End-side vehicle dispatching guidance instruction generation method and system considering safety efficiency comfort targets

Abstract

The application provides a kind of end side vehicle scheduling guidance instruction generation method and system considering safety efficiency comfort target, method includes: obtaining target area vehicle trajectory related data and real-time traffic signal light state data, road construction regulation data, meteorological environment data and special event announcement, and data is time-space alignment and feature fusion;Based on physical constraint design includes safety, efficiency, comfort index target function, construct trajectory prediction model based on LSTM and train;Combined with real-time environment data, generate future trajectory probability distribution results that meet multi-objective optimization;According to vehicle operation scene, future trajectory probability distribution is encoded into executable scheduling guidance instruction sequence, and is distributed to target vehicle.The application combines data-driven and physical driving, significantly improves the safety, traffic efficiency and ride comfort of scheduling instruction in complex traffic scene, and provides reliable technical support for dynamic management and control of intelligent transportation system.

Description

Technical Field

[0001] This invention belongs to the field of vehicle dispatching technology, and relates to a method and system for generating vehicle dispatching guidance instructions, and more particularly to a method and system for generating end-side vehicle dispatching guidance instructions that takes into account safety, efficiency and comfort goals. Background Technology

[0002] Faced with dynamic and complex traffic scenarios and diverse driving demands, existing traffic control measures are no longer sufficient to meet the needs of efficient scheduling and safe management. The increasing complexity of modern urban traffic environments and the continuous growth of traffic flow pose challenges to existing traffic management methods. Traditional traffic management systems typically employ fixed signal control and linear flow guidance, which are ill-suited to effectively adapt to the dynamic travel needs of individual vehicles. The rapid development of the Internet of Vehicles (IoV) offers new possibilities for improving traffic control. Through collaborative information exchange between connected vehicles, real-time scheduling and control of traffic flow is expected to become more efficient.

[0003] To perform real-time scheduling and guidance of connected vehicles, it is necessary to collaboratively extrapolate and predict vehicle trajectories, selecting the optimal trajectories for vehicles within the scene, and thereby generating scheduling instructions. Trajectory prediction models are mainly divided into three categories: traditional model methods, deep learning methods, and reinforcement learning methods.

[0004] Traditional modeling methods are based on vehicle dynamics and kinematic rules (such as acceleration and steering angle constraints), using methods like Kalman filtering and particle filtering to predict trajectories. Their advantages include high computational efficiency, strong interpretability, and robustness to sparse sensor data or noisy environments. However, traditional models assume independent behavior of traffic participants, making it difficult to model complex interactions, and they rely on manually designed motion equations. If the actual scenario deviates from the pre-designed model, accuracy will significantly decrease. Therefore, traditional models are suitable for short-term predictions on structured roads such as highways, or as auxiliary calibration modules for deep learning models, but their limitations are evident in dynamic scenarios such as urban intersections.

[0005] Deep learning methods automatically learn the spatiotemporal patterns of trajectories through data-driven approaches (such as LSTM, Transformer, and graph neural networks), capturing complex features like multi-vehicle interactions and traffic rules, and generating multimodal probabilistic predictions. Their advantages lie in high accuracy, end-to-end training, and multi-source data fusion capabilities. However, deep learning models require a large amount of labeled data, and their decision-making process is a "black box," making security verification difficult and prone to failure in scenarios with out-of-data distribution.

[0006] Reinforcement learning treats trajectory prediction as a sequential decision-making problem, guiding the agent to optimize its strategy through reward functions (such as safety, comfort, and efficiency) and its interaction with the environment. Its advantage lies in its ability to simulate driving behavior driven by long-term gains and to adapt to real-time traffic changes online. However, training reinforcement learning models relies on high-fidelity simulation environments, and the design of reward functions needs to balance multiple conflicting objectives. Oversimplification can lead to strategies deviating from real human driving habits. Therefore, reinforcement learning is suitable for dynamic scheduling scenarios, or for inferring driver intent through inverse reinforcement learning. However, in scenarios with stringent safety requirements, it needs to be combined with traditional physical constraints to reduce risks.

[0007] Therefore, there is an urgent need for a vehicle dispatching and guidance scheme that takes into account safety, efficiency, and comfort goals, and can effectively predict the optimal operating trajectory of vehicles and generate corresponding guidance and dispatching instructions. Summary of the Invention

[0008] To address the aforementioned problems, this invention provides a method and system for generating end-side vehicle dispatching and guidance instructions that considers safety, efficiency, and comfort objectives.

[0009] The technical solution adopted in this invention is as follows:

[0010] A method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives includes the following steps:

[0011] S1. Acquire real-time trajectory data, historical trajectory data, vehicle type and driving status collected by vehicle sensors within the target area, and perform preprocessing to obtain an initial multi-dimensional feature matrix;

[0012] S2. Acquire real-time traffic signal status data, road construction control data, meteorological environment data, and special event announcements, and perform spatiotemporal alignment and feature fusion with the initial multi-dimensional feature matrix to obtain an enhanced dataset containing environmental constraints;

[0013] S3. Design an objective function that includes safety, efficiency, and comfort indicators based on real physical rules, construct a trajectory prediction model based on a Long Short-Term Memory (LSTM) network, and train the trajectory prediction model based on the enhanced dataset;

[0014] S4. Combine real-time environmental data and use the trained trajectory prediction model to generate a future trajectory probability distribution that satisfies multi-objective optimization;

[0015] S5. Determine the vehicle operation scenario based on the future trajectory probability distribution result, encode the future trajectory probability distribution into an executable scheduling guidance instruction sequence based on the vehicle operation scenario, and distribute it to the target vehicle.

