A driving behavior-based dynamic calculation method for state of charge of an electric vehicle

Abstract

This invention relates to the field of battery management technology and discloses a dynamic calculation method for the state of charge (SOC) of electric vehicles based on driving behavior. The method includes: S1, real-time acquisition of battery data and driving behavior time-series signals; constructing a driving behavior feature vector; S2, classifying driving behavior and outputting classification confidence scores; calling a matching sub-model based on the classification results; S3, real-time accumulation of charge, predicting and correcting the battery operating voltage, and fusing a temperature compensation coefficient; S4, establishing an error compensation model; performing dynamic weighted fusion based on confidence scores; S5, outputting and displaying the SOC value after a reasonableness check; and storing it in a historical database for periodic training of the driving style classification model and the SOC sub-model. This application integrates ampere-hour integration, adaptive filtering, and behavior compensation, combined with a closed-loop optimization mechanism, to achieve a stable SOC estimation error within 3% under complex operating conditions, solving the problems of low estimation accuracy and inaccurate range prediction caused by the lack of quantification of driving behavior in traditional methods.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and more specifically to a method for dynamically calculating the state of charge of an electric vehicle based on driving behavior. Background Technology

[0002] In the field of electric vehicles, accurate estimation of state of charge (SOC) is a core function of battery management systems, directly affecting vehicle range prediction, battery health assessment, and safe optimization of charging and discharging strategies. Existing SOC estimation techniques can be mainly categorized into three types: empirical lookup table methods based on open-circuit voltage, ampere-hour integration methods, and filtering algorithms based on battery models.

[0003] However, these methods face significant challenges under actual dynamic driving conditions. The open-circuit voltage method relies heavily on static or quasi-static voltage measurements and cannot be applied in real time while the vehicle is in motion. Although the ampere-hour integration method can achieve continuous estimation, its inherent flaw lies in the continuous accumulation of initial SOC error and current sensor measurement error, causing the estimation result to deviate significantly from the true value over time. While filtering algorithms based on equivalent circuit models improve accuracy to some extent, their model parameters are usually calibrated based on standard operating conditions and are difficult to adapt to the drastic current and temperature fluctuations caused by complex conditions such as acceleration, deceleration, and hill climbing during real driving.

[0004] Ultimately, existing technologies generally treat driving behavior as an unpredictable external disturbance, failing to treat it as a quantifiable and modelable system input. As a result, they cannot accurately compensate for the dynamic response of the battery caused by differences in driving style. This leads to a decrease in the accuracy of SOC estimation and inaccurate range prediction under complex and ever-changing real-world road conditions, seriously affecting the user's driving experience and driving safety. Summary of the Invention

[0005] The present invention aims to provide a dynamic calculation method for the state of charge (SOC) of electric vehicles based on driving behavior, which solves the problems of decreased accuracy of existing electric vehicle SOC estimation and inaccurate range prediction under complex and variable real road conditions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for dynamically calculating the state of charge of an electric vehicle based on driving behavior, comprising the following steps: S1 collects battery data and driving behavior time series signals, including drive motor power, accelerator pedal opening, and vehicle speed, in real time; based on the set sliding time window, it calculates multi-dimensional feature indicators and constructs the driving behavior feature vector of the current time slice. S2 uses a K-means clustering model to classify the current driving behavior into three categories: aggressive, mild, or standard, and outputs the classification confidence score. Based on the real-time classification results, it calls the matching sub-model from the model library to provide algorithm parameters adapted to the driving style for SOC estimation. S3 accumulates the charge in real time through the ampere-hour integration method. At the same time, based on the equivalent circuit model, it uses an adaptive unscented Kalman filter to predict and correct the battery operating voltage and integrates the temperature compensation coefficient. S4. Based on driving style, a SOC error compensation model is established to calculate the SOC offset caused by driving behavior. The ampere-hour integral result, UKF output value, and behavior compensation amount are dynamically weighted and fused according to the classification confidence to suppress result jumps. S5, after a reasonableness check, outputs the SOC value to the instrument panel display and provides it to the vehicle control system for range prediction and energy management; it stores complete time-series data, classification results, estimated SOC values ​​and actual values ​​in the historical database for periodic retraining of the driving style classification model and SOC sub-model.

