A multi-intersection bus priority real-time optimization method based on vehicle-road cloud cooperation

Abstract

The application discloses a kind of based on multi-intersection bus priority real-time optimization method of car-road cloud cooperation, obtains global car-road state information by V2X communication interface, and constructs multi-intersection bus priority control optimization model according to global car-road state information, the model utilizes hybrid initialization based on space-time constraint reverse propagation, solves the initial optimization lag problem caused by sparse solution space.After crossing and variation, introduce the deterministic repair operator based on boundary projection, solve the calculation time jitter problem caused by non-deterministic random repair.Adopt the adaptive adjustment of evolution parameter and the adaptive variation of evolution state perception, solve the dynamic imbalance problem of exploration and development capability in whole cycle.The method reduces the invalid calculation of single iteration, improves the optimization efficiency and the stability of calculation time of algorithm, realizes the matching of algorithm optimization speed and C-V2X communication speed, meets the real-time and robustness requirements of intelligent transportation system.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration. Background Technology

[0002] With the high-quality development of my country's economy and the rise of the new energy vehicle industry, the number of motor vehicles has experienced explosive growth. Statistics show that the number of civilian vehicles in China exceeded 360 million in 2023. Driven by policies such as purchase tax reductions and trade-in programs, the penetration rate of new energy vehicles continues to climb, but the resulting contradiction between urban road supply and demand is becoming increasingly acute. According to Gaode Maps' traffic big data monitoring, the average speed of vehicles on the road network during morning and evening rush hours in key cities has fallen below 26 km / h, and traffic congestion has become a bottleneck restricting urban operational efficiency. Against this backdrop, building an efficient Intelligent Transportation System (ITS) has become a key path to solving the congestion problem.

[0003] In recent years, with the deep integration of 5G communication technology and transportation systems, vehicle-to-everything (V2X) technology is undergoing a leapfrog evolution from single-vehicle intelligence to vehicle-road-cloud collaboration, and then to integrated vehicle-road-cloud systems. Since 2019, my country has led the establishment of a 5G-based C-V2X standard system, which integrates the perception data and computing resources of vehicles, roadside infrastructure, and cloud control platforms to achieve millisecond-level ultra-low latency data interaction. This provides a solid physical and link layer foundation for multi-intersection collaborative bus priority passage, making dynamic and real-time traffic flow optimization within a region possible.

[0004] Although C-V2X technology has solved the real-time problem of information interaction, the overall decision-making response speed of the vehicle-road-cloud integrated system is still limited by the computational efficiency of the core optimization algorithm. In the multi-intersection bus priority cooperative control problem, the Genetic Algorithm (GA) is widely used due to its parallel search capability. However, existing research mostly adopts the traditional paradigm of "random generation first, then simulation verification," which exposes the problems of blind initialization and defects in the evolutionary iteration mechanism when dealing with bus priority problems with strict spatiotemporal window constraints.

[0005] First, theoretical flaws in the initialization process cause optimization delays in the early stages of the algorithm. In multi-intersection cooperative control, the solution space expands exponentially with the number of intersections, while the feasible region satisfying bus priority is extremely sparse within the entire solution space. The forward random sampling strategy used in traditional GA ignores the inherent strong spatiotemporal constraints of signal control, resulting in the vast majority of individuals in the initial population being invalid solutions. This forces the algorithm to consume a large amount of computing power in the early stages of iteration to evaluate and eliminate invalid individuals, slowing down the optimization process.

[0006] Besides initialization issues, traditional genetic algorithms suffer from dual defects in timeliness and effectiveness under the strong constraints of vehicle-road-cloud integrated systems. First, there's the random fluctuation in computation time: existing algorithms often employ rejection sampling strategies to handle infeasible mutations (i.e., a "mutation-verification-failure-retry" loop). This non-deterministic retry logic leads to drastic fluctuations in the time consumed per iteration, making it difficult to meet the stringent millisecond-level real-time response requirements of vehicle-road-cloud collaborative control. Second, there's an imbalance between the supply and demand of search capabilities: traditional crossover and mutation operators typically use fixed parameters or simple random strategies. In the early stages of evolution, a fixed low mutation rate limits the algorithm's ability to explore the global optimum and escape local optima; while in the later stages, large-step random perturbations disrupt the stability of converged solutions, lacking refined local exploration capabilities. This operator mechanism cannot adapt to the dynamic changes in the population's evolutionary state, resulting in low convergence accuracy and long optimization generations. Summary of the Invention

[0007] The purpose of this invention is to provide a real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration, which can significantly reduce invalid computation, achieve matching between algorithm optimization speed and C-V2X communication speed, meet the real-time and robustness requirements of intelligent transportation systems, and solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration includes the following steps:

[0010] S1: Obtain global vehicle-road status information through the V2X communication interface, including multi-intersection traffic light timing schemes, current phase and duration of traffic lights, and real-time movement status of buses;

[0011] S2: Construct a multi-intersection bus priority control optimization model based on a genetic algorithm according to global vehicle-road state information, as detailed below;

[0012] S201: Define decision variables: Assume the multi-way signal control system has a total of At the intersection, for the first intersection ,Include Each signal phase defines the green light duration decision vector for this intersection. for

[0013] (1)

[0014] in Indicates the first The first intersection The green light duration for each phase, assuming the yellow / red light loss time during each phase transition is a fixed value. Then the total loss for a single signal cycle is Signal period Defined as:

[0015] (2)

[0016] S202: Constructing the objective function: Based on the hard constraints of "reverse initialization" and "repair operator" to ensure bus priority, the optimization objective is to achieve bus priority while minimizing the negative impact on background traffic flow.