[0016] Furthermore, in step S1, real-time trajectory data and historical trajectory data are collected through roadside sensors, including longitude, latitude, speed, lateral acceleration, longitudinal acceleration, and heading angle; the vehicle type and driving status are obtained through the terminal side, wherein the vehicle type includes manual driving and machine intelligent driving, and the driving status includes following instructions and arbitrary driving.

[0017] Furthermore, in step S2, the traffic signal status data includes the current light color and duration, the road construction control data includes the coordinate range and duration of the construction control area, and the meteorological environment data includes temperature, humidity, rainstorm and fog conditions.

[0018] Furthermore, in step S3, designing the objective function that includes safety, efficiency, and comfort indicators specifically includes:

[0019] S31. Construct a safety objective function based on the driving safety field:

[0020] ,

[0021] in, This indicates that the vehicle's driving safety is consistently strong. This indicates the field strength of the dynamic potential energy field. This indicates that safe behavior is consistently strong.

[0022] S32. Construct an efficiency objective function by comparing the difference in average traffic flow between affected and unaffected areas:

[0023] ,

[0024] in, Indicates traffic flow efficiency. Indicates the area affected by the vehicle Average traffic flow This represents the average traffic flow in the area u where vehicles do not affect the area.

[0025] S3.3. Construct a comfort objective function by calculating the standard deviation between the vehicle's real-time acceleration and average acceleration:

[0026]

[0027] in, This represents the standard deviation of vehicle acceleration from average acceleration. Indicates the total length of the time range covered. Represents the acceleration at time t. Indicates average acceleration;

[0028] S3.4. The objective reward function of the trajectory prediction model is constructed by weighting the objective functions of safety, efficiency, and comfort together with the objective function of accuracy:

[0029] ,

[0030] Where Y represents the target reward function, and a, b, c, and d represent adjustable weight parameters. This indicates that the vehicle's driving safety is consistently strong. Represents traffic flow efficiency. This represents the standard deviation of vehicle acceleration from average acceleration. Represents the trajectory prediction value. This represents the actual value of the trajectory.

[0031] Furthermore, the driving safety field is a physical field model used to describe and quantify potential risks during vehicle operation;

[0032] The dynamic potential energy field described above is a model representing the degree of risk that object i poses to the overall traffic environment under certain road conditions. The formula is as follows:

[0033] ,

[0034] in, Let represent the dynamic potential energy field strength of vehicle i at coordinate point j. This represents the equivalent mass of vehicle i. This represents the speed of vehicle i. This represents the acceleration of vehicle i. This represents the angle between the direction vector of the vehicle relative to coordinate point j and the direction of the object's velocity. This represents the influencing factor of the road conditions in the z-th segment where the vehicle is located. Indicates the safe distance between the vehicle and an object in front. Indicates the centroid coordinates of vehicle i. Indicates vehicle i and Vector distance between them This indicates the direction in which the potential energy decreases along the gradient;

[0035] The aforementioned safe behavior field is a behavior field model formed by vehicle i under certain road conditions, expressed by the formula:

[0036] ,

[0037] in, This represents the field strength of the safe behavior field of vehicle i at coordinate point j. This represents the driving risk factor for vehicle i, and the other letters have the same meaning as in the dynamic potential energy field calculation formula.

[0038] Furthermore, the training process of the trajectory prediction model based on the Long Short-Term Memory network includes:

[0039] The temporal feature matrix from the enhanced dataset is input into the trajectory prediction model based on the long short-term memory network. The data is processed sequentially through multiple cell units. In each cell unit, the temporal data is processed through a forget gate, an input gate, and an output gate to update the cell state and hidden state.

[0040] After passing through a fully connected layer and a softmax layer, a hybrid density network is used to output a trajectory probability distribution that follows a Gaussian mixture distribution.

[0041] Based on the trajectory probability distribution of the output, the target reward function is transformed into a log-likelihood loss form, and iterative optimization is performed on the training set until the model converges on the validation set.

[0042] Furthermore, the vehicle operation scenarios include conflict scenarios, signal scenarios, and following scenarios; the scenario judgment method specifically includes: establishing scoring formulas for conflict scenarios, signal scenarios, and following scenarios respectively, and taking the scenario with the highest score as the current scenario category of the vehicle;

[0043] The scoring formula is constructed based on the following principles: when a vehicle is operating in an intersection area and there is a risk of its trajectory intersecting with that of other vehicles, conflict scenarios should be given priority; when a vehicle is in a traffic light-controlled intersection area and its operation is directly affected by the traffic lights, it should be classified as a signal scenario; when a vehicle is in a straight road or continuous traffic flow scenario and mainly interacts with the vehicle in front, it should be classified as a following scenario.

[0044] Furthermore, in conflict scenarios, the time and location of the conflict are first calculated, the priority probability is evaluated based on the trajectory statistical characteristics of the vehicle and other vehicles, and the comprehensive safety probability of each vehicle is calculated by combining the priority probabilities of the vehicle and other vehicles. When the comprehensive safety probability is greater than the preset safety threshold, a priority instruction is sent to the vehicle; otherwise, a yield instruction is sent.

