A method for constructing a digital twin system of an intelligent driving vehicle

Abstract

The present invention relates to the field of vehicle driving technology, and in particular to a method for constructing a digital twin system of an intelligent driving vehicle. The method comprises: collecting multi-dimensional data of a vehicle in real time and using it as raw data, preprocessing the raw data to obtain preprocessed data; using a multi-scale dynamic feature enhancement and reconstruction algorithm to perform enhancement processing to obtain enhanced data; constructing and training a digital twin model in combination with the multi-dimensional historical data of the vehicle to obtain a trained digital twin model; combining real-time environmental perception data with driver's behavior habit data, using a multi-level adaptive variational optimization algorithm to intelligently optimize the vehicle's driving state, driving decision, and path planning to obtain the optimal path planning. The method solves the technical problems of inaccurate processing of real-time vehicle data and inability to accurately and efficiently obtain driving states under different environmental conditions during the construction of a digital twin system for intelligent driving vehicles.

Description

Technical Field

[0001] The present invention relates to the field of vehicle driving technology, and in particular to a method for constructing a digital twin system of an intelligent driving vehicle. Background Art

[0002] With the rapid development of intelligent driving technology, how to improve the decision-making and adaptability of autonomous driving systems has become a core issue. Intelligent driving systems mainly rely on a variety of sensors and real-time data to perceive the environment and optimize driving decisions through complex algorithms. However, existing technologies still face some challenges, especially in real-time data processing, environmental adaptability, and optimization of comprehensive decision-making strategies.

[0003] Traditional intelligent driving systems mainly rely on a single data source, such as lidar, cameras, radar and other sensors to obtain environmental information. This environmental information is processed through rule-based algorithms or deep learning models to make driving decisions. However, these intelligent driving systems usually have certain limitations. On the one hand, the complexity of real-time data processing is high, especially in high-speed or complex environments. Existing algorithms are often unable to process a large amount of information in a short period of time, resulting in slow decision-making. On the other hand, although deep learning models perform well in some specific scenarios, they lack adaptability to diverse and dynamically changing traffic environments, and often fail to make accurate decisions in emergencies or extreme situations.

[0004] However, the existing methods for constructing digital twin systems for intelligent driving vehicles have the following technical problems: in the process of constructing digital twin systems for intelligent driving vehicles, the processing of real-time vehicle data is not accurate enough, and the driving status cannot be accurately and efficiently obtained under different environmental conditions. Summary of the invention

[0005] The present invention provides a method for constructing a digital twin system of an intelligent driving vehicle, so as to solve the technical problems that, during the construction process of the digital twin system of an intelligent driving vehicle, real-time vehicle data is not processed accurately and efficiently, and the driving status cannot be accurately and efficiently acquired under different environmental conditions.

[0006] A method for constructing a digital twin system of an intelligent driving vehicle of the present invention specifically includes the following technical solutions:

[0007] A method for constructing a digital twin system of an intelligent driving vehicle comprises the following steps:

[0008] S1. Collect multi-dimensional data of the vehicle in real time and use it as raw data, pre-process the raw data to obtain pre-processed data; enhance the pre-processed data using a multi-scale dynamic feature enhancement and reconstruction algorithm to obtain enhanced data; build and train a digital twin model based on the enhanced data and combined with the multi-dimensional historical data of the vehicle to obtain a trained digital twin model;

[0009] S2. Based on the enhanced data and the trained digital twin model, combined with the real-time environmental perception data and the driver's behavior habit data, the vehicle's driving status, driving decision-making, and path planning are intelligently optimized using a multi-level adaptive variational optimization algorithm to obtain the optimal path planning.

[0010] Preferably, the S1 specifically includes:

[0011] The multi-scale dynamic feature enhancement and reconstruction algorithm adopts multi-scale feature fusion, nonlinear mapping, multi-dimensional feature cross processing and adaptive feature reconstruction and optimization to obtain enhanced data.

