Intelligent Driving - Human-Driver Mode Switching Method for Urban Logistics Vehicles Considering Driving Risks and Energy Consumption

Abstract

This application relates to the field of intelligent driving technology and discloses a method for switching between intelligent driving and human-driven modes for urban logistics vehicles, considering driving risks and energy consumption. The method includes extracting driving data for driver mode switching, obtaining the mode switching interval considering driver style and weekly vehicle trajectory prediction, and optimizing the mode switching timing based on driving risks and maximum energy consumption. This method uses historical feature data of the driver mode switching process and weekly vehicle trajectory prediction, combined with style recognition results, to obtain the risk interval and risk change curve for mode switching decisions using a risk assessment model. Based on these two factors, the time interval for mode switching decisions is obtained. Finally, an optimization model is established based on human-driven mode speed prediction to optimize the switching timing. This application comprehensively considers driving risks and driving energy consumption, effectively solving the problems of high driving energy consumption and insufficient safety caused by unreasonable human-machine co-driving mode switching in complex urban scenarios for intelligent driving logistics vehicles, thereby improving the traffic efficiency of urban logistics vehicles.

Description

Technical Field

[0001] This application relates to the field of intelligent driving technology, and in particular to a method for switching between intelligent driving and human driving modes for urban logistics vehicles that takes into account driving risks and energy consumption. Background Technology

[0002] In the complex urban road environment of logistics and transportation, driving scenarios are diverse and highly variable. Maintaining driving safety while optimizing energy consumption in human-machine co-driving vehicles is a significant challenge. Therefore, under the premise of safe driving, research is urgently needed to develop a predictive switching method for intelligent driving-human driving modes based on driving risks and energy consumption. This method aims to optimize the switching between intelligent driving and manual driving modes through real-time prediction and assessment of driving risks and energy consumption, thereby minimizing energy consumption.

[0003] However, existing intelligent driving-human driving mode switching methods mainly focus on one aspect of driving safety or energy consumption optimization, failing to comprehensively consider the balance between the two. For example, prior art published on November 5, 2021 (application publication number CN 113602284 A) specifically discloses a human-machine co-driving mode decision-making method, device, equipment, and storage medium. The method uses a watch to sense the driver's physiological characteristic indicators; determines the corresponding driving mode based on the physiological characteristic indicators using a preset adaptive learning model; acquires vehicle information within a preset range ahead; generates a corresponding vehicle intent model based on the vehicle information using a preset behavior prediction algorithm; determines the current risk level of the vehicle based on the vehicle intent model; and determines the target mode switching decision based on the driving mode and risk level. While this method addresses the lag in human-machine co-driving mode switching by predicting the driver's objective physiological state and driving risk level, considering only the objective state leads to a poor driving experience for different drivers due to their different subjective driving styles. Furthermore, the driving decision does not take vehicle energy consumption into account, resulting in problems such as poor driving experience and high energy consumption in the target mode switching decision. Summary of the Invention

[0004] To address the aforementioned issues, this application provides a method for switching between intelligent driving and human driving modes for urban logistics vehicles, taking into account driving risks and energy consumption. This method aims to resolve the problems of poor driving experience and high energy consumption in target mode switching decisions when dealing with complex urban road conditions and scenarios involving multiple vehicles interfering with each other.

[0005] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0006] This application provides a method for switching between intelligent driving and human driving modes for urban logistics vehicles, taking into account driving risks and energy consumption. The method includes:

[0007] Based on the driving characteristic data of drivers in different driving scenarios, the drivers are classified using a clustering algorithm, and the driver style recognition model based on deep learning is used to identify the driver style, thereby obtaining driving data on the mode switching process of the current driver style category.

[0008] Based on driving data during the mode switching process of the current driver style category, the driving risk value during mode switching is calculated based on the risk field model. The left interval of the mode switching decision corresponding to the driver style is obtained based on the driving risk value during mode switching. The future trajectories of the vehicle and surrounding vehicles are obtained based on the EA-Net trajectory prediction model of the environmental attention network. The future driving risk change curve is obtained using the driving risk assessment method based on the risk field model. The driving risk value at the collision location is used as the right interval of the mode switching decision.

[0009] Based on the combination of future risk curves and mode switching decision risk intervals, the time interval for drivers to switch driving modes is obtained. The speed prediction model of driver-driver mode based on LSTM recurrent neural network is used to obtain the speed change prediction sequence of driver-driver mode. Based on the speed prediction and mode switching time interval constraints, a mode switching optimization model based on limit energy consumption is established, and the optimal switching time is obtained by solving the problem.

[0010] Furthermore, the different driving scenarios include cruising, following other vehicles, crossing intersections, and queuing at intersections;

[0011] The driving characteristic data of the driver under different driving scenarios includes accelerator pedal opening, steering wheel angle, steering wheel speed, vehicle longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, vehicle roll angle and heading angle. The average, maximum and variance of the absolute values ​​of the driver's driving characteristic data under different driving scenarios are used as the driver style characteristic indicators.