[0045] In signal scenarios, the vehicle's position distribution at the red light time is obtained based on the future trajectory probability distribution. The vehicle's passage probability is calculated based on the relative position of the vehicle and the stop line. When the vehicle's passage probability is greater than the preset passage threshold, a passage permission instruction is sent to the vehicle. When the vehicle's passage probability is less than the preset stop threshold, a passage prohibition instruction is sent to the vehicle. When the vehicle's passage probability is between the preset stop threshold and the preset passage threshold, a deceleration instruction is sent to the vehicle.

[0046] In a car-following scenario, a safety threshold is first calculated based on the mean and standard deviation of the distance between vehicles. Then, a comprehensive risk index is calculated based on the safety threshold and the relative distance between the vehicle in front and the vehicle following. When the risk index is less than 0, a deceleration command is issued to the vehicle. When the risk index and the relative speed are greater than the preset threshold, an acceleration command is issued to the vehicle. Otherwise, a speed-maintaining command is issued.

[0047] Furthermore, in step S5, the dispatching guidance instructions include standardized control commands and natural language guidance prompts, which are distributed to the target vehicle through the V2X communication system. At the same time, a feedback mechanism is established to continuously optimize the dispatching guidance instruction generation strategy.

[0048] An end-side vehicle dispatching and guidance instruction generation system considering safety, efficiency, and comfort objectives, used in the above method, includes:

[0049] Data acquisition and processing module: used to acquire real-time trajectory data, historical trajectory data, vehicle type and driving status collected by vehicle sensors within the target area, and to preprocess the data to obtain an initial multi-dimensional feature matrix;

[0050] Environmental constraint fusion module: used to acquire real-time traffic light status data, road construction control data, meteorological environment data and special event announcements, and perform spatiotemporal alignment and feature fusion with the initial multi-dimensional feature matrix to obtain an enhanced dataset containing environmental constraints;

[0051] Model building and training module: used to design an objective function that includes safety, efficiency and comfort metrics, build a trajectory prediction model based on a long short-term memory network, and train the trajectory prediction model based on the enhanced dataset;

[0052] Trajectory distribution prediction module: This module combines real-time environmental data with a trained trajectory prediction model to generate a future trajectory probability distribution that satisfies multi-objective optimization.

[0053] Instruction generation and feedback module: Based on the future trajectory probability distribution result, determine the vehicle operation scenario, encode the future trajectory probability distribution into an executable scheduling and guidance instruction sequence according to the vehicle operation scenario, and distribute it to the target vehicle.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0055] 1. This invention combines data-driven and physics-driven approaches, adding physical constraints to the deep learning model architecture to collaboratively perform trajectory prediction and scheduling instruction generation. This ensures the physical rationality of the model while improving the fine-grainedness and accuracy of predictions. This invention balances security, interpretability, and adaptability to complex scenarios, making it particularly suitable for real-time decision-making systems with high reliability requirements.

[0056] 2. This invention uses a weighted index based on safety, efficiency, and comfort to construct a target reward function. The predicted target is not the traditional absolute accuracy but to maximize the overall benefit, avoiding the result of local optima but global suboptimal, and can effectively improve traffic efficiency.

[0057] 3. This invention generates vehicle dispatch instructions based on trajectory prediction results and adds a feedback mechanism to continuously optimize the instruction generation strategy based on the feedback results. It can provide appropriate dispatch guidance for different driving groups, maximize the overall benefits of the scenario, and effectively regulate traffic flow. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall method in an embodiment of the present invention. Detailed Implementation

[0059] The technical solution of the present invention will be further described clearly and in detail below with reference to the accompanying drawings and specific embodiments.

[0060] LSTM, a special type of recurrent neural network, is specifically designed for processing sequential data and effectively alleviates the gradient vanishing problem of traditional networks. This invention uses LSTM as a foundation, constructing an objective function based on multiple indicators of safety, efficiency, and comfort. It abandons the excessive pursuit of accuracy inherent in traditional prediction models, prioritizing the overall benefit of vehicles on a road segment as the primary objective, with prediction accuracy as a secondary objective. Simultaneously, physical constraints are added to the model to ensure the authenticity of the generated trajectories. An instruction conversion engine is developed for the generated candidate trajectory set, converting it into traffic or speed suggestions for drivers to support vehicle scheduling guidance for specific road segments.

[0061] like Figure 1 As shown, a method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives includes the following steps:

[0062] S1. Acquire real-time trajectory data, historical trajectory data, vehicle type and driving status collected by vehicle sensors within the target area, and preprocess them to obtain an initial multi-dimensional feature matrix. At the same time, extract the road network topology and dynamic traffic flow information as environmental state input.

[0063] Real-time and historical trajectory data are collected by roadside sensors, including longitude, latitude, speed, lateral acceleration, longitudinal acceleration, and heading angle; among them, real-time vehicle trajectory data is collected by roadside sensors at 10Hz.

[0064] The vehicle type and driving status are obtained by matching the vehicle ID on the terminal side. The vehicle ID is used to identify characteristics such as vehicle type (human driving, machine intelligent driving) and vehicle driving status (following instructions, driving at will).

[0065] In one specific embodiment of the present invention, vehicle trajectory data is standardized using the Min-Max normalization method. For vehicle trajectory data X, standardized input is obtained after normalization. As shown in the following formula:

[0066]

[0067] Where X represents the input vector. Represents the standardized input vector. This represents the minimum value of the input vector. This represents the maximum value of the input vector.

