cloud fusion battery state of charge estimation method and estimation system based on expansion force provided in the present application perform Kalman filtering on the vehicle side to estimate the battery state of charge using battery state data including battery expansion force. Since Kalman filtering can estimate the state at the next moment based on the state at the previous moment, a relatively accurate SOC estimation result can be obtained without consuming too many vehicle-side computing resources. At the same time, a deep learning joint prediction model is constructed on the cloud side using powerful computing and storage capabilities. By conducting in-depth analysis on the massive battery state data uploaded by the vehicle, expansion force characteristic parameters that are more sensitive to changes in the battery state of charge are extracted to improve the accuracy of the computational SOC estimation. Finally, the estimation results on the vehicle side and the cloud side are integrated to estimate the battery SOC more quickly and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the architecture of a vehicle-cloud fusion battery state of charge estimation system based on expansion force provided according to an embodiment of the present application;

[0054] Figure 2 A flowchart of a method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to an embodiment of the present application;

[0055] Figure 3 A specific flow chart of the IAEKF algorithm executed by the vehicle side in the embodiment of the present application;

[0056] Figure 4 A flow chart for performing parameter identification on a battery in an embodiment of the present application;

[0057] Figure 5 Schematic diagram of the experimental conditions for performing HPPC on a battery in an embodiment of the present application;

[0058] Figure 6 is a schematic diagram of battery voltage changes obtained through HPPC experiments in the embodiments of the present application;

[0059] Figure 7 Schematic diagram of the change of battery expansion force obtained through HPPC experiment in the embodiment of the present application;

[0060] Figure 8 A schematic diagram of the architecture of a GA-CNN-LSTM joint prediction model provided according to an embodiment of the present application;

[0061] Fig. 9 A flow chart of a training algorithm for a GA-CNN-LSTM joint prediction model provided according to an embodiment of the present application;

[0062] Fig.10 A schematic diagram of the architecture of an LSTM model provided according to an embodiment of the present application;

[0063] Fig.11 A genetic algorithm flow chart of a GA model provided according to an embodiment of the present application;

[0064] Fig.12 A schematic diagram of the architecture of a CNN model provided according to an embodiment of the present application;

[0065] Fig.13 The estimation results of the battery SOC using different algorithms under the BJDST working condition provided by the embodiment of the present application;

[0066] Fig.14 It is the estimation error of the battery SOC using different algorithms under the BJDST working condition provided by the embodiment of the present application;

[0067] Fig.15 The estimation results of the battery SOC using different algorithms under the FUDS working condition provided by the embodiment of the present application;

[0068] Fig.16 The estimation error of the battery SOC using different algorithms under the FUDS working condition provided by the embodiment of the present application;

[0069] Fig.17 The estimation results of the battery SOC using different algorithms under the US06 working condition provided in the embodiment of the present application;

[0070] Fig.18 The estimated error of the battery SOC using different algorithms under the US06 working condition provided in the embodiment of the present application;

[0071] Fig.19 The estimation results of the battery SOC using different algorithms under the DST condition provided by the embodiment of the present application;

[0072] Fig. 20It is the estimation error of the battery SOC using different algorithms under the DST condition provided in the embodiment of the present application. DETAILED DESCRIPTION

[0073] Hereinafter, the present application will be further described based on preferred embodiments with reference to the accompanying drawings.

[0074] In the description of the embodiments of the present application, it should be noted that if the terms "upper", "lower", "inside", "outside", etc. indicate an orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the products of the embodiments of the present application are usually placed when in use, it is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. In addition, in the description of the present application, in order to distinguish different units, the words first, second, etc. are used in this specification, but these are not limited by the order of manufacture, nor can they be understood as indicating or implying relative importance, and their names may be different in the detailed description and claims of the present application.

[0075] The vocabulary in this specification is used to illustrate the embodiments of the present application, but is not intended to limit the present application. It should also be noted that, unless otherwise clearly specified and limited, the terms "disposed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, an indirect connection through an intermediate medium, or a connection between the two elements. For those skilled in the art, the specific meanings of the above terms in this application can be specifically understood.

[0076] Figure 1 The schematic diagram of the architecture of the vehicle-cloud fusion battery state of charge estimation system based on expansion force provided according to an embodiment of the present application is as follows: Figure 1 As shown, the system includes a vehicle-side estimation unit arranged inside the vehicle, a cloud-side estimation unit arranged on a cloud server, and a fusion estimation unit arranged on a cloud server (or inside the vehicle). Among them, the vehicle-side estimation unit and the cloud-side estimation unit can communicate bidirectionally through a wireless network to realize the transmission of data and instructions, and the fusion estimation unit can select the connection and communication method with the vehicle-side estimation unit and the cloud-side estimation unit according to its location. For example, when the fusion estimation unit is arranged on the cloud server side, it can directly communicate with the cloud-side estimation unit by wire, and communicate with the vehicle-side estimation unit by wireless. The above-mentioned wired / wireless data transmission and communication methods are known to those skilled in the art and will not be repeated here.