[0007] The principles and advantages of this scheme are: Existing State of Charge (SOC) estimation techniques have significant limitations under dynamic driving conditions. The fundamental problem lies in treating driving behavior as an unpredictable external disturbance, failing to transform it into a quantifiable system input. Among traditional methods, the open-circuit voltage method relies on static measurements and cannot be applied in real time; the ampere-hour integration method suffers from error accumulation; and while filtering algorithms based on equivalent circuit models improve accuracy, their parameters are calibrated under standard operating conditions, making it difficult to cope with the drastic fluctuations in current and temperature caused by complex conditions such as acceleration and deceleration. This leads to decreased SOC estimation accuracy and inaccurate range prediction under real-world driving conditions.

[0008] This solution first collects battery data such as voltage, current, and temperature in real time, and simultaneously records driving behavior signals such as drive motor power and accelerator pedal opening. It then uses a sliding time window to extract feature vectors such as average acceleration and frequency of rapid acceleration, and uses a K-means clustering model to dynamically classify driving styles. Based on the classification results, it calls the corresponding SOC estimation sub-model. Aggressive driving is matched with parameters that enhance the response to sudden current changes, while mild driving focuses on long-term stability parameters. Simultaneously, it integrates ampere-hour integral, adaptive unscented Kalman filtering, and driving behavior compensation, and dynamically weights the output based on classification confidence. Furthermore, the system has an online parameter update mechanism, fine-tuning model parameters through recursive least squares and periodically retraining the model using historical data to form a closed-loop optimization.

[0009] This solution transforms driving behavior from a disturbance variable into a key input, constructing a heterogeneous computing framework and a full-stack solution that addresses the poor adaptability of traditional single-model approaches. In practical applications, the SOC estimation error is consistently controlled within 3%, a reduction of over 60% compared to traditional methods, significantly improving the reliability of range prediction and alleviating users' range anxiety. Simultaneously, a closed-loop optimization mechanism enables high-precision estimation throughout the entire battery lifecycle, extending the service life of the battery management system, reducing maintenance costs, and driving technological progress in the industry. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a method for dynamically calculating the state of charge of an electric vehicle based on driving behavior, according to the present invention. Figure 2 This is a logic block diagram of a dynamic calculation method for the state of charge of an electric vehicle based on driving behavior, according to the present invention. Detailed Implementation

[0011] The following detailed description illustrates the specific implementation method: This embodiment presents a dynamic calculation method for the state of charge (SOC) of electric vehicles based on driving behavior. By collecting and dynamically classifying driving behavior features in real time, calling the corresponding SOC estimation sub-model, and integrating ampere-hour integral, adaptive filtering, and behavior compensation, combined with online parameter updates and closed-loop optimization mechanisms, the method achieves stable control of SOC estimation error within 3% under complex operating conditions. This solves the problems of low estimation accuracy and inaccurate range prediction caused by the lack of quantification of driving behavior in traditional methods.

[0012] A method for dynamically calculating the state of charge of an electric vehicle based on driving behavior is provided, as shown in the appendix. Figure 1 As shown, it includes the following steps: S1 collects battery data in real time, as well as time-series signals of driving behavior, including drive motor power, accelerator pedal opening, and vehicle speed; based on a set sliding time window, it calculates multi-dimensional feature indicators and constructs the driving behavior feature vector for the current time slice.

[0013] In this embodiment, combined with the appendix Figure 2 As shown, before data acquisition, the process also includes initializing and configuring core parameters, including measuring the open-circuit voltage (OCV) of the battery in a static state using a high-precision voltage sensor and calculating the initial SOC value based on the calibration curve; simultaneously, the coulombic efficiency coefficient of the ampere-hour integration method is preset. Initialize the state vector of the extended Kalman filter, denoted as: ; and the corresponding error covariance matrix P.

[0014] In this embodiment, the core calculation formula is: ;(1).

[0015] Simultaneously, a driving style classification model trained on historical data is loaded from flash memory, along with parameter libraries for SOC estimation sub-models optimized for aggressive, mild, and standard driving styles. All parameters are validated and verified to ensure their integrity, establishing a stable foundation for subsequent real-time data processing.

