Method for estimating SOC (State of Charge) of power battery pack driven by big data under actual operation condition

Abstract

The invention discloses a method for estimating the SOC (state of charge) of a power battery pack driven by big data under an actual operation condition. The method comprises the following steps: acquiring a multi-dimensional time sequence monitoring signal of an electric vehicle and a battery pack thereof under the actual operation condition; selecting a signal dimension of which the correlation with the SOC value of the battery pack is greater than a preset value; determining that terminal voltage distribution in the battery pack accords with Weibull distribution and temperature distribution accords with normal distribution; estimating a Weibull distribution parameter and a normal distribution parameter by using a maximum likelihood method; extracting Weibull distribution characteristics; fusing the signal dimension, the Weibull distribution characteristics and the normal distribution parameters to form fused data; fusing the data as the input of a deep learning model, and carrying out offline training; and inputting online data into the trained deep learning model, and outputting an estimated SOC value of the battery pack by the deep learning model. According to the estimation method, the SOC value of the power battery pack under the actual operation condition can be effectively estimated, and the estimation precision of the SOC value of the battery pack is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power battery packs, and in particular to a method for estimating the SOC of a power battery pack under actual operating conditions driven by big data. Background Art

[0002] In the process of large-scale promotion and application of new energy vehicles, problems such as low mileage estimation accuracy, poor environmental adaptability, and sudden fire accidents still pose certain obstacles to the rapid and healthy development of the industry. Among these problems, the SOC (State of Charge) in the electric vehicle battery management system (BMS) is a crucial indicator. Accurate online estimation of the SOC value of the battery pack can effectively prevent safety accidents caused by overcharging and over-discharging of the battery, and provide high-value reference for the charging and discharging strategy and efficient energy utilization of the whole vehicle. With the development of advanced communication technology and information technology, big data and deep learning have provided new opportunities for solving battery management problems. The introduction of these new technologies will help optimize battery management strategies, improve battery life and safety, and promote the sustainable and healthy development of the industry.

[0003] By building a full life cycle database based on battery information monitored and uploaded by the vehicle's BMS, and combining big data and artificial intelligence algorithms to conduct intelligent analysis and research on battery performance and status under massive driving data, intelligent battery management can be achieved for different application scenarios, working conditions, climates, and regional conditions. Effectively improving the performance of new energy vehicle products and improving product quality is not only an important means to achieve high-quality development of the new energy industry, but also an inevitable direction for jointly promoting the transformation and upgrading of my country's automobile industry.

[0004] At present, battery pack SOC estimation is completed by BMS. However, the SOC estimation model in BMS is usually fixed, not updated with battery aging, and cannot accurately reflect the evolution trend of battery inconsistency. Although the traditional ampere-hour integration method has low implementation cost and strong real-time performance, its estimation accuracy is greatly affected by factors such as initial SOC, capacity decay, current sensor accuracy drift and noise. The open circuit voltage-based method requires the lithium-ion battery to be disconnected from the load and experience a sufficiently long relaxation time, but this limits the real-time performance of the method. Although the model-based method overcomes the error accumulation problem of the ampere-hour integration method, its performance depends largely on the accuracy of battery behavior modeling. Given the high nonlinearity and time variability of the battery system, it is almost impossible to perfectly incorporate various influencing factors into complex mathematical equations. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method for estimating the SOC of a power battery pack under actual operating conditions driven by big data, which can more accurately estimate the SOC value of the power battery pack.

[0006] The technical solution adopted by the present invention to solve the above technical problems is: a method for estimating the SOC of a power battery pack under actual operating conditions driven by big data, comprising the following steps:

[0007] S1. Obtain multi-dimensional timing monitoring signals of electric vehicles and their battery packs under actual operating conditions;

[0008] S2. Perform correlation analysis on the multi-dimensional time series monitoring signal, and select the signal dimension whose correlation with the battery pack SOC value is greater than a preset value;

[0009] S3. extracting a list of terminal voltages of the series-connected cells of the battery pack at each sampling moment, and determining that the distribution of the terminal voltages of the series-connected cells in the battery pack conforms to the Weibull distribution according to the distribution characteristics of the terminal voltages;