[0077] The present application also provides a method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force through an embodiment. The estimation method can be performed by Figure 1 The illustrated estimation system is implemented to achieve an accurate estimation of the battery state of charge. Figure 2 A flow chart of the estimation method is shown in FIG. Figure 2 As shown, the method comprises the following steps:

[0078] Step 100, obtaining battery status data at the vehicle end and executing the IAEKF algorithm to obtain a first estimated value of the battery state of charge;

[0079] Step 200, obtaining battery status data in the cloud and using a joint prediction model including a GA model, a CNN model, and a LSTM model to obtain a second estimated value of the battery state of charge;

[0080] Step 300 , performing one-dimensional Kalman filtering on the first estimated value and the second estimated value of the battery state of charge to obtain a final estimated value of the battery state of charge.

[0081] It is understandable that the steps 100 to 300 of the estimation method can be performed after the vehicle is started and connected to the cloud. Figure 1 The vehicle-side estimation unit, the cloud-side estimation unit and the fusion estimation unit are repeatedly executed at a preset estimation time interval until the condition for stopping the estimation is met (such as receiving an instruction to stop the estimation, the vehicle is turned off or the connection with the cloud is disconnected).

[0082] The specific implementation of the estimation method and the estimation system will be described in detail below with reference to the accompanying drawings.

[0083] <1. Data Collection>

[0084] In some optional embodiments, such as Figure 1 As shown, the sensor unit arranged on the vehicle battery collects battery status data at a preset sampling time interval (or at a preset sampling frequency) through the various sensors it contains. The collected battery status data can be directly transmitted to the vehicle-side estimation unit, and wirelessly transmitted to the cloud estimation unit on the cloud server side for estimation of the battery SOC.

[0085] The battery status data includes the battery excitation current I, momentum voltage U, momentum expansion force p and momentum temperature T (in the embodiments of the present application, momentum refers to the real-time value of each variable) obtained by the sensor at each sampling moment. Obviously, the above battery status data outputs a 1*4 vector at each sampling moment, which can be expressed as [I, U, p, T] without loss of generality.

[0086] Among them, the methods of obtaining and using the excitation current, momentum voltage and momentum temperature of the battery are known to those skilled in the art. Different from the above-mentioned prior art, in the embodiment of the present application, the collected battery status data also includes momentum expansion force p. Specifically, a thin film pressure sensor can be set on the outer surface of the battery to obtain the expansion force p between batteries, or between batteries and their containing shells. The above-mentioned momentum expansion force p obtained will be involved in the estimation of SOC together with I, U, and T.

[0087] The reason for introducing expansion force in the vehicle-side estimation and cloud-side estimation of the battery state of charge in the present application is that under the dynamic conditions of battery charging and discharging, the force is more sensitive to changes in the state of charge than the voltage. This is because during the charging and discharging process of automotive batteries such as lithium-ion batteries, lithium insertion and lithium extraction reactions occur at the positive and negative electrodes, and the thickness of the electrode sheets will change accordingly. As the lithium ion content of the negative electrode of the battery increases, the structure of the lithium-containing polymer material inside the electrode changes (generally, it is difficult to accurately and quickly reflect the extent of this change from the voltage data), and the battery's state of charge will also increase, converting the change in the battery volume into a change in the battery's expansion force.

[0088] It can be seen that the momentum expansion force of the battery is closely related to the battery state of charge, and the battery expansion force response can better explain the microscopic changes of the battery than the voltage response. Therefore, the battery expansion force data can be introduced in the process of estimating the battery state of charge, and the characteristic parameters related to the battery expansion force can be further extracted to improve the accuracy of SOC estimation.

[0089] refer to Figure 1 and Figure 2 After the vehicle-side estimation unit obtains the above battery status data, it executes the IAEKF (Iterative Adaptive Extended Kalman Filter) algorithm to obtain the first estimated value of the SOC of the battery at each time k (in the embodiment of the present application, the first estimated value of the SOC at time k is recorded as ); After obtaining the above battery status data, the cloud estimation unit uses a joint prediction model including a GA model, a CNN model and a LSTM model to obtain a second estimated value of the battery state of charge (in the embodiment of the present application, the second estimated value of the SOC at time k is recorded as ); Finally, the fusion estimation unit set up in the cloud or vehicle side and A one-dimensional Kalman filter is performed to obtain the final estimated value of the battery SOC. Preferably, the final estimated value of the battery SOC can be stored in a data storage device on the vehicle side and / or in the cloud (such as a storage device of an onboard computer, a database of a cloud server, etc.), and displayed in real time on a display terminal such as a dashboard or control panel of the vehicle as needed. In addition, it can also be synchronously stored and / or displayed in an APP used in conjunction with the onboard computer.

[0090] The above-mentioned vehicle-cloud fusion battery state of charge estimation method and estimation system based on expansion force provided in the present application use battery state data including battery expansion force to perform Kalman filtering on the vehicle side to estimate the battery state of charge. Since Kalman filtering can estimate the state at the next moment based on the state at the previous moment, a relatively accurate SOC estimation result can be obtained without consuming too many vehicle-side computing resources. At the same time, a deep learning joint prediction model is constructed on the cloud side using powerful computing and storage capabilities. By conducting in-depth analysis of the massive battery state data uploaded by the vehicle, characteristic parameters that are more sensitive to changes in the battery state of charge are extracted to improve the accuracy of the computational SOC estimation. Finally, the estimation results on the vehicle side and the cloud side are integrated to estimate the battery SOC more quickly and accurately.

