Electric vehicle battery soc prediction method and system based on improved ensemble learning

Abstract

The present application relates to the technical field of electric vehicle battery, more particularly, to an electric vehicle battery SOC prediction method and system based on improved ensemble learning.The prediction method of the present application is based on an XGBoost model, a reference prediction value is predicted through ensemble learning, and an adaptive Kalman filter (ACKF) is introduced for optimization, combined with a conventional prediction value predicted by the ampere-hour integral method, a series connection of processes is realized, and a closed-loop estimation framework is constructed, thereby improving the estimation accuracy and stability of the electric vehicle SOC.Through simulation comparison, the prediction method of the present application has good robustness and convergence.The adaptive Kalman filter (ACKF) used in the present application can adaptively update the noise covariance and measurement noise covariance, further optimizing the convergence ability and estimation accuracy, and improving the generalization ability.The present application solves the problems of low prediction accuracy and large error of the existing open-loop estimation method.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle battery technology, and more specifically, to an electric vehicle battery SOC prediction method based on improved ensemble learning, and a prediction system using the prediction method. Background Technology

[0002] With the rapid development of new energy vehicle technology, more and more technological bottlenecks have been overcome. Lithium-ion batteries, with their long cycle life, high energy density, and high power, are widely used in various electric vehicle fields. However, due to the highly nonlinear electrochemical characteristics of the battery's internal system, it is difficult to obtain the battery's real-time state through direct measurement. Therefore, a battery management system capable of accurately estimating the battery's real-time state plays a crucial role in electric vehicles. The accuracy of SOC prediction not only affects the driver's assessment of the vehicle's condition but also has profound implications for predicting remaining driving range, extending battery life, and improving energy efficiency.

[0003] Existing methods may use a single dataset for prediction, such as the traditional ampere-hour integration method. This approach is an open-loop estimation, resulting in low prediction accuracy and large errors. Summary of the Invention

[0004] Therefore, it is necessary to provide an improved ensemble learning-based method and system for predicting the state of charge (SOC) of electric vehicle batteries, addressing the problems of low prediction accuracy and large errors in existing open-loop estimation methods.

[0005] This invention is achieved using the following technical solution: In a first aspect, this invention discloses an electric vehicle battery SOC prediction method based on improved ensemble learning, used to obtain the SOC prediction result of a target electric vehicle at a target time. t k Final predicted value of battery SOC SOC z ( k ).

[0006] The electric vehicle battery SOC prediction method based on improved ensemble learning includes the following steps: Step 1, obtain t k Battery voltage value U k Battery current value I k Battery temperature value T k ; Step 2, use the trained XGBoost model to... U k , Ik , T k The process is performed to obtain the battery SOC reference prediction value. SOC X ( k ); The XGBoost model is trained using a historical dataset; the historical dataset includes... n Groups of data, with a time interval between adjacent groups of data. Δt ; Step 3: Construct the state-space equations and use the Adaptive Kalman Filter (ACKF) to... SOC X ( k Process it to obtain SOC z ( k ); The state-space equations are as follows: ; In the formula, SOC A ( k () indicates the result obtained using the ampere-hour integration method. t k Battery SOC conventional prediction value; SOC A ( k -1) indicates the result obtained using the ampere-hour integration method. t k-1 Battery SOC conventional prediction value; t k-1 for t k The previous moment, t k - t k-1 = Δt ; I k-1 express t k-1 The battery current value; Q n Indicates the rated capacity of the battery; Q k Indicates the first k Measurement noise covariance in each iteration; R k Indicates the first k Measurement noise covariance of the next iteration.

[0007] This improved ensemble learning-based electric vehicle battery SOC prediction method implements the method or process according to embodiments of this disclosure.

[0008] Secondly, the present invention discloses an electric vehicle battery SOC prediction system based on improved ensemble learning, which uses the electric vehicle battery SOC prediction method based on improved ensemble learning of the first aspect.

[0009] The electric vehicle battery SOC prediction system based on improved ensemble learning includes: a data acquisition module, an XGBoost model processing module, and an adaptive Kalman filter (ACKF) module.