[0012] Preferably, the S1 specifically includes:

[0013] In the process of implementing the multi-scale dynamic feature enhancement and reconstruction algorithm, multi-scale feature fusion is performed, and the weighted sliding window method is used to perform time series enhancement processing on the preprocessed data. The weighted sliding window method obtains the local feature data of the preprocessed data by weighted averaging the time series information of each preprocessed data. The specific implementation formula is:

[0014] ,

[0015] in, is the preprocessed data in time The Local feature data, that is, Local feature data of each sensor; is the weighting coefficient; is the size of the sliding window; is the index variable in the sliding window; is the Gaussian weighting function; is the standard deviation parameter of the Gaussian weighting function; It is Sensors at time The preprocessed data; is the weighting coefficient of the preprocessed data at the current moment; It is Sensors at time The preprocessed data.

[0016] Preferably, the S1 specifically includes:

[0017] In the process of implementing the multi-scale dynamic feature enhancement and reconstruction algorithm, nonlinear mapping and multi-dimensional feature cross processing are performed based on the local feature data of the preprocessed data; the local feature data is nonlinearly mapped, and the local feature data is mapped in combination with a multi-layer nonlinear activation function. The specific mapping process is as follows:

[0018] ,

[0019] in, is the preprocessed data in time The The feature data after nonlinear mapping transformation, that is, The characteristic data after nonlinear mapping transformation corresponding to each sensor; is the number of sensors collecting data; It is The sensor pairs The weighting coefficients of the sensors; It is for Local feature data of each sensor After non-linear activation function The processed result, nonlinear activation function The choice of hyperparameters Sure, Hyperparameters indicating the type of nonlinear activation function; is a hyperparameter that controls the strength of the nonlinear mapping; is the regularization coefficient; is a constant.

[0020] Preferably, the S1 specifically includes:

[0021] In the process of implementing the multi-scale dynamic feature enhancement and reconstruction algorithm, adaptive feature reconstruction and optimization are performed; the feature data after nonlinear mapping transformation is optimized through an adaptive mechanism, and the enhanced data is obtained by using the feature reconstruction process based on weighted regularization. The specific implementation formula is:

[0022] ,

[0023] in, After reconstruction, Feature data after nonlinear mapping transformation, that is, enhanced data; is the weighted mapping matrix; is the total number of feature data after nonlinear mapping transformation; It is The weighting coefficients of the feature data after nonlinear mapping transformation; is the regularization coefficient; is the regularization term; is the L2 norm; It is The standard deviation of the feature data after nonlinear mapping transformation.

[0024] Preferably, the S2 specifically includes:

[0025] In the implementation process of the multi-level adaptive variational optimization algorithm, a multi-level vehicle state space is constructed based on the enhanced processed data, real-time environmental perception data and the driver's behavior habit data; based on the vehicle state space, the variational optimization method is used to define the optimization objective function, and constraints are introduced to optimize the vehicle's driving state, driving decision-making and path planning.

[0026] Preferably, the S2 specifically includes:

[0027] Define the optimization objective function In the following form:

[0028] ,

[0029] in, , , They are the weight coefficients of the vehicle's energy consumption, comfort and vehicle dynamic characteristics; is the initial moment; is the end time, and the integral sign indicates the entire time interval from arrive Perform integration, that is, accumulate the calculation results at each moment within the time range; The vehicle is at time speed; The vehicle is at time acceleration; The vehicle is at time The displacement change of The vehicle is at time energy consumption; The vehicle is at time The total time elapsed; The vehicle is at time The turning angle change; , , , , , , , is the constant coefficient of the control term.

[0030] Preferably, the S2 specifically includes:

[0031] In the implementation process of the multi-level adaptive variational optimization algorithm, the constraints are integrated into the optimization problem by introducing the Lagrange multiplier method to obtain the optimal vehicle driving state, that is, the optimization parameters; through the dynamic feedback mechanism, the constraints and objective functions in the optimization process are adjusted in real time. The specific implementation formula is:

[0032] ,

[0033] in, is the optimization parameter at the current moment; Indicates the optimization parameters of the previous moment; is the constraint adjustment parameter; and Respectively represent the environmental variables of the current and previous moments, i.e., real-time environmental perception data; It is a modulo operation.