[0091] <2. SOC estimation on the vehicle side>

[0092] In the embodiment of the present application, the vehicle-side estimation unit can be set in the vehicle-mounted computer, which obtains the [I, U, p, T] data collected by each sensor at each sampling time through a communication method well known to those skilled in the art, and then executes the IAEKF algorithm (iterative adaptive extended Kalman filter algorithm) to estimate the first estimated value of the battery state of charge at each time. Obviously, the sampling time and the estimation time can use the same time interval or frequency, so the battery state data at time k can be recorded as [I, U, p, T] k , k = 1, 2, 3, ..., the first estimated value of the battery state of charge at time k is recorded as

[0093] Figure 3 FIG. 4 shows a flow chart of SOC estimation by executing the IAEKF algorithm on the vehicle side in some preferred embodiments of the present application, such as Figure 3 As shown in the figure, the algorithm mainly includes the main process shown in the orange part, namely the extended Kalman filter part (EKF), the iterative Kalman filter operation part (Iteration) shown in the green part, and the adaptive update filter parameter (Adaptive module) part shown in the blue part.

[0094] 2.1) Obtain IAEKF algorithm parameters.

[0095] During the execution of the IEAKF algorithm, it is necessary to obtain the internal parameters that characterize the battery characteristics in advance, and the form and quantity of the battery internal parameters depend on the specific form of the equivalent model that describes the battery performance.

[0096] In some embodiments of the present application, when the internal parameters of the battery required for SOC estimation are known (for example, stored as a retrievable parameter table), the vehicle-side estimation unit can directly read and start executing the IEAKF algorithm; in other embodiments of the present application, when SOC estimation of a battery of unknown properties is required, it is necessary to first obtain the internal parameters of the battery through modeling and measurement. The process of obtaining the internal parameters of the battery is called parameter identification.

[0097] In some optional embodiments, such as Figure 4 As shown, the battery parameters can be identified by the following steps:

[0098] Step 410: Establish a second-order RC model of the battery taking into account the expansion force.

[0099] The existing battery equivalent circuit model generally uses a second-order RC network, such as Figure 5 As shown, it includes two RC branches and an internal resistance R0, wherein one RC branch is used to represent the concentration polarization resistance R1 and the capacitance C1, and the other RC branch is used to represent the electrochemical polarization resistance R2 and the capacitance C2. The above parameters are the internal parameters of the battery that need to be parameterized.

[0100] Different from the general battery equivalent circuit, the circuit model established in the present application introduces an expansion force parameter p on this basis, that is, the internal parameters of the battery change with the change of p. For this reason, in the embodiments of the present application, the above-mentioned internal parameters are recorded as R0(p), R1(p), C1(p), R2(p), and C2(p).

[0101] The calculation formula of battery SOC is shown in formula (1):

[0102]

[0103] In the formula, t is the time variable, SOC0 represents the initial state of charge, η is the charge and discharge efficiency of the battery, i(t) is the current density, Q n is the actual capacity of the battery.

[0104] Through the above-mentioned battery second-order RC model considering the expansion force parameter p, its state expression (2) can be obtained:

[0105]

[0106] Among them, s is a complex variable, U L , U OC,I L They are the terminal voltage, open circuit voltage and excitation current of the battery respectively.

[0107] Step 420, discretize the second-order RC model of the battery taking the expansion force into consideration and determine its least squares recursive expression.

[0108] Specifically, the battery internal resistance voltage is U I (s)=U OC (s)-U L (s), the circuit impedance is Z T (s), whose expression is shown in formula (3):

[0109]

[0110] Among them, τ1=R1C1, τ2=R2C2. Obviously, τ1 and τ2 are also functions of p, which can be expressed as τ1(p) and τ2(p).

[0111] Z T (s) is the system transfer function, which can be mapped from the s domain to the z domain and discretized by the bilinear transformation shown in equation (4), thereby obtaining the battery discrete transfer function shown in equation (5):

[0112]

[0113] Among them, z is a complex variable in the discrete time domain, β0, β1, β2, α1, α2 represent the transformed internal parameters of the battery, and ΔT is the sampling period.

[0114] definition represents the observation vector of the system at each discrete k moment, θ = [β0β1β2α1α2], represents the internal parameter vector of the battery to be identified, then the discrete difference equation can be expressed in the form of a recursive least squares equation, as shown in formula (6):

[0115]

[0116] After sorting, the least squares recursive expression of the model is finally obtained:

[0117]

[0118] Among them, P and K represent the covariance matrix and gain matrix in the recursive process respectively. Represents the estimation result of θ, λ is the forgetting factor, and I is the unit matrix.

[0119] Step 430, perform HPPC test on the battery.

[0120] The internal parameters R0(p), R1(p), C1(p), R2(p), C2(p) required for the execution of the IEAKF algorithm include the open circuit voltage U OCV The above parameters can be obtained by performing HPPC test (Hybrid Pulse Power Characterization) on the battery, performing curve fitting on the test results and combining them with the recursive formula of formula (7).

[0121] In the embodiments of the present application, as described above, the estimation of the battery charge state introduces the expansion force factor. Therefore, when performing HPPC testing on the battery, in addition to considering different initial temperatures, different initial expansion forces also need to be considered.