[0010] S4. Estimate Weibull distribution parameters using maximum likelihood method;

[0011] S5. Extracting Weibull distribution features based on the deterministic Weibull distribution model;

[0012] S6. extracting a temperature list of the battery pack series cells at each sampling moment, and determining that the distribution conforms to a normal distribution according to the distribution characteristics of the temperature of the series cells in the battery pack;

[0013] S7. Estimate normal distribution parameters using maximum likelihood method;

[0014] S8. Fusing the signal dimension in step S2, the Weibull distribution feature in step S5, and the normal distribution parameter in step S7 to form fused data;

[0015] S9. The fused data is used as the input of the deep learning model for offline training;

[0016] S10. Online data is input into the deep learning model trained in step S9, and the deep learning model outputs an estimated battery pack SOC value.

[0017] Preferably, in step S1, a multi-dimensional timing monitoring signal of the electric vehicle and its battery pack under actual operating conditions is obtained from the electric taxi operating platform, and the multi-dimensional timing monitoring signal includes vehicle speed, vehicle driving status, vehicle charging status, battery pack total output voltage, battery pack total output current, vehicle mileage, battery pack SOC value, battery pack unit maximum output voltage, battery pack unit minimum output voltage, battery pack temperature, battery pack temperature, battery pack temperature, battery pack insulation resistance, and DCDC converter status.

[0018] Preferably, in step S2, a correlation analysis is performed on the multi-dimensional timing monitoring signal, and signal dimensions having a correlation with the battery pack SOC value greater than 0.5 are selected, wherein the signal dimensions include vehicle speed, total battery pack output voltage, total battery pack output current, maximum battery pack unit output voltage, minimum battery pack unit output voltage, local maximum temperature within the battery pack, and local minimum temperature within the battery pack.

[0019] Preferably, in step S3, based on the characteristic that the voltage distribution at the terminals of the series-connected cells of the battery pack is right-skewed compared to the normal distribution, it is determined that the distribution conforms to the Weibull distribution.

[0020] Preferably, in step S6, according to the characteristics that the temperature distribution of the battery pack series-connected cells is centered at the mean and symmetrical on both sides, it is determined that the distribution conforms to the normal distribution.

[0021] Preferably, in step S4, the method of estimating Weibull distribution parameters using the maximum likelihood method comprises the following steps:

[0022] S4-1. Establish the likelihood function, assuming n observations x1, x2, ..., x n , then the likelihood function is defined as:

[0023]

[0024] In the formula, v i is a random variable, λ is a scale parameter, k is a shape parameter, and γ0 is an initial position parameter;

[0025] S4-2. Take the logarithm of both sides of the likelihood function to obtain the log-likelihood function:

[0026]

[0027] S4-3. Calculate the partial derivatives of the parameters k and λ of the above log-likelihood function and set them to zero to obtain the equation system:

[0028]

[0029] S4-4. By setting the initial location parameter γ0 in the set of equations to 0 and solving them, the maximum likelihood estimates of the initial shape parameters and scale parameters can be obtained; then, they are substituted into the unbiased estimate of the location parameters to obtain new location parameters.

[0030] Preferably, in step S7, the method of estimating normal distribution parameters using the maximum likelihood method comprises the following steps:

[0031] S6-1. Establish the likelihood function. For the given sample data x1, x2, ..., x n, assume that these data are independently and randomly sampled from a normal distribution N(μ,σ 2 ), then the likelihood function of the sample can be expressed as:

[0032]

[0033] where, Π represents the product symbol, and x i represents the i-th observed value.

[0034] S6-2. Take the logarithm of both sides of the likelihood function to obtain the log-likelihood function:

[0035]

[0036] S4-3. Take the partial derivatives of the above log-likelihood function with respect to the mean μ and variance σ respectively, and set them to zero to obtain a system of equations:

[0037]

[0038] Solve to obtain:

[0039] Preferably, in the step S5, when extracting Weibull distribution characteristics, in order to characterize the symmetry and tail characteristics of the discrete data histogram of the terminal voltage of the battery pack series units, skewness and kurtosis are used. The mean μ1 and variance σ1 of the Weibull distribution 2 can be calculated by the following formulas:

[0040]

[0041] where λ and k are the scale parameter and shape parameter in the step S4 respectively.