[0010] The data acquisition module is used to acquire t k Battery voltage value U k Battery current value I k Battery temperature value T k The XGBoost model processing module is used to load the trained XGBoost model and process it. U k , I k , T k The process is performed to obtain the battery SOC reference prediction value. SOC X ( k The Adaptive Kalman Filter (ACKF) module is used to construct the state-space equations and apply the ACKF algorithm to the state-space equations. SOC X ( k Process it to obtain SOC z ( k ).

[0011] This electric vehicle battery SOC prediction system based on improved ensemble learning implements the method or process according to embodiments of this disclosure.

[0012] Compared with the prior art, the present invention has the following beneficial effects: 1. The prediction method of this invention is based on the XGBoost model. It predicts a reference value through ensemble learning and introduces an adaptive Kalman filter (ACKF) for optimization. Combined with the conventional prediction value predicted by the ampere-hour integral method, the process is serialized and a closed-loop estimation framework is constructed, thereby improving the estimation accuracy and stability of the electric vehicle's State of Charge (SOC) and obtaining an accurate final prediction value. Simulation comparisons show that the prediction method of this invention has good robustness and convergence.

[0013] 2. This invention uses the XGBoost model, which does not require considering the internal mechanism of the battery or building a battery model. Instead, it relies on historical battery experimental data, selects relevant features and battery SOC to build a black box model, and directly outputs the estimated results. It is efficient and convenient overall, and at the same time greatly reduces the model training time and the risk of overfitting.

[0014] 3. The adaptive Kalman filter ACKF used in this invention can adaptively update the noise covariance and measurement noise covariance, thereby further optimizing the convergence ability and estimation accuracy, while improving the generalization ability. Attached Figure Description

[0015] Figure 1 This is a simplified flowchart of the electric vehicle battery SOC prediction method based on improved ensemble learning in Embodiment 1 of the present invention; Figure 2 This is a data flow diagram of the electric vehicle battery SOC prediction method based on improved ensemble learning in Embodiment 1 of the present invention; Figure 3 This is a comparison chart of simulation results of the three methods in Embodiment 2 of the present invention under 0℃ and UDDS conditions; Figure 4 This is a comparison chart of simulation results of the three methods in Embodiment 2 of the present invention under 25℃ and UDDS conditions; Figure 5 This is a comparison chart of simulation results of the three methods in Embodiment 2 of the present invention under 40℃ and UDDS conditions; Figure 6 This is a comparison chart of simulation results of the three methods in Embodiment 2 of the present invention under 0℃ and ECE conditions; Figure 7 This is a comparison chart of simulation results of the three methods in Embodiment 2 of the present invention under 25°C and ECE conditions; Figure 8 This is a comparison chart of simulation results of the three methods in Embodiment 2 of the present invention under 40℃ and ECE conditions. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] It should be noted that when a component is said to be "installed on" another component, it can be directly on the other component or it may be in a component that is centered on it. When a component is said to be "set on" another component, it can be directly set on the other component or it may also be in a component that is centered on it. When a component is said to be "fixed to" another component, it can be directly fixed to the other component or it may also be in a component that is centered on it.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the specification of this invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0019] Example 1 This embodiment 1 provides an improved ensemble learning-based method for predicting the State of Charge (SOC) of an electric vehicle battery, used to obtain the SOC of the target electric vehicle at a target time. t k Final predicted value of battery SOC SOC z ( k ).

[0020] Please see Figure 1 , Figure 1 Here is a simplified flowchart of the electric vehicle battery SOC prediction method based on improved ensemble learning in this invention, which includes the following steps: Step 1, obtain t k Battery voltage value U k Battery current value I k Battery temperature value T k .

[0021] The aforementioned parameter data of the electric vehicle battery can be collected here using onboard sensors and the BMS system. It should be noted that this method is a real-time prediction, and theoretically, the data collection time interval... Δt The smaller the value, the better the prediction performance of the entire method, but this will result in excessive runtime and model training load. Therefore, Δt The value of should take into account the actual computational capabilities. Generally, Δt Set the time to 10 seconds, with a possible fluctuation of 5 seconds.