[0122] Figure 5 FIG. 1 shows a schematic diagram of the experimental conditions for performing HPPC on a certain type of soft-pack lithium cobalt oxide battery in a specific embodiment, such as Figure 5 As shown, the variables that need to be controlled in the HPPC experiment in this application include initial temperature, initial expansion force and experimental conditions.

[0123] Taking into account the temperature range of battery operation, three initial temperature characteristic points of 5°C, 25°C and 45°C are selected.

[0124] In order to show the effect of different initial expansion forces on the battery SOC, the initial expansion forces are selected as 200N, 400N, and 800N to simulate the low pressure, normal pressure, and high pressure states of the battery. pre The choice of is not only related to the battery itself and the final EOL, but also to the contact area of ​​the force sensor. The specific calculation formula is shown in formula (8):

[0125] S pre =P0*A (8),

[0126] Where P0 is the initial pressure and A is the loading area.

[0127] Considering the specific application scenarios of vehicle batteries, BJDST, FUDS, US06, and DST of the Maryland O dataset are selected as the experimental conditions of this article.

[0128] After determining the experimental variables, the current, voltage, temperature, and expansion force data were measured under different initial temperature, initial expansion force, and working conditions. Figure 6 and Figure 7The voltage change and expansion force change data of a certain model of soft-pack lithium cobalt oxide battery subjected to HPPC in a specific embodiment are shown respectively. In the figure, L stands for Low, M stands for Middle, H stands for High, T stands for temperature, P stands for expansion force, and V stands for voltage. LTLP-V means the voltage curve under low temperature and low pressure conditions, that is, the battery voltage under 5°C and 20Kg test conditions. LTMP-P means the expansion force curve under low temperature and medium pressure conditions, that is, the expansion force of the battery under 5°C and 40Kg test conditions. The meanings of other curves are similar to those explained above.

[0129] pass Figure 6 , Figure 7 It can be seen that during the battery charging and discharging process, the pressure change trend is consistent with the voltage change trend, and the inflection point of the expansion force is earlier than the inflection point of the voltage. This phenomenon is due to the change in the material structure causing the change in potential, which further proves the necessity of introducing the expansion force to estimate the battery charging state.

[0130] Step 440, experimental data fitting.

[0131] After completing the HPPC experiment, by fitting the open circuit voltage and terminal voltage data and using formula (7) to perform FFRLS (Fast Forgetting Recursive Least Squares) processing on the experimental data, the R0(p), R1(p), C1(p), R2(p), C2(p) and SOC-OCV fitting curves corresponding to different temperatures and expansion forces can be obtained, and finally the parameter identification is completed.

[0132] 2.2) Battery state space equation and observation equation.

[0133] For nonlinear systems where the state value and observation value are continuously differentiable at any time, we can use Taylor series expansion to linearize the state function and observation value at any state estimation point. By ignoring high-order terms and residual terms, we can obtain an approximate linearized model. After obtaining the linearized model, we can combine it with the Kalman filter to obtain the EKF state equation of the battery, as shown in equation (9):

[0134]

[0135] (9) In the formula, the subscripts k-1 and k are two adjacent estimation moments, and x k-1 、x k are the battery state vectors at time k-1 and time k respectively, are the estimated values ​​of the battery state vector at time k-1 and time k, respectively. The battery state vector includes the battery SOC and the voltages of the two RC loops, u k-1is the system input at time k-1 (in the embodiment of the present application, the system input is the excitation current of the battery), z k is the system observation quantity at time k (in the embodiment of the present application, the system observation quantity is the battery momentum voltage), f() and h() are the battery state function and the observation function respectively, Φ k-1 is the battery state transfer matrix at time k-1, H k is the battery observation matrix at time k, w k-1 is the process noise at time k-1, v k is the observation noise at time k.

[0136] Comparing the expressions of EKF and conventional linear Kalman filter, we can get Among them, A k , B k , C k They represent the state transfer matrix, input matrix and observation matrix in the linear Kalman filter respectively.

[0137] 2.3) Execute the IAEKF algorithm considering the expansion force.

[0138] After establishing the EKF state equation and observation equation as shown in formula (9), the vehicle-side estimation unit can execute the IAEKF algorithm to continuously estimate the SOC at time k through the estimated value of the SOC at time k-1 and the measured data. The estimated result is used as the first estimated value of the battery state of charge at time k. As mentioned above, it can be recorded as

[0139] refer to Figure 3 As shown in the algorithm flow chart, the execution process of the IAEKF algorithm includes the EKF process in the orange part, the iterative update process in the green part, and the adaptive update process in the blue part.

[0140] a.EKF process.

[0141] like Figure 3 As shown, the EKF process in orange includes the following steps:

[0142] First, perform algorithm initialization operation.

[0143] Specifically, for any k-time EKF, the initialization operation includes obtaining the a posteriori estimate of the battery state vector at time k-1: (For the starting time, the initial value can be set based on experience ), obtain the battery status data at the current moment (including excitation current, momentum voltage, momentum expansion force and momentum temperature), and calculate Φ according to the parameter identification results k , B k and Hk .

[0144] After the initialization operation is completed, the prior estimate of the battery state vector at time k is determined by equations (10) and (11) respectively. and the a priori error d of the battery terminal voltage k (i.e., prior observation error):

[0145]

[0146] in, and That is, u k-1 Substitute into the state transfer matrix and The specific value obtained after substituting into the observation function.

