A multi-objective optimization method and system for tunnel low-carbon driving behavior

Abstract

The application provides a tunnel low-carbon driving behavior multi-objective optimization method and system. In view of the unique scene characteristics of the urban road tunnel, such as the closed space, the sharp light-dark transition and the complex line shape, relying on the multi-source driving data synchronous collection and fusion technology, a three-dimensional evaluation index system of safety-visual comfort-energy consumption special for tunnels is constructed. Through an unsupervised clustering algorithm, the fine portrait classification of drivers is completed. Taking the time headway (THW) as the core control parameter, combining a high-adaptability machine learning prediction model and a multi-objective constraint optimization algorithm, the individualized and targeted driving behavior optimization for different types of drivers is realized, and finally the collaborative optimization goals of improving the tunnel driving safety, improving the driving visual comfort and reducing the vehicle driving energy consumption are achieved.

Description

Technical Field

[0001] This invention relates to the technical field of intelligent traffic control, specifically to a multi-objective optimization method and system for low-carbon driving behavior in tunnels. Background Technology

[0002] In recent years, the number and total mileage of tunnels in China have shown rapid growth, with the proportion of urban road tunnels continuing to increase. With the rapid development of technologies such as intelligent connected vehicles and vehicle-road cooperation, tunnel traffic control has evolved from traditional passive safety management to active driving behavior guidance and personalized adaptation optimization. At the same time, the low-carbon transformation of urban transportation has become a core trend in industry development, and ecological driving and energy consumption optimization in tunnel scenarios have become important research directions in the field of tunnel traffic control.

[0003] Urban road tunnels possess unique characteristics such as enclosed spaces, abrupt transitions between light and dark at entrances and exits, complex alignments, dense traffic flow, and mixed vehicle types. Compared to open roads, drivers' visual perception, spatial judgment, and following behavior significantly deviate within tunnels. Safety, comfort, and energy consumption are the three core objectives of tunnel traffic operation and driving behavior management. These three aspects are interdependent and dynamically balanced, jointly determining the operational quality and overall benefits of tunnel traffic. Existing technologies struggle to achieve synergistic optimization of these three aspects. Optimizing driving behavior in tunnel scenarios faces core challenges, including safety management failing to consider comfort and energy consumption, comfort optimization focusing solely on passive adjustments to the external environment, energy consumption optimization not adapting to the characteristics of the tunnel environment, and the lack of multi-objective collaborative mechanisms. The deviated driving behavior caused by the tunnel environment exacerbates driver visual fatigue, increases driving safety risks, and significantly increases vehicle energy consumption. Therefore, conducting multi-objective synergistic optimization of driving behavior in urban tunnel scenarios has significant engineering and application value for improving tunnel operational safety, enhancing driver travel experience, and promoting low-carbon transformation in urban transportation.

[0004] Currently, the following technical means are mainly used for driving behavior optimization and traffic control in tunnel scenarios: (1) Tunnel driving safety optimization technology: This type of technology takes car-following risk prevention and control as its core. Most of them are based on classic car-following models such as headway and safety margin. They collect car-following data through millimeter-wave radar and vehicle-road cooperative equipment to realize real-time early warning of driving risks and speed control in tunnels. Some technologies regulate vehicle driving behavior in tunnels through tunnel alignment optimization and speed limit sign setting. This type of method only takes safety as the single optimization goal and does not consider the linkage effect of driving behavior adjustment on comfort and energy consumption. It is easy to sacrifice energy conservation and comfort for safety. At the same time, the generalized safety control strategy is adopted without combining the individual behavioral differences and profile characteristics of drivers in tunnel scenarios. It is impossible to achieve personalized and targeted safety optimization. Moreover, it can only achieve passive risk warning and does not reduce the risk of car-following in tunnels from the root through active driving behavior guidance. The optimization effect is limited. (2) Tunnel driving comfort optimization technology: This type of technology is mainly divided into two categories. One category is to improve the driving visual environment in the tunnel through intelligent dimming of the tunnel lighting system, deployment of visual guidance facilities, and optimization of the spatial environment. The other category is to improve the driving comfort of the vehicle by optimizing the smoothness of acceleration and deceleration based on vehicle dynamic parameters. Most of these methods focus on the passive optimization of the external environment of the tunnel and the vehicle hardware, without improving the driver's visual comfort experience from the perspective of active control of driving behavior. The optimization dimension is singular. At the same time, it is impossible to establish a precise mapping relationship between driving behavior and comfort experience. The optimization effect lacks quantitative support, and comfort indicators are not included in the same optimization system as safety and energy consumption indicators, so it is impossible to achieve the coordinated adaptation of multiple objectives. (3) Tunnel ecological driving and energy consumption optimization technology: This type of technology focuses on reducing vehicle driving energy consumption. By constructing vehicle speed guidance models and fuel consumption prediction models, it provides drivers with economical vehicle speed and smooth acceleration and deceleration strategies in the tunnel. Some technologies combine tunnel slope and linear characteristics to optimize the speed planning of autonomous vehicles to achieve ecological driving. Most of these methods take reducing energy consumption as the optimization goal, without fully considering the hard constraints of driving safety in tunnel scenarios. Some aggressive energy-saving speed strategies may increase the risk of rear-end collisions. At the same time, they do not take into account the visual comfort needs of drivers in tunnel scenarios. The optimized driving behavior may increase the driver's visual load and reduce driving comfort. Moreover, the fuel consumption prediction model does not integrate the scene-specific features such as tunnel illumination, slope, and following status. The prediction accuracy and generalization ability of the model in tunnel scenarios are insufficient. (4) Multi-objective collaborative optimization technology for driving behavior: This type of technology is designed for open road scenarios. It constructs multi-dimensional optimization objective functions such as safety, energy consumption, and smoothness. It solves the optimal driving strategy through model predictive control and intelligent algorithms to achieve multi-objective collaborative optimization. This type of method does not design an optimization framework for the unique environmental characteristics of urban tunnel scenarios. It cannot adapt to the special requirements of driving behavior caused by tunnel light-dark transitions, enclosed spaces, and slope changes. The scene adaptability is poor.Meanwhile, the core visual comfort indicators of the tunnel scenario were not included in the optimization system, the optimization objectives were incomplete, and the coordination of safety, comfort and energy consumption in the tunnel scenario could not be achieved; also, the optimization was not differentiated based on driver characteristics, and a general optimization strategy was adopted, resulting in inconsistent optimization effects for different types of drivers.