[0017]

[0018] S203: Define constraints: decision variables The following two types of constraints must be satisfied simultaneously:

[0019] 1) Basic physical constraints for signal control, used to ensure that traffic light timings comply with basic traffic engineering specifications, and to prevent safety hazards caused by extremely short green lights or queue overflow caused by extremely long red lights, i.e., phase duration boundary constraints:

[0020] (5)

[0021] in, To allow pedestrians to cross the street and vehicles to start, a minimum green light time of 10-15 seconds is adopted. This refers to the maximum green light time limit;

[0022] 2) Hard constraints on bus priority in time and space: When a bus arrives at the stop line at an intersection, the target phase must be within a valid green light window. Assume the bus is expected to arrive at the [number missing] stop line. The time at each intersection is If the target phase is Then the following is required:

[0023] (6)

[0024] In formula (6): This indicates that buses will be in the future. One signal cycle, and The first The start and end times of the green light for cyclical goals are decision variables. The function; This is a safety buffer period to ensure that vehicles do not pass through at the edge of the green light's start / stop point, thus avoiding the risk of running a yellow light;

[0025] S3: Based on the multi-intersection bus priority control optimization model, output the globally optimal traffic light phase timing scheme to the roadside controller for execution.

[0026] Furthermore, the reverse initialization in S202 specifically involves reverse spatial reconstruction and population initialization: after obtaining the constraint boundary of the objective function, a hybrid initialization stage based on the reverse propagation of spatiotemporal constraints is entered. By decoupling the decision variables, the feasible interval of the key phase that satisfies the bus priority constraint is derived in reverse using a system of linear inequalities. The initial population generated by sampling within this interval ensures that 100% of the individuals fall within the feasible region.

[0027] Furthermore, the hybrid initialization process based on spatiotemporal constraint backpropagation is as follows:

[0028] 1) Variable decoupling and non-target phase adaptive boundary shrinkage sampling: First, the intersection... The phase set is divided into target phase and non-target phase sets. If directly in the boundary constraints Random sampling of non-target phases may result in an unsolvable situation because the time allocated to non-target phases is too long, which may squeeze the available time when deriving the target phase later.

[0029] To address this, an adaptive boundary contraction mechanism for non-target phases based on kinematic prediction is introduced before sampling. Taking the instant when the system triggers collaborative decision-making as the origin of the time coordinate, to ensure that the bus can enter the green light window of the target phase without deceleration, the total green light duration of all non-target phases experienced during the transition from the current phase to the target phase must have a physical upper limit.

[0030] Let the set of non-target phases that need to be crossed be . Phase index is Target phase The number of times the yellow light / all-red light switch was experienced was Under extreme conditions, when all non-target phases are taken as the lower bound of the minimum green light duration allowed by traffic engineering. At that time, the earliest activation time of the target phase :

[0031] (7)

[0032] In equation (7) For the first The remaining green light duration for the currently executing phase at each intersection is used to define the first... Free time budget for signal cycles at each intersection Under the condition of meeting green wave traffic requirements, the vehicle-road-cloud cooperative control system can allocate a maximum of extra green light time surplus to non-target phases. :

[0033] (8)

[0034] Using this time budget, a global search upper bound for non-target phases is established. Dynamic clipping is performed to derive an adaptive new upper bound after spatiotemporal constraint clipping. :

[0035] (9)

[0036] After boundary contraction is completed, the algorithm model randomly generates the green light duration for non-target phases within the new safety boundary constraints. :

[0037] (10)

[0038] Through non-target phase adaptive boundary contraction sampling based on kinematic prediction, the first The total duration of the non-target phases at each intersection is converted into a physically determined known constant. :

[0039] (11)

[0040] 2) Reverse derivation of the feasible interval of the critical phase: taking the intersection as an example. Target phase For example, let its decision variables be... Each phase ends with a fixed yellow / red light duration, recorded as follows: Assume that there are a total of If there are 1 phase, then the fixed total loss time within a single signal period is 1 / 2. According to the bus arrival time Remaining time with the current phase The relationship divides the timeline into different periodic intervals. Let's discuss:

[0041] Case A ( If the bus catches the current green light and arrives within the current cycle, that is... And meets safety buffer requirements. If the timing is not adjusted for the next cycle, the bus can pass directly. The value of only needs to satisfy the basic boundary constraints, and the feasible interval is:

[0042] (12)

[0043] Case B ( ): Public transport in the future Each cycle passes, when When the distance is large, the bus will miss the current green light and will need to wait until the next... It passes through the target phase window of one signal cycle, where... Indicates the next cycle. This indicates the period after the next, and so on. In this context, the first... The moment the green light for the secondary target is activated. It consists of three parts: the remaining time of the current phase. ,forward Total green light duration for the next non-target phase ,forward Green light duration for the secondary target phase and Accumulated yellow light time over a period of time The details are as follows:

[0044] 1) When At this time, the bus arrives in the next cycle: the start time of the target phase is no longer affected by the future. The influence is only affected by the remaining time of the current cycle and the non-target phase, therefore there is no upper bound constraint determined by the start time, and it is only limited by the physical upper bound:

[0045] (13)

[0046] 2) When hour:

[0047] (14)

[0048] No. The end of the green light for the secondary target This is the start time plus the duration of the current green light. :

[0049] (15)

[0050] According to the criteria for determining bus priority, the following must be met:

[0051] (16)

[0052] Will and Substituting into equation (13), we can obtain the following about The system of bilateral linear inequalities: The upper bound constraint is determined by the green light start time: ensuring that the green light is on when the bus arrives:

[0053] (17)

[0054] when Solving for:

[0055] (18)

[0056] The lower bound constraint is determined by the green light end time: ensuring that the green light has not yet ended when the bus arrives.