[0147] Then, the posterior estimate of the battery state vector at time k is updated by iterative Kalman filtering and from Extract the first estimated value of the battery state of charge at time k

[0148] In completion After the update of , the battery terminal voltage z at time k is further determined by formula (12) k The residual ε k (i.e., a posteriori observation error):

[0149]

[0150] Finally, the k time is updated and the IAEKF algorithm is returned and re-executed until the estimation of the battery state of charge is completed.

[0151] b. Iterative Kalman filter update

[0152] like Figure 3 As shown, each time The update is achieved through iterative Kalman filtering, which mainly includes three parts: initialization, measurement update, and state and covariance maintenance.

[0153] The iterative algorithm used for measurement update is shown in (13) and (14):

[0154]

[0155]

[0156] Where superscript T is the matrix transpose, superscript i and i+1 represent two consecutive iterations, i≥0, are the i-th and i+1-th iteration results at time k, respectively. is the Kalman gain of the i-th iteration at time k. Obviously, when i = 0, That is, using a priori estimates Initialize the iterative update of the posterior estimate.

[0157] like Figure 3 As shown, the end condition of iterative Kalman filtering is Where δ is the preset accuracy threshold. When the error of two consecutive iterations is less than δ, the optimization work can be considered nonlinear. As This completes the update of the posterior estimate.

[0158] Prior error covariance matrix Posterior error covariance matrix The update of is performed at the beginning and end of each iterative Kalman filter by equation (15):

[0159]

[0160] Among them, Φ k-1 (or as Figure 3 As shown in ), are the process transfer matrix and the posterior error covariance matrix at time k-1, and I is the unit matrix. Obviously, it is used to calculate of and The values ​​are taken at the end of the iteration.

[0161] c. Adaptive update R k , Q k .

[0162] In the conventional EKF prediction step, R can be determined by (16): k :

[0163]

[0164] Among them, S k Represents the new information covariance matrix at time k. However, using formula (16) cannot guarantee that its estimated value is a positive definite matrix, because it is estimated by subtracting two positive definite matrices. Therefore, in the embodiment of the present application, an adaptive method based on residuals is used to perform R through formula (17). t estimate.

[0165]

[0166] Where E[] is the expected value function, is the observation matrix corresponding to the end of the iteration at time k (that is, the number of iterations i at the end of the iteration corresponding to ).

[0167] More preferably, the relationship between the previous and next errors is realized by introducing the forgetting factor α instead of the time window, and the R shown in formula (18) is obtained. k Update formula:

[0168]

[0169] Among them, R k-1 , ε k-1 are the measurement noise covariance and the residual of the battery terminal voltage at time k-1, respectively. Obviously, the value of α is determined by the update of the error information before and after two adjacent moments. Therefore, R in (18) k The update operation means that if the error at a certain point suddenly diverges, the error at that moment is discarded and the error at the previous moment is used. This adaptive update method can effectively improve the stability of SOC estimation and avoid the influence of mutation points on the estimation results.

[0170] In order to adaptively estimate Q k-1 , the process noise can be calculated by equation (19):

[0171] w k-1 =x k -Φ(x k-1 ,u k-1 ) (19).

[0172] The update formula of the extended Kalman filter can be obtained as formula (20):

[0173]

[0174] In the embodiments of the present application, k The process is the same as that of the previous one. We introduce the forgetting factor β and make the same assumptions. Finally, we get Q as shown in formula (21): k The update formula is:

[0175]

[0176] Among them, Q k-1 ,d k-1 are the process noise covariance at time k-1 and the prior error of the battery terminal voltage, is the Kalman gain corresponding to the end of the iteration at time k (that is, the number of iterations i corresponding to the end of the iteration ).

[0177] In the embodiments of the present application, Figure 3 As shown, the above Rk , Q k Updates are implemented through an adaptive process.

[0178] <3. SOC estimation in the cloud>

[0179] 3.1) Overall structure and process.

[0180] Figure 8 FIG. 4 shows a schematic diagram of the architecture of a cloud estimation unit in some embodiments of the present application, such as Figure 8 As shown, the cloud estimation unit can be set in the cloud server, and its core for battery SOC estimation is the GA-CNN-LSTM joint prediction model. In addition, preferably, the unit also includes a communication module and a storage module. The communication module communicates with the vehicle end through a wireless network to obtain the battery status data sent by the vehicle end, and sends the battery SOC estimation result to the vehicle end. The storage module is used to store the battery status data and the battery SOC estimation result.

[0181] Furthermore, if Figure 8 As shown, in the embodiment of the present application, the LSTM model estimates the battery SOC at time k based on the battery status data obtained at multiple historical moments. In the embodiment of the present application, the estimation result is recorded as The GA model and the CNN model are respectively connected to the LSTM model for feedback. The GA model is used to optimize the hyperparameters of the LSTM model based on genetic algorithms to improve the network structure of the LSTM model. The CNN model is used to generate the expansion force characteristic parameters of the battery that are fed back to the LSTM model. The expansion force characteristic parameters can effectively characterize how the expansion force characteristics of the battery change with the SOC state. Using them together with the collected battery normal state data as the input of the LSTM model can greatly enhance the ability to train the LSTM network and use the LSTM network for SOC estimation.