[0005] In summary, existing tunnel driving behavior optimization technologies have limitations in terms of optimization target integrity, scenario adaptability, targeted optimization capability, and multi-objective collaboration. There is an urgent need for a multi-objective optimization method and system for low-carbon driving behavior in tunnels that can achieve long-term, continuous, accurate, wide-area adaptability, high reliability, and easy implementation, so as to achieve a synergistic improvement in driving safety, driving comfort, and low-carbon energy consumption in urban tunnel scenarios. Summary of the Invention

[0006] This invention is made to solve the above-mentioned problems, and aims to provide a multi-objective optimization method and system for low-carbon driving behavior in tunnels.

[0007] This invention provides a multi-objective optimization method for low-carbon driving behavior in tunnels, characterized by the following steps: S1: Multi-source driving data acquisition step, collecting multi-source driving data in a tunnel scenario and aligning the multi-source driving data over time. The multi-source driving data includes vehicle dynamic data, tunnel environment data, and driver eye movement physiological data; S2: Data preprocessing and fusion step, cleaning and fusing the time-aligned multi-source driving data to obtain a standardized time-series feature dataset; S3: Three-dimensional index quantification calculation step, constructing a three-dimensional evaluation index system for tunnel scenarios based on the time-series feature dataset, and quantifying and calculating the safety dimension index based on car-following safety margin, the visual comfort index based on pupil area change rate grading, and the energy consumption dimension index based on instantaneous fuel consumption normalization; S4: Driver profile classification step, based on... Unsupervised clustering analysis is performed on the global mean data of the three-dimensional evaluation index system for each driver to divide the driver group into different types of profiles and determine the target driver type to be optimized; S5: Multi-objective optimization solution step, for the target driver type, using the headway time-of-flight (THW) as the control parameter, candidate optimization speed sequences that meet the constraints of the tunnel scenario are generated based on the THW increment. Combined with the pre-trained visual comfort prediction model and instantaneous fuel consumption prediction model, the changes in the three-dimensional evaluation index under different THW increments are predicted. Multi-objective optimization solution is performed on the candidate optimization speed sequences to obtain the optimal speed sequence that achieves the best synergy between safety, comfort and energy consumption, which serves as the driving behavior optimization strategy; S6: Optimization strategy output and guidance step, the driving behavior optimization strategy is output through the vehicle guidance and prompting device to provide personalized driving behavior guidance to the target driver to be optimized.

[0008] The multi-objective optimization method for low-carbon driving behavior in tunnels provided by this invention may also have the following features: the method for cleaning and fusing the time-aligned multi-source driving data in S2 is as follows: screening non-congested and non-over-limit driving data with vehicle speeds in the range of 20km / h-100km / h, removing outliers based on the 3σ principle, filling missing data with linear interpolation, and uniformly resampling multi-source driving data with different sampling frequencies to the same frequency to complete spatiotemporal alignment and feature fusion.

[0009] The multi-objective optimization method for low-carbon driving behavior in tunnels provided by this invention may also have the following features: The quantitative calculation method for the safety dimension index in S3 is as follows: The following state is determined by a continuous detection time of ≥15s for the preceding vehicle and THW < 7s. The following safety level is quantified using the Safety Margin SM model, where the value of SM ranges from 0 to 1; a higher value indicates a lower driving risk.

[0010] The headway is obtained by dividing the relative distance between the vehicle in front and the vehicle ahead by the vehicle's speed, and is expressed as:

[0011] (1)

[0012] In the formula, The following distance (m) is the following distance. The vehicle speed is (m / s).

[0013] The formula for calculating SM at time t is expressed as:

[0014] (2)

[0015] In the formula, For the vehicle's speed, To maintain following distance, The speed of the vehicle in front. This is the acceleration due to gravity.

[0016] The multi-objective optimization method for low-carbon driving behavior in tunnels provided by this invention may also have the following feature: The quantitative calculation method for the visual comfort index in S3 is as follows: calculate the driver's pupil area change rate, and perform visual comfort grading and assignment based on a preset threshold for the pupil area change rate. The assignment range is 0~1, with higher values ​​representing better visual comfort.

[0017] The rate of change of pupil area is calculated from pupil area data at consecutive time points and is expressed as:

[0018] (3)

[0019] In the formula, Let be the rate of change of pupil area at time i (%). is the pupil area of the driver at the i-th moment (mm 2 ). is the pupil area of the driver at the (i - 1)-th moment (mm 2 ). is the average pupil area of the driver in the tunnel section (mm 2 ).

[0020] In the multi-objective optimization method for low-carbon driving behavior in tunnels provided by the present invention, it can also have the following characteristics: Among them, the visual comfort level classification and assignment follow the following rules: When U ≤ 6.62%, it indicates that the driver's visual adaptation state is the best, which is defined as the "extremely comfortable" level and assigned H = 1; when 6.62% < U ≤ 19.74%, it indicates that the driver has a good visual experience, which is defined as the "comfortable" level and assigned H = 0.7; when 19.74% < U ≤ 41.51%, it indicates that the driver has obvious visual fatigue or discomfort, which is defined as the "uncomfortable" level and assigned H = 0.3; when U > 41.51%, it indicates that the driver's vision bears a large load, which is defined as the "extremely uncomfortable" level and assigned H = 0.

[0021] In the multi-objective optimization method for low-carbon driving behavior in tunnels provided by the present invention, it can also have the following characteristics: Among them, the quantization calculation method of the energy consumption dimension index in S3 is: Perform maximum-minimum normalization processing on the instantaneous fuel consumption data of the vehicle to obtain the energy consumption dimension index in the range of 0 - 1. The higher the value, the higher the energy consumption of the vehicle during driving, which is expressed as:

[0022] (4)

[0023] In the formula, is the instantaneous fuel consumption value of the driver at time t (L / 100km), and are respectively the maximum and minimum instantaneous fuel consumption values (L / 100km) in all driving data.