[0057] (19)

[0058] Solving for:

[0059] (20)

[0060] (twenty one)

[0061] In summary, based on the derived theoretical upper limit... and the lower bound of the theory For a given periodicity, Target phase Constructive feasible interval for:

[0062] (twenty two)

[0063] in, and These are the derived lower bound constraint value and upper bound constraint value, respectively;

[0064] 3) Adaptive construction generation: If the calculated lower bound constraint value is greater than the upper bound constraint value, then Algorithm traversal , To determine the maximum number of prediction periods, typically 2-3, calculate the set of all non-empty intervals. ;like Then, randomly select an interval and uniformly sample integers from it. Ensure the generated decision vector It enables buses to pass smoothly through a future green light window, and strictly conforms to the physical law of the periodic wear and tear of traffic lights.

[0065] Furthermore, the repair operator in S202 is specifically based on the standard simulated binary crossover SBX operator, introducing a phased adaptive control of the distribution index, and coordinating the regulation of the SBX crossover operator and the mutation operator based on deterministic spatiotemporal constraints.

[0066] Furthermore, the control method for the SBX crossover operator is as follows:

[0067] 1) The principle of the SBX operator is as follows:

[0068] For the selected parent Simulate binary crossover to generate offspring For each decision variable First, generate random numbers. Calculate the expansion factor :

[0069] (twenty three)

[0070] Offspring generation formula:

[0071] (twenty four)

[0072] 2) Phased adaptive parameter adjustment:

[0073] The process is divided into three stages—early, middle, and late—based on the number of evolutionary generations, and the distribution index is dynamically adjusted. ,because The variance of the offspring distribution is inversely proportional to the variance of the offspring distribution, gradually transitioning from a wide distribution in the early stages to a narrow distribution in the later stages, using the variation with the number of generations. Distribution index of change :

[0074] (25)

[0075] in, The maximum number of generations, the smaller It will produce offspring that are distantly related to their parents, and larger offspring. This will produce offspring that are similar to their parents;

[0076] 3) Adaptive crossover probability: The crossover probability is dynamically calculated based on the fitness value of the parent individual relative to the average fitness of the population. To protect superior individuals and accelerate the elimination of low-fit individuals, we assume... For the maximum fitness of the population, The average fitness of the population. The larger fitness value among the two parent generations to be crossovered:

[0077] (26)

[0078] in, and For adaptive parameters that decrease with evolutionary stage, if the fitness value of the two parents to be crossbred is larger... When the fitness of a species is greater than the average fitness of the population, a higher crossover probability is adopted; otherwise, a lower crossover probability is adopted.

[0079] Furthermore, the feature is that the control method for the mutation operator employs evolutionary state-aware adaptive mutation, specifically as follows:

[0080] 1) Evolutionary state perception mutation probability: Construct a dynamic mutation probability model based on population fitness distribution and evolutionary process, assuming the... The first generation of the population The fitness of each individual is The average fitness of the population is Maximum fitness is Define evolutionary process factors ,in For the maximum number of iterations, the th The probability of variation for each individual Defined as:

[0081] (27)

[0082] in, These are the upper and lower bounds of the basic mutation rate and the radical mutation rate, respectively. The damping coefficient decreases with each generation of evolution and is used to reduce the perturbation frequency in the later stages of convergence, thus protecting the obtained superior gene patterns.

[0083] 2) Multi-strategy hybrid perturbation model: based on evolutionary process factors Dynamic switching of mutation operators Mathematical form:

[0084] (28)

[0085] This represents the mutated gene value. The gene values ​​before mutation are used; this mechanism initially works through a uniform distribution. Perform large-scale jumps, and later utilize variance. Gaussian distribution with algebraic decay A small neighborhood search is performed to achieve adaptive adjustment of the search step size.

[0086] Furthermore, deterministic spatiotemporal constraints utilize mathematical projection methods to forcibly map out-of-bounds individuals back to the nearest constraint boundary, eliminating the uncertainty caused by random retries, as detailed below:

[0087] Under strong constraints, intermediate solutions generated by mutation and crossover operators are considered. May violate minimum / maximum green light duration constraints Instead of using a random retry strategy, a repair operator is adopted. Project it to the nearest constraint boundary using the following piecewise function:

[0088] (29)

[0089] in, This is the final solution after processing by the repair operator. If an individual falls within the feasible region, the operator does not intervene and preserves the diversity characteristics of genetic evolution; if an individual falls to the left of the feasible region, the operator forcibly pulls it back to the lower boundary. If an individual falls to the right of the feasible region, the operator forcibly pulls it back to the upper boundary. .

[0090] Compared with the prior art, the beneficial effects of the present invention are:

[0091] To address the stringent requirements of real-time performance and robustness in multi-intersection bus priority control within a vehicle-road-cloud integrated environment, this invention presents a real-time optimization method for multi-intersection bus priority based on vehicle-road-cloud collaboration. This method not only achieves a several-fold increase in optimization speed but also demonstrates strong determinism in the distribution of computation time. It alleviates the time-consuming problem caused by random trial and error and can meet the standard of millisecond-level rigid response in intelligent transportation systems under 5G-CV2X communication environments. Attached Figure Description

[0092] Figure 1 This is a flowchart of the real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration according to the present invention.

[0093] Figure 2 This is a schematic diagram of the principle of the present invention based on spatiotemporal constraint reverse propagation;

[0094] Figure 3 The simulation test scene image used in the simulation experiment of this invention is located on Dagongshan Road, Yijiang District, Wuhu City.

[0095] Figure 4 This is a line graph comparing the average computation time of the algorithm before and after the improvement of this invention.