[0068] S2. Acquire real-time traffic signal status data, road construction control data, meteorological environment data, and special event announcements, and perform spatiotemporal alignment and feature fusion with the initial multi-dimensional feature matrix to obtain an enhanced dataset containing environmental constraints.

[0069] The data includes traffic signal status data, including the current light color and its duration; road construction control data, including the coordinate range and duration of the construction control area; and meteorological environmental data, including temperature, humidity, heavy rain, fog, and other conditions.

[0070] Environmental features such as traffic light status, road construction control information, and meteorological data can affect vehicle driving status. Therefore, it is necessary to align environmental data with the aforementioned features in time and space and incorporate them into the variable range of the prediction model.

[0071] S3. Design an objective function that includes safety, efficiency, and comfort indicators based on real-world physical rules, construct a trajectory prediction model based on a long short-term memory network, and train the trajectory prediction model using the enhanced dataset. Specifically, this includes:

[0072] S3.1. Construction of the security objective function

[0073] The safety objective function is constructed using the vehicle's driving safety field, a physical field model used to describe and quantify potential risks during vehicle operation. It combines vehicle dynamics, traffic environment, and driver behavior to construct a virtual "risk field" for predicting collision probabilities, assessing driving safety, and supporting decision-making. Its model can be expressed as:

[0074]

[0075] in, This indicates that the vehicle's driving safety is consistently strong. This indicates the field strength of the dynamic potential energy field. This indicates that safe behavior is strong in every situation.

[0076] The dynamic potential energy field model represents the degree of risk that vehicle i poses to the overall traffic environment under certain road conditions, and can be expressed as:

[0077]

[0078] in, Let represent the dynamic potential energy field strength of vehicle i at coordinate point j. This represents the equivalent mass of vehicle i. This represents the speed of vehicle i. This represents the acceleration of vehicle i. This represents the angle between the direction vector of the vehicle relative to coordinate point j and the direction of the object's velocity. This represents the influencing factor of the road conditions in the z-th segment where the vehicle is located. Indicates the safe distance between the vehicle and an object in front. Indicates the centroid coordinates of vehicle i. Indicates vehicle i and Vector distance between them This indicates the direction in which the potential energy decreases along the gradient.

[0079] The safe behavior field model represents the safe behavior field formed by vehicle i under certain road conditions, and it can be expressed as:

[0080]

[0081] in, This represents the field strength of the safe behavior field of vehicle i at coordinate point j. This represents the driving risk factor for vehicle i, and the other letters have the same meaning as in the dynamic potential energy field calculation formula.

[0082] S3.2. Construction of the efficiency objective function

[0083] The efficiency objective function is defined using a generalized definition method, which uses xt to calculate the key parameters of the basic traffic flow map, measures the average traffic flow within a certain time and space range, and evaluates the impact of vehicles on the efficiency of surrounding traffic flow by comparing the difference in average flow between regions.

[0084] Its flow calculation can be expressed as:

[0085]

[0086] in, Indicates the region Average flow, Indicates the region The total number of vehicles in the country Indicates that vehicle n is in region The driving distance in the middle, Indicates the region The area.

[0087] area The process of calculating the area can be specifically represented as follows:

[0088]

[0089] in, Indicates the region area, Indicates the region The upper bound of the time t coordinate. Indicates the region The lower bound of the time t coordinate. Indicates the region The upper bound of the x-coordinate. Indicates the region The lower bound of the distance from the x-coordinate.

[0090] Quantifying the impact of vehicles on traffic flow efficiency will affect the region. The difference between the average flow rate in the unaffected area and the average flow rate in the unaffected area is used as an efficiency indicator, which can be specifically expressed as:

[0091]

[0092] in, Represents traffic flow efficiency. Indicates the area affected by the vehicle. Average traffic flow This indicates the average traffic flow in area u where vehicles do not affect the area.

[0093] S33. Construction of the Comfort Objective Function

[0094] The comfort objective function is constructed by analyzing acceleration disturbances. The vehicle's acceleration constantly changes during operation, thus affecting the comfort of the driver and passengers. This is measured by the standard deviation of the vehicle's acceleration from its mean acceleration, and can be specifically expressed as:

[0095]

[0096] in, Let represent the standard deviation of the vehicle's acceleration from its average acceleration, T represent the total length of the covered time range, and a(t) represent the acceleration at time t. It represents the average acceleration.

[0097] The average acceleration can be expressed as:

[0098]

[0099] Where T represents the total length of the covered time range, and a(t) represents the acceleration at time t.

[0100] S34. Determining the Target Reward Function

[0101] The objective functions of safety, efficiency, and comfort are combined with the objective of accuracy in a weighted manner to construct the objective reward function of the LSTM prediction model, which is the overall benefit. This can be specifically expressed as:

[0102]

[0103] Where Y represents the target reward function, and a, b, c, and d represent adjustable weight parameters. This indicates that the vehicle's driving safety is consistently strong. Represents traffic flow efficiency. This represents the standard deviation of vehicle acceleration from average acceleration. Represents the trajectory prediction value. This represents the actual value of the trajectory.

[0104] Different weights are assigned to safety, efficiency, comfort and accuracy indicators to comprehensively measure their overall benefits and ensure that the model can output a candidate trajectory set that maximizes the overall vehicle benefits in the region.