[0042] The calculation formulas for the skewness and kurtosis of the Weibull distribution are:

[0043]

[0044] In the formula, N is the sample size, and x i is the i-th sample value.

[0045] Preferably, in the step S8, the moving window method is used to resample the fusion data. The length of the moving window is T, and the window moves s steps each time, where s < T. The input-output mapping scale of the deep learning model in the step S7 is denoted as T→s, and the resampled data is used as the input of the deep learning model.

[0046] Preferably, in the step S9, the deep learning model adopts a convolutional neural network-bi-directional long short-term memory network model integrating an attention mechanism.

[0047] Preferably, in step S10, a data set of the electric vehicle and its battery pack during actual operation is collected, and the collected data set during actual operation is resampled using a moving window method and used as online data to input the deep learning model trained in step S7; the collected data set during actual operation includes the following data of the electric vehicle during actual operation: vehicle speed, total output voltage of the battery pack, total output current of the battery pack, maximum output voltage of the battery pack unit, minimum output voltage of the battery pack unit, local maximum temperature in the battery pack, and local minimum temperature in the battery pack.

[0048] The beneficial effects of the present invention are as follows: the method for estimating the SOC of a power battery pack under actual operating conditions driven by big data of the present invention can effectively estimate the SOC value of a power battery pack under actual operating conditions, train the model through periodic actual operating condition data, and update the model parameters to make it better adapt to the changes and evolution of the battery system, maintain the effectiveness and reliability of the model, and ensure that the model still has a high predictive ability under different operating conditions.

[0049] The present invention adds terminal voltage distribution characteristics and temperature distribution characteristics to the input dimensions of the deep learning model, which compensates for the influence of cell inconsistency on the estimation accuracy and greatly improves the estimation accuracy of the SOC value of the battery pack. Compared with the existing battery pack SOC estimation method, the present invention can avoid the previous deep learning-based methods that do not consider the influence of single cell inconsistency in the battery pack, and significantly improves the estimation accuracy of the SOC value of the actual operating battery pack. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the estimation method of the present invention;

[0051] Figure 2 is the multi-dimensional time series monitoring signal used in step S2 of the verification case adopted by the present invention;

[0052] Figure 3 These are the eight signal dimensions selected based on correlation analysis in step S3 of the verification case adopted by the present invention;

[0053] Figure 4 It is a discrete data histogram of the terminal voltage of the series-connected cells of the battery pack in the verification case adopted by the present invention and a fitted Weibull distribution curve;

[0054] Figure 5 It is a histogram of discrete data of the temperature of the series-connected cells of the battery pack and a fitted normal distribution curve in the verification case adopted by the present invention;

[0055] Figure 6 is a schematic diagram of the structure of the deep learning model used in step S9 of the present invention;

[0056] Figure 7 It is a final result diagram of estimating the SOC value of a battery pack in a verification case using the estimation method of the present invention. DETAILED DESCRIPTION

[0057] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments only represent a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0058] like Figure 1 As shown, a method for estimating the SOC of a power battery pack under actual operating conditions driven by big data includes the following steps:

[0059] S1. Obtain multi-dimensional timing monitoring signals of electric vehicles and their battery packs under actual operating conditions;

[0060] S2. Perform correlation analysis on the multi-dimensional time series monitoring signal, and select the signal dimension whose correlation with the battery pack SOC value is greater than a preset value;

[0061] S3. extracting a list of terminal voltages of the series-connected cells of the battery pack at each sampling moment, and determining that the distribution of the terminal voltages of the series-connected cells in the battery pack conforms to the Weibull distribution according to the distribution characteristics of the terminal voltages;

[0062] S4. Estimate Weibull distribution parameters using maximum likelihood method;

[0063] S5. Extracting Weibull distribution features based on the deterministic Weibull distribution model;

[0064] S6. extracting a temperature list of the battery pack series cells at each sampling moment, and determining that the distribution conforms to a normal distribution according to the distribution characteristics of the temperature of the series cells in the battery pack;

[0065] S7. Estimate normal distribution parameters using maximum likelihood method;

[0066] S8. Fuse the signal dimension described in step S2, the Weibull distribution characteristics described in step S5, and the normal distribution parameters described in step S7 to form fused data; the fused data is a higher-dimensional time series signal data.