[0096] Figure 5 This is a line graph comparing the standard deviation of the algorithm's single-run time before and after the improvement of this invention. Detailed Implementation

[0097] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0098] This invention provides a real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration. It obtains global vehicle-road status information through a V2X communication interface, including the traffic light timing scheme at multiple intersections, the current phase and duration of the traffic lights, and the real-time motion status of buses. Based on the global vehicle-road status information, it constructs a multi-intersection bus priority control optimization model based on a genetic algorithm, as detailed below.

[0099] 1) Define decision variables: Assume the multi-intersection signal control system has a total of At the intersection, for the first intersection ,Include Each signal phase defines the green light duration decision vector for this intersection. for

[0100] (1)

[0101] in Indicates the first The first intersection The green light duration for each phase, assuming the yellow / red light loss time during each phase transition is a fixed value. Then the total loss for a single signal cycle is Signal period Defined as:

[0102] (2)

[0103] 2) Constructing the objective function: Considering that bus priority control may interfere with other vehicles, the optimization objective of this invention is to maintain the original traffic flow steady state as much as possible, reduce drastic disturbances to non-priority phases, and minimize the negative impact on background traffic flow while achieving bus priority, based on the hard constraints of ensuring bus priority passage through "reverse initialization" and "repair operators".

[0104]

[0105] 3) Analysis of Constraints: Bus priority signal control at multiple intersections is a typical strongly constrained optimization problem. To ensure the right-of-way for buses and maintain the basic logic of signal control, the decision variables... The following two types of constraints must be satisfied simultaneously:

[0106] 1) Basic physical constraints for signal control: These constraints are used to ensure that signal timings comply with basic traffic engineering specifications and prevent safety hazards caused by extremely short green lights or queue overflow caused by extremely long red lights, i.e., phase duration boundary constraints:

[0107] (5)

[0108] in, To allow pedestrians to cross the street and vehicles to start, the minimum green light time is typically set at 10-15 seconds. This refers to the maximum green light time limit;

[0109] 2) Bus Priority Spatiotemporal Hard Constraints: This is the core constraint that the algorithm of this invention must strictly satisfy. When a bus arrives at the stop line at an intersection, the target phase must be within a valid green light window. Assume the bus is expected to arrive at the... The time at each intersection is If the target phase is Then the following is required:

[0110] (6)

[0111] In formula (6): This indicates that buses will be in the future. One signal cycle, and The first The start and end times of the green light for cyclical goals are decision variables. The function; This is a safety buffer period used to ensure that vehicles do not pass through at the edge of the green light's start / stop point, thus avoiding the risk of running a yellow light.

[0112] The aforementioned inequalities constitute an extremely narrow feasible region, which conventional genetic algorithms often struggle to directly target with random initialization. This invention introduces a backpropagation strategy, and the entire execution process consists of a closed-loop system comprised of three stages: data awareness, initial construction, and iterative evolution. To further explain the construction principle of the above model, as follows... Figure 1 As shown, the details are as follows:

[0113] S1: First, global vehicle-road status information is obtained through the V2X communication interface, including the traffic light timing scheme at multiple intersections, the current phase and duration of the traffic lights, and the real-time motion status (position, speed, acceleration) of buses. Based on this data, the system uses kinematic equations to predict the arrival time of buses and calculates the safe green light time window that the target phase must satisfy, providing rigid spatiotemporal constraints for subsequent optimization.

[0114] S2: Reverse Space Reconstruction and Population Initialization. After obtaining the constraint boundaries, the algorithm enters a hybrid initialization phase based on the reverse propagation of spatiotemporal constraints. Addressing the pain point of extremely sparse feasible solutions in the high-dimensional solution space, the algorithm abandons the traditional forward random generation mode and instead adopts a "reverse construction" strategy. By decoupling the decision variables, the algorithm uses a system of linear inequalities to deduce the key phase feasible interval that satisfies the bus priority constraint in reverse. The initial population sampled within this interval ensures that 100% of individuals fall within the feasible region, thereby alleviating the problem of ineffective iterations in the initialization phase and solving the problem of initial optimization lag in the algorithm.

[0115] S3: Fitness Assessment and Benchmark Establishment: Calculate the fitness value of each individual in the initial population and record the optimal chromosome of the current generation. Simultaneously, set the maximum number of generations as the criterion for algorithm termination;

[0116] S4: Adaptive Adjustment of Evolutionary Parameters (Loop Start Point) The first step in entering the evolutionary loop is the adaptive adjustment of evolutionary parameters. The algorithm's built-in evolutionary state controller dynamically calculates the evolutionary stage factor based on the current iteration progress and population fitness distribution. This factor directly determines the magnitude of subsequent crossover and mutation probabilities, providing parameter support for the algorithm to focus on "exploration" or "development" at different stages.

[0117] S5: Survival of the fittest selection is based on calculated fitness. It uses tournament selection and elite retention to select superior individuals from the parents, builds a mating pool, and preserves excellent gene patterns.

[0118] S6: Simulated binary crossover (SBX) is performed on the selected individuals using adaptive parameter-based SBX crossover. In this case, the distribution exponent of the crossover operator is no longer a fixed value, but is adaptively adjusted based on the stage factor calculated in S4. A wide distribution is used in the early stages of evolution to maintain diversity, while a narrow distribution is used in the later stages to accelerate convergence.

[0119] S7: Deterministic Repair Operator Based on Boundary Projection: For intermediate solutions that may violate period conservation or green light duration constraints during crossover operations, the algorithm immediately executes a deterministic repair operator based on boundary projection. Using mathematical projection methods, out-of-bounds individuals are mapped back to the nearest constraint boundary. This step eliminates the uncertainty introduced by random retries.