[0012] Furthermore, the classification of drivers using a clustering algorithm includes:

[0013] Based on driver style characteristic indicators, factor analysis was used to reduce dimensionality, resulting in two factors: a horizontal factor and a vertical factor.

[0014] Based on the aforementioned horizontal and vertical factors, a clustering algorithm is used to classify drivers into three categories: aggressive, average, and conservative.

[0015] Furthermore, the deep learning-based driver style recognition model identifies driver style and obtains driving data on the current driver style category's mode switching process, including:

[0016] A learning sample dataset is constructed based on classification features and category data. A deep learning-based driver style recognition model is then established. Based on this model, the processed driver feature index data is used as input to obtain the driver style recognition result. The calculation process is as follows:

[0017]

[0018] Where X is an input vector containing driver characteristic index data, X = {x1, x2}, where x1 represents the horizontal factor of driver characteristics, x2 represents the vertical factor of driver characteristics, and W i and b i is the weight matrix and bias of the i-th layer of the multilayer perceptron, n is the number of layers of the multilayer perceptron, and Y is the driver style category output;

[0019] Based on the clustering results, we analyze the driving data of drivers with three driving styles during the driving mode switching process in four driving scenarios, including the vehicle's position coordinates, vehicle speed, ambient vehicle position coordinates, and ambient vehicle speed. We then use the driver style recognition results to obtain the driving dataset N representing the current driver style category during the mode switching process. a , is represented as;

[0020]

[0021] in To identify the driving data during the mode switching process of the j-th driver in driving style a, For the driving data of the i-th environment vehicle, Let x, y, and y represent the x-coordinate, y-coordinate, and speed of vehicle i, respectively, and let I be the set of vehicle IDs for the surrounding environment. For the driving data of the j-th driver, Let x, y, and y be the x-coordinate of the vehicle j, and y be the vehicle speed.

[0022] Furthermore, the step of calculating the driving risk value during mode switching based on the driving data of the current driver style category and the risk field model includes:

[0023] Based on the influence of Z environmental vehicles covering all directions around the vehicle, the field strength E of the driving risk field generated by the surrounding moving vehicles at the vehicle's location is calculated. j The calculation formula is:

[0024]

[0025] Among them, E V_ij For surrounding vehicles i(x) i ,y i ) in the vehicle j(x j ,yj The resulting risk field strength, (x) i ,y i ), (x j ,y j Let ) represent the coordinates of the vehicle's center of mass, where k1, k2, and G are constants greater than 0, and R... i M represents the road condition influencing factor at location i for surrounding vehicles. i Given the equivalent mass of surrounding vehicle i, r ij =(x j -x i ,y j -y i ), representing the distance vector between surrounding vehicle i and vehicle j, with the direction being the same as the field strength direction, v i Let θ be the speed of the surrounding vehicle i. i Let r be the velocity direction of the surrounding vehicle i and the distance vector r. ij The angle between them, where exp is an exponential function with the natural constant e as its base;

[0026] The force acting on the vehicle is calculated using the formula for force in an electric field, and this force is used as the vehicle's driving risk. j The formula for calculating the force in the field strength is expressed as follows:

[0027] risk j =E j M j R j exp[-k2v j cos(θ j )](1+Dr j (4)

[0028] Among them, M j For the equivalent mass of the vehicle, R j v is the factor affecting road conditions at the vehicle location. j Let θ be the vehicle's speed. j The direction of the vehicle's velocity and the field strength E j The angle between directions, D rj Risk factors for drivers of private vehicles.

[0029] Furthermore, the left interval for obtaining the mode switching decision corresponding to the driver's style based on the driving risk value during mode switching includes:

[0030] Based on driving data of the current driver style category during mode switching, a driving risk assessment method based on a risk field model is used to calculate the driving risk value of each driver when switching driving modes in the driving dataset. The minimum value among these values ​​is taken as the driving risk value for mode switching of that driving style and is used as the left endpoint of the driving risk interval for mode switching decisions. The calculation formula is as follows:

[0031]

[0032] in, This represents the left endpoint of the driving risk range for the mode switching decision of driver type A. Let represent the driving risk value of the j-th driver in class a in the dataset when switching driver modes, and min represents the minimum value function.