[0067] S9. The fused data is used as the input of the deep learning model for offline training;

[0068] S10. Online data is input into the deep learning model trained in step S9, and the deep learning model outputs an estimated battery pack SOC value.

[0069] Specifically, in an optional implementation, in the step S1, a multi-dimensional timing monitoring signal of the electric vehicle and its battery pack under actual operating conditions is obtained from the electric taxi operating platform, and the multi-dimensional timing monitoring signal includes vehicle speed, vehicle driving status, vehicle charging status, battery pack total output voltage, battery pack total output current, vehicle mileage, battery pack SOC value, battery pack unit maximum output voltage, battery pack unit minimum output voltage, battery pack temperature, battery pack temperature, battery pack temperature, battery pack insulation resistance, and DCDC converter status.

[0070] like Figure 2 As shown in the figure, the multi-dimensional time series monitoring signal obtained from the electric taxi operation platform contains 13 dimensions: vehicle speed speed, vehicle driving state vehicle_state, vehicle charging state charging_status, battery pack output total voltage total_volt, battery pack output total current total_current, vehicle mileage mileage, battery pack SOC value standard_SOC, battery pack unit maximum output voltage max_cell_volt, battery pack unit minimum output voltage min_cell_volt, battery pack maximum temperature point temperature max_temp, battery pack minimum temperature point temperature min_temp, battery pack insulation resistance isulate_r, DCDC converter state dcdc_stat. The sampling interval of the signal is 10s, and the sampling interval greater than 10s is filled with a gray background. Among these 13 dimensions, the values ​​of three dimensions are predefined: (1) vehicle_state, where 01 indicates vehicle start and 02 indicates vehicle shutdown; (2) charging_status, where 01 indicates parking charging, 02 indicates driving charging, 03 indicates not charging, and 04 indicates charging completed; (3) dcdc_stat, where 01 indicates working and 02 indicates interrupted.

[0071] Specifically, in an optional implementation, in step S2, correlation analysis is performed on the multi-dimensional time series monitoring signal to select the signal dimension with a correlation greater than 0.5 with the battery pack SOC value, such as Figure 3 As shown, the signal dimensions include vehicle speed, total battery output voltage, total battery output current, maximum battery unit output voltage, minimum battery unit output voltage, local maximum temperature in the battery pack, and local minimum temperature in the battery pack. The above correlation analysis is based on Pearson correlation analysis.

[0072] Specifically, in an optional implementation manner, in step S3, if Figure 4As shown, according to the characteristic that the voltage distribution of the battery pack terminals in series is right-skewed compared to the normal distribution, it is determined that its distribution conforms to the Weibull distribution.

[0073] Specifically, in an optional implementation manner, in step S4, the method for estimating Weibull distribution parameters using the maximum likelihood method includes the following steps:

[0074] S4-1. Establish the likelihood function, assuming n observations x1, x2, ..., x n , then the likelihood function is defined as:

[0075]

[0076] In the formula, v i is a random variable, λ is a scale parameter, k is a shape parameter, and γ0 is an initial position parameter;

[0077] S4-2. Take the logarithm of both sides of the likelihood function to obtain the log-likelihood function:

[0078]

[0079] S4-3. Calculate the partial derivatives of the parameters k and λ of the above log-likelihood function and set them to zero to obtain the equation system:

[0080]

[0081] S4-4. By setting the initial position parameter γ0 in the equation group to 0 and solving it, the maximum likelihood estimates of the initial shape parameters and scale parameters can be obtained; then, they are substituted into the unbiased estimate of the position parameter to obtain a new position parameter, and substituted into the unbiased estimate of the scale parameter to obtain a new scale parameter. Repeat the solution of the equation group in step S4-3. In this process, the shape parameter k and the scale parameter λ gradually decrease, and the position parameter γ gradually increases, and finally stable shape parameters, scale parameters, and position parameters are obtained.