[0120] S8: Evolutionary State-Aware Adaptive Mutation: Perform mutation operations. The mutation strategy here is driven by the evolutionary state-aware mechanism: if the population is detected to be trapped in a local optimum (premature convergence), the mutation rate is increased and a large step size perturbation is used; if it is in the convergence phase, Gaussian perturbation is used. This solves the problem of the imbalance between supply and demand for search capabilities.

[0121] S9: Deterministic Spatiotemporal Constraint Repair of Mutated Operators: Mutation operations can also violate constraints (e.g., causing the sum of phase and time to be unequal to the period). Therefore, Figure 1 The deterministic spatiotemporal constraint repair operator is invoked again to perform a second correction on the mutated individuals, ensuring that all individuals entering the next generation are strictly feasible both physically and logically.

[0122] S10: Termination determination and output check whether the maximum number of iterations has been reached. If the condition is met, output the globally optimal traffic light phase timing scheme to the roadside controller for execution; otherwise, return to S4 and enter the next iteration.

[0123] In the aforementioned construction principle, this invention constructs a hybrid initialization strategy based on the backward propagation of spatiotemporal constraints. Addressing the pain point that feasible solutions satisfying the hard spatiotemporal constraint of bus priority (Equation 6) are extremely sparse in the high-dimensional solution space, the traditional initialization mode of "random generation first, then verification and filtering" leads to the algorithm getting stuck in the problem of initial optimization lag. Therefore, this embodiment proposes a hybrid initialization strategy based on "variable decoupling-backward mapping." This strategy does not rely on blind trial and error, but instead uses a system of linear inequalities to directly solve for the feasible intervals of key variables, as detailed below:

[0124] 1) Non-target phase adaptive boundary contraction sampling based on kinematic prediction: The first algorithm first samples the intersection The phase set is divided into target phase and non-target phase sets. If directly in the boundary constraints Random sampling of non-target phases may result in an unsolvable situation because the time allocated to non-target phases is too long, which may squeeze the available time when deriving the target phase later.

[0125] To address this, an adaptive boundary contraction mechanism for non-target phases based on kinematic prediction is introduced before sampling. Taking the instant when the system triggers collaborative decision-making as the origin of the time coordinate, to ensure that the bus can enter the green light window of the target phase without deceleration, the total green light duration of all non-target phases experienced during the transition from the current phase to the target phase must have a physical upper limit.

[0126] Let the set of non-target phases that need to be crossed be . Phase index is Target phase The number of times the yellow light / all-red light switch was experienced was Under extreme conditions, when all non-target phases are taken as the lower bound of the minimum green light duration allowed by traffic engineering. At that time, the earliest activation time of the target phase :

[0127] (7)

[0128] In formula (7) No. The remaining green light duration for the current phase at each intersection.

[0129] Accordingly, the first is defined Free time budget for signal cycles at each intersection The characterization system represents the maximum amount of extra green time surplus that can be allocated to non-target phases under the condition of satisfying green wave passage. :

[0130] (8)

[0131] Using this time budget, a global search upper bound for non-target phases is established. Dynamic clipping is performed to derive an adaptive new upper bound after spatiotemporal constraint clipping. :

[0132] (9)

[0133] After completing the boundary contraction, the algorithm randomly generates the green light duration for non-target phases within the new safety boundary constraints. :

[0134] (10)

[0135] Through this non-target phase adaptive boundary contraction sampling based on kinematic prediction, the first... The total duration of the non-target phases at each intersection is converted into a physically determined known constant. :

[0136] (11)

[0137] 2) Reverse derivation of the feasible interval of the critical phase: taking the intersection as an example. Target phase For example, let its decision variables be... Each phase ends with a fixed yellow / red light duration, recorded as follows: Assume that there are a total of If there are 1 phase, then the fixed total loss time within a single signal period is 1 / 2. According to the bus arrival time Remaining time with the current phase The relationship divides the timeline into different periodic intervals. Let's discuss:

[0138] Case A ( If the bus catches the current green light and arrives within the current cycle, that is... And meets safety buffer requirements. If the timing is not adjusted for the next cycle, the bus can pass directly. The value of only needs to satisfy the basic boundary constraints, and the feasible interval is:

[0139] (12)

[0140] Case B ( ): Public transport in the future Each cycle passes, when When the distance is large, the bus will miss the current green light and will need to wait until the next... It passes through the target phase window of one signal cycle, where... Indicates the next cycle. This indicates the period after the next, and so on. In this context, the first... The moment the green light for the secondary target is activated. It consists of three parts: the remaining time of the current phase. ,forward Total green light duration for the next non-target phase ,forward Green light duration for the secondary target phase and Accumulated yellow light time over a period of time The details are as follows:

[0141] a. When Time (i.e., the bus arrives in the next cycle): At this point, the start time of the target phase is no longer affected by the future. The influence of the current cycle's remaining time and non-target phase is only affected, therefore there is no upper bound constraint determined by the start time, and it is only limited by the physical upper bound:

[0142] (13)

[0143] b. When hour:

[0144] (14)

[0145] No. The end of the green light for the secondary target This is the start time plus the duration of the current green light. :

[0146] (15)

[0147] According to the criteria for determining bus priority, the following must be met:

[0148] (16)

[0149] Will and Substituting into equation (13), we can obtain the following about The system of bilateral linear inequalities: The upper bound constraint is determined by the green light start time: ensuring that the green light is on when the bus arrives:

[0150] (17)

[0151] when Solving for:

[0152] (18)

[0153] The lower bound constraint is determined by the green light end time: ensuring that the green light has not yet ended when the bus arrives.