[0016] After completing the initial configuration settings, the battery system's battery data, including time-series data streams of voltage U(t), current I(t), and temperature T(t), is synchronously collected at a fixed frequency of 10Hz via the vehicle's CAN bus network and distributed sensor array. Simultaneously, the real-time power of the drive motor, reflecting driving behavior, is recorded. Accelerator pedal opening Speed Dynamic parameters, etc.

[0017] In this embodiment, data acquisition employs a hardware timestamp mechanism to ensure that the synchronization error of multi-source signals is less than 10ms. The raw data undergoes Kalman filtering preprocessing and is represented as follows: (2) This process effectively suppresses sensor noise, providing a high-quality data source for subsequent analysis. Simultaneously, the acquisition module has an automatic diagnostic function, detecting signal anomalies in real time and triggering a data retransmission mechanism.

[0018] In this embodiment, a 60-second sliding time window is used to calculate multidimensional feature indicators in real time. These multidimensional feature indicators include the average acceleration within the window. Rapid acceleration frequency Motor power fluctuation variance These features constitute the driving behavior feature vector for the current time slice. It also includes synchronous calculation of the accelerator pedal change rate. Maximum speed and average power Together they form a six-dimensional feature vector. Then it can be expressed as ;(3)

[0019] In this embodiment, the feature vector is updated every 6 seconds to dynamically reflect driving behavior characteristics. The feature calculation uses an incremental update algorithm, which significantly reduces computational complexity.

[0020] S2 uses a K-means clustering model to classify the current driving behavior into three categories: aggressive, mild, or standard, and outputs the classification confidence score. Based on the real-time classification results, it calls the matching sub-model from the model library to provide algorithm parameters adapted to the driving style for SOC estimation.

[0021] In this embodiment, the real-time feature vector is input into the pre-trained K-means clustering model. By calculating the Mahalanobis distance between the real-time feature vector and the centers of its various clusters, it can be expressed as: (4) The category corresponding to the minimum distance is selected as the current driving style classification result, and the membership confidence score is calculated simultaneously. Furthermore, when the confidence level is less than the set threshold, fuzzy logic reasoning is initiated for secondary classification.

[0022] In this embodiment, fuzzy logic reasoning is initiated for secondary classification when p < 0.7, comprehensively considering the style change trends of the most recent 10 time slices to improve classification stability. Simultaneously, the system records classification history for analysis of the long-term evolution of driving habits.

[0023] Based on the real-time driving style classification results, the corresponding SOC estimator parameter set is dynamically invoked.

[0024] For aggressive driving models, which are matched with enhanced current surge response capabilities, the time constant is invoked. A first-order hysteresis loop with a time delay of 10 seconds provides a rapid response to sudden current changes; for mild driving scenarios, models prioritizing long-term stability are used. The filtering parameters are set to 60s to improve long-term stability. Simultaneously, the model's internal resistance parameters are dynamically adjusted based on real-time current, which can be expressed as... (5) in, This is a proportionality coefficient related to the current. Total resistance; I is the reference resistor; I is the current. In this embodiment, The settings are dynamically adjusted based on driving style. Simultaneously, the diffusion resistance parameter is adjusted in real-time based on battery temperature to ensure the model's adaptability to different environments.

[0025] S3 accumulates the charge in real time using the ampere-hour integration method. At the same time, based on the equivalent circuit model, it uses an adaptive unscented Kalman filter to predict and correct the battery operating voltage, and integrates a temperature compensation coefficient.

[0026] In this embodiment, the real-time accumulated charge is expressed as: (6) The state equation of the Unscented Kalman Filter (UKF) algorithm is expressed as follows: (7) The prediction equation is expressed as follows (8) The temperature compensation coefficient, applied to capacity correction to eliminate the effect of temperature on the battery's usable energy, can be expressed as: (9) In equation (6), The initial state of charge; For charge and discharge efficiency; Let k be the state of charge at time k; It represents the current quantity; This refers to the battery's rated capacity. In equation (7), , These are the two voltage states at time k; This is the state vector from the previous moment; Let be the input current at time k; This is process noise; In equation (8), Let be the battery terminal voltage at time k; The internal resistance of the battery is ohmic. This refers to model error or noise. In equation (9), These are the coefficient values. Through parallel execution of multiple models, the output results are temporarily stored in a buffer.

[0027] S4 establishes a SOC error compensation model based on driving style to calculate the SOC offset caused by driving behavior; and dynamically weights and fuses the ampere-hour integral result, UKF output value, and behavior compensation amount according to classification confidence to suppress result jumps.