[0182] Fig. 9 FIG. 4 shows an overall algorithm flow chart for training the above GA-CNN-LSTM joint prediction model in some preferred embodiments, as shown in FIG. Fig. 9 As shown, after importing the collected battery status data, the theoretical state of charge SOC of the battery SOC can be obtained by the ampere-hour integration method. real , which is used as the evaluation benchmark for the training effect of each model, and then continuously drives each model to perform its own training. Among them, the LSTM model continuously outputs the estimated value SOC of the battery state of charge during its training process. pre , and use the above SOC real and SOC preEvaluate the training effects of the GA model and the CNN model (it should be noted that the SOC of the GA model and the CNN model input pre It is not the second estimate of SOC output by the LSTM model. ).

[0183] Understandably, Fig. 9 The process shown can be used for training the joint prediction model, and can also be implemented in the process of using the LSTM model for SOC estimation. That is, the above-mentioned GA model and CNN model can optimize the architecture and input amount of the LSTM model in real time according to the battery status data during the cloud-based estimation of the battery SOC.

[0184] The specific implementation methods of the three models are described in detail below.

[0185] 3.2) LSTM model.

[0186] LSTM is a commonly used time series prediction model, which mainly solves the gradient vanishing or gradient exploding problems faced by traditional RNNs when processing long sequence data. LSTM effectively captures long-term dependencies by introducing three "gate" structures (forget gate, input gate, output gate) to control the information flow; forget gate: determines which information should be forgotten or discarded. Input gate: determines which new information should be stored in the cell state; input gate: determines which new information should be stored in the cell state.

[0187] Fig.10 The schematic diagram of the architecture of the LSTM model is shown. By optimizing the hyperparameters of the LSTM model and expanding the dimension and feature representation capabilities of the input data, the ability of the LSTM model to estimate the battery SOC can be effectively improved.

[0188] Specifically, the hyperparameters of the LSTM model include the time window length, the number of hidden layer neurons, the learning rate and the training batch size of the LSTM model; its basic input data includes [I, U, p, T] collected at each moment, on this basis, characteristic parameters reflecting the changes in expansion force affected by SOC can be further superimposed.

[0189] 3.3) Optimization of LSTM hyperparameters by GA model.

[0190] In an embodiment of the present application, the optimization of the LSTM model hyperparameters is implemented by the GA model through the execution of a genetic algorithm (GA). The genetic algorithm is a method for finding the optimal solution by simulating the natural evolution process. It uses computer simulation operations to convert the problem-solving process into a process similar to the crossover and mutation of chromosome genes in biological evolution.

[0191] Fig.11 In some specific embodiments, a flowchart of the implementation of the genetic algorithm is shown, such as Fig.11 As shown in Figure 2, the genetic algorithm takes the current hyperparameters of the LSTM model as input, first selects a suitable population, then uses the norm of the SOC estimation error shown in formula (22) as the fitness function L, selects the optimal indicator through the selection of intermediate parameters, crossover and mutation of encoding and decoding, and finally obtains the optimal time window, number of hidden layer units, learning rate and training batch size:

[0192] L=||SOC real -SOC pre || (22).

[0193] 3.4) Optimization of LSTM input by CNN model.

[0194] Convolutional neural network is a commonly used feature extraction method in the field of machine vision, which can effectively highlight and extract specific features in raw data, such as Fig.12 As shown, in some specific embodiments of the present application, the CNN model includes an input layer, a convolutional layer, a pooling layer and a fully connected layer.

[0195] Specifically, the input layer concatenates the battery status data at each moment (i.e., the [I, U, p, T] vector at each moment), and then performs a convolution operation with a 1*3 convolution kernel. Through the above 1*3 one-dimensional convolution kernel, two characteristic parameters related to the expansion force can be extracted from each [I, U, p, T] vector, namely Fig.12 The result of the convolution of [I,U,p] and [U,p,T] is obtained, and then the time window length M optimized by the GA model and the length N of the time series for SOC estimation are intercepted at each moment, and finally a convolution layer of size M*2*N is obtained.

[0196] Furthermore, the above convolutional layer is compressed by the pooling layer to obtain a pooling layer of size 1*2*N, wherein each moment corresponds to a 1*2 dimensional vector, and the two parameters in the two-dimensional vector are called expansion force characteristic parameters, which reflect the characteristic values ​​of the three passive quantities of momentum voltage, momentum expansion force and momentum temperature under a specific excitation current. The 1*2 dimensional vector is further input into LSTM, thereby realizing the dimensional expansion of the LSTM input, and the input at each moment will become 4M+2, that is, M [I,V,p,T] vectors obtained according to the time window length M, plus the two newly generated expansion force characteristic parameters.

[0197] In addition, the training result output by the output layer is an estimated value of the battery SOC, which can be compared with the SOC input by the LSTM model. preConstruct a cost function to train the CNN model.

[0198] <IV. SOC estimation based on vehicle-cloud fusion>

[0199] The first estimated value of SOC is obtained by estimating it through the vehicle side and the cloud side respectively. and the second estimate After that, the fusion estimation unit is A one-dimensional Kalman filter is performed to obtain the final estimate of the battery state of charge.