[0154] (19)

[0155] Solving for:

[0156] (20)

[0157] (twenty one)

[0158] In summary, based on the derived theoretical upper limit... and the lower bound of the theory For a given periodicity, Target phase Constructive feasible interval for:

[0159] (twenty two)

[0160] in, and These are the derived lower bound constraint value and upper bound constraint value, respectively;

[0161] 3) Adaptive construction generation: If the calculated lower bound is greater than the upper bound, then Algorithm traversal ( To determine the maximum number of prediction periods (usually 2-3), calculate the set of all non-empty intervals. .like Then, randomly select an interval and uniformly sample integers from it. This process ensures that the generated decision vectors are accurate. It enables buses to pass smoothly through a future green light window, and strictly conforms to the physical law of the periodic wear and tear of traffic lights.

[0162] Figure 2 The image intuitively demonstrates how the above algorithm works from the vehicle arrival time. Reverse derivation of the target phase green light duration The feasible range, combined with Figure 2 visible:

[0163] 1. Spatiotemporal trajectory projection: The blue solid line represents the spatiotemporal trajectory of the bus. According to the kinematic equations, the trajectory intersects the intersection stop line (vertical axis). The intersection of the points determines the estimated arrival time. (Red dot in the illustration).

[0164] 2. Safe passage window construction: To absorb random disturbances and avoid running yellow lights, the algorithm... Safety buffers are introduced at both the front and back. (The area shown is outlined in orange dashed lines). This window The minimum time range that the target green light must cover is defined.

[0165] 3. Phase Composition and Loss Accumulation: The signal status bar at the top displays the three rigid components that constitute the start time of the next cycle:

[0166] (Pink): The remaining green light time that cannot be changed at present;

[0167] (Gray): The sum of non-target phases randomly sampled in the first step of the algorithm, which is now used as a fixed parameter;

[0168] (yellow): Accumulated yellow / red light loss time over a period of time .

[0169] Elastic component: (Green dashed line section): This is the front The duration of the target phase accumulation within each cycle. Core logic: Decision variables. The value not only determines how long the current green light is, but also... The magnification by a factor of 1 determines the future number of times. When does the green light start?

[0170] In the above embodiments, the present invention also constructs an evolutionary strategy of stage adaptation and evolutionary state perception. Addressing the imbalance between the supply and demand of "global exploration capability" and "local development capability" in traditional genetic algorithms under the strong constraints of vehicle-road-cloud integration, the present invention constructs a dual adaptive mechanism including an evolutionary state controller. This mechanism coordinates the regulation of the SBX crossover operator and mutation operator by real-time perception of the evolutionary process and population fitness distribution. It includes:

[0171] 1. Stage-Adaptive SBX Crossover Strategy: Simulated Binary Crossover (SBX) utilizes the probability distribution to simulate the characteristics of single-point crossover, exhibiting excellent search space capabilities. To achieve a smooth transition between global exploration and local development, this invention introduces stage-adaptive control of the distribution exponent based on the standard SBX operator:

[0172] (1) The principle of the SBX operator is as follows: for the selected parent generation Simulate binary crossover to generate offspring For each decision variable First, generate random numbers. Calculate the expansion factor :

[0173] (twenty three)

[0174] Offspring generation formula:

[0175] (twenty four)

[0176] (2) Stage-based adaptive parameter adjustment: The process is divided into three stages—early, middle, and late—based on the number of generations of evolution, and the distribution index is dynamically adjusted. ,because It is inversely proportional to the variance of the generated offspring distribution ( The smaller the size, the wider the distribution. The larger the size, the narrower the distribution (from a wide distribution in the early stages to a narrow distribution in the later stages). This is achieved by using the number of generations of evolution... Distribution index of change :

[0177] (25)

[0178] in, This represents the maximum number of generations. Smaller... This will produce offspring that are distantly related to their parents (maintaining diversity), and larger offspring. This results in offspring that are similar to their parents (fine-grained search);

[0179] (3) Adaptive crossover probability: For minimization optimization problems, in order to avoid destroying the excellent pattern with high fitness, the crossover probability is... The crossover probability should decrease as the fitness value decreases, and should be dynamically calculated based on the relative fitness values ​​of the parent individuals compared to the population average fitness. This is to protect superior individuals and accelerate the elimination of low-fit individuals. Let... For the maximum fitness of the population, The average fitness of the population. The larger fitness value among the two parent generations to be crossovered:

[0180] (26)

[0181] in, and For adaptive parameters that decrease with evolutionary stages (e.g., in the early stages) =0.2, =0.9 (decreases later), if the fitness value of the two parent generations to be crossovered is larger. A higher crossover probability is used when the fitness of the population is greater than the average fitness of the population, and a lower crossover probability is used when the fitness of the population is less than the average fitness of the population. The aim is to gradually reduce the degree of population disturbance as convergence progresses.

[0182] 2. Evolutionary State-Aware Adaptive Mutation: To address the computational latency jitter and low search efficiency issues caused by the single perturbation mode of traditional mutation operators, this embodiment proposes a novel mutation strategy that integrates state awareness, as detailed below:

[0183] 1) Evolutionary state-aware mutation probability: To balance the algorithm's global exploration capability in the early stages of evolution with its local exploitation capability in the later stages, a dynamic mutation probability model based on population fitness distribution and evolutionary progress is constructed. Considering that the optimization objective is to minimize, let the... The first generation of the population The fitness of each individual is The average fitness of the population is Maximum fitness is Define evolutionary process factors. ,in This represents the maximum number of iterations. The probability of variation for each individual Defined as:

[0184] (27)

[0185] in, These are the upper and lower bounds of the basic mutation rate and the radical mutation rate, respectively. The damping coefficient decreases with each generation of evolution and is used to reduce the perturbation frequency in the later stages of convergence, thus protecting the obtained superior gene patterns.