[0033] Furthermore, the trajectory prediction model based on the Environmental Attention Network (EA-Net) obtains the future trajectories of the vehicle and surrounding vehicles, and uses a driving risk assessment method based on a risk field model to derive the future driving risk change curve. The driving risk value at the collision location is used as the right interval for the mode switching decision, including:

[0034] A training dataset was constructed based on historical time-domain information of the vehicle and surrounding vehicles from the experimental data. This dataset includes lateral position, longitudinal position, lateral velocity, longitudinal velocity, lateral acceleration, longitudinal acceleration, and heading angle relative to the lane lines. Inter-vehicle interaction characteristics were analyzed, and a trajectory prediction model based on the Environmental Attention Network (EA-Net) was established. This EA-Net trajectory prediction model was then used to predict the trajectories of the vehicle and multiple surrounding vehicles in real time. The trajectory prediction input was the past T... l The historical time-domain information of the vehicle and surrounding vehicles is used to output the future T. p The predicted trajectory data is of a certain duration; the trajectory prediction model based on the EA-Net environmental attention network is represented as follows:

[0035]

[0036] in, T represents the predicted future trajectory of the vehicle. l T represents the length of the historical time domain. g T represents the interval between the historical time domain and the future time domain. p To predict the time domain length in the future, f (t) ={x (t) ,y (t)}, f (t) Let x be the position coordinate at time t. (t) Let y be the x-coordinate of the position at time t. (t) Let t be the ordinate of the position at time t; The input feature matrix is ​​constructed using historical time-domain information of the vehicle to be predicted and vehicles in the surrounding environment. A is the spatiotemporal correlation matrix, according to W i With W j Established based on the interaction and influence at a certain moment. This indicates whether there is an interaction between surrounding vehicle i and vehicle j at a certain moment. If there is an interaction between surrounding vehicle i and vehicle j at a certain moment, then... If there is no interaction between surrounding vehicle i and vehicle j at any time, then W i This represents the historical time-domain information of the vehicle to be predicted and its surrounding environment. i = 0, 1, 2, ..., n; the feature information of the i-th vehicle at time t is represented as: x (t) ,y (t) , θ (t) Let represent the lateral position, longitudinal position, lateral velocity, longitudinal velocity, lateral acceleration, longitudinal acceleration, and heading angle relative to the lane line at time t, respectively.

[0037] Based on the trajectory prediction results, the future position coordinates and velocities of the vehicle and multiple surrounding vehicles are obtained. Then, using a driving risk assessment method based on a risk field model, the future T-wave velocity of the vehicle is calculated. p The driving risk at each moment is used to derive the future driving risk change curve, and the driving risk value at the location where the vehicle will collide with the future trajectories of surrounding vehicles is used as the right endpoint of the driving risk interval for mode switching decisions.

[0038] Furthermore, based on the combination of future risk curves and mode-switching decision risk intervals, the time intervals for drivers to switch driving modes are obtained, including:

[0039] Driving risk range based on driver mode switching decision The curve of future driving risk changes determines the time range for drivers to make decision recommendations on switching driving modes. in, This represents the left endpoint of the driving risk range for the mode switching decision of driver type A. This represents the right endpoint of the driving risk range for the mode switching decision of driver type A. This indicates the earliest recommended time to switch driving modes. This indicates the latest recommended time to switch driving modes.

[0040] Furthermore, the human-driving mode speed prediction model based on LSTM recurrent neural network obtains the human-driving mode speed change prediction sequence.

[0041] Based on the characteristics of vehicle speed changes before and after driving mode switching under different driving styles in different driving scenarios, a human-driving mode speed prediction model based on LSTM recurrent neural network is established. The model input is driving style and time length T. s The intelligent driving mode speed time series and driving risk time series are output with a time length of T. e The human-driving mode speed change prediction sequence, the human-driving mode speed prediction model based on LSTM recurrent neural network is expressed as follows:

[0042] V hu =w*LSTM(V in ,R,S,h)+b (7)

[0043] in, This represents the output predicted velocity sequence. Representing the future T e Predicted velocity at different times within a time period, The planned speed sequence for vehicles in intelligent driving mode. These represent the vehicle's historical T data in intelligent driving mode. s The planning speed at different times within the time frame. As a driving risk sequence, Representing history T respectively s Driving risk at different times within a time interval, where S represents driver style, h represents hidden layer state, w and b represent the weights and biases of the fully connected layer, and the time interval T represents driving risk at different times. s +T e ≥t f -t0, LSTM represents the Long Short-Term Memory network structure.

[0044] Furthermore, the mode switching optimization model based on extreme energy consumption is expressed as:

[0045]

[0046]

[0047] Where T represents the timing of the decision variable switching, and v in (t) represents the speed of the intelligent driving mode at time t, v hu v(t) represents the predicted speed of the driver mode at time t, where t is the time, v is the vehicle speed, v(t) is the vehicle speed at time t, a is the driver mode switching decision category, denergy is the instantaneous energy consumption calculation function obtained by fitting, and J is the energy consumption target value.

[0048] This application has at least the following beneficial effects:

[0049] (1) Based on driver characteristic data analysis and driving risk prediction, this application determines the time interval for driver mode switching decisions. This interval combines the earliest switching time considering the driver's subjective style and the latest switching time considering driving safety. It can more accurately characterize the time range for the driver to make future mode switching decisions while ensuring driving safety, thus laying the foundation for optimizing the switching time to provide drivers with more accurate decision-making suggestions.