[0024] In the multi-objective optimization method for low-carbon driving behavior in tunnels provided by the present invention, it can also have the following characteristics: Among them, the generation method of the candidate optimization speed sequence in S5 is: Divide the tunnel into segments at a preset interval, set the iteration range and step size of the THW increment, and generate the initial optimization speed of each segment based on the following distance and the THW increment, which is expressed as:

[0025] (5)

[0026] In the formula,<{ is the following distance, is the THW increment, is the time headway before optimization For initial speed optimization,

[0027] Segmented velocity fluctuation constraints and smoothing constraints on velocity changes in adjacent time steps are applied to the initial optimized velocity to complete the velocity sequence calibration; the tunnel scene features are mapped to the corresponding mileage points of the calibrated velocity sequence to obtain the candidate optimized velocity sequence.

[0028] The multi-objective optimization method for low-carbon driving behavior in tunnels provided by this invention may also have the following features: In S5, the visual comfort prediction model is an XGBoost model, with inputs including vehicle speed, tunnel illumination, and pupil area change rate at a previous preset time step, and output being the pupil area change rate at the current moment. The XGBoost model completes hyperparameter optimization through temporal cross-validation and grid search. The instantaneous fuel consumption prediction model is an integrated stack of the XGBoost model and the random forest model, with inputs including weighted composite acceleration, vehicle speed-slope coupling features, and instantaneous fuel consumption value at a previous preset time step, and output being the instantaneous fuel consumption value at the current moment. The XGBoost model in the instantaneous fuel consumption prediction model completes hyperparameter optimization through Bayesian optimization.

[0029] The multi-objective optimization method for low-carbon driving behavior in tunnels provided by this invention may also have the following features: In S4, the unsupervised clustering analysis uses the k-means unsupervised clustering algorithm. Different driver profiles include: safety-oriented, efficiency-comfort, and passive-conservative. The passive-conservative type is identified as the target driver type to be optimized. In S5, the driving behavior optimization strategy is obtained by constructing a multi-objective optimization objective function J, while setting constraints a>0, b>0, and c<0, expressed as:

[0030] (10)

[0031] In the formula, a represents the average increment of the safety dimension index, b represents the average increment of the visual comfort index, and c represents the average increment of the energy consumption dimension index.

[0032] Iterate through all THW increments, select the THW increments that satisfy the constraints and maximize the objective function J as the optimal control parameters, and obtain the corresponding collaborative optimal speed sequence as the driving behavior optimization strategy.

[0033] This invention also provides a multi-objective optimization system for low-carbon driving behavior in tunnels, characterized by: a multi-source driving data acquisition module, which acquires multi-source driving data in a tunnel scenario and performs time alignment on the multi-source driving data, including vehicle dynamic data, tunnel environment data, and driver eye movement physiological data;

[0034] The data preprocessing and fusion module cleans and fuses the time-aligned multi-source driving data to obtain a standardized time-series feature dataset.

[0035] The three-dimensional index quantification and calculation module constructs a three-dimensional evaluation index system for tunnel scenarios based on a time-series feature dataset. It quantifies and calculates safety dimension indicators based on car-following safety margin, visual comfort indicators based on pupil area change rate grading, and energy consumption dimension indicators based on instantaneous fuel consumption normalization. The driver profile classification module performs unsupervised clustering analysis based on the global mean data of each driver's three-dimensional evaluation index system, classifying the driver group into different types of profiles and identifying the target driver type to be optimized. The multi-objective optimization solution module, for the target driver type, uses the time-to-head distance (THW) as a control parameter. Based on the THW increment, it generates candidate optimization speed sequences that meet the constraints of the tunnel scenario. Combining a pre-trained visual comfort prediction model and an instantaneous fuel consumption prediction model, it predicts the changes in three-dimensional evaluation indicators under different THW increments, and performs multi-objective optimization on the candidate optimization speed sequences to obtain the optimal speed sequence that achieves the best synergy between safety, comfort, and energy consumption, serving as a driving behavior optimization strategy. The optimization strategy output and guidance module outputs the driving behavior optimization strategy through a vehicle guidance and prompting device, providing personalized driving behavior guidance to the target driver.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] This invention addresses the shortcomings of existing tunnel driving behavior optimization technologies, such as single optimization objectives, poor scenario adaptability, lack of multi-objective collaborative mechanisms, and insufficient targeted optimization capabilities. It proposes a multi-objective optimization method and system for low-carbon driving behavior in tunnels. This invention constructs a tunnel-specific three-dimensional evaluation index system encompassing safety, visual comfort, and energy consumption, and completes refined driver profile classification. It uses vehicle headway as the core control parameter to achieve multi-objective collaborative optimization. The driving behavior optimization strategy obtained by this invention is adapted to the unique scenario characteristics of urban tunnels, such as enclosed spaces and transitions between light and dark areas. It balances the three core objectives of driving safety, driving comfort, and low-carbon energy conservation, enabling personalized targeted optimization for different types of drivers. The method has clear logic and strong system feasibility, effectively solving problems such as high safety risks, poor visual comfort, and high driving energy consumption caused by distorted tunnel driving behavior. It provides core technical support for intelligent traffic management and low-carbon operation in tunnels. Attached Figure Description

[0038] Figure 1 This is a flowchart of a multi-objective optimization method for low-carbon driving behavior in tunnels, as described in an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram illustrating the construction logic of the three-dimensional evaluation index system of safety, visual comfort, and energy consumption in an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of the optimized speed sequence generation process based on THW incremental iteration in an embodiment of the present invention.

[0041] Figure 4 This is a schematic diagram comparing the three-dimensional evaluation index of a passive and conservative driver before and after optimization in an embodiment of the present invention.