[0022] Step 2, use the trained XGBoost model to... U k , Ik , T k The process is performed to obtain the battery SOC reference prediction value. SOC X ( k ).

[0023] The XGBoost model is trained using a historical dataset; the historical dataset includes... n Groups of data, with a time interval between adjacent groups of data. Δt .

[0024] Specifically, the XGBoost model training method includes the following steps: S101, Obtain historical data values ​​of the power battery of the target electric vehicle; the types of historical data values ​​include battery voltage, battery current, battery temperature, and battery SOC.

[0025] S102, process the historical data values ​​to obtain the historical dataset.

[0026] The processing method in S102 includes data filtering, data cleaning, and interpolation. Specifically, historical data values ​​are filtered to identify data occurring during the vehicle's discharge process. The filtered data is then cleaned to remove outliers and duplicates, resulting in cleaned data. Finally, the cleaned data is interpolated to fill in missing values, yielding the historical dataset.

[0027] In other words, S102 is actually about optimizing historical data values ​​to obtain a high-quality historical dataset for model training.

[0028] S103, sort the historical dataset according to time intervals Δt Divided into n Group data; Among them, the i The data set includes the first set of data. i Each sampling time t i Battery voltage value Battery current value Battery temperature value Battery SOC value .

[0029] S104, Obtain input from historical dataset The XGBoost model is trained several times to make its output approximate... ; When the number of training iterations exceeds the preset number or the XGBoost model has converged, save the current XGBoost model parameters to obtain the trained XGBoost model.

[0030] In other words, As input, each training iteration adds a new tree model to approximate the previous predicted output value. This can be expressed by the formula: ; In the formula, k For the number of iterations, t Indicates the number of tree models. for t SOC prediction output value obtained from the iteration of the tree model; for t -1 SOC prediction output value obtained from tree model iteration; f for Z A subfunction, Z The set of all possible tree models can be represented by the formula: ; In the formula, w Represents the weight vector of the leaf node; q This represents the mapping relationship of the leaf nodes. R T This represents the set of weight vectors for the leaf nodes. T Indicates the number of leaves. d The number of features. R T Represents the set of weight vectors of the leaf nodes. R d Indicates the first d A set of weight vectors for each leaf node.

[0031] when t Training stops when the maximum prediction accuracy is reached or the required accuracy is met.

[0032] During the training of the XGBoost model, in order to suppress the complexity of the XGBoost model and prevent overfitting, an objective function is constructed to minimize the objective function.

[0033] The objective function is composed of the loss function and the regularization function, and the formula is as follows: ; In the formula, F obj Let be the objective function. l (.) is the loss function, and Ω(.) is the regularization function; n For the sample size, The actual value of SOC , Predict the output value for SOC. γ, λ is the penalty coefficient for the regularization term. w j For leaf nodes j The weight of the position.

[0034] Therefore, to minimize the objective function, the minor expression in each leaf node must be minimized. Thus, for w j Find the first derivative and set it to 0, then we get: ; In the formula, w j * express w j The first derivative, F obj * This represents the optimal function of the objective function.

[0035] I j To fall into a leaf node j The sample set at that location, ; g i For loss function pairs The first-order partial derivative; h i For loss function pairs The second-order partial derivative.

[0036] In summary, the training process starts with a tree model with a depth of 0, continuously enumerates all feature splitting schemes, determines the optimal splitting position by calculating the difference in the objective function before and after the split, repeats this operation, continuously optimizes the structure of the tree model, and iterates to obtain the optimal XGBoost model.

[0037] Step 3: Construct the state-space equations and use the Adaptive Kalman Filter (ACKF) to... SOC X ( k Process it to obtain SOC z ( k ).

[0038] This step combines the predictions from the XGBoost model and the ampere-hour integral method, and constructs a closed-loop estimation framework using the Adaptive Kalman Filter (ACKF).