[0050] (2) Based on the time interval of the driver mode switching decision and the speed prediction of the human-driving mode after the switching time, this application obtains the optimal switching time by establishing a switching time optimization model with energy consumption as the target. Furthermore, in the speed prediction, the driver style and driving risk are comprehensively considered, which can more accurately predict the speed change of the human-driving process.

[0051] (3) Compared with existing mode switching timing suggestions that only consider driving safety, this application takes into account the balance between safety and energy consumption, and provides drivers with the optimal switching timing suggestions with the lowest energy consumption while ensuring driving safety. Attached Figure Description

[0052] Figure 1 A flowchart illustrating the intelligent driving-human driving mode switching method for urban logistics vehicles, which considers driving risks and energy consumption, provided in this application embodiment.

[0053] Figure 2 This is a schematic diagram of a complex urban driving scenario provided in an embodiment of this application.

[0054] Figure 3 This is a schematic diagram of the positions of eight environmental vehicles surrounding the vehicle provided in an embodiment of this application.

[0055] Figure 4 This is a schematic diagram illustrating the generation of the driving mode switching decision time interval provided in an embodiment of this application.

[0056] Figure 5 This is a schematic diagram illustrating the human-driving mode speed prediction after the mode switching timing provided in an embodiment of this application. Detailed Implementation

[0057] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0058] The specific implementation methods of this application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0059] This application provides a method for switching between intelligent driving and human driving modes for urban logistics vehicles, taking into account driving risks and energy consumption. Figure 1 The flowchart shown is for this method. This method for switching between intelligent driving and human driving modes for urban logistics vehicles, which considers driving risks and energy consumption, includes the following steps S1 to S3, which are described in detail below.

[0060] S1: Feature data extraction for the mode switching process considering driver style analysis.

[0061] In this embodiment, by analyzing the driving characteristic data of drivers in different driving scenarios, clustering algorithms are used to classify drivers, a deep learning-based driver style recognition model is used to identify driver styles, and driving data of the current driver style category mode switching process is obtained accordingly.

[0062] In some embodiments, different driving scenarios include cruising, following, intersection crossing, and intersection queuing. The driver's driving characteristic data in different driving scenarios include accelerator pedal opening, steering wheel angle, steering wheel speed, vehicle longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, vehicle roll angle, and heading angle. The average, maximum, and variance of the absolute values ​​of these data are calculated as driver style characteristic indicators.

[0063] like Figure 2 The diagram shown is a schematic of a complex urban driving scenario provided in an embodiment of this application. The method provided in this embodiment can be applied to scenarios such as... Figure 2 In the scene shown. Figure 2 Four scenarios are shown: cruising, following another vehicle, queuing at an intersection, and passing through an intersection.

[0064] In some embodiments, a clustering algorithm is used to classify drivers. Specifically, the obtained driver style characteristic indicators are reduced in dimensionality using factor analysis to eliminate the correlation between features, resulting in two factors (horizontal factors and vertical factors). Then, the drivers are classified into three categories using the reduced-dimensional characteristic factor data: aggressive, average, and conservative.

[0065] In some embodiments, a deep learning-based driver style recognition model identifies driver style, including:

[0066] A learning sample dataset is constructed based on classification features and category data. A driver style recognition model based on deep learning is established. The input is the horizontal factor and the vertical factor, and the output is the category, which is expressed as Equation (1). Based on the driver style recognition model, the processed driver feature index data is used as input to obtain the driver style recognition result as class a.

[0067]

[0068] Where X is an input vector containing driver characteristic index data, X = {x1, x2}, where x1 represents the horizontal factor of driver characteristics, x2 represents the vertical factor of driver characteristics, and W i and b i is the weight matrix and bias of the i-th layer of the multilayer perceptron, n is the number of layers in the multilayer perceptron, and Y is the driver style category output.

[0069] In some embodiments, driving data during the mode switching process of the current driver style category is obtained through the following method:

[0070] Based on the clustering results, we analyze the driving data of drivers with three driving styles during the driving mode switching process in four driving scenarios, including the vehicle's position coordinates, vehicle speed, ambient vehicle position coordinates, and ambient vehicle speed. We then use the driver style recognition results to obtain the driving dataset N representing the current driver style category during the mode switching process. a (Equation (2)).

[0071]

[0072] in To identify the driving data during the mode switching process of the j-th driver in driving style a, For the driving data of the i-th environment vehicle, Let x, y, and y represent the x-coordinate, y-coordinate, and speed of vehicle i, respectively, and let I be the set of vehicle IDs for the surrounding environment. For the driving data of the j-th driver, Let x, y, and y be the x-coordinate of the vehicle j, and y be the vehicle speed.

[0073] S2: Determine the mode switching decision range considering driver subjective style and weekly vehicle trajectory prediction.