[0105] S35. Establish an LSTM trajectory prediction model

[0106] The input vehicle trajectory time sequence is in a matrix, and it passes through multiple cell units sequentially. Each cell goes through a forget gate, an input gate, and an output gate. Finally, the hidden state is mapped to the predicted trajectory distribution through a fully connected layer. The model gating parameters are corrected through training. , The specific calculation process of the model is as follows:

[0107] Each cell in an LSTM contains a forget gate, an input gate, and an output gate. The input to the cell is the cell state from the previous time step. and hidden state and the input of the current time step .

[0108] The forget gate determines the portion of the cell state to be retained, allowing for selective forgetting of information from the previous time step. Specifically, it can be represented as:

[0109]

[0110] in, Represents the forget gate vector. The weight matrix represents the forget gate. The bias term representing the forget gate. This indicates the hidden state at the previous moment. Indicates the current input. This represents the Sigmoid activation function.

[0111] After the forgotten information is determined, it enters the input gate for processing. The input gate is mainly used to determine the content stored in the new cell state. First, the probability of each variable recorded in the cell state is calculated, which can be expressed as:

[0112]

[0113] in, Represents the input gate vector. This represents the weight matrix of the input gate. This represents the bias term of the input gate. This indicates the hidden state at the previous moment. Indicates the current input. This represents the Sigmoid activation function.

[0114] Based on this selection, the candidate cell state vectors can be represented as:

[0115]

[0116] in, Represents the state vector of the candidate unit. The weight matrix representing the candidate states. The bias term represents the candidate state. This indicates the hidden state at the previous moment. Indicates the current input. This represents the hyperbolic tangent function.

[0117] Next, the cell state is updated, and the decision-making process, combining the forget gate and the input gate, can be specifically represented as follows:

[0118]

[0119] in, Indicates the current state of the cell. This indicates the cell state at the previous moment. Represents the state vector of the candidate unit. Represents the forget gate vector. This represents the input gate vector.

[0120] The last part of a cell is for output. First, determine the part of the cell state that needs to be output, which can be represented as:

[0121]

[0122] in, Represents the output gate vector. This represents the Sigmoid activation function. This represents the weight matrix of the output gate. This represents the bias term of the output gate. This indicates the hidden state at the previous moment. This indicates the current input.

[0123] Next, selective output of the cell state can be represented as follows:

[0124]

[0125] in, This indicates the output at the current moment. Indicates the current state of the cell. Represents the output gate vector. This represents the hyperbolic tangent function.

[0126] The final output is processed through a fully connected layer and a softmax layer to obtain the final candidate trajectory probability distribution. A hybrid density network is used to output a trajectory probability distribution that follows a Gaussian mixture distribution, and its probability can be expressed as:

[0127]

[0128] in, Let represent the conditional probability distribution of the output trajectory y given input data x, where K represents the number of Gaussian components in the mixture distribution, and k represents the index of the k-th Gaussian component. This represents the mixing weight of the k-th Gaussian component. This represents the probability density function of a Gaussian distribution.

[0129] The probability density function of the Gaussian distribution can be specifically expressed as:

[0130]

[0131] in, Let the mean function of the k-th Gaussian component be denoted as . Let represent the standard deviation function of the k-th Gaussian component.

[0132] c36, Model Training and Validation

[0133] The augmented dataset is divided into training, validation, and test sets in a specific ratio (7:2:1). The model is trained using the training set and validated using the validation set. Based on the probability distribution output, the objective function is transformed into a log-likelihood loss form, and the model eventually converges through continuous iteration.

[0134] S4. Combining real-time vehicle location information and road condition updates, the trained trajectory prediction model is used to generate a future trajectory probability distribution that satisfies multi-objective optimization.

[0135] S5. Based on the future trajectory probability distribution results and the aforementioned environmental state data, determine the vehicle operation scenario, encode the future trajectory probability distribution into an executable scheduling guidance instruction sequence, including standardized control commands and natural language guidance prompts, and distribute them to the target vehicle through the V2X communication system. At the same time, establish a feedback mechanism to continuously optimize the scheduling guidance instruction generation strategy.

[0136] The vehicle operation scenarios mentioned include conflict scenarios, signal scenarios, and following scenarios; the specific scenario judgment methods include:

[0137] By establishing scoring formulas for conflict scenarios, signal scenarios, and following scenarios respectively, and comparing these formulas, the final scenario category to which the vehicle currently belongs is output. The overall discrimination rule can be expressed as:

[0138]

[0139] Where S represents the vehicle operation scenario, and argmax represents taking the maximum value. This represents the score indicating the degree of matching between conflict scenarios. This indicates the signal scene matching score. This indicates the matching score for the driving scenario.

[0140] When a vehicle is operating in an intersection area and there is a risk of its trajectory intersecting with that of other vehicles, conflict scenarios should be given priority. The formula for the conflict scenario matching score can be expressed as:

[0141]

[0142] in, This represents the score indicating the degree of matching between conflict scenarios. Let M represent the minimum time difference between the arrival of the vehicle and other vehicles at the potential conflict point, and let M represent the number of other vehicles with potential trajectory conflicts. R represents the normalized upper bound of the number of conflicting vehicles, R=i indicates that the road type is an intersection, and I represents the indicator function. , , This represents the weighting coefficient.

[0143] When a vehicle is in a traffic light-controlled intersection area and its operation is directly affected by the traffic lights, it must be classified as a traffic light scenario. The formula for the traffic light scenario matching score can be expressed as:

[0144]

[0145] in, This represents the signal scene matching score. Indicates the distance from the car to the parking line. To influence the normalized upper limit of the distance, L represents the remaining duration of the traffic light, L=1 indicates that traffic light control is in place, and I represents the indication function. , , This represents the weighting coefficient.