[0039] The state-space equations are as follows: ; As an observation equation; As a measurement equation.

[0040] In the formula, SOC A ( k () indicates the result obtained using the ampere-hour integration method. t k Battery SOC conventional prediction value; SOC A ( k -1) indicates the result obtained using the ampere-hour integration method. t k-1 Battery SOC conventional prediction value; t k-1 for t k The previous moment, t k - t k-1 = Δt ; I k-1 express t k-1 The battery current value; Q n Indicates the rated capacity of the battery; Q k Indicates the first k Measurement noise covariance in each iteration; R k Indicates the first k Measurement noise covariance of the next iteration.

[0041] It should be noted that in practice, the process noise covariance and measurement noise covariance cannot be stable constants. However, the standard CKF treats both as constants, which can lead to large estimation errors. The method in this embodiment 1 calculates the process noise covariance and measurement noise covariance for each iteration, thereby reducing the estimation error.

[0042] Step 3 is to extract... Q k , R k The process includes the following steps: Step 301: Initialize the parameters required for the Adaptive Kalman Filter (ACKF).

[0043] This step is fundamental to the Adaptive Kalman Filter (ACKF) and includes: S3011, Set the initial value of the state vector x 0. Initial value of process noise covariance Q0. Measure the initial value of the noise covariance. R 0. Covariance matching window size L w .

[0044] in, ; U 0、 I 0、 T 0、 SOC X (0) represents the initial time. t Battery voltage, battery current, battery temperature, and battery SOC reference prediction value (0).

[0045] S3012, Calculate the initial matrix of error covariance. P 0: ; in, for x The mean of 0, .

[0046] Step 302: Update the time based on the initialized parameters to obtain the first... k Error covariance matrix correction value for the next iteration .

[0047] Specifically, step 302 includes: S3021, calculate the... k The prior volume point of the next iteration S k : ; in, P k For the first k The prior error covariance matrix of the next iteration chol (.) indicates Cholesky decomposition; S3022, Calculate the... k The state vector of the next iteration x k The i Volume point : ; in, ; For the first i A set of volume points; ; m The number of volume points; Indicates the first i One volume point; S3023, dissemination S k And calculate the first k The state vector correction value of the next iteration The i Each component : ; in, u k for x k The first derivative of ; g(.) represents the state function of a nonlinear system; S3024, calculate the... k Error covariance matrix correction value for the next iteration : ; in, for x k+1 The mean, x k+1 Indicates the first k+ The state vector of the first iteration Q k Indicates the first k The process noise covariance of each iteration.

[0048] Step 303, based on Perform measurement updates to obtain the first... k Kalman gain after +1 iterations .

[0049] Specifically, step 303 includes: S3031, calculate the... k The posterior volume point of the next iteration : ; S3032, calculate the... k The state vector correction value of the next iteration The i Volume point : ; S3033, dissemination And calculate the first k The output vector correction value of the next iteration The i Each component : ; in, ; for The first derivative; S3034, calculate the... k Error covariance of the next iteration Mutual covariance : ; ; in, for The mean, ; for The mean, ; Calculate the first k Kalman gain after +1 iterations : ; S3035, Update No. k The state vector after +1 iterations Error covariance matrix : ; ; in, y k+1 For the first k+ The output vector of the first iteration; for y k+1 The mean; , U k+1 , I k+1 , T k+1 They represent t k+1 Battery voltage, battery current, and battery temperature; SOC A ( k +1) indicates that the result was obtained using the ampere-hour integration method. t k+1 Battery SOC conventional prediction value; t k+1 for t k The next moment, t k+1 - t k = Δt .

[0050] Step 304, adaptive update of the first... k The process noise covariance of the next iteration Measurement of noise covariance .

[0051] Specifically, step 304 includes: S3041, calculate the... k Covariance of SOC residual sequence in the next iteration : ; in, e i For the first i SOC residuals of the step; S3042, Update No. k The process noise covariance of the next iteration Measurement of noise covariance : ; .