[0074] In this embodiment, based on the driving data of the current driver style category during the mode switching process, the driving risk value at the time of mode switching is calculated based on the risk field model to obtain the left interval of the mode switching decision corresponding to the driver style; the future trajectories of the vehicle and surrounding vehicles are obtained based on the EA-Net trajectory prediction model of the environmental attention network, and the future driving risk change curve is obtained using the driving risk assessment method based on the risk field model. The driving risk value at the collision location is used as the right interval of the mode switching decision; thus, the mode switching decision interval considering the driver's subjective style and the prediction of surrounding vehicle trajectories is determined.

[0075] In some embodiments, such as Figure 3 The diagram shown is a schematic representation of the positions of eight environmental vehicles surrounding the vehicle provided in this embodiment of the application. Figure 3 In this context, EV represents the vehicle itself, and LF, L, LE, Re, RR, R, and RF represent eight environmental vehicles in all directions (eight directions) surrounding the vehicle. Based on driving data during the mode switching process according to the current driver's style category, the driving risk value during mode switching is calculated using a risk field model, specifically including:

[0076] Considering the influence of eight surrounding vehicles on the vehicle from all directions, calculate the field strength E of the driving risk field generated by the surrounding moving vehicles at the vehicle's location. j (Equation (3), where Z = 8), and then use the formula for calculating the force in the field strength to calculate the force on the vehicle, that is, the driving risk of the vehicle. j (Equation (4));

[0077]

[0078] Among them, E V_ij For surrounding vehicles i(x) i ,y i ) in the vehicle j(x j ,y j The resulting risk field strength, (x) i ,y i ), (x j ,y j Let ) represent the coordinates of the vehicle's center of mass, where k1, k2, and G are constants greater than 0, and R... i M represents the road condition influencing factor at location i for surrounding vehicles. i Given the equivalent mass of surrounding vehicle i, r ij =(x j -x i ,y j -y i ), representing the distance vector between surrounding vehicle i and vehicle j, with the direction being the same as the field strength direction, v i Let θ be the speed of the surrounding vehicle i. iLet r be the velocity direction of the surrounding vehicle i and the distance vector r. ij The angle between the two sides, where exp is an exponential function with the natural constant e as its base.

[0079] risk j =E j M j R j exp[-k2v j cos(θ j )](1+Dr j (4)

[0080] Among them, M j For the equivalent mass of the vehicle, R j v is the factor affecting road conditions at the vehicle location. j Let θ be the vehicle's speed. j The direction of the vehicle's velocity and the field strength E j The angle between directions, D rj Risk factors for drivers of private vehicles.

[0081] In some embodiments, the left interval for mode switching decisions corresponding to a driver's style is obtained through the following method:

[0082] For the driving data of the current driver style category mode switching process, the driving risk assessment method based on the risk field model is used to calculate the driving risk value of each driver in the driving data when switching driving modes. The minimum value is taken as the driving risk value when switching the mode of this type of driving style (Equation (5)), which is the left endpoint of the driving risk interval of the mode switching decision.

[0083]

[0084] in, This represents the left endpoint of the driving risk range for the mode switching decision of driver type A. Let represent the driving risk value of the j-th driver in class a in the dataset when switching driver modes, and min represents the minimum value function.

[0085] In some embodiments, the specific implementation process for obtaining the future trajectories of the vehicle and surrounding vehicles based on the EA-Net trajectory prediction model of the environmental attention network is as follows:

[0086] A training dataset was constructed based on the historical time-domain information of the vehicle and surrounding vehicles in the experimental data, including lateral position, longitudinal position, lateral velocity, longitudinal velocity, lateral acceleration, longitudinal acceleration, and heading angle relative to the lane line. The interaction characteristics between vehicles were analyzed, and a trajectory prediction model based on the EA-Net environmental attention network (Equation (6)) was established. The trajectory prediction model was used to predict the trajectory of the vehicle and eight surrounding vehicles in real time. The trajectory prediction input was the past Tl The historical time-domain information of the vehicle and surrounding vehicles is used to output the future T. p Predicted trajectory data of duration;

[0087]

[0088] in, T represents the predicted future trajectory of the vehicle. l T represents the length of the historical time domain. g T represents the interval between the historical time domain and the future time domain. p To predict the time domain length in the future, f (t) ={x (t) ,y (t)}, f (t) Let x be the position coordinate at time t. (t) Let y be the x-coordinate of the position at time t. (t) Let t be the ordinate of the position at time t; The input feature matrix is ​​constructed using historical time-domain information of the vehicle to be predicted and vehicles in the surrounding environment. A is the spatiotemporal correlation matrix, according to W i With W j Established based on the interaction and influence at a certain moment. This indicates whether there is an interaction between surrounding vehicle i and vehicle j at a certain moment. If there is an interaction between surrounding vehicle i and vehicle j at a certain moment, then... If there is no interaction between surrounding vehicle i and vehicle j at any time, then W i This represents the historical time-domain information of the vehicle to be predicted and its surrounding environment. i = 0, 1, 2, ..., n; the feature information of the i-th vehicle at time t is represented as: x (t) ,y (t) , θ (t) These represent the lateral position, longitudinal position, lateral velocity, longitudinal velocity, lateral acceleration, longitudinal acceleration, and heading angle relative to the lane line at time t, respectively.