[0042] Figure 5 This is a schematic diagram of the architecture of the multi-objective optimization system for low-carbon driving behavior in tunnels, as described in an embodiment of the present invention. Detailed Implementation

[0043] To make the technical means, creative features, objectives and effects of the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate a multi-objective optimization method and system for low-carbon driving behavior in tunnels according to the present invention.

[0044] This embodiment provides a multi-objective optimization method for low-carbon driving behavior in tunnels, including the following steps:

[0045] Figure 1 This is a flowchart of a multi-objective optimization method for low-carbon driving behavior in tunnels, as described in an embodiment of the present invention.

[0046] like Figure 1 As shown, step S1 is the multi-source driving data acquisition step, which collects multi-source driving data in an urban tunnel scenario and performs time alignment on the multi-source driving data. Specifically:

[0047] Urban road tunnels were selected as the experimental setting. Drivers of different genders, ages, driving experience, and occupations were recruited to conduct natural driving experiments. Three types of core data were collected synchronously through onboard equipment: first, vehicle dynamic data, including vehicle speed, lateral / longitudinal acceleration, pitch angle, instantaneous fuel consumption, and relative distance and speed to the vehicle in front; second, tunnel environmental data, including real-time illumination, road slope, and alignment parameters; and third, driver eye movement physiological data, including real-time pupil area, fixation, and saccades. All collected data were synchronized and aligned at the millisecond level based on hardware CPU timestamps, providing raw data support for subsequent analysis.

[0048] Step S2 is the data preprocessing and fusion step, which cleans and fuses the time-aligned multi-source driving data to obtain a standardized time-series feature dataset, specifically:

[0049] The time-aligned multi-source driving data is standardized, cleaned, and fused: First, valid data is selected, retaining non-congested and non-overspeed driving data within the vehicle speed range of 20km / h-100km / h; second, outliers are removed based on the 3σ principle, and linear interpolation is used to fill in a small number of consecutive missing data to ensure data integrity; finally, sensor data with different sampling frequencies are uniformly resampled to 2Hz to complete the spatiotemporal alignment and feature fusion of multi-source driving data, resulting in a standardized time-series feature dataset.

[0050] Figure 2 This is a schematic diagram illustrating the construction logic of the three-dimensional evaluation index system of safety, visual comfort, and energy consumption in an embodiment of the present invention.

[0051] Step S3 is the three-dimensional index quantification calculation step, such as... Figure 2 As shown, a three-dimensional evaluation index system for tunnel scenarios is constructed based on a time-series feature dataset. This system quantifies and calculates safety dimension indicators based on car-following safety margin, visual comfort indicators based on pupil area change rate grading, and energy consumption dimension indicators based on instantaneous fuel consumption normalization. Specifically:

[0052] Based on the characteristics of urban tunnel scenarios, a three-dimensional evaluation index system specifically for urban tunnels—safety, visual comfort, and energy consumption—is constructed, and the above three dimensions are quantitatively evaluated:

[0053] (1) Safety dimension index SM: Based on the car-following theory, the dual rule of "continuous detection time of the preceding vehicle ≥ 15s, THW < 7s" is first used as the car-following state judgment condition, and then the safety margin SM model is used to quantify the car-following safety level. The value range of SM is 0~1, and the higher the value, the lower the driving risk.

[0054] The headway can be obtained by dividing the relative distance between the vehicle in front and the vehicle itself by the vehicle's speed, expressed as:

[0055] (1)

[0056] In the formula, The following distance (m) is the following distance. The vehicle speed is (m / s).

[0057] The formula for calculating SM at time t can be expressed as:

[0058] (2)

[0059] In the formula, For the vehicle's speed, To maintain following distance, The speed of the vehicle in front. This is the acceleration due to gravity.

[0060] (2) Visual comfort index H: Focusing on the core visual comfort in urban tunnel scenarios, the change rate of pupil area is used to quantify the driver's visual load. Calculate the change rate of the driver's pupil area, and based on the change rate of pupil area, complete the 4-level visual comfort classification and standard assignment. The assignment range is 0-1, and the higher the value, the better the visual comfort.

[0061] The change rate of pupil area is calculated from the pupil area data at consecutive moments and is expressed as:

[0062] (3)

[0063] In the formula, is the change rate of pupil area at the i-th moment (%), is the pupil area of the driver at the i-th moment (mm 2 ), is the pupil area of the driver at the (i-1)-th moment (mm 2 ), is the average pupil area of the driver in the tunnel section (mm 2 ).

[0064] The 4-level visual comfort classification and assignment follow the following rules: When U≤6.62%, it means that the driver's visual adaptation state is the best, which is defined as the "extremely comfortable" classification and is assigned H = 1; when 6.62% < U≤19.74%, it means that the driver has a good visual experience, which is defined as the "comfortable" classification and is assigned H = 0.7; when 19.74% < U≤41.51%, it means that the driver shows obvious visual fatigue or discomfort, which is defined as the "uncomfortable" classification and is assigned H = 0.3; when U>41.51%, it means that the driver's vision bears a large load, which is defined as the "extremely uncomfortable" classification and is assigned H = 0.

[0065] (3) Energy consumption dimension index E: Perform maximum-minimum normalization on the vehicle's instantaneous fuel consumption data to obtain the standard energy consumption dimension index in the range of 0-1. The higher the value, the higher the vehicle's driving energy consumption, which is expressed as:

[0066] (4)

[0067] In the formula, is the instantaneous fuel consumption value of the driver at time t (L / 100km), and are the maximum and minimum instantaneous fuel consumption values (L / 100km) in all driving data respectively.

[0068] After the calculation is completed, the time-series data and global mean data of the three-dimensional evaluation index system for each driver's full-tunnel journey are finally obtained.