[0052] S3043, , SOC X ( k Substituting into the state-space equations, we get SOC z ( k ).

[0053] Following the steps outlined above, the data flow can be seen in the documentation. Figure 2 That is, to obtain SOC z ( k Furthermore, a battery management system can be integrated to set control strategies for battery charging and discharging control.

[0054] Example 2 This embodiment 2 discloses a first electric vehicle battery SOC prediction system based on improved ensemble learning, which uses the electric vehicle battery SOC prediction method based on improved ensemble learning in embodiment 1.

[0055] The first electric vehicle battery SOC prediction system based on improved ensemble learning includes: a data acquisition module, an XGBoost model processing module, and an adaptive Kalman filter (ACKF) module.

[0056] The data acquisition module is used to acquire t k Battery voltage value U k Battery current value Ik Battery temperature value T k The XGBoost model processing module is used to load the trained XGBoost model and process it. U k , I k , T k The process is performed to obtain the battery SOC reference prediction value. SOC X ( k The Adaptive Kalman Filter (ACKF) module is used to construct the state-space equations and apply the ACKF algorithm to the state-space equations. SOC X ( k Process it to obtain SOC z ( k ).

[0057] Of course, the model training function can also be integrated into the electric vehicle battery SOC prediction system based on improved ensemble learning, which is the second type of electric vehicle battery SOC prediction system based on improved ensemble learning. The structure of the second type of electric vehicle battery SOC prediction system based on improved ensemble learning is similar to that of the first type, but the difference is: A model training module has been added. The model training module is used to train the XGBoost model to obtain a trained XGBoost model.

[0058] Example 3 This embodiment 3 introduces two other methods: combining the XGBoost model with the Extended Kalman Filter (EKF) (abbreviated as XGBoost+EKF); and combining the GBDT network model with the Adaptive Kalman Filter (ACKF) (abbreviated as GBDT+ACKF). These methods were then used together with the electric vehicle battery SOC prediction method based on improved ensemble learning in embodiment 1 (abbreviated as XGBoost+ACKF) for simulation experiments under the same environmental parameters.

[0059] The simulation experiment used a battery experimental platform, including battery charging and discharging equipment, temperature control chamber, host computer, battery module, BMS master and slave board and other components.

[0060] The battery module, used as the experimental subject, employed a ternary lithium-ion battery with a rated capacity of 94Ah, a nominal voltage of 3.570V, and charge / discharge cutoff voltages of 4.12V / 3.416V, and maximum charge / discharge currents of 2.5C / 3C. The battery charging / discharging equipment was used to charge and discharge the battery module, with a charging / discharging current up to 300A and a maximum voltage up to 600V. It also supported constant voltage, constant current, and constant power charging / discharging, meeting the needs of most operating conditions. A temperature control chamber was used to simulate operating conditions at different temperatures, with a temperature control range of -20°C to 100°C. A host computer was used to control the battery charging / discharging equipment and simultaneously connected to the BMS motherboard for data acquisition.

[0061] Example 3 simulates two electric vehicle driving conditions (UDDS, ECE) and tests the conditions at temperatures of 0℃, 25℃, and 40℃ respectively.

[0062] See Figure 3 , Figure 4 and Figure 5 The image shows the battery SOC prediction results and prediction error curves for the three methods under UDDS conditions at 0℃, 25℃, and 40℃. It can be seen that the battery SOC prediction result curve of XGBoost+ACKF has the highest fit with the Reference curve, which represents the actual battery SOC value, intuitively demonstrating the accuracy of XGBoost+ACKF.

[0063] Referring to Table 1, it shows the comparison of MAE and RMSE of the three methods at 0℃, 25℃, and 40℃ under UDDS conditions.

[0064] Table 1. Comparison of MAE and RMSE of the three methods under UDDS operating conditions. It can be seen that the method in Example 1 has the lowest MAE and RMSE under UDDS conditions.