[0034] The beneficial effects of the technical solution of the present invention are:

[0035] 1. In order to improve the timing and dynamic characteristics of the data, a multi-scale dynamic feature enhancement and reconstruction algorithm is adopted. Through multi-scale feature fusion and nonlinear mapping, more accurate vehicle status and environmental information can be extracted from the pre-processed data. This method can enhance the sensitivity to vehicle status and environmental changes, improve the real-time and accuracy of driving decisions, and thus effectively optimize driving behavior.

[0036] 2. The trained data twin model can continuously adapt to complex road conditions and different driving scenarios through incremental learning and periodic dynamic updates. This self-optimization mechanism enables the trained digital twin model to always maintain a high prediction accuracy, especially in the face of extreme weather or sudden traffic incidents, its robustness is enhanced.

[0037] 3. Based on the dynamically updated data twin model and multi-level adaptive variational optimization algorithm, it can intelligently optimize driver behavior, vehicle driving status, path planning, etc., which enables the vehicle to make more intelligent decisions in complex traffic environments, optimize driving paths, reduce unnecessary vehicle energy consumption, and improve driving experience and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a method for constructing a digital twin system of an intelligent driving vehicle described in the present invention. DETAILED DESCRIPTION

[0039] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Unless defined otherwise, 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 belongs.

[0041] The following is a detailed description of a method for constructing a digital twin system of an intelligent driving vehicle provided by the present invention in conjunction with the accompanying drawings.

[0042] Refer to the attached Figure 1 , which shows a flow chart of a method for constructing a digital twin system of an intelligent driving vehicle provided by an embodiment of the present invention, the method comprising the following steps:

[0043] S1. Collect multi-dimensional data of the vehicle in real time and use it as raw data, pre-process the raw data to obtain pre-processed data; enhance the pre-processed data using a multi-scale dynamic feature enhancement and reconstruction algorithm to obtain enhanced data; build and train a digital twin model based on the enhanced data and combined with the multi-dimensional historical data of the vehicle to obtain a trained digital twin model;

[0044] The multi-dimensional data of the vehicle is collected in real time through sensors such as LiDAR, millimeter-wave radar, camera, GPS, IMU, etc., and the multi-dimensional data of the vehicle is used as the original data, including vehicle status data such as vehicle speed, acceleration, steering wheel angle, fuel consumption, braking status, etc., sensor data such as LiDAR scanning point cloud data, camera images, millimeter-wave radar echo data, etc., and environmental data such as road conditions, traffic flow, weather (such as temperature, humidity, rainfall, etc.), road signs, etc.;

[0045] The original data is preprocessed, such as denoising, data cleaning, normalization and standardization, and dimensional unification, to obtain preprocessed data. The technical means used in the preprocessing are well known to those skilled in the art and will not be described in detail here.

[0046] The pre-processed data is further enhanced using a multi-scale dynamic feature enhancement and reconstruction algorithm, which uses multi-scale feature fusion, nonlinear mapping, multi-dimensional feature cross-processing, and adaptive feature reconstruction and optimization to adapt to the diversity, dynamics, and complexity of intelligent driving vehicle data, and achieve efficient and accurate vehicle feature extraction and enhancement. The specific implementation process is as follows:

[0047] The first step is multi-scale feature fusion. First, the weighted sliding window method is used to perform time series enhancement on the preprocessed data to obtain local features at different times and scales, and the Gaussian weighting function is used to integrate the importance of the preprocessed data at different time points.