[0069] Step S4 is the driver profile classification step. Based on the global mean data of each driver's three-dimensional evaluation index system, unsupervised cluster analysis is performed to divide the driver group into different profile types and determine the target driver type to be optimized. Specifically:

[0070] Using the k-means unsupervised clustering algorithm, the global mean data of the existing three-dimensional evaluation index system of safety-visual comfort-energy consumption for each driver are standardized and clustered. The driving group is divided into three typical profiles: the first type is safety-oriented, characterized by high safety, moderate comfort, and high energy consumption; the second type is high-efficiency comfort, characterized by above-average safety, high comfort, and low energy consumption, which is the ideal driving mode; the third type is passive and conservative, characterized by low safety and comfort, and high energy consumption, which is the core target of driving behavior optimization, i.e., the target driver type to be optimized.

[0071] Step S5 is the multi-objective optimization solution step. For the target driver type, the headway THW is used as the core control parameter to construct a multi-objective driving behavior optimization model in order to obtain the driving behavior optimization strategy.

[0072] The multi-objective driving behavior optimization model obtains driving behavior optimization strategies through the following sub-steps:

[0073] Figure 3 This is a schematic diagram of the optimized speed sequence generation process based on THW incremental iteration in an embodiment of the present invention.

[0074] S5-1: As Figure 3 As shown, candidate optimized velocity sequences that satisfy the constraints of the tunnel scenario are generated incrementally based on THW, specifically as follows:

[0075] An optimized speed sequence is generated and constrained by incremental iteration of the Thickness-Wheel (THW) method. The tunnel is divided into segments at 100m intervals. The THW incremental iteration range is set to (0, 1.0] with a step size of 0.1s. An initial optimized speed is generated based on the following distance and the THW increment. The initial optimized speed is then calibrated by applying a fluctuation limit of ±5km / h for segmented speeds and a smoothing constraint of ≤0.5km / h for speed changes between adjacent time steps. Finally, scene features such as tunnel slope, illumination, and following distance data are mapped to the corresponding mileage points of the optimized speed sequence, achieving accurate matching between scene features and the speed sequence, resulting in a candidate optimized speed sequence. The above steps are represented as follows:

[0076] (5)

[0077] (6)

[0078] (7)

[0079] In the formula, To maintain following distance, For THW increment, The headway before optimization, For initial speed optimization, The average of the initial optimized speed, The speed after fluctuation limitation The velocity after the fluctuation is limited in the i-th step. Let be the smoothed velocity at step i. For the first The speed after the step is smoothed out.

[0080] S5-2: Construct a visual comfort prediction model based on the XGBoost model. This visual comfort prediction model takes multidimensional features consisting of vehicle speed, tunnel illumination, and pupil area change rate at the previous four time steps as input, and outputs the pupil area change rate at the current time. Temporal cross-validation and grid search are used to complete hyperparameter optimization.

[0081] The trained visual comfort prediction model is used to predict the pupil area change rate under different optimized velocity sequences, and then the average increment b of the visual comfort index is calculated.

[0082] The determination coefficient R² of the visual comfort prediction model was verified on the test set and reached over 0.82, indicating that the model has high prediction accuracy and good robustness for the rate of change of driver pupil area in urban tunnel scenarios, and can provide reliable quantitative support for multi-objective optimization solutions.

[0083] The visual comfort prediction model outputs the predicted pupil change rate at time t. , represented as:

[0084] (8)

[0085] in, Let be the multidimensional input feature vector at time t.

[0086] S5-3: Construct a two-layer prediction framework based on the stacked integration of the XGBoost model and the Random Forest (RF) model, namely the instantaneous fuel consumption prediction model. This instantaneous fuel consumption prediction model takes as input a multi-dimensional feature consisting of weighted composite acceleration, vehicle speed-gradient coupling features, and instantaneous fuel consumption values ​​from the previous four time steps, and outputs the instantaneous fuel consumption value at the current moment. The XGBoost model performs hyperparameter optimization through Bayesian optimization.

[0087] The trained instantaneous fuel consumption prediction model is used to predict instantaneous fuel consumption values ​​under different optimized speed sequences, and then the average increment c of the energy consumption dimension index is calculated.

[0088] The instantaneous fuel consumption prediction model was validated on the test set, and the coefficient of determination R² reached over 0.82. This indicates that the stacked ensemble model has high prediction accuracy and good generalization ability for instantaneous fuel consumption of vehicles under urban tunnel conditions, and can provide reliable energy consumption quantification support for multi-objective optimization solutions.

[0089] The instantaneous fuel consumption prediction model outputs the instantaneous fuel consumption prediction value at time t. , represented as:

[0090] (9)

[0091] in, The XGBoost model optimized for Bayesian methods. For the random forest model, For stacking integration strategy, Let be the multidimensional input feature vector at time t.

[0092] S5-4: Optimal THW Incremental Solution under Multi-Objective Constraints. Construct a multi-objective optimization objective function J, while setting constraints a>0, b>0, and c<0, expressed as:

[0093] (10)

[0094] In the formula, a represents the average increment of the safety dimension indicator SM, b represents the average increment of the visual comfort indicator H, and c represents the average increment of the energy consumption dimension indicator E.

[0095] Iterate through all THW increments, select the THW increments that satisfy the constraints and maximize the objective function J as the optimal control parameters, and obtain the corresponding optimized speed sequence that achieves optimal safety, comfort, and energy consumption, as the driving behavior optimization strategy.

[0096] Step S6 is the optimization strategy output and guidance step. The driving behavior optimization strategy is output through the vehicle guidance and prompting device to provide personalized driving behavior guidance to the target driver. Specifically:

[0097] Personalized driving behavior optimization strategy output and vehicle-road cooperative guidance. Corresponding guidance strategies are output for different types of drivers: For passive and conservative drivers, the optimal THW increment and driving behavior optimization strategy are delivered to the driver through vehicle-road cooperative roadside equipment, in-vehicle intelligent terminals, and intelligent cockpit voice interaction systems, providing visualized and real-time following distance and speed adjustment prompts. For safety-oriented and efficiency-comfort-oriented drivers, prompts to maintain the current driving behavior are output.

[0098] Figure 4This is a schematic diagram comparing the three-dimensional evaluation index of a passive and conservative driver before and after optimization in an embodiment of the present invention.