[0065] See Figure 6 , Figure 7 and Figure 8 The image shows the battery SOC prediction results and prediction error curves for the three methods under ECE conditions at 0℃, 25℃, and 40℃. It can be seen that the battery SOC prediction result curve of XGBoost+ACKF has the highest fit with the Reference curve, and the Reference curve represents the actual battery SOC value, which also intuitively reflects the accuracy of XGBoost+ACKF.

[0066] Referring to Table 2, it shows the comparison of MAE and RMSE of the three methods at 0℃, 25℃, and 40℃ under ECE conditions.

[0067] Table 2 Comparison of MAE and RMSE of the three methods under ECE conditions It can be seen that the method in Example 1 has the lowest MAE and RMSE under ECE conditions.

[0068] In summary, compared to the other two methods, the method in Example 1 has better estimation accuracy, stability, and convergence ability.

[0069] Example 4 This embodiment also discloses a readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the improved ensemble learning-based electric vehicle battery SOC prediction method of Embodiment 1 is executed.

[0070] When applying the method of Example 1, it can be applied in the form of software, such as by designing it as a program that can run independently on a computer-readable storage medium, such as a USB flash drive or a USB security token. The program is designed to start the entire method through an external trigger using a USB flash drive or USB security token.

[0071] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0072] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. An improved ensemble learning-based method for predicting the State of Charge (SOC) of an electric vehicle battery at a target time. t k Final predicted value of battery SOC SOC z ( k ), characterized in that: The electric vehicle battery SOC prediction method based on improved ensemble learning Includes the following steps, Step 1, obtain t k Battery voltage value U k Battery current value I k Battery temperature value T k ; Step 2, use the trained XGBoost model to... U k , I k , T k The process is performed to obtain the battery SOC reference prediction value. SOC X ( k ); The XGBoost model is trained using a historical dataset; the historical dataset includes... n Groups of data, with a time interval between adjacent groups of data. Δt ; Step 3: Construct the state-space equations and use the Adaptive Kalman Filter (ACKF) to... SOC X ( k Process it to obtain SOC z ( k ); The state-space equation is as follows: ； In the formula, SOC A ( k () indicates the result obtained using the ampere-hour integration method. t k Battery SOC conventional prediction value; SOC A ( k -1) indicates the result obtained using the ampere-hour integration method. t k-1 Battery SOC conventional prediction value; t k-1 for t k The previous moment, t k - t k-1 = Δt ; I k-1 express t k-1 The battery current value; Q n Indicates the rated capacity of the battery; Q k Indicates the first k The process noise covariance of each iteration; R k Indicates the first k Measurement noise covariance in each iteration; Step 301: Initialize the parameters required for the Adaptive Kalman Filter (ACKF); Step 302: Update the time based on the initialized parameters to obtain the first... k Error covariance matrix correction value for the next iteration ; Step 303, based on Perform measurement updates to obtain the first... k Kalman gain after +1 iterations ; Step 304, adaptive update of the first... k The process noise covariance of the next iteration Q k Measurement of noise covariance R k and calculate SOC z ( k ); Step 304 includes: S3041, calculate the... k Covariance of SOC residual sequence in the next iteration : ; in, e i For the first i SOC residuals of the step; S3042, Update No. k The process noise covariance of the next iteration Q k Measurement of noise covariance R k : ； ； In the formula, For the first k The output vector correction value of the next iteration The i One component; y k For the first k The output vector of the next iteration; S3043, R k , SOC X ( k Substituting into the state-space equations, we get SOC z ( k ).

2. The electric vehicle battery SOC prediction method based on improved ensemble learning according to claim 1, characterized in that, Step 301 includes: S3011, Set the initial value of the state vector x 0. Initial value of process noise covariance Q 0. Measure the initial value of the noise covariance. R 0. Covariance matching window size L w ; in, ; U 0、 I 0、 T 0、 SOC X (0) represents the initial time. t Battery voltage, battery current, battery temperature, and battery SOC reference prediction value (0); S3012, Calculate the initial matrix of error covariance. P 0: ； in, for x The mean of 0, .