[0048] The weighted sliding window method obtains local feature data of the preprocessed data by weighted averaging the time series information of each preprocessed data. The local feature data is defined as:

[0049] ,

[0050] in, is the preprocessed data in time The Local feature data, that is, Local feature data of each sensor; is the weighting coefficient used to control the The preprocessed data is compared with the The weight of the preprocessed data in the local feature data is used to adjust the time The influence degree of the pre-processed data is obtained according to the expert experience method; is the size of the sliding window, that is, the weighted average of the preprocessed data in the past time steps, which determines the local feature data calculation at a specific time. The time series range of the preprocessed data is set by the time series characteristics of the preprocessed data or the application requirements; Is the index variable in the sliding window, indicating that from the current moment The time step to look back; is a Gaussian weighting function, which is used for weighting to smooth the time The impact of preprocessed data on current local feature data; is the standard deviation parameter of the Gaussian weighting function, which is used to control the width of the Gaussian weighting function and thus adjust the The influence range of the preprocessed data on the current local feature data is determined according to the expert experience method; It is Sensors at time The preprocessed data; It is the weighting coefficient of the preprocessed data at the current moment, which is used to control the weight of the preprocessed data at the current moment in the final local feature data calculation, and is determined according to the expert experience method; It is Sensors at time The preprocessed data;

[0051] The above formula smoothes and weights the preprocessed data, so that the preprocessed data can fuse multi-scale information in the time dimension, thereby enhancing the temporal characteristics of the preprocessed data.

[0052] The second step is to cross-process nonlinear mapping and multi-dimensional features. After extracting local feature data, further nonlinear mapping is performed on the local feature data to capture the complex interaction between sensors. Through multiple layers of nonlinear activation functions, the local feature data is deeply transformed and mapped to extract higher-order implicit features. The specific mapping process is as follows:

[0053] ,

[0054] in, is the preprocessed data in time The The feature data after nonlinear mapping transformation, that is, The characteristic data after nonlinear mapping transformation corresponding to each sensor; is the number of sensors collecting data; It is The sensor pairs The weighting coefficient of each sensor is used to express the influence weight between different sensors and is learned through the existing training algorithm; It is for Local feature data of each sensor After non-linear activation function The processed result, nonlinear activation function The choice of (e.g. ReLU, Sigmoid, Tanh) is determined by the hyperparameters Sure, Hyperparameters representing the type of nonlinear activation function, used to specify the The nonlinear activation mode of each sensor is determined according to the expert experience method; It is a hyperparameter that controls the strength of nonlinear mapping. It is used to adjust the nonlinearity of the result after being processed by the nonlinear activation function and is determined by experimental methods. is the regularization coefficient, which is used to balance the smoothness of local feature data and the retention of nonlinear features to prevent overfitting. It is determined through experiments or automatic parameter adjustment mechanisms; is a constant that prevents division by zero errors in the logarithmic function and is used to control the smoothing effect in local feature data processing;

[0055] The above formula extracts higher-dimensional implicit information from local feature data through nonlinear activation functions and logarithmic smoothing operations, so as to facilitate subsequent deep feature fusion and optimization.

[0056] The third step is adaptive feature reconstruction and optimization. In order to avoid redundant and irrelevant information in the feature data after nonlinear mapping transformation, the feature data after nonlinear mapping transformation is optimized through an adaptive mechanism. The feature reconstruction process based on weighted regularization is used to select the most representative features from all features and compress redundant features. The core formula of feature reconstruction is as follows:

[0057] ,

[0058] in, After reconstruction, Feature data after nonlinear mapping transformation, that is, enhanced data; is a weighted mapping matrix, which defines how to extract the first feature data after all nonlinear mapping transformations and reconstruct The information related to the feature data after the nonlinear mapping transformation is obtained by training the parameter matrix using the existing gradient descent method; is the total number of feature data after nonlinear mapping transformation; It is The weighting coefficient of the feature data after nonlinear mapping transformation represents the weight in the final reconstructed feature, which determines the influence of each feature on the enhanced data and is determined by experimental method; It is the regularization coefficient, which is used to control the smoothness of the feature data after nonlinear mapping transformation. It determines the influence of the correlation between features in the calculation. The purpose is to avoid overfitting between features. It is automatically determined by cross-validation or training data. is the regularization term; is the L2 norm; It is The standard deviation of the feature data after the nonlinear mapping transformation measures the fluctuation range of the feature. Through the feature reconstruction process based on weighted regularization, it is possible to dynamically weight and select features, remove redundant information and retain the most predictive features. The introduction of regularization terms effectively suppresses overfitting while retaining the diversity of features.