[0200] Specifically, the result output by the vehicle-side estimation unit can be As input, the output of the cloud estimation unit As the output, the above assumptions are rearranged into the state equation and observation equation of the one-dimensional Kalman filter shown in equation (23):

[0201]

[0202] in, The final estimated value of the battery state of charge at time k+1 is displayed on the vehicle's dashboard or control panel. Represents the measurement error of the vehicle-side sensor and the vehicle-side output The sum of the errors with the true value (the true value can also be the result of the ampere-hour integration method), It represents the cloud output The error from the true value.

[0203] Formula (23) can be used to construct an iterative expression of the one-dimensional Kalman filter, thereby continuously passing through the k-th moment get This enables battery state of charge estimation for vehicle-cloud fusion.

[0204] <V. Specific Examples>

[0205] In order to verify the effectiveness of the estimation method provided in the present application, through specific embodiments, the battery state of charge under different working conditions is estimated respectively, wherein the initial stress is 20Kg.

[0206] The specific steps are:

[0207] First, the RC network was constructed based on the data collected by HPPC to verify the accuracy of the model;

[0208] Then, the traditional time window AEKF algorithm and the vehicle-side IAEKF algorithm proposed in this application are used to verify the four working conditions of BJDST, FUDS, US06 and DST;

[0209] At the same time, the [I, V, p, T] data at each moment is obtained and processed in the cloud. After optimization by the CNN model, the data input dimension is 4M+2 (M is the optimal time window length). In order to verify the effect of the GA-CNN-LSTM joint prediction model, simulation calculations are carried out under the above four working conditions and compared with the traditional LSTM and optimized prediction results;

[0210] Finally, high-precision SOC is calculated by Kalman fusion.

[0211] Fig.13 , Fig.14 The estimation results and errors of battery SOC using different algorithms under BJDST conditions are shown respectively; Fig.15 , Fig.16 The estimation results and errors of battery SOC using different algorithms under FUDS conditions are shown respectively; Fig.17 , Fig.18 The estimation results and errors of battery SOC using different algorithms under US06 conditions are shown respectively. Fig.19 , Fig. 20 The estimation results and errors of battery SOC using different algorithms under DST conditions are shown respectively. The RMSE and MAE of various algorithms under four conditions are listed in Tables 1 to 4 respectively.

[0212] Table 1 RMSE and MAE of each algorithm under BJDST condition

[0213]

[0214] Table 2 RMSE and MAE of each algorithm under FUDS condition

[0215]

[0216]

[0217] Table 3 RMSE and MAE of each algorithm under US06 condition

[0218]

[0219] Table 4 RMSE and MAE of each algorithm under DST conditions

[0220]

[0221] like Figures 13 to 20As shown in Tables 1 to 4, under the BJDST, FUDS, US06 and DST conditions, the RMSE of AEKF is 1.92%, 2.1%, 2.54% and 1.9% respectively, and the RMSE of the IAEKF algorithm used on the vehicle side in this application is 1.39%, 1.15%, 1.71% and 1.52% respectively, which shows the effectiveness of using the IAEKF algorithm considering the expansion force on the vehicle side; the RMSE of LSTM on the cloud side is 1.25%, 1.93%, 1.72% and 1.68% respectively, the RMSE of GA-LSTM is 1.14%, 0.98%, 1.15% and 0.92% respectively, and the RMSE of GA-CNN-LSTM is 0.91%, 0.66%, 0.80% and 0.76% respectively, which proves that the optimization effect of the CNN-GA-LSTM joint prediction model provided in this application is obvious.

[0222] Finally, the root mean square errors of SOC estimation by IAEKF fused with GA-CNN-LSTM are 0.57%, 0.58%, 0.58% and 0.55%, respectively, indicating that under the dynamic conditions of actual vehicle driving, the charge state estimation of vehicle-cloud fusion is more accurate.

[0223] The above is a detailed introduction to the specific implementation methods of the present application. For those skilled in the art, several improvements and modifications may be made to the present application without departing from the principles of the present application. These improvements and modifications also fall within the scope of protection of the claims of the present application.

Claims

1. A method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force, characterized in that: The following steps are involved: Acquire battery status data at the vehicle end and execute the IAEKF algorithm to obtain a first estimated value of the battery state of charge, wherein the battery status data includes an excitation current, a momentum voltage, a momentum expansion force, and a momentum temperature of the battery; Acquire battery status data in the cloud and use a joint prediction model including a GA model, a CNN model, and an LSTM model to obtain a second estimated value of the battery state of charge, wherein the GA model is used to optimize the hyperparameters of the LSTM model, the CNN model is used to generate an expansion force characteristic parameter of the battery that is fed back to the LSTM model, and the LSTM model outputs a second estimated value of the battery state of charge based on the battery status data and the expansion force characteristic parameter; A one-dimensional Kalman filter is performed on the first estimated value and the second estimated value of the battery state of charge to obtain a final estimated value of the battery state of charge.

2. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 1 is characterized in that: The parameters required to execute the IAEKF algorithm are identified by the following steps: Establish a second-order RC model of the battery considering the expansion force and determine its state expression: Among them, s is a complex variable, U L , U OC ,I L are the terminal voltage, open circuit voltage and excitation current of the battery respectively, p is the expansion force variable of the battery, R0(p), R1(p), C1(p), R2(p) and C2(p) are the internal resistance, concentration polarization resistance, concentration polarization capacitance, electrochemical polarization resistance and electrochemical polarization capacitance of the battery respectively; Discretize the battery second-order RC model considering the expansion force to obtain its least squares recursive expression; HPPC experiments were conducted on the battery under different initial expansion forces, initial temperatures and test conditions; The HPPC experimental data were recursively fitted to achieve parameter identification, where the parameter identification results were the R0(p), R1(p), C1(p), R2(p), C2(p) and SOC-OCV fitting curves corresponding to different expansion forces and temperatures.

3. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 2 is characterized in that: The IAEKF algorithm includes the following steps: Initialize the algorithm; The prior estimate of the battery state vector at time k is determined based on the following formula: in, is the a posteriori estimate of the battery state vector at time k-1, u k-1 is the battery excitation current at time k-1, Φ is the battery state transfer matrix; The a priori error d of the battery terminal voltage at time k is determined based on the following formula: k : Among them, z k is the battery terminal voltage at time k, h() is the battery observation function; Determine the a posteriori estimate of the battery state vector at time k by iterative Kalman filtering and from The first estimated value of the battery state of charge at time k is extracted from the iterative Kalman filter, wherein the iterative Kalman filter adaptively updates the measurement noise covariance R at time k. k , process noise covariance Q k , prior error covariance matrix and the posterior error covariance matrix The residual ε of the battery terminal voltage at time k is determined based on the following formula: k : Update the k time and re-execute the IAEKF algorithm until the estimation of the battery state of charge is completed.

4. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 3 is characterized in that: The iterative Kalman filter determines the a posteriori estimate of the battery state vector at time k by the following formula: Where superscript T is the matrix transpose, superscript i and i+1 represent two consecutive iterations, i≥0, are the i-th and i+1-th iteration results at time k, respectively, and is the Kalman gain of the i-th iteration at time k, v k is the observation noise at time k, δ is the preset accuracy threshold; And, the iterative Kalman filter determines the prior error covariance matrix at time k by the following formula: Posterior error covariance matrix Among them, Φ k-1 , are the process transfer matrix and the posterior error covariance matrix at time k-1 respectively, and I is the unit matrix.

5. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 4 is characterized in that: The measurement noise covariance R at time k k Determined based on the following formula: Among them, α is the forgetting factor, E[] is the expected value function, R k-1 , ε k-1 are the measurement noise covariance and the residual of the battery terminal voltage at time k-1, is the observation matrix corresponding to the end of the iteration at time k; The process noise covariance Q at time k k Determined based on the following formula: Among them, β is the forgetting factor, Q k-1 d k-1 are the process noise covariance at time k-1 and the prior error of the battery terminal voltage, is the Kalman gain corresponding to the end of the iteration at time k.

6. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 1, characterized in that: The GA model uses a genetic algorithm to optimize the hyperparameters of the LSTM model, wherein the hyperparameters include the time window length, the number of hidden layer neurons, the learning rate and the training batch size of the LSTM model, and the fitness function L of the genetic algorithm is shown as follows: L=||SOC real -SOC pre ||, Among them, SOC real SOC is the estimated value of the battery state of charge obtained by the ampere-hour integration method. pre It is an estimated value of the battery state of charge output by the LSTM model, and the estimated value is used to train the GA model and the CNN model.

7. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 6 is characterized in that: The input data of the CNN model are the excitation current I, momentum voltage U, momentum expansion force p and momentum temperature T of the battery at each moment, and the output data are two expansion force characteristic parameters of the battery at each moment, wherein the two expansion force characteristic parameters are respectively used to characterize the [I, U, p] related characteristics and [U, p, T] related characteristics of the battery at each moment; The convolution kernel size of the CNN model is 1*3, and the convolution layer size is M*2*N, where M is the time window length of the LSTM model, and N is the length of the time series for battery state of charge estimation; The CNN model is based on the SOC output of the LSTM model pre Conduct training.

8. The method for estimating the state of charge of a vehicle-cloud fusion battery based on expansion force according to claim 7 is characterized in that: The input data of the LSTM model are M groups of [I, U, p, T] data of the battery at each time, and two expansion force characteristic parameters output by the CNN model. The output data is the estimated SOC of the battery state of charge at each time. pre and a second estimate of the battery state of charge.

9. A vehicle-cloud fusion battery state of charge estimation system based on expansion force, characterized in that: The following operations are included: A vehicle-side estimation unit disposed inside the vehicle, used to obtain battery status data and execute an IAEKF algorithm to obtain a first estimated value of the battery state of charge, wherein the battery status data includes an excitation current, a momentum voltage, a momentum expansion force, and a momentum temperature of the battery; A cloud estimation unit provided in a cloud server, for acquiring battery status data and using a joint prediction model including a GA model, a CNN model and an LSTM model to obtain a second estimated value of the battery state of charge, wherein the GA model is used to optimize the hyperparameters of the LSTM model, the CNN model is used to generate an expansion force characteristic parameter of the battery which is fed back to the LSTM model, and the LSTM model predicts and outputs a second estimated value of the battery state of charge based on the battery status data and the expansion force characteristic parameter; The fusion estimation unit is used to perform a one-dimensional Kalman filter on the first estimation value and the second estimation value of the battery state of charge to obtain a final estimation value of the battery state of charge.

Images