[0099] At the same time, it can be like Figure 4 The comparison results of the three-dimensional evaluation indicators before and after optimization are simultaneously pushed to the vehicle terminal, providing drivers with an intuitive reference for behavior optimization, improving the acceptance and execution of guidance strategies, and ultimately achieving proactive and personalized optimization guidance for driving behavior in urban tunnels.

[0100] Specifically, for the core target group—passive and conservative drivers—the optimization effect of this embodiment is as follows: Figure 4 As shown, when the THW increment is 0.2s, the safety index can be improved from 0.579 to 0.597, the comfort index from 0.664 to 0.794, and the energy consumption index from 0.324 to 0.306, achieving a synergistic improvement in safety, comfort, and energy consumption, thus verifying the effectiveness of the method of the present invention.

[0101] This embodiment also provides a multi-objective optimization system for low-carbon driving behavior in tunnels. The system's modules are connected via bidirectional communication to collaboratively complete multi-objective optimization of driving behavior in tunnels, including:

[0102] Figure 5 This is a schematic diagram of the architecture of the multi-objective optimization system for low-carbon driving behavior in tunnels, as described in an embodiment of the present invention.

[0103] like Figure 5 As shown, module one, the multi-source driving data acquisition module, is used to implement step S1, namely: to acquire multi-source driving data in the tunnel scenario and to perform time alignment on the multi-source driving data. The multi-source driving data includes vehicle dynamic data, tunnel environment data and driver eye movement physiological data.

[0104] Module 1 is the basic data unit of this system, including the vehicle OBD diagnostic system, six-axis inertial navigation attitude sensor, millimeter-wave radar, SmartEye eye tracker, illuminance meter, driving recorder and other acquisition units. It is responsible for the synchronous acquisition of vehicle dynamic data, tunnel environment data and driver eye movement physiological data in urban tunnel scenarios. At the same time, it completes the hardware timestamp synchronization and alignment of multi-source driving data to provide raw data support for subsequent analysis.

[0105] Module 2, the data preprocessing and fusion module, is used to implement step S2, namely: cleaning and fusing the time-aligned multi-source driving data to obtain a standardized time-series feature dataset.

[0106] Module 2 communicates with the multi-source data acquisition module and is responsible for cleaning and standardizing the raw acquired data. Its core functions include filtering effective data, removing outliers, filling missing values, unifying the sampling frequency of multi-source data, and feature engineering. Finally, it outputs a standardized fused time-series feature dataset for subsequent modules to use.

[0107] Module 3, the three-dimensional index quantification calculation module, is used to implement step S3, namely: constructing a three-dimensional evaluation index system for tunnel scenarios based on time-series feature datasets, and quantifying and calculating the safety dimension index based on car-following safety margin, the visual comfort index based on pupil area change rate grading, and the energy consumption dimension index based on instantaneous fuel consumption normalization.

[0108] Module 3 communicates with the data preprocessing and fusion module, and has a built-in three-dimensional evaluation index calculation model of safety, visual comfort and energy consumption. It completes intelligent judgment of following state, quantitative calculation of safety margin SM, calculation of pupil area change rate and visual comfort level assignment, and calculation of instantaneous fuel consumption normalized energy consumption index. It outputs the time series data and global mean data of the three-dimensional evaluation index system for each driver's entire tunnel journey.

[0109] Module 4, Driver Profile Classification Module, is used to implement step S4, namely: performing unsupervised cluster analysis based on the global mean data of each driver's three-dimensional evaluation index system to divide the driver group into different types of profiles and determine the target driver type to be optimized.

[0110] Module 4 communicates with the 3D index quantification calculation module and incorporates a built-in k-means unsupervised clustering model. Based on existing driver profile datasets, it standardizes and clusters the global mean data of the 3D evaluation index system for new drivers, classifying them into one of three categories: safety-oriented, efficiency-comfort, or passive-conservative. This completes the refined driver profile classification and accurately identifies key targets for driving behavior optimization.

[0111] Module 5, the multi-objective optimization solution module, is used to implement step S5, namely: for the target driver type, using the headway time-of-flight (THW) as the control parameter, generating candidate optimized speed sequences that meet the constraints of the tunnel scenario based on the THW increment, combining the pre-trained visual comfort prediction model and instantaneous fuel consumption prediction model, predicting the changes in the three-dimensional evaluation indicators under different THW increments, and performing multi-objective optimization solution on the candidate optimized speed sequences to obtain the optimized speed sequence that achieves the best synergy between safety, comfort, and energy consumption, as the driving behavior optimization strategy.

[0112] Module 5 communicates with the 3D index quantification calculation module and the driver profile classification module, respectively. It is the core processing module of this system, and includes a speed sequence generation subunit based on THW incremental iteration, an XGBoost visual comfort prediction subunit, an XGBoost+RF stacked integrated instantaneous fuel consumption prediction subunit, and a multi-objective optimization subunit. It is responsible for optimizing the driver's performance based on objectives, iteratively generating candidate optimized speed sequences under different THW increments, predicting the corresponding 3D evaluation index increments, and selecting the optimal THW increment and corresponding optimized driving strategy through objective functions and constraints. The optimization results of this module can be verified through... Figure 5 The three-dimensional evaluation index can be intuitively verified by comparing the curves before and after optimization. It can output the comparison results of the index before and after optimization for different types of drivers, providing quantitative support for personalized guidance.

[0113] The optimization strategy output and guidance module is used to implement step S6, namely: outputting the driving behavior optimization strategy through the vehicle guidance and prompting device to provide personalized driving behavior guidance to the target driver to be optimized.

[0114] Module 6 communicates with the multi-objective optimization solution module and is responsible for transforming driving behavior optimization strategies into executable guidance instructions in the vehicle-road cooperative scenario. Through vehicle-road cooperative roadside equipment, in-vehicle intelligent terminals, variable information signs in tunnels, and intelligent cockpit voice interaction systems, it issues personalized following distance and speed guidance prompts to the driver, realizing proactive and precise optimization guidance of driving behavior in tunnels.