[0059] Based on the enhanced processed data, combined with the multi-dimensional historical data of the vehicle obtained from the existing database, select the existing modeling method to build and train the digital twin model, such as a neural network based on deep learning (such as a convolutional neural network, a recurrent neural network) or a regression model based on traditional machine learning, support vector machine, etc. The specific choice should be determined according to the data type and model goal. For example, when processing time series data (such as the historical driving data of the vehicle), you can choose a deep learning method such as RNN or LSTM. The digital twin model is trained by the enhanced processed data and multi-dimensional historical data, and verified using existing methods such as cross-validation, and finally the trained digital twin model is obtained.

[0060] According to periodic evaluation, such as re-evaluating the precision and accuracy of the trained digital twin model at regular intervals or when the performance of the digital twin model deteriorates. If the prediction accuracy of the trained digital twin model decreases in some cases, such as in extreme weather or complex traffic environments, it needs to be updated and adjusted through the latest collected data. For example, incremental learning or transfer learning technology can be used to optimize the existing trained digital twin model by continuously absorbing new data, so that the trained digital twin model is continuously improved and adapted to new environments and scenarios, and periodic dynamic updates are achieved;

[0061] S2. Based on the enhanced data and the trained digital twin model, combined with the real-time environmental perception data and the driver's behavior habit data, the vehicle's driving status, driving decision-making, and path planning are intelligently optimized using a multi-level adaptive variational optimization algorithm to obtain the optimal path planning.

[0062] Based on the trained digital twin model, the vehicle's environment is identified through real-time perception of factors such as traffic signals, road conditions, and weather changes, and real-time environmental perception data is obtained. For example, there is a red light or traffic accident ahead; at the same time, combined with the vehicle's historical driving data extracted from the existing database, the driver's historical behavior is analyzed through existing machine learning algorithms (such as cluster analysis and behavior prediction models), and the driver's behavior habit data is analyzed, such as sudden acceleration, sudden braking, and turning frequency.

[0063] Furthermore, based on the enhanced processed data and the driver's behavior habit data, a multi-level adaptive variational optimization algorithm is introduced. The multi-level adaptive variational optimization algorithm can intelligently optimize the vehicle's driving state, driving decision and path planning in a complex and dynamically changing environment according to the real-time environmental perception data and the driver's behavior habit data through multi-level adaptive modeling, variational optimization, dynamic feedback adjustment and multi-objective optimization technology; the specific implementation process is as follows:

[0064] First, based on the enhanced processed data, real-time environmental perception data and driver behavior habit data, a multi-level vehicle state space is constructed. The dimensions of the vehicle state space include real-time vehicle data such as position, speed, acceleration, and real-time environmental perception data such as traffic flow, weather conditions, and road conditions. The vehicle state space is defined as , is a vector where each element represents the state variable of the vehicle at a certain moment. This can include vehicle dynamic variables (such as speed, acceleration) and environmental variables (such as road slope, traffic flow), etc. Further, define the vehicle at time The status is:

[0065] ,

[0066] in, For vehicles at time speed; The vehicle is at time acceleration; The vehicle is at time The road slope; The vehicle is at time of traffic flow.

[0067] Further model the driver's behavior habit data, and use the existing regression analysis method to derive the state transfer function of the driver's behavior habit data. , describes the change relationship of the vehicle state at the current moment based on the state at the previous moment:

[0068] ,

[0069] in, is the vehicle state at the current moment after being processed by the state transfer function; is the vehicle state at the previous moment after being processed by the state transfer function; Indicates at time Real-time environmental perception data, such as traffic conditions, weather, etc. It is a state transfer function, which is used to predict the state change of the vehicle in a given environment.

[0070] Furthermore, the variational optimization method is used to optimize the vehicle's driving state, driving decision and path planning. The goal of variational optimization is to minimize the vehicle's total energy consumption, driving time, comfort and other indicators by solving a series of optimization functions, taking into account various constraints in the path planning and driving decision-making process.