[0146] When a vehicle is on a straight road or in continuous traffic flow and primarily interacts with the vehicle in front, it should be classified as a car-following scenario. The formula for the matching score of a car-following scenario can be expressed as:

[0147]

[0148] in, This indicates the matching score with the driving scenario. This indicates the relative speed between the vehicle in front and the vehicle itself. This represents the upper limit of the normalized relative velocity. Indicates the average distance between vehicles. R=s indicates the safe following distance, and R=s indicates that the road type is a straight road. , , This represents the weighting coefficient.

[0149] After the aforementioned score calculation is completed, the system takes the scenario corresponding to the maximum score as the final judgment result, thereby realizing the automatic identification of vehicle operation scenarios and providing input for the generation of subsequent scheduling and guidance instructions.

[0150] (1) Conflict scenarios

[0151] Conflict scenarios typically occur at unsignalized intersections, where conflicting traffic flows may lead to collisions. When a conflict occurs, vehicles must yield.

[0152] First, it is necessary to determine when the conflict might occur, which can be represented as:

[0153]

[0154] in Indicates the point in time when conflict may occur. This represents the position of the vehicle at time t. This represents the location of his car at time t. This represents the Euclidean distance between the two vehicles at time t. This indicates the time point at which the distance between the two vehicles is minimized.

[0155] Determining the location of a conflict based on its timing can be expressed as:

[0156]

[0157] in, Indicates the location where a conflict may occur. Represents t c The current location of this vehicle Represents t c The location of his car at that moment.

[0158] Multiple trajectories are sampled from the probability distribution of the trajectories, and the statistical characteristics of the vehicle and other vehicles are recorded. This can be represented as:

[0159]

[0160]

[0161] in, This represents the time when the i-th trajectory of the vehicle reaches the conflict point. This represents the time it takes for the i-th trajectory of the vehicle to reach the conflict point, and N represents the number of trajectories. This represents the average time it takes for the vehicle to reach the point of conflict. This indicates the average time it took for his vehicle to reach the point of conflict. This represents the variance of the time it takes for the vehicle to reach the conflict point. This represents the variance of the time it takes for the other vehicle to reach the point of conflict.

[0162] The probability of a vehicle arriving first, based on statistical characteristics, can be specifically expressed as:

[0163]

[0164] in, T represents the probability that one car arrives at the point of conflict before another car. e Indicates the time his vehicle arrived at the point of conflict, T o Indicates the time when the vehicle arrived at the point of conflict. This represents the average time it takes for the vehicle to reach the point of conflict. This indicates the average time it took for his vehicle to reach the point of conflict. This represents the variance of the time it takes for the vehicle to reach the conflict point. This represents the variance of the time it takes for the other vehicle to reach the point of conflict.

[0165] The cumulative distribution function of the standard normal distribution can be expressed as:

[0166]

[0167] Because multiple vehicle collisions may occur, statistics are collected for each other vehicle, and the overall safety probability is calculated, which can be specifically expressed as:

[0168]

[0169] in, Let P represent the overall safety probability, M represent the number of other vehicles involved in the conflict, and P represent the overall safety probability. k This represents the probability that the vehicle corresponding to the kth other vehicle arrives first.

[0170] When the overall safety probability is greater than 0.7, a priority instruction is sent to the vehicle; otherwise, a yield instruction is sent to the vehicle.

[0171] (2) Signal scenario

[0172] The main conflict in signal scenarios arises at the timing of the green light turning red, requiring a determination of whether a vehicle can cross the stop line. First, the vehicle's positional distribution during the red light's onset time is obtained based on trajectory prediction results, which can be specifically represented as:

[0173]

[0174] in, This indicates the time t when the vehicle's red light turns on. red The position, K represents the number of components in the mixture Gaussian distribution, This represents the weight of the k-th Gaussian distribution in the mixed distribution. This represents the k-th Gaussian distribution. This represents the mean of a Gaussian distribution. This represents the variance of the Gaussian distribution.

[0175] By comparing its positional distribution with the relative position of the stop line, the probability of it passing the stop line can be determined, which can be specifically expressed as:

[0176]

[0177] in, This represents the probability that a vehicle can pass the stop line. Indicates the location of the parking line. Let K represent the cumulative distribution function of the standard normal distribution, and K represent the number of components in the Gaussian mixture distribution. This represents the weight of the k-th Gaussian distribution in the mixed distribution. This represents the mean of a Gaussian distribution. This represents the variance of the Gaussian distribution.

[0178] When the probability of passage is greater than 0.8, a passage permission instruction is issued to the vehicle; when it is less than 0.2, a passage prohibition instruction is issued; and when it falls between these two values, the vehicle is guided to decelerate. The recommended acceleration can be expressed as:

[0179]

[0180] in, This indicates a suggested acceleration command. This represents the minimum acceleration that the vehicle can achieve. Indicates the vehicle's current speed. Indicates the vehicle's current location. This indicates the safe distance that a vehicle should maintain from the stop line when it is stopped.

[0181] (3) Following the car scenario

[0182] In car-following scenarios, the distance between vehicles is used to determine whether the following vehicle should accelerate or decelerate, thereby improving the overall traffic efficiency of the road network. First, a safety threshold is calculated based on the mean and standard deviation of the predicted distance, which can be expressed as follows:

[0183]

[0184] in, Indicates the safety threshold. This represents the mean of the predicted vehicle distances. This represents the standard deviation of the predicted vehicle distance. This indicates the safe distance that vehicles should maintain while driving.