[0186] 2) Multi-strategy hybrid perturbation model: Abandoning the single random perturbation method, it is based on the evolutionary process factor. Dynamic switching of mutation operators Mathematical form:

[0187] (28)

[0188] This represents the mutated gene value. The gene values ​​before mutation are used; this mechanism initially works through a uniform distribution. Perform large-scale jumps, and later utilize variance. Gaussian distribution with algebraic decay A small neighborhood search is performed to achieve adaptive adjustment of the search step size.

[0189] Although the aforementioned adaptive crossover and mutation strategies significantly improve the algorithm's search efficiency, random perturbations can still generate infeasible solutions that violate spatiotemporal constraints under strong constraints. Therefore, a deterministic repair mechanism must be introduced to post-process the evolved individuals, namely, a deterministic repair operator based on boundary projection:

[0190] In the crossover and mutation operations of genetic algorithms, the resulting offspring may exceed the physical boundaries of signal control, leading to excessively short or long green light durations. To address this issue, this invention abandons time-consuming random retry strategies and employs a deterministic repair method based on boundary projection to map infeasible solutions back into the feasible region: Under strong constraints, for intermediate solutions generated by mutation and crossover operators... May violate minimum / maximum green light duration constraints Instead of using a random retry strategy, a repair operator is adopted. Project it to the nearest constraint boundary using the following piecewise function:

[0191] (29)

[0192] in, This is the final solution after processing by the repair operator. If an individual falls within the feasible region, the operator does not intervene and preserves the diversity characteristics of genetic evolution; if an individual falls to the left of the feasible region, the operator forcibly pulls it back to the lower boundary. If an individual falls to the right of the feasible region, the operator forcibly pulls it back to the upper boundary. This deterministic projection mechanism ensures strict physical feasibility for all individuals with minimal modification cost, avoiding the waste of computational resources caused by traditional rejection sampling strategies.

[0193] In summary, this invention utilizes a hybrid initialization strategy based on spatiotemporal constraint backpropagation: During the initialization phase, the feasible intervals of key phases are derived inversely using the vehicle kinematics equations, replacing forward random sampling. This ensures high feasibility of the initial population from the outset, addressing the algorithm initialization optimization delay caused by solution space sparsity. Secondly, a deterministic spatiotemporal constraint repair operator is employed: abandoning the traditional random retry mechanism, a deterministic repair method based on boundary projection is designed to forcibly correct infeasible solutions to the constraint boundaries, eliminating computational time uncertainty and resolving the computational time jitter problem caused by "rejection sampling." Furthermore, in the adaptive mutation of stage-adaptive SBX crossover and evolutionary state perception, an evolutionary state controller is introduced. This dynamically adjusts the SBX crossover distribution index and mutation probability according to the iteration progress, achieving an adaptive balance between the algorithm's global exploration and local development capabilities, resolving the dynamic imbalance between exploration and development capabilities throughout the entire cycle. Through these improvements, invalid computation can be significantly reduced, achieving a match between algorithm optimization speed and C-V2X communication speed, meeting the real-time and robustness requirements of intelligent transportation systems.

[0194] To further verify the feasibility of the method of the present invention, the following simulation experiments and result analysis are provided for further explanation:

[0195] Simulation Environment Setup: To verify the effectiveness of the proposed real-time bus priority optimization method based on vehicle-road-cloud collaboration in a vehicle-road-cloud integrated environment, SUMO (Simulation of UrbanMobility) was selected as the microscopic traffic simulation platform. Using the osmWebWizard.py tool included with SUMO, geographic topology data of Dagongshan Road (the real-world road network) in Yijiang District, Wuhu City was extracted and imported into the simulation environment. Real-time interaction between the algorithm and the simulation environment was implemented using Python based on the TraCI (TrafficControl Interface) interface.

[0196] Road Scenario Description: The selected road section is 1805 meters long and includes two consecutive signalized intersections (Intersection 1 and Intersection 2). The bus travels from east to west, and the key node distances are distributed as follows: distance from the starting point to the stop line at Intersection 1: 810 meters; distance from the starting point to the destination at Intersection 2: 1576 meters. This is a simulation test scenario on Dagongshan Road, Yijiang District, Wuhu City. Figure 3 As shown.

[0197] Experimental parameter settings: The bus travels from east to west, with a maximum speed of 11 m / s and an acceleration of... deceleration After the bus accelerates to its maximum speed, it maintains a constant speed. The traffic lights have eight phases: north-south straight green light, north-south straight yellow light, north-south left turn green light, north-south left turn yellow light, east-west straight green light, east-west straight yellow light, east-west left turn green light, and east-west left turn yellow light. The yellow light duration for both traffic lights number one and number two is 3 seconds. The green light durations for north-south straight, north-south left turn, east-west straight, and east-west left turn on both traffic lights number one and number two are 18 seconds, 18 seconds, 18 seconds, and 18 seconds respectively.

[0198] Algorithm parameters: In global planning algorithms, the population size... The maximum number of iterations is 100. The value range for the green light duration of each phase at the dual intersection is set to 100. The maximum time for straight-ahead traffic (east-west) on main roads has been increased to 50 seconds. (Safe buffer time) The fitness function is defined in the form of equation (4), where the weight parameter is set as follows: The evolutionary stages are divided into exploratory phases based on the process factor. transition period Convergence period During the exploration phase: transition period Convergence period: .

[0199] To verify the computational stability of the improved genetic algorithm based on spatiotemporal constraint backpropagation and deterministic repair under complex spatiotemporal conditions, a dual-intersection full-phase combination test set was constructed. Since each signal-controlled intersection contains four key phases (north-south straight, north-south left turn, east-west straight, and east-west left turn), the dual-intersection system has a total of 16 phase combination states. The experimental setup defines the phase set of a single intersection as follows: The subscripts represent "east-west straight," "east-west left turn," "north-south straight," and "north-south left turn," respectively. The experiment covered all possible combinations of upstream and downstream intersection phases, totaling 16 test scenarios. For each scenario, the algorithm was run 20 times consecutively, recording the CPU time for each optimization and calculating the average time. Due to the large number of experiments, the two tables below only show data from 10 runs, with all units in milliseconds.