[0071] The specific contents include: defining the optimization objective function In the following form:

[0072] ,

[0073] in, , , They are the weight coefficients of the vehicle's energy consumption, comfort and vehicle dynamic characteristics, which are used to balance the relative importance of different optimization objectives and are obtained according to the expert experience method; is the initial moment; is the end time, and the integral sign indicates the entire time interval from arrive Perform integration, that is, accumulate the calculation results at each moment in the time range, and the time interval represents all the moments in the vehicle's driving process; The vehicle is at time speed; The vehicle is at time acceleration; The vehicle is at time The displacement change of The vehicle is at time The energy consumption is calculated by the existing vehicle dynamics model; The vehicle is at time The total time taken is obtained by calculating the path time; The vehicle is at time The turning angle change is obtained through the existing vehicle-mounted computing system; the above vehicle state variables , , , , , All based on get; , , , , , , , It is the constant coefficient of the control item, which is used to express the contribution of each factor (such as speed, acceleration, displacement change, energy consumption, total time, etc.) to the optimization objective function and is determined by experimental method.

[0074] In order to optimize the above objective function, constraints are introduced to ensure that the state of the vehicle during the optimization process is always within the physical and safe boundaries. For example, the speed of the vehicle cannot exceed the maximum limit , the acceleration cannot exceed the maximum value , and the path planning time must be within a reasonable range; under given constraints, the vehicle's speed, acceleration and path planning time must always be kept within a safe range.

[0075] By introducing the Lagrange multiplier method, the constraints are integrated into the optimization problem, the variational equation is obtained, and the optimal vehicle driving state is solved by the existing numerical method, that is, the optimization parameters are obtained. .

[0076] Furthermore, the environment in which the vehicle is located may change at any time, such as road traffic conditions, weather conditions, etc. Through the dynamic feedback mechanism, the constraints and objective functions in the optimization process are adjusted in real time. Define constraint adjustment parameters , which indicates the degree of influence of environmental changes on the optimization process. The core formula of the dynamic feedback adjustment mechanism is as follows:

[0077] ,

[0078] in, is the optimization parameter at the current moment; Represents the optimization parameters of the previous moment; It is a constraint adjustment parameter used to control the impact of environmental changes on the update of optimization parameters and is determined based on expert experience. and Respectively represent the environmental variables of the current and previous moments, that is, real-time environmental perception data, including real-time traffic flow, weather conditions, road friction coefficient, traffic signals, obstacle information, etc.; This is a modulo operation. In this way, we can respond to changes in the external environment in real time and dynamically adjust the optimization strategy.

[0079] By using the existing multi-objective optimization algorithm, the path is replanned based on traditional goals such as shortest time and minimum energy consumption, as well as factors such as road conditions and traffic flow. The constraints are determined according to the specific scenario and solved using the existing multi-objective optimization algorithm to obtain the optimal path planning.

[0080] Through the above process, intelligent optimization of the vehicle's driving status, driving decisions, and path planning is achieved.

[0081] In summary, a method for constructing a digital twin system of an intelligent driving vehicle has been completed.