[0082] Specifically, in an optional implementation manner, in step S6, if Figure 5 As shown, according to the characteristics that the temperature distribution of the battery pack series cells is centered at the mean and symmetrical on the left and right sides, it is determined that its distribution conforms to the normal distribution.

[0083] Specifically, in an optional implementation, in step S7, the method of estimating normal distribution parameters using the maximum likelihood method includes the following steps:

[0084] S6-1. Establish the likelihood function. For the given sample data x1, x2, ..., x n , we assume that these data are from a normal distribution N(μ,σ 2), then the likelihood function of the sample can be expressed as:

[0085]

[0086] Among them, Π represents the multiplication symbol, x i represents the i-th observation.

[0087] S6-2. Take the logarithm of both sides of the likelihood function to obtain the log-likelihood function:

[0088]

[0089] S6-3. Calculate the partial derivatives of the mean μ and variance σ of the above log-likelihood function and set them to zero to obtain the equation system:

[0090]

[0091] Solving for this yields:

[0092] Specifically, in an optional implementation, in step S5, when extracting the Weibull distribution features, skewness and kurtosis are used to characterize the symmetry and tail of the discrete data histogram of the battery pack series cell terminal voltage. Specifically, the mean μ1 and variance σ1 of the Weibull distribution are 2 It can be calculated by the following formula:

[0093]

[0094] Wherein λ and k are the scale parameter and shape parameter in step S4 respectively.

[0095] The skewness and kurtosis calculation formulas for the Weibull distribution are:

[0096]

[0097] In the formula, N is the sample size, x i is the i-th sample value.

[0098] Specifically, in an alternative embodiment, in step S8, the initial signal dimension in step S2, the Weibull distribution feature dimension in step S5, and the normal distribution parameter dimension in step S7 are aligned in time to perform splicing in dimensions, forming fused data; the fused data is higher-dimensional time series signal data. The resampling of the fused data is performed using the moving window method, the length of the moving window is T, and the window moves s steps each time, where s < T. The input-output mapping scale of the deep learning model in step S7 is denoted as T→s, and the data formed by resampling is used as the input of the deep learning model. For the time series window data [X t-T+1 , …, X t of length T, X t-T+1 is the input data at time t - T + 1, and X t is the input data at time t. Correspondingly, the array of the battery pack SOC values estimated by the model is [SOC t-s , …, SOC t , where SOC t-s is the battery pack SOC value at time t - s, and SOC t is the battery pack SOC value at time t.

[0099] Specifically, in an alternative embodiment, in step S9, as Figure 6 shows, the deep learning model uses a convolutional neural network - bidirectional long short-term memory network model with a fusion attention mechanism. The bidirectional long short-term memory network (BiLSTM) and the convolutional neural network (CNN) are combined to construct the deep learning model. The former processes the time features of the multi-dimensional time series, and the latter processes the spatial features of the multi-dimensional time series. In order to focus on the more critical and important information in the output features, an attention mechanism (Attention) module is added to the BiLSTM-CNN combined model, which can help improve its information processing ability and obtain better results. Specifically, in an alternative embodiment, a training set, a validation set, and a test set are set up. The model is trained using the training set to optimize the hyperparameters of the network, and the performance of the model is evaluated through the validation set. The validated model is used to predict the test set, and the trained model is saved and waiting to be called during online estimation.

[0100] Specifically, in an optional implementation, in step S10, a data set of the electric vehicle and its battery pack during actual operation is collected, and the collected data set during actual operation is resampled using a moving window method and input into the deep learning model trained in step S7 as online data, and the deep learning model estimates the battery pack SOC value online. The collected data set during actual operation includes the following data during the actual operation of the electric vehicle: vehicle speed, total output voltage of the battery pack, total output current of the battery pack, maximum output voltage of the battery pack unit, minimum output voltage of the battery pack unit, local maximum temperature in the battery pack, and local minimum temperature in the battery pack.