3. The electric vehicle battery SOC prediction method based on improved ensemble learning according to claim 2, characterized in that, Step 302 includes: S3021, calculate the... k The prior volume point of the next iteration S k : ； in, P k For the first k The prior error covariance matrix of the next iteration chol (.) indicates Cholesky decomposition; S3022, Calculate the... k The state vector of the next iteration x k The i Volume point : ； in, ; For the first i A set of volume points; S3023, dissemination S k And calculate the first k The state vector correction value of the next iteration The i Each component : ； in, u k for x k The first derivative of ; g(.) represents the state function of a nonlinear system; S3024, calculate the... k Error covariance matrix correction value for the next iteration : ； in, for x k+1 The mean, x k+1 Indicates the first k+ The state vector of the first iteration Q k Indicates the first k The process noise covariance of each iteration.

4. The electric vehicle battery SOC prediction method based on improved ensemble learning according to claim 3, characterized in that, Step 303 includes: S3031, calculate the... k The posterior volume point of the next iteration : ； S3032, calculate the... k The state vector correction value of the next iteration The i Volume point : ； S3033, dissemination And calculate the first k The output vector correction value of the next iteration The i Each component : ； in, ; for The first derivative; S3034, calculate the... k Error covariance of the next iteration Mutual covariance : ； ； in, for The mean, ; for The mean, ; Calculate the first k Kalman gain after +1 iterations : ； S3035, Update No. k The state vector after +1 iterations Error covariance matrix : ； ； in, y k+1 For the first k+ The output vector of the first iteration; for y k+1 The mean; , U k+1 , I k+1 , T k+1 They represent t k+1 Battery voltage, battery current, and battery temperature; SOC A ( k +1) indicates that the result was obtained using the ampere-hour integration method. t k+1 Battery SOC conventional prediction value; t k+1 for t k The next moment, t k+1 - t k = Δt .

5. The electric vehicle battery SOC prediction method based on improved ensemble learning according to claim 3, characterized in that, In step 1, the XGBoost model training method includes the following steps: S101, Obtain historical data values ​​of the power battery of the target electric vehicle; the types of the historical data values ​​include battery voltage, battery current, battery temperature, and battery SOC; S102, Process the historical data values ​​to obtain the historical dataset; S103, sort the historical dataset according to time intervals Δt Divided into n Group data; Among them, the i The data set includes the first set of data. i Each sampling time t i Battery voltage value Battery current value Battery temperature value Battery SOC value ; S104, Obtain input from historical dataset The XGBoost model is trained several times to make its output approximate... ; When the number of training iterations exceeds the preset number or the XGBoost model has converged, save the current XGBoost model parameters to obtain the trained XGBoost model.

6. The electric vehicle battery SOC prediction method based on improved ensemble learning according to claim 5, characterized in that, The processing method for S102 includes: Historical data values ​​are filtered to identify data that is in the process of vehicle discharge, and this data is then used as the filter data. The filtered data is cleaned to remove outliers and duplicates, resulting in cleaned data. The cleaned data is interpolated and filled with missing values ​​to obtain the historical dataset.

7. An electric vehicle battery SOC prediction system based on improved ensemble learning, characterized in that, It uses the electric vehicle battery SOC prediction method based on improved ensemble learning as described in any one of claims 1-6; The electric vehicle battery SOC prediction system based on improved ensemble learning includes: The data acquisition module is used to acquire... t k Battery voltage value U k Battery current value I k Battery temperature value T k ; The XGBoost model processing module is used to load pre-trained XGBoost models and process them... U k , I k , T k The process is performed to obtain the battery SOC reference prediction value. SOC X ( k ); And the Adaptive Kalman Filter (ACKF) module, which is used to construct the state-space equations and use the ACKF to... SOC X ( k Process it to obtain SOC z ( k ).

8. The electric vehicle battery SOC prediction system based on improved ensemble learning according to claim 7, characterized in that, The electric vehicle battery SOC prediction system based on improved ensemble learning also includes: The model training module is used to train the XGBoost model and obtain the trained XGBoost model.

Images