[0028] Based on the current driving style classification results, a preset compensation coefficient matrix is ​​read to calculate the SOC offset caused by driving behavior. In this embodiment, the SOC error compensation model is represented as follows: (10) In the formula, This is the driving compensation coefficient; is the effective value of the current; T is the temperature. Among them, the compensation coefficient for the aggressive driving type... =1.2, corresponding to a higher compensation coefficient, used to correct estimation errors caused by instantaneous high current; the compensation coefficient for mild driving type. =0.8. Wherein, the effective value of the current is... (11) The compensation function can be expressed as: (12) In the formula, is the activation energy; R is the molar gas constant; a and b are empirical constants obtained through regression analysis of historical data to accurately reflect the capacity loss characteristics under high current conditions; T is the thermodynamic temperature. In this embodiment, the compensation amount is updated every 30 seconds to ensure a smooth transition and avoid abrupt changes.

[0029] Weighted fusion is performed based on classification confidence level p. At high confidence levels, behavioral compensation weights increase, while at low confidence levels, regression and traditional methods dominate. Kalman filtering is used during the fusion process to smooth the results and suppress abrupt changes. Therefore, its dynamic weighted fusion can be expressed as: (13) In the formula; , , These are the weighting coefficients; SOC estimated by the ampere-hour integration method; SOC estimated by unscented Kalman filtering; This is a SOC correction item.

[0030] Its specific weight allocation can be expressed as: (14) and satisfy =1; In the formula, This represents the i-th confidence value; The dynamic weighting strategy is as follows: when p > 0.8, To highlight driving behavior compensation; when p < 0.5, ; To strengthen UKF's dominant position.

[0031] S5, after a reasonableness check, outputs the SOC value to the instrument panel display and provides it to the vehicle control system for range prediction and energy management; it stores complete time-series data, classification results, estimated SOC values ​​and actual values ​​in the historical database for periodic retraining of the driving style classification model and SOC sub-model.

[0032] In this embodiment, the final fused SOC value is checked for reasonableness, specifically by performing a moving average filter on the fusion result, which can be expressed as follows: (15) Further boundary protection of the output value can be expressed as follows: (16) Simultaneously calculating the remaining driving range, it can be expressed as: (17) In the formula, The original display state of charge value; This refers to the average energy conversion efficiency; Average power consumption; The average speed is used to calculate the device's theoretical driving range using the battery's available energy, system efficiency, average power consumption, and speed.

[0033] Once the inspection is passed, the final result is displayed on the dashboard and updated every 5 seconds to ensure real-time information and avoid frequent refreshes. It is also provided to the vehicle control system for range prediction and energy management.

[0034] The system stores complete time-series data, classification results, SOC estimates, and actual values ​​in a historical database for periodic retraining of the driving style classification model and SOC sub-model, achieving closed-loop self-optimization.

[0035] In this embodiment, establishing a complete data link can be represented as follows: (18) If the model is retrained monthly using accumulated data, the objective function can be expressed as follows: (19) In the formula, The estimated state of charge; This represents the true state of charge; The hyperparameter for controlling the regularization strength; These are the parameters to be optimized in the model. By adjusting the parameters... By minimizing the sum of the squared estimation errors and the regularization term, a SOC estimation model that balances estimation accuracy and model robustness is obtained.

[0036] The particle swarm optimization algorithm is used to optimize the cluster center positions and SOC sub-model parameters, while simultaneously updating the onboard model parameter library. The optimization process is completed in the cloud, with update packages distributed via OTA (Over-The-Air) updates, enabling continuous self-learning optimization of the system.

[0037] S6: The system continuously monitors the SOC estimation error. When the continuous error exceeds the threshold, the equivalent circuit model parameters are finely adjusted online using the recursive least squares method.

[0038] In this embodiment, the estimation error of 10 consecutive sampling points A time-triggered parameter update mechanism is implemented, employing a recursive least squares method with a forgetting factor to update model parameters online, thereby enhancing environmental adaptability. This is represented as follows: (20) in, ;(twenty one) In the formula, This is the state estimation vector at time k; This is the state estimation vector at time k-1; Let be the gain matrix at time k; Let be the observation value at time k; ; The state covariance matrix at the previous time step; This is the state transition matrix; As the forgetting factor, in this embodiment, The update process is executed in a background thread and does not affect real-time estimation performance.