[0101] The SOC value of the power battery pack can be estimated more accurately by the method of estimating the SOC of the power battery pack under actual operating conditions driven by big data of the present invention. The model is trained by periodic actual operating condition data, and the model parameters are updated to better adapt to the changes and evolution of the battery system, maintain the validity and reliability of the model, and ensure that the model still has a high prediction ability under different operating conditions.

[0102] The present invention adds the terminal voltage distribution characteristics to the input dimension of the deep learning model, which compensates for the influence of cell inconsistency on the estimation accuracy and greatly improves the estimation accuracy of the SOC value of the battery pack. Compared with the existing battery pack SOC estimation method, the present invention can avoid the fact that the previous deep learning-based method does not consider the influence of cell inconsistency in the battery pack, and significantly improves the SOC estimation accuracy of the actual operating battery pack.

[0103] like Figure 7 The figure shows the final effect of estimating the SOC value of the battery pack using the method of the present invention, including the tracking curve and the error curve, where the error curve is the actual SOC value (blue curve) minus the estimated SOC value (orange curve). The estimated value is very close to the actual value, indicating that the model has good fitting and strong generalization ability.

[0104] The above description is only a basic implementation mode of the present invention. The protection scope of the present invention is not limited to the above embodiments. Any improvement, equivalent substitution and modification of the present invention without departing from the principle of the present invention shall be regarded as the protection scope of the present invention.

Claims

1. A method for estimating the SOC of a power battery pack under actual operating conditions driven by big data, characterized in that: The steps include: S1. Obtain multi-dimensional timing monitoring signals of electric vehicles and their battery packs under actual operating conditions; S2. Perform correlation analysis on the multi-dimensional time series monitoring signal, and select the signal dimension whose correlation with the battery pack SOC value is greater than a preset value; S3. extracting a list of terminal voltages of the series-connected cells of the battery pack at each sampling moment, and determining that the distribution of the terminal voltages of the series-connected cells in the battery pack conforms to the Weibull distribution according to the distribution characteristics of the terminal voltages; S4. Estimate Weibull distribution parameters using maximum likelihood method; S5. Extracting Weibull distribution features based on the deterministic Weibull distribution model; S6. extracting a temperature list of the battery pack series cells at each sampling moment, and determining that the distribution conforms to a normal distribution according to the distribution characteristics of the temperature of the series cells in the battery pack; S7. Estimate normal distribution parameters using maximum likelihood method; S8. Fusing the signal dimension in step S2, the Weibull distribution feature in step S5, and the normal distribution parameter in step S7 to form fused data; S9. The fused data is used as the input of the deep learning model for offline training; S10. Online data is input into the deep learning model trained in step S9, and the deep learning model outputs an estimated battery pack SOC value.

2. The method for estimating the SOC of a power battery pack under actual operating conditions driven by big data according to claim 1, characterized in that: In the step S1, a multi-dimensional timing monitoring signal of the electric vehicle and its battery pack under actual operating conditions is obtained from the electric taxi operating platform, and the multi-dimensional timing monitoring signal includes vehicle speed, vehicle driving status, vehicle charging status, battery pack total output voltage, battery pack total output current, vehicle mileage, battery pack SOC value, battery pack unit maximum output voltage, battery pack unit minimum output voltage, battery pack temperature, battery pack temperature, battery pack temperature, battery pack insulation resistance, and DCDC converter status.

3. The method for estimating the SOC of a power battery pack under actual operating conditions driven by big data according to claim 1, characterized in that: In the step S2, correlation analysis is performed on the multi-dimensional time series monitoring signal to select signal dimensions with a correlation greater than 0.5 with the battery pack SOC value, the signal dimensions including vehicle speed, battery pack total output voltage, battery pack total output current, battery pack unit maximum output voltage, battery pack unit minimum output voltage, local maximum temperature in the battery pack, and local minimum temperature in the battery pack.