[0115] The role and effect of the embodiments

[0116] The multi-objective optimization method and system for low-carbon driving behavior in tunnels according to the present invention have the following beneficial effects:

[0117] (1) Multi-objective collaborative optimization: Steps S3 to S5 of this invention construct a tunnel-specific three-dimensional collaborative optimization system of safety, visual comfort and energy consumption. Taking the headway THW as the core control parameter, an optimization objective function with multi-objective constraints is established. By solving the function, the driving behavior optimization strategy that achieves the best synergy between safety improvement, comfort improvement and energy consumption reduction can be obtained, thus realizing the synergistic optimization of the three.

[0118] (2) Strong adaptability to tunnel scenarios: This invention is designed specifically for the unique scenario characteristics of urban road tunnels, such as enclosed space, sharp transition between light and dark at entrances and exits, complex alignment, and dense traffic flow. It integrates the scenario-specific features such as tunnel illumination, slope, alignment, and following state into all aspects of the process, including the construction of the three-dimensional evaluation index system, prediction model training, and speed sequence optimization in steps S3 to S5. This overcomes the problem of poor scenario adaptability when existing open road optimization schemes are directly applied to tunnel scenarios.

[0119] (3) Personalized targeted optimization capability: In step S4 of the present invention, a driver refined profile classification system is constructed based on the three-dimensional evaluation index system of safety-comfort-energy consumption through k-means clustering algorithm. It accurately divides three typical driving groups: safety-oriented, high-efficiency and comfort, and passive and conservative. Differentiated optimization strategies can be matched for the behavioral characteristics of different types of drivers to achieve personalized and targeted driving behavior guidance.

[0120] (4) High accuracy of indicator prediction: In step S5 of the present invention, for the time series data characteristics of tunnel scene, an XGBoost model adapted to visual comfort prediction and an XGBoost+RF stacked integrated model adapted to instantaneous fuel consumption prediction are constructed respectively. They integrate the environment, vehicle and time series characteristics specific to tunnel scene. The test set determination coefficient R² of the two models both reach more than 0.82, with high prediction accuracy and strong robustness, providing reliable quantitative support for multi-objective optimization solution.

[0121] (5) Proactive prevention of safety management: In step S5 of the present invention, the headway THW, a core parameter of following behavior, is selected as the core control variable. By actively optimizing the following distance and guiding the smooth speed sequence, the upgrade from "post-event risk warning" to "pre-event proactive prevention" is realized, which greatly improves the effectiveness of tunnel driving safety management.

[0122] (6) High feasibility and easy promotion of the solution: The method and system of the present invention are fully compatible with existing mature commercial hardware such as vehicle-mounted OBD, millimeter-wave radar, eye tracker, vehicle-road cooperative roadside equipment, and vehicle-mounted intelligent terminal. There is no need to carry out large-scale modification of tunnel infrastructure and vehicle body. The modular system architecture facilitates functional expansion, daily maintenance and iterative upgrades. The engineering implementation threshold is low and the scope of application is wide. It can be directly promoted to intelligent traffic management scenarios of various urban road tunnels and highway tunnels.

[0123] (7) Significant comprehensive benefits in multiple dimensions: Under the premise of ensuring the safety of driving in tunnels and improving the driving experience of drivers, this invention can significantly reduce the energy consumption of vehicles driving in tunnels. It not only meets the core needs of the low-carbon transformation of urban transportation, but also reduces the tunnel traffic accident rate and improves the tunnel operation efficiency, thus possessing outstanding engineering application value and social and economic benefits.

[0124] Those skilled in the art should understand that this invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to this invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A multi-objective optimization method for low-carbon driving behavior in tunnels, characterized in that, Includes the following steps: S1: Multi-source driving data acquisition step, acquiring multi-source driving data in a tunnel scenario, and performing time alignment on the multi-source driving data, which includes vehicle dynamic data, tunnel environment data, and driver eye movement physiological data; S2: Data preprocessing and fusion step, which cleans and fuses the time-aligned multi-source driving data to obtain a standardized time-series feature dataset; S3: Three-dimensional index quantification calculation step: Based on the time-series feature dataset, construct a three-dimensional evaluation index system for the tunnel scenario, and quantify and calculate the safety dimension index based on the following safety margin, the visual comfort index based on the pupil area change rate classification, and the energy consumption dimension index based on instantaneous fuel consumption normalization. S4: Driver profile classification step, based on the global mean data of the three-dimensional evaluation index system of each driver, performs unsupervised cluster analysis to divide the driver group into different types of profiles and determine the target driver type to be optimized; S5: Multi-objective optimization solution step: For the target driver type, using the headway time-of-flight (THW) as the control parameter, a candidate optimized speed sequence that satisfies the tunnel scenario constraints is generated based on the THW increment. Combining the pre-trained visual comfort prediction model and instantaneous fuel consumption prediction model, the change in three-dimensional evaluation indicators under different THW increments is predicted. The candidate optimized speed sequence is then subjected to multi-objective optimization solution to obtain the optimized speed sequence that achieves the best synergy between safety, comfort, and energy consumption, which serves as the driving behavior optimization strategy. S6: Optimize strategy output and guidance steps, output the driving behavior optimization strategy through the vehicle guidance and prompting device, and provide personalized driving behavior guidance to the target driver to be optimized.

2. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 1, characterized in that: in, The method for cleaning and fusing the time-aligned multi-source driving data in S2 is as follows: Non-congested and non-overspeed driving data within the vehicle speed range of 20km / h-100km / h were selected, outliers were removed based on the 3σ principle, missing data were filled using linear interpolation, and multi-source driving data with different sampling frequencies were uniformly resampled to the same frequency to complete spatiotemporal alignment and feature fusion.

3. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 1, characterized in that: in, The quantitative calculation method for the security dimension indicators in S3 is as follows: The following vehicle following condition is determined by a continuous detection time of ≥15s and a total safety margin (THW) of <7s. The safety margin SM model is used to quantify the following safety level. The value of SM ranges from 0 to 1, with a higher value indicating a lower driving risk. The headway is obtained by dividing the relative distance between the vehicle in front and the vehicle ahead by the vehicle's speed, and is expressed as: (1) In the formula, The following distance (m) is the following distance. The vehicle speed is (m / s). The formula for calculating SM at time t is expressed as: (2) In the formula, For the vehicle's speed, To maintain following distance, The speed of the vehicle in front. This is the acceleration due to gravity.

4. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 1, characterized in that: in, The quantitative calculation method for the visual comfort index in S3 is as follows: Calculate the change rate of the driver's pupil area, and perform visual comfort grading and assignment based on a preset threshold of the pupil area change rate. The assignment range is 0-1, and the higher the value, the better the visual comfort. The change rate of the pupil area is calculated from the pupil area data at consecutive moments and is expressed as: (3) In the formula, Let be the rate of change of pupil area at time i (%). Let be the pupil area (mm²) of the driver at time i. 2 ), The pupil area (mm) of the driver at time i-1. 2 ), The average pupil area (mm²) of drivers in tunnel sections 2 ).

5. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 4, wherein: in, The visual comfort grading and the assignment follow the following rules: When U≤6.62%, it indicates that the driver's visual adaptation state is optimal, which is defined as the "extremely comfortable" grading, and the assignment is H = 1; When 6.62% < U≤19.74%, it indicates that the driver has a good visual experience, which is defined as the "comfortable" grading, and the assignment is H = 0.7; When 19.74% < U≤41.51%, it indicates that the driver has obvious visual fatigue or discomfort, which is defined as the "uncomfortable" grading, and the assignment is H = 0.3; When U>41.51%, it indicates that the driver's vision bears a large load, which is defined as the "extremely uncomfortable" grading, and the assignment is H = 0.

6. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 1, wherein: in, The quantization calculation method of the energy consumption dimension index in S3 is: For vehicles Instantaneous fuel consumption data is subjected to max-min normalization to obtain an energy consumption dimension index in the range of 0 to 1. The higher the value, the higher the vehicle's energy consumption, as shown below: (4) In the formula, This represents the driver's instantaneous fuel consumption at time t (L / 100km). and These are the maximum and minimum instantaneous fuel consumption values ​​(L / 100km) among all driving data.

7. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 1, wherein: Among them, the generation method of the candidate optimization speed sequence in S5 is: The tunnel is segmented at a preset interval, the iteration range and step size of the THW increment are set, and the initial optimization speed of each segment is generated based on the following distance and the THW increment, which is expressed as: (5) In the formula, To maintain following distance, For THW increment, The headway before optimization, For initial speed optimization, Apply the segment speed fluctuation limit and the adjacent time step speed change smoothing constraint to the initial optimization speed to complete the speed sequence calibration; Map the tunnel scene features to the corresponding mileage points of the calibrated speed sequence to obtain the candidate optimization speed sequence.

8. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 1, wherein: in, The visual comfort prediction model in S5 is an XGBoost model. The inputs include: vehicle speed, tunnel illumination, and the change rate of the pupil area at the previous preset time step. The output is the change rate of the pupil area at the current moment. The XGBoost model completes hyperparameter optimization through time series cross-validation and grid search. The instantaneous fuel consumption prediction model is stacked and integrated by an XGBoost model and a random forest model. The inputs include: weighted composite acceleration, vehicle speed-slope coupling characteristics, and the instantaneous fuel consumption value at the previous preset time step. The output is the instantaneous fuel consumption value at the current moment. The XGBoost model in the instantaneous fuel consumption prediction model completes hyperparameter optimization through Bayesian optimization.

9. The multi-objective optimization method for low-carbon driving behavior in tunnels according to claim 7, wherein: in, The unsupervised clustering analysis in S4 uses the k-means unsupervised clustering algorithm. Different types of portraits include: safety-oriented, efficient and comfortable, and passive and conservative. The passive and conservative type is determined as the target driver type to be optimized. The method for obtaining the driving behavior optimization strategy in S5 is: Construct a multi-objective optimization objective function J, and at the same time set the constraint conditions a>0, b>0, c<0, which is expressed as: (10) In the formula, a represents the average increment of the safety dimension index, b represents the average increment of the visual comfort index, and c represents the average increment of the energy consumption dimension index. Iterate through all THW increments, select the THW increments that satisfy the constraints and maximize the objective function J as the optimal control parameters, and obtain the corresponding collaborative optimal speed sequence as the driving behavior optimization strategy.

10. A multi-objective optimization system for low-carbon driving behavior in tunnels, characterized in that, include: The multi-source driving data acquisition module collects multi-source driving data in tunnel scenarios and performs time alignment on the multi-source driving data, which includes vehicle dynamic data, tunnel environment data, and driver eye movement physiological data. The data preprocessing and fusion module cleans and fuses the time-aligned multi-source driving data to obtain a standardized time-series feature dataset. The three-dimensional index quantification calculation module constructs a three-dimensional evaluation index system for the tunnel scenario based on the time-series feature dataset, and respectively quantifies and calculates the safety dimension index based on the car-following safety margin, the visual comfort index based on the pupil area change rate grading, and the energy consumption dimension index based on instantaneous fuel consumption normalization. The driver profile classification module performs unsupervised clustering analysis based on the global mean data of the three-dimensional evaluation index system for each driver, divides the driver group into different types of profiles, and determines the target driver type to be optimized. The multi-objective optimization solution module, for the target driver type, uses the headway time-of-flight (THW) as the control parameter, generates candidate optimized speed sequences that meet the constraints of the tunnel scenario based on the THW increment, and combines a pre-trained visual comfort prediction model and an instantaneous fuel consumption prediction model to predict the changes in three-dimensional evaluation indicators under different THW increments. The module then performs multi-objective optimization solution on the candidate optimized speed sequences to obtain the optimized speed sequence that achieves the best synergy between safety, comfort, and energy consumption, which serves as a driving behavior optimization strategy. The optimization strategy output and guidance module outputs the driving behavior optimization strategy through the vehicle guidance and prompting device to provide personalized driving behavior guidance to the target driver to be optimized.

Images