[0039] In this embodiment, driving behavior, traditionally considered an external disturbance, is transformed for the first time into a quantifiable and modelable key system input. By extracting dynamic features such as the accelerator pedal change rate and motor power fluctuation variance in real time, and utilizing a pre-trained machine learning model to identify and classify driving styles online, the system can match customized SOC estimation sub-model parameters for different driving modes, such as aggressive and mild driving. This deep coupling mechanism transforms SOC estimation from a passive response to an active perception of the driving situation, fundamentally solving the pain point of inaccuracy caused by drastic changes in driving conditions in traditional single models.

[0040] In real-world driving conditions involving frequent starts and stops and rapid acceleration, the SOC estimation error can be stably controlled within 3%, which is more than 60% lower than the error of traditional methods. This greatly improves the reliability of range prediction and effectively alleviates users' range anxiety.

[0041] This solution overcomes the limitations of relying on a single model by innovatively constructing a heterogeneous computing framework that allows multiple estimation methods to run in parallel. The ampere-hour integration method, adaptive filtering algorithm, and driving behavior compensation model are calculated simultaneously. The final SOC result is not simply a matter of selecting one method, but rather a dynamic weighted fusion based on the real-time confidence level of driving style recognition. This design allows the system to fully leverage the adaptability of the behavior compensation model at high confidence levels, while automatically reverting to a robust mode centered on the filtering algorithm at low confidence levels, ensuring the reasonableness of the output results. The advantage of this architecture lies in combining the strengths of various methods, guaranteeing millisecond-level response speed while maintaining extremely high accuracy and robustness under complex conditions such as high current surges and voltage plateaus, thus solving the problem of traditional methods "sacrificing one aspect for another."

[0042] Meanwhile, this solution endows the system with the ability to continuously self-optimize, evolving it from a static model into a "living" learning system. Through a built-in online parameter identification algorithm, the system can monitor estimation errors in real time and automatically fine-tune key model parameters when performance deviations exceed thresholds, thus adapting to the performance degradation that occurs as batteries age. Furthermore, the massive amounts of data accumulated over long-term operation are used for periodic retraining and global optimization of the entire algorithm model. The direct effect of this closed-loop optimization mechanism is that the SOC estimation system can maintain high accuracy throughout its entire lifecycle, effectively overcoming the common industry problem of gradual accuracy decline due to battery aging in traditional methods, greatly extending the effective service life of the battery management system, and reducing maintenance costs.

[0043] Finally, this solution is not just an algorithm, but a full-stack solution covering data acquisition, real-time vehicle-side computing, cloud-based analysis, and model iteration. The vehicle-side handles high-real-time computing tasks, ensuring the stability and reliability of core functions; the cloud platform aggregates massive amounts of data, performs in-depth mining and large-scale model training, generating a more accurate globally optimized model, which is then deployed to the vehicle via OTA (Over-The-Air) updates. The advantage of this architecture lies in forming a virtuous cycle of "data-driven model iteration, and model-enhanced user experience." It not only provides high-precision SOC estimation products but also builds a sustainable technological and commercial ecosystem, laying the foundation for developing value-added services and driving technological progress across the entire industry.

[0044] The above descriptions are merely embodiments of the present invention, and common knowledge such as specific technical solutions and / or characteristics are not described in detail here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the technical solutions of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for dynamically calculating the state of charge of an electric vehicle based on driving behavior, characterized in that, Includes the following steps: S1 collects battery data and driving behavior time series signals, including drive motor power, accelerator pedal opening, and vehicle speed, in real time; based on the set sliding time window, it calculates multi-dimensional feature indicators and constructs the driving behavior feature vector of the current time slice. S2 uses a K-means clustering model to classify the current driving behavior into three categories: aggressive, mild, or standard, and outputs the classification confidence score. Based on the real-time classification results, it calls the matching sub-model from the model library to provide algorithm parameters adapted to the driving style for SOC estimation. S3 accumulates the charge in real time through the ampere-hour integration method. At the same time, based on the equivalent circuit model, it uses an adaptive unscented Kalman filter to predict and correct the battery operating voltage and integrates the temperature compensation coefficient. S4. Based on driving style, a SOC error compensation model is established to calculate the SOC offset caused by driving behavior. The ampere-hour integral result, UKF output value, and behavior compensation amount are dynamically weighted and fused according to the classification confidence to suppress result jumps. S5, after a reasonableness check, outputs the SOC value to the instrument panel display and provides it to the vehicle control system for range prediction and energy management; Complete time-series data, classification results, SOC estimates, and actual values ​​are stored in a historical database for periodic retraining of the driving style classification model and SOC sub-model.