[0185] The relative distance between the vehicle in front and the vehicle following can be expressed as:

[0186]

[0187]

[0188] in, This represents the average relative speed between the vehicle in front and the vehicle itself. This represents the average speed of the vehicle in front. This represents the average speed of the vehicle. The standard deviation of relative velocity, The variance of the speed of the vehicle in front. This represents the variance of the vehicle's speed.

[0189] The comprehensive risk index is calculated based on the above indicators and can be specifically expressed as follows:

[0190]

[0191] in, This represents a comprehensive risk indicator. Indicates the safety threshold. This represents the average relative speed between the vehicle in front and the vehicle itself. The standard deviation of relative velocity.

[0192] When the risk indicator is less than 0, a deceleration command is issued to the vehicle; when the risk indicator is greater than 2 and... When the vehicle is in a certain condition, an acceleration command is issued; otherwise, the vehicle can maintain its original speed.

[0193] The present invention also provides an end-side vehicle dispatching guidance instruction generation system that considers safety, efficiency, and comfort objectives, for use in the above method, comprising:

[0194] Data acquisition and processing module: used to acquire real-time trajectory data, historical trajectory data, vehicle type and driving status collected by vehicle sensors within the target area, and to preprocess the data to obtain an initial multi-dimensional feature matrix;

[0195] Environmental constraint fusion module: used to acquire real-time traffic light status data, road construction control data, meteorological environment data and special event announcements, and perform spatiotemporal alignment and feature fusion with the initial multi-dimensional feature matrix to obtain an enhanced dataset containing environmental constraints;

[0196] Model building and training module: used to design an objective function that includes safety, efficiency and comfort metrics, build a trajectory prediction model based on a long short-term memory network, and train the trajectory prediction model based on the enhanced dataset;

[0197] Trajectory distribution prediction module: This module combines real-time environmental data with a trained trajectory prediction model to generate a future trajectory probability distribution that satisfies multi-objective optimization.

[0198] Instruction generation and feedback module: Based on the future trajectory probability distribution result, determine the vehicle operation scenario, encode the future trajectory probability distribution into an executable scheduling and guidance instruction sequence according to the vehicle operation scenario, and distribute it to the target vehicle.

[0199] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0200] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0201] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0202] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0203] The above description is merely a preferred embodiment of the present invention. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the technical solutions of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall still fall within the protection scope of the technical solutions of the present invention.

Claims

1. A method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives, characterized in that, Includes the following steps: S1. Acquire real-time trajectory data, historical trajectory data, vehicle type and driving status collected by vehicle sensors within the target area, and perform preprocessing to obtain an initial multi-dimensional feature matrix; S2. Acquire real-time traffic signal status data, road construction control data, meteorological environment data, and special event announcements, and perform spatiotemporal alignment and feature fusion with the initial multi-dimensional feature matrix to obtain an enhanced dataset containing environmental constraints; S3. Design an objective function that includes safety, efficiency, and comfort indicators based on physical constraints, construct a trajectory prediction model based on a long short-term memory network, and train the trajectory prediction model based on the enhanced dataset; S4. Combine real-time environmental data and use the trained trajectory prediction model to generate a future trajectory probability distribution that satisfies multi-objective optimization; S5. Determine the vehicle operation scenario based on the future trajectory probability distribution result, encode the future trajectory probability distribution into an executable scheduling guidance instruction sequence based on the vehicle operation scenario, and distribute it to the target vehicle.

2. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, In step S1, real-time trajectory data and historical trajectory data are collected by roadside sensors, including longitude, latitude, speed, lateral acceleration, longitudinal acceleration, and heading angle; the vehicle type and driving status are obtained by the terminal side, wherein the vehicle type includes manual driving and machine intelligent driving, and the driving status includes following instructions and driving at will.

3. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, In step S2, the traffic signal status data includes the current light color and duration, the road construction control data includes the coordinate range and duration of the construction control area, and the meteorological environment data includes temperature, humidity, rainstorm and fog conditions.

4. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, In step S3, designing the objective function that includes safety, efficiency, and comfort indicators specifically includes: S31. Construct a safety objective function based on the driving safety field: ， in, This indicates that the vehicle's driving safety is consistently strong. This indicates the field strength of the dynamic potential energy field. This indicates that safe behavior is consistently strong. S32. Construct an efficiency objective function by comparing the difference in average traffic flow between affected and unaffected areas: ， in, Indicates traffic flow efficiency. Indicates the area affected by the vehicle Average traffic flow This represents the average traffic flow in the area u where vehicles do not affect the area. S3.

3. Construct a comfort objective function by calculating the standard deviation between the vehicle's real-time acceleration and average acceleration: ， in, This represents the standard deviation of vehicle acceleration from average acceleration. Indicates the total length of the time range covered. Represents the acceleration at time t. Indicates average acceleration; S3.

4. The objective reward function of the trajectory prediction model is constructed by weighting the objective functions of safety, efficiency, and comfort together with the objective function of accuracy: ， Where Y represents the target reward function, and a, b, c, and d represent adjustable weight parameters. This indicates that the vehicle's driving safety is consistently strong. Represents traffic flow efficiency. This represents the standard deviation of vehicle acceleration from average acceleration. Represents the trajectory prediction value. This represents the actual value of the trajectory.

5. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, The driving safety field is a physical field model used to describe and quantify potential risks during vehicle operation; The dynamic potential energy field described above is a model representing the degree of risk posed by vehicle i to the overall traffic environment under certain road conditions, and is expressed by the formula: ， in, Let represent the dynamic potential energy field strength of vehicle i at coordinate point j. This represents the equivalent mass of vehicle i. This represents the speed of vehicle i. This represents the acceleration of vehicle i. This represents the angle between the direction vector of the vehicle relative to coordinate point j and the direction of the object's velocity. This represents the influencing factor of the road conditions in the z-th segment where the vehicle is located. Indicates the safe distance between the vehicle and the object in front. Indicates the centroid coordinates of vehicle i. Indicates vehicle i and Vector distance between them This indicates the direction in which the potential energy decreases along the gradient; The aforementioned safe behavior field is a behavior field model formed by vehicle i under certain road conditions, expressed by the formula: ， in, This represents the field strength of the safe behavior field of vehicle i at coordinate point j. This represents the driving risk factor for vehicle i.

6. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, The training process of the trajectory prediction model based on long short-term memory network includes: The temporal feature matrix from the enhanced dataset is input into the trajectory prediction model based on the long short-term memory network. The data is processed sequentially through multiple cell units. In each cell unit, the temporal data is processed through a forget gate, an input gate, and an output gate to update the cell state and hidden state. After passing through a fully connected layer and a softmax layer, a hybrid density network is used to output a trajectory probability distribution that follows a Gaussian mixture distribution. Based on the trajectory probability distribution of the output, the target reward function is transformed into a log-likelihood loss form, and iterative optimization is performed on the training set until the model converges on the validation set.

7. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, The vehicle operation scenarios include conflict scenarios, signal scenarios, and following scenarios; the scenario judgment method specifically includes establishing scoring formulas for conflict scenarios, signal scenarios, and following scenarios respectively, and taking the scenario with the highest score as the current scenario category of the vehicle; The scoring formula is constructed based on the following principles: when a vehicle is operating in an intersection area and there is a risk of its trajectory intersecting with that of other vehicles, conflict scenarios should be given priority; when a vehicle is in a traffic light-controlled intersection area and its operation is directly affected by the traffic lights, it should be classified as a signal scenario; when a vehicle is in a straight road or continuous traffic flow scenario and mainly interacts with the vehicle in front, it should be classified as a following scenario.

8. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 7, characterized in that, In conflict scenarios, the time and location of the conflict are first calculated. The priority probability is evaluated based on the trajectory statistical characteristics of the vehicle and other vehicles. The comprehensive safety probability of each vehicle is calculated by combining the priority probabilities of the vehicle and other vehicles. When the comprehensive safety probability is greater than the preset safety threshold, a priority instruction is sent to the vehicle; otherwise, a yield instruction is sent. In signal scenarios, the vehicle's position distribution at the red light time is obtained based on the future trajectory probability distribution. The vehicle's passage probability is calculated based on the relative position of the vehicle and the stop line. When the vehicle's passage probability is greater than the preset passage threshold, a passage permission instruction is sent to the vehicle. When the vehicle's passage probability is less than the preset stop threshold, a passage prohibition instruction is sent to the vehicle. When the vehicle's passage probability is between the preset stop threshold and the preset passage threshold, a deceleration instruction is sent to the vehicle. In a car-following scenario, a safety threshold is first calculated based on the mean and standard deviation of the distance between vehicles. Then, a comprehensive risk index is calculated based on the safety threshold and the relative distance between the vehicle in front and the vehicle following. When the risk index is less than 0, a deceleration command is issued to the vehicle. When the risk index and the relative speed are greater than the preset threshold, an acceleration command is issued to the vehicle. Otherwise, a speed-maintaining command is issued.

9. The method for generating end-side vehicle dispatching guidance instructions considering safety, efficiency, and comfort objectives according to claim 1, characterized in that, In step S5, the dispatch guidance instructions include standardized control commands and natural language guidance prompts, which are distributed to the target vehicle through the V2X communication system. At the same time, a feedback mechanism is established to continuously optimize the dispatch guidance instruction generation strategy.

10. A vehicle dispatching and guidance instruction generation system considering safety, efficiency, and comfort objectives, for implementing the method as described in any one of claims 1-9, characterized in that, include: Data acquisition and processing module: used to acquire real-time trajectory data, historical trajectory data, vehicle type and driving status collected by vehicle sensors within the target area, and to preprocess the data to obtain an initial multi-dimensional feature matrix; Environmental constraint fusion module: used to acquire real-time traffic light status data, road construction control data, meteorological environment data and special event announcements, and perform spatiotemporal alignment and feature fusion with the initial multi-dimensional feature matrix to obtain an enhanced dataset containing environmental constraints; Model building and training module: used to design an objective function that includes safety, efficiency and comfort metrics, build a trajectory prediction model based on a long short-term memory network, and train the trajectory prediction model based on the enhanced dataset; Trajectory distribution prediction module: This module combines real-time environmental data with a trained trajectory prediction model to generate a future trajectory probability distribution that satisfies multi-objective optimization. Instruction generation and feedback module: Based on the future trajectory probability distribution result, determine the vehicle operation scenario, encode the future trajectory probability distribution into an executable scheduling and guidance instruction sequence according to the vehicle operation scenario, and distribute it to the target vehicle.

Images