[0089] In some embodiments, the right interval for mode switching decisions is determined as follows: based on trajectory prediction results, the future position coordinates and velocities of the vehicle and eight surrounding vehicles are obtained; using a driving risk assessment method based on a risk field model, the future T-wave distance of the vehicle is calculated. p The driving risk at each moment is used to derive a curve showing the future driving risk change. The driving risk value at the point where the vehicle collides with the surrounding vehicles in the future is used as the right endpoint of the driving risk range for mode switching decisions.

[0090] S3: Optimize the timing of mode switching decisions based on driving risks and extreme energy consumption.

[0091] In this embodiment, the time interval for the driver to switch driving modes is obtained by combining the future risk curve and the risk interval for mode switching decision. The speed prediction sequence of the driver-driving mode is obtained based on the speed prediction and mode switching time interval constraint. Based on the speed prediction and mode switching time interval constraint, a mode switching optimization model based on limit energy consumption is established to solve for the optimal switching time.

[0092] In some embodiments, such as Figure 4 The diagram shown illustrates the generation of the driving mode switching decision time interval according to an embodiment of this application. The time interval for the driver to switch driving modes is obtained using the following method:

[0093] Driving risk range based on driver mode switching decision The curve of future driving risk changes determines the time range for drivers to make decision recommendations on switching driving modes. in, This represents the left endpoint of the driving risk range for the mode switching decision of driver type A. This represents the right endpoint of the driving risk range for the mode switching decision of driver type A. This indicates the earliest recommended time to switch driving modes. This indicates the latest recommended time to switch driving modes.

[0094] In some embodiments, such as Figure 5 The diagram shown illustrates the human-driving mode speed prediction after the mode switching timing provided in this embodiment. The human-driving mode speed prediction sequence is obtained through a human-driving mode speed prediction model based on an LSTM recurrent neural network, specifically including:

[0095] By analyzing the characteristics of vehicle speed change before and after driving mode switching for different driving styles under different driving scenarios, a human-driving mode speed prediction model based on LSTM recurrent neural network is established (Equation (7)). The model input is driving style and time length is T. s The intelligent driving mode speed time series and driving risk time series are output with a time length of T. e Predicted sequence of speed changes in human driving mode;

[0096] V hu =w*LSTM(V in ,R,S,h)+b (7)

[0097] in, This represents the output predicted velocity sequence. Representing the future T e Predicted velocity at different times within a time period, The planned speed sequence for vehicles in intelligent driving mode. These represent the vehicle's historical T data in intelligent driving mode. s The planning speed at different times within the time frame. As a driving risk sequence, Representing history T respectively s Driving risk at different times within a time interval, where S represents driver style, h represents hidden layer state, w and b represent the weights and biases of the fully connected layer, and the time interval T represents driving risk at different times. s +T e ≥t f -t0, LSTM represents the Long Short-Term Memory network structure.

[0098] In some embodiments, a mode switching optimization model based on limit energy consumption is established (Equation (8)) according to speed prediction, mode switching time interval constraints and fitted energy consumption calculation model, and the model is solved to determine the driver's optimal switching time T. opt .

[0099]

[0100]

[0101] Where T represents the timing of the decision variable switching, and v in (t) represents the speed of the intelligent driving mode at time t, v hu v(t) represents the predicted speed of the driver mode at time t, where t is the time, v is the vehicle speed, v(t) is the vehicle speed at time t, a is the driver mode switching decision category, denergy is the instantaneous energy consumption calculation function obtained by fitting, and J is the energy consumption target value.

[0102] The above embodiments are only used to illustrate this application and are not intended to limit this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this application. Therefore, all equivalent technical solutions also fall within the scope of this application, and the patent protection scope of this application should be defined by the claims.

Claims

1. A method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, characterized in that... The method includes; Based on the driving characteristic data of drivers in different driving scenarios, the drivers are classified using a clustering algorithm, and the driver style recognition model based on deep learning is used to identify the driver style, thereby obtaining driving data on the mode switching process of the current driver style category. Based on the driving data of the current driver style category during mode switching, the driving risk value during mode switching is calculated based on the risk field model, and the left interval of the mode switching decision corresponding to the driver style is obtained based on the driving risk value during mode switching. The future trajectories of the vehicle and surrounding vehicles are obtained based on the EA-Net trajectory prediction model of the environmental attention network. The future driving risk change curve is obtained by using the driving risk assessment method based on the risk field model. The driving risk value at the collision location is used as the right interval for the mode switching decision. Based on the combination of future risk curves and mode switching decision risk intervals, the time interval for drivers to switch driving modes is obtained. The speed prediction model of driver-driver mode based on LSTM recurrent neural network is used to obtain the speed change prediction sequence of driver-driver mode. Based on the speed prediction and mode switching time interval constraints, a mode switching optimization model based on limit energy consumption is established, and the optimal switching time is obtained by solving the problem.

2. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 1, is characterized in that... The different driving scenarios include cruising, following other vehicles, passing through intersections, and queuing at intersections. The driving characteristic data of the driver under different driving scenarios includes accelerator pedal opening, steering wheel angle, steering wheel speed, vehicle longitudinal speed, lateral speed, longitudinal acceleration, lateral acceleration, vehicle roll angle and heading angle. The average, maximum and variance of the absolute values ​​of the driver's driving characteristic data under different driving scenarios are used as the driver style characteristic index.

3. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 2, is characterized in that... The classification of drivers using clustering algorithms includes: Based on driver style characteristic indicators, dimensionality reduction was performed using factor analysis to obtain two factors: a horizontal factor and a vertical factor. Based on the aforementioned horizontal and vertical factors, a clustering algorithm is used to classify drivers into three categories: aggressive, average, and conservative.

4. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 3, is characterized in that... The deep learning-based driver style recognition model identifies driver style and obtains driving data on the mode switching process of the current driver style category, including: A learning sample dataset is constructed based on classification features and category data. A deep learning-based driver style recognition model is then established. Based on this model, the processed driver feature index data is used as input to obtain the driver style recognition result. The recognition model is represented as follows: Where X is an input vector containing driver characteristic index data, X = {x1, x2}, where x1 represents the horizontal factor of driver characteristics, x2 represents the vertical factor of driver characteristics, and W i and b i is the weight matrix and bias of the i-th layer of the multilayer perceptron, n is the number of layers of the multilayer perceptron, and Y is the driver style category output; Based on the clustering results, we analyze the driving data of drivers with three different driving styles during the driving mode switching process in four driving scenarios. This includes the vehicle's position coordinates, vehicle speed, ambient vehicle position coordinates, and ambient vehicle speed. We then use the driver style recognition results to obtain the driving dataset N representing the current driver style category during the mode switching process. a , is represented as; in To identify the driving data during the mode switching process of the j-th driver in driver style a, For the driving data of the i-th environment vehicle, Let x, y, and y represent the x-coordinate, y-coordinate, and speed of vehicle i, respectively, and let I be the set of vehicle IDs for the surrounding environment. For the driving data of the j-th driver, Let x, y, and y be the x-coordinate of the vehicle j, and y be the vehicle speed.

5. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 4, is characterized in that... The driving risk value during mode switching, calculated based on the driving data during the current driver style category and the risk field model, includes: Based on the influence of Z environmental vehicles covering all directions around the vehicle, the field strength E of the driving risk field generated by the surrounding moving vehicles at the vehicle's location is calculated. j , the calculation formula is: Among them, E V_ij For surrounding vehicles i(x) i ,y i ) in the vehicle j(x j ,y j The resulting risk field strength, (x) i ,y j ), (x j ,y j Let ) represent the coordinates of the vehicle's center of mass, where k1, k2, and G are constants greater than 0, and R... i M represents the road condition influencing factor at location i for surrounding vehicles. i Given the equivalent mass of surrounding vehicle i, r ij =(x j -x i ,y j -y i ), representing the distance vector between surrounding vehicle i and vehicle j, with the direction being the same as the field strength direction, v i Let θ be the speed of the surrounding vehicle i. i Let r be the velocity direction of the surrounding vehicle i and the distance vector r. ij The angle between them, where exp is an exponential function with the natural constant e as its base; The force acting on the vehicle is calculated using the formula for force in an electric field, and this force is used as the vehicle's driving risk. j The formula for calculating the force in the field strength is expressed as follows: risk j =E j M j R j exp[-k2v j cos(θ j )](1+Dr j ) (4) Among them, M j For the equivalent mass of the vehicle, R j v is the factor affecting road conditions at the vehicle location. j Let θ be the vehicle's speed. j The direction of the vehicle's velocity and the field strength E j The angle between directions, D rj Risk factors for drivers of private vehicles.

6. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 5, is characterized in that... The left interval for determining the mode switching decision based on the driving risk value during mode switching, corresponding to the driver's style, includes: Based on driving data of the current driver style category during mode switching, a driving risk assessment method based on a risk field model is used to calculate the driving risk value of each driver in the driving dataset when switching driving modes. The minimum value is taken as the driving risk value of this type of driver style when switching modes, and it is used as the left endpoint of the driving risk interval for mode switching decisions. The calculation formula is as follows: in, This represents the left endpoint of the driving risk range for the mode switching decision of driver type A. Let represent the driving risk value of the j-th driver in class a in the dataset when switching driver modes, and min represents the minimum value function.

7. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 6, is characterized in that... The trajectory prediction model based on the Environmental Attention Network (EA-Net) obtains the future trajectories of the vehicle and surrounding vehicles. A driving risk assessment method based on a risk field model is used to derive the future driving risk change curve. The driving risk value at the collision location is used as the right interval for the mode switching decision, including: A training dataset was constructed based on historical time-domain information of the vehicle and surrounding vehicles from the experimental data. This dataset includes lateral position, longitudinal position, lateral velocity, longitudinal velocity, lateral acceleration, longitudinal acceleration, and heading angle relative to the lane lines. Inter-vehicle interaction characteristics were analyzed, and a trajectory prediction model based on the Environmental Attention Network (EA-Net) was established. This EA-Net trajectory prediction model was then used to predict the trajectories of the vehicle and multiple surrounding vehicles in real time. The trajectory prediction input was the past T... l The historical time-domain information of the vehicle and surrounding vehicles is used to output the future T. p The predicted trajectory data is of a certain duration; the trajectory prediction model based on the EA-Net environmental attention network is represented as follows: in, T represents the predicted future trajectory of the vehicle. l T represents the length of the historical time domain. g T represents the interval between the historical time domain and the future time domain. p To predict the time domain length in the future, f (t) ={x (t) ,y (t) }, f (t) Let x be the position coordinate at time t. (t) Let y be the x-coordinate of the position at time t. (t) Let t be the ordinate of the position at time t; The input feature matrix is ​​constructed using historical time-domain information of the vehicle to be predicted and vehicles in the surrounding environment. A is the spatiotemporal correlation matrix, according to W i With W j Established based on the interaction and influence at a certain moment. This indicates whether there is an interaction between surrounding vehicle i and vehicle j at a certain moment. If there is an interaction between surrounding vehicle i and vehicle j at a certain moment, then... If there is no interaction between surrounding vehicle i and vehicle j at any time, then W i This represents the historical time-domain information of the vehicle to be predicted and its surrounding environment. The feature information of the i-th vehicle at time t is represented as follows: x (t) ,y (t) , θ (t) Let represent the lateral position, longitudinal position, lateral velocity, longitudinal velocity, lateral acceleration, longitudinal acceleration, and heading angle relative to the lane line at time t, respectively. Based on the trajectory prediction results, the future position coordinates and velocities of the vehicle and multiple surrounding vehicles are obtained. Then, using a driving risk assessment method based on a risk field model, the future T-wave velocity of the vehicle is calculated. p The driving risk at each moment is used to derive the future driving risk change curve, and the driving risk value at the location where the vehicle will collide with the surrounding vehicles in the future is used as the right endpoint of the driving risk interval for mode switching decisions.

8. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 1, is characterized in that... Based on the combination of future risk curves and mode-switching decision risk intervals, the time intervals for drivers to switch driving modes are obtained, including: Driving risk range based on driver mode switching decision The curve of future driving risk changes determines the time range for drivers to make decision recommendations on switching driving modes. in, This represents the left endpoint of the driving risk range for the mode switching decision of driver type A. This represents the right endpoint of the driving risk range for the mode switching decision of driver type A. This indicates the earliest recommended time to switch driving modes. This indicates the latest recommended time to switch driving modes.

9. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 8, is characterized in that... The human-driving mode speed prediction model based on LSTM recurrent neural network obtains a human-driving mode speed change prediction sequence, including: Based on the characteristics of vehicle speed changes before and after driving mode switching under different driving scenarios and different driver styles, a human-driving mode speed prediction model based on LSTM recurrent neural network is established. The model input is the driver style and the time length is T. s The intelligent driving mode speed time series and driving risk time series are output with a time length of T. e The human-driving mode speed change prediction sequence, the human-driving mode speed prediction model based on LSTM recurrent neural network is expressed as follows: V hu =w*LSTM(V in ,R,S,h)+b (7) in, This represents the output predicted velocity sequence. Representing the future T e Predicted velocity at different times within a time period, The planned speed sequence for vehicles in intelligent driving mode. These represent the vehicle's historical T data in intelligent driving mode. s The planning speed at different times within the time frame. As a driving risk sequence, Representing history T respectively s Driving risk at different times within a time interval, where S represents driver style, h represents hidden layer state, w and b represent the weights and biases of the fully connected layer, and the time interval T represents driving risk at different times. s +T e ≥t f -t0, LSTM represents the Long Short-Term Memory network structure.

10. The method for switching between intelligent driving and human driving modes for urban logistics vehicles, considering driving risks and energy consumption, as described in claim 9, is characterized in that... The mode switching optimization model based on extreme energy consumption is expressed as follows: Where T represents the timing of the decision variable switching, and v in (t) represents the speed of the intelligent driving mode at time t, v hu v(t) represents the predicted speed of the driver mode at time t, where t is the time, v is the vehicle speed, v(t) is the vehicle speed at time t, a is the driver mode switching decision category, denergy is the fitted instantaneous energy consumption calculation function, and J is the energy consumption target value.

Images