2. The method for dynamic calculation of the state of charge of an electric vehicle based on driving behavior according to claim 1, characterized in that: It also includes S6, which continuously monitors the SOC estimation error. When the continuous error exceeds the threshold, the equivalent circuit model parameters are finely adjusted online using the recursive least squares method.

3. The method for dynamically calculating the state of charge of an electric vehicle based on driving behavior according to claim 1, characterized in that: It also includes initial configuration settings, which involve measuring the open-circuit voltage of the battery in a static state and calculating the initial SOC value based on the calibration curve; at the same time, it presets the coulomb efficiency coefficient of the ampere-hour integral method and initializes the state vector and the corresponding error covariance matrix of the extended Kalman filter.

4. The method for dynamic calculation of the state of charge of an electric vehicle based on driving behavior according to claim 1, characterized in that: In S1, the multidimensional feature index includes average acceleration. Rapid acceleration frequency Motor power fluctuation variance The driving behavior feature vector also includes synchronously calculating the accelerator pedal change rate. Maximum speed and average power , represented as 。 5. The method for dynamically calculating the state of charge of an electric vehicle based on driving behavior according to claim 1, characterized in that: In S2, the Mahalanobis distance between the real-time feature vector and its various cluster centers is calculated, and the category corresponding to the minimum distance is selected as the current driving style classification result. At the same time, the membership confidence is calculated. When the confidence is less than a set threshold, fuzzy logic reasoning is initiated for secondary classification.

6. The method for dynamically calculating the state of charge of an electric vehicle based on driving behavior according to claim 1, characterized in that: In S2, for the aggressive driving class, the time constant is called. A first-order hysteresis loop with a time delay of 10 seconds provides a rapid response to sudden current changes; mild driving applications employ... =Filtering parameters for 60s; The model's internal resistance parameters are dynamically adjusted based on the real-time current.

7. The method for dynamically calculating the state of charge of an electric vehicle based on driving behavior according to claim 1, characterized in that: In S3, the real-time accumulated charge is expressed as ； The state equation of the unscented Kalman filter algorithm is expressed as follows: ； The prediction equation is expressed as follows ； The temperature compensation coefficient is expressed as ； In the formula, The initial state of charge; For charge and discharge efficiency; Let k be the state of charge at time k; It represents the current quantity; This refers to the battery's rated capacity. , These are the two voltage states at time k; This is the state vector from the previous moment; Let be the input current at time k; This is process noise; Let be the battery terminal voltage at time k; The internal resistance of the battery is ohmic. This refers to model error or noise. This is the coefficient value.

8. The method for dynamically calculating the state of charge of an electric vehicle based on driving behavior according to claim 7, characterized in that: In S4, the SOC error compensation model is expressed as: ； Its dynamic weighted fusion is represented as ； In the formula, This is the driving compensation coefficient; The current is the effective value; T is the temperature. , , These are the weighting coefficients; SOC estimated by the ampere-hour integration method; SOC estimated by unscented Kalman filtering; This is a SOC correction item.

9. The method for dynamic calculation of the state of charge of an electric vehicle based on driving behavior according to claim 8, characterized in that: The weight allocation is as follows: And satisfy =1; In the formula, This represents the i-th confidence value; Its dynamic weighting strategy is as follows: when p > 0.8, ; When p < 0.5, ; .

10. The method for dynamically calculating the state of charge of an electric vehicle based on driving behavior according to claim 2, characterized in that: When the estimation error of 10 consecutive sampling points Time-triggered parameter update, represented as ； in, ; In the formula, This is the state estimation vector at time k; This is the state estimation vector at time k-1; Let be the gain matrix at time k; Let be the observation value at time k; ; The state covariance matrix at the previous time step; This is the state transition matrix; It is a forgetting factor.

Images