4. The method for estimating the SOC of a power battery pack under actual operating conditions driven by big data according to claim 1, characterized in that: In step S3, based on the characteristic that the voltage distribution at the terminals of the series-connected cells of the battery pack is right-skewed compared to the normal distribution, it is determined that the distribution conforms to the Weibull distribution. In the step S6, according to the characteristics that the temperature distribution of the battery pack series cells is centered at the mean and symmetrical on both sides, it is determined that the distribution conforms to the normal distribution.

5. The method for estimating the SOC of a power battery pack under actual operating conditions driven by big data according to claim 1, characterized in that: In step S4, the method of estimating the parameters of the Weibull distribution using the maximum likelihood method comprises the following steps: S4-1. Establish the likelihood function, assuming n observations x1, x2, ..., x n , then the likelihood function is defined as: In the formula, v i is a random variable, λ is a scale parameter, k is a shape parameter, and γ0 is an initial position parameter; S4-2. Take the logarithm of both sides of the likelihood function to obtain the log-likelihood function: S4-3. Take the partial derivatives of the above log-likelihood function with respect to the parameters k and λ respectively, and set them to zero to obtain a system of equations: S4-4. By setting the initial position parameter γ0 in the system of equations to 0 and solving, the maximum likelihood estimates of the initial shape parameter and scale parameter can be obtained; then, substitute them into the unbiased estimate of the position parameter to obtain the new position parameter.

6. The estimation method of the SOC of the power battery pack under the actual operating conditions driven by big data according to claim 1, wherein: In the step S7, the method for estimating the normal distribution parameters by the maximum likelihood method includes the following steps: S6-1. Establish the likelihood function. For the given sample data x1, x2, ..., x n , assuming that these data are from a normal distribution N(μ,σ 2 ), the likelihood function of the sample can be expressed as: Among them, Π represents the multiplication symbol, x i represents the i-th observation. S6-2. Take the logarithm of both sides of the likelihood function to obtain the log-likelihood function: S4-3. Take the partial derivatives of the above log-likelihood function with respect to the mean μ and variance σ respectively, and set them to zero to obtain a system of equations: Solving for this yields:

7. The estimation method of the SOC of the power battery pack under the actual operating conditions driven by big data according to claim 1, wherein: In step S5, when extracting the Weibull distribution features, in order to characterize the symmetry and tail of the discrete data histogram of the battery pack series cell terminal voltage, the skewness and kurtosis are used, and the mean μ1 and variance σ1 of the Weibull distribution are 2 It can be calculated by the following formula: Where λ and k are the scale parameter and shape parameter in the step S4 respectively. The calculation formulas for the skewness and kurtosis of the Weibull distribution are: In the formula, N is the sample size, x i is the i-th sample value.

8. The estimation method of the SOC of the power battery pack under the actual operating conditions driven by big data according to claim 1, wherein: In the step S8, the moving window method is used to resample the fused data. The length of the moving window is T, and the window moves s steps each time, s < T. The input-output mapping scale of the deep learning model in the step S7 is denoted as T→s, and the resampled data is used as the input of the deep learning model.

9. The estimation method of the SOC of the power battery pack under the actual operating conditions driven by big data according to claim 1, wherein: In the step S9, the deep learning model adopts a convolutional neural network-bi-directional long short-term memory network model integrating an attention mechanism.

10. The method for estimating the SOC of a power battery pack under actual operating conditions driven by big data according to claim 1, characterized in that: In the step S10, collect the dataset of the electric vehicle and its battery pack during actual operation. After resampling the collected dataset of actual operation by the moving window method, use it as the online data to input into the deep learning model trained in the step S7; the collected dataset of actual operation includes the following data during the actual operation of the electric vehicle: vehicle speed, total output voltage of the battery pack, total output current of the battery pack, maximum output voltage of the battery pack unit, minimum output voltage of the battery pack unit, local highest temperature inside the battery pack, local lowest temperature data inside the battery pack.

Images