[0082] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0083] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0084] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A method for constructing a digital twin system of an intelligent driving vehicle, characterized in that: The following steps are involved: S1. Collect multi-dimensional data of vehicles in real time and use it as raw data, pre-process the raw data to obtain pre-processed data; The preprocessed data is enhanced using a multi-scale dynamic feature enhancement and reconstruction algorithm, which uses multi-scale feature fusion, nonlinear mapping, multi-dimensional feature cross-processing, and adaptive feature reconstruction and optimization to obtain enhanced data; based on the enhanced data, a digital twin model is constructed and trained in combination with multi-dimensional historical data of the vehicle to obtain a trained digital twin model; S2. Based on the enhanced processed data and the trained digital twin model, combined with the real-time environmental perception data and the driver's behavior habit data, the multi-level adaptive variational optimization algorithm is used to intelligently optimize the vehicle's driving state, driving decision, and path planning to obtain the optimal path planning. In the implementation process of the multi-level adaptive variational optimization algorithm, a multi-level vehicle state space is constructed based on the enhanced processed data, real-time environmental perception data, and the driver's behavior habit data; Based on the vehicle state space, the variational optimization method is used to define the optimization objective function and define the optimization objective function. In the following form: ， in, , , They are the weight coefficients of the vehicle's energy consumption, comfort and vehicle dynamic characteristics; is the initial moment; is the end time, and the integral sign indicates the entire time interval from arrive Perform integration; The vehicle is at time speed; The vehicle is at time acceleration; The vehicle is at time The displacement change of The vehicle is at time energy consumption; The vehicle is at time The total time elapsed; The vehicle is at time The turning angle change; , , , , , , , is the constant coefficient of the control term; And introduce constraints to optimize the vehicle's driving status, driving decisions and path planning; By introducing the Lagrange multiplier method, the constraints are integrated into the optimization problem to obtain the optimal vehicle driving state, that is, the optimization parameters. Through the dynamic feedback mechanism, the constraints and objective functions in the optimization process are adjusted in real time. The specific implementation formula is: ， in, is the optimization parameter at the current moment; Represents the optimization parameters of the previous moment; is the constraint adjustment parameter; and Respectively represent the environmental variables of the current and previous moments, i.e., real-time environmental perception data; It is a modulo operation.

2. The method for constructing a digital twin system of an intelligent driving vehicle according to claim 1, characterized in that: The S1 specifically includes: In the process of implementing the multi-scale dynamic feature enhancement and reconstruction algorithm, multi-scale feature fusion is performed, and the weighted sliding window method is used to perform time series enhancement processing on the preprocessed data. The weighted sliding window method obtains the local feature data of the preprocessed data by weighted averaging the time series information of each preprocessed data. The specific implementation formula is: ， in, is the preprocessed data in time The Local feature data, that is, Local feature data of each sensor; is the weighting coefficient; is the size of the sliding window; is the index variable in the sliding window; is the Gaussian weighting function; is the standard deviation parameter of the Gaussian weighting function; It is Sensors at time The preprocessed data; is the weighting coefficient of the preprocessed data at the current moment; It is Sensors at time The preprocessed data.

3. The method for constructing a digital twin system of an intelligent driving vehicle according to claim 2, characterized in that: The S1 specifically includes: In the process of implementing the multi-scale dynamic feature enhancement and reconstruction algorithm, nonlinear mapping and multi-dimensional feature cross processing are performed based on the local feature data of the preprocessed data; the local feature data is nonlinearly mapped, and the local feature data is mapped in combination with a multi-layer nonlinear activation function. The specific mapping process is as follows: ， in, is the preprocessed data in time The The feature data after nonlinear mapping transformation, that is, The characteristic data after nonlinear mapping transformation corresponding to each sensor; is the number of sensors collecting data; It is The sensor pairs The weighting coefficients of the sensors; It is for Local feature data of each sensor After non-linear activation function The processed result, nonlinear activation function The choice of hyperparameters Sure, Hyperparameters indicating the type of nonlinear activation function; is a hyperparameter that controls the strength of the nonlinear mapping; is the regularization coefficient; is a constant.

4. The method for constructing a digital twin system of an intelligent driving vehicle according to claim 3, characterized in that: The S1 specifically includes: In the process of implementing the multi-scale dynamic feature enhancement and reconstruction algorithm, adaptive feature reconstruction and optimization are performed; the feature data after nonlinear mapping transformation is optimized through an adaptive mechanism, and the enhanced data is obtained by using the feature reconstruction process based on weighted regularization. The specific implementation formula is: ， in, After reconstruction, Feature data after nonlinear mapping transformation, that is, enhanced data; is the weighted mapping matrix; is the total number of feature data after nonlinear mapping transformation; It is The weighting coefficients of the feature data after nonlinear mapping transformation; is the regularization coefficient; is the regularization term; is the L2 norm; It is The standard deviation of the feature data after nonlinear mapping transformation.

Images