[0200] Table 1. Runtime of the standard genetic algorithm on the full-phase combinatorial test set (unit: ms)

[0201] Table 2. Runtime of the improved genetic algorithm on the full-phase combinatorial test set (unit: ms)

[0202]

[0203] The method of the present invention is analyzed based on the above data collection table:

[0204] 1. Computational Efficiency Analysis (Average Time): To visually compare the performance advantages of the improved algorithm, the experiment introduced the standard genetic algorithm as a baseline control group. The standard GA adopts the traditional "random initialization + rejection sampling" strategy.

[0205] From Table 3 and Figure 4 As can be seen, the average computation time of the two algorithms is compared across 16 test scenarios:

[0206] Table 3 Comparison of average algorithm time and performance improvement under different phase combinations

[0207]

[0208] Limitations of the benchmark algorithm: The standard genetic algorithm has an overall average execution time of up to 4470ms across all test scenarios, and under complex phase combinations, the average execution time even exceeds 14s (14161ms). This is mainly because the random initialization strategy of the standard GA is difficult to hit the sparse feasible region, causing the algorithm to get stuck in a large number of invalid trial and error and retry loops in the early stage of iteration, which seriously hinders the real-time response of vehicle-road-cloud cooperative control.

[0209] Advantages of the improved algorithm: In comparison, the improved algorithm proposed in this invention has an average execution time of only 710ms across all scenarios. Thanks to the backpropagation initialization strategy based on spatiotemporal constraints, the algorithm compresses the sparse solution space at the algebraic level and avoids the time consumption caused by random retries during the evolution process through a repair operator based on boundary projection.

[0210] Performance Improvement: Data shows that the improved algorithm achieves an average improvement of approximately 6.3 times in computational efficiency. For the most extreme scenarios, the performance improvement reaches as high as 16.6 times. This means that the improved algorithm can stably complete the collaborative optimization of two intersections within sub-second (<1s) timeframes, fully meeting the stringent real-time requirements of intelligent transportation systems.

[0211] 2. Calculate stability and time delay jitter analysis (standard deviation):

[0212] In practical intelligent traffic control systems, not only is a fast average computation speed of the algorithm required, but also "determinism" in the time of a single computation (i.e., extremely low latency jitter). Therefore, the experiment introduced the standard deviation of the time taken for a single operation as a metric, and the comparison results are shown in Table 4:

[0213] Table 4 Comparison of Algorithm Standard Deviation under Different Phase Combinations

[0214]

[0215] From Table 4 and Figure 5 The data comparison shows that the method proposed in this invention has an advantage in the determinism of computation time:

[0216] In extreme scenarios, the standard deviation of single-computation time is reduced: In complex phase combinations with extremely narrow feasible regions (such as "62", "02", "00", etc.), the genetic algorithm exhibits severe random oscillations in computation time (standard deviation as high as 7000~10500 ms) due to its nondeterministic retry logic. In contrast, the method of this invention introduces a deterministic repair operator based on boundary projection to forcibly pull outbound individuals back to the nearest constraint boundary, successfully compressing the standard deviation of the aforementioned core complex scenarios to about 250ms, reducing latency uncertainty by more than 95%.

[0217] A significant leap in global robustness: From a macro-statistical perspective, the original algorithm had an overall average standard deviation of 3076 ms across 16 scenarios. The method of this invention significantly reduces the standard deviation of single-computation time to 219 ms, with an average jitter reduction of over 92%. This rigorously demonstrates statistically that the method of this invention effectively overcomes the "random jitter" defect of heuristic algorithms, achieving convergence and stability of solution time.

[0218] Reasonable perturbations under extremely low time base: In a few simple phase combinations (such as "22", "24", "44"), although the standard deviation of the method in this invention rebounds slightly, it must be pointed out that its absolute average time has been compressed to the sub-second range of 580~750ms. Under this extremely low time base, the fluctuations at the hundred-millisecond level mainly originate from normal perturbations during the scheduling of the underlying operating system and the early population inverse construction of the algorithm. On the time scale of the overall system, such small perturbations can be completely absorbed and do not affect the rigid deployment of the real-time control scheme.

[0219] In summary, the multi-intersection bus priority real-time optimization method based on vehicle-road-cloud collaboration provided by this invention not only effectively improves optimization efficiency but also enhances the consistency of computation time distribution. This method alleviates the problem of excessive time consumption caused by random trial and error by mapping infeasible solutions to the feasible region boundary, fully meeting the stringent standard of millisecond-level rigid response for intelligent transportation systems in the 5G-CV2X communication environment.

[0220] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration, characterized in that, Includes the following steps: S1: Obtain global vehicle-road status information through the V2X communication interface, including multi-intersection traffic light timing schemes, current phase and duration of traffic lights, and real-time movement status of buses; S2: Construct a multi-intersection bus priority control optimization model based on a genetic algorithm according to global vehicle-road state information, as detailed below; S201: Define decision variables: Assume the multi-way signal control system has a total of At the intersection, for the first intersection ,Include Each signal phase defines the green light duration decision vector for this intersection. for （1） in Indicates the first The first intersection The green light duration for each phase, assuming the yellow / red light loss time during each phase transition is a fixed value. Then the total loss for a single signal cycle is Signal period Defined as: （2） S202: Constructing the objective function: Based on the hard constraint of ensuring bus priority passage through the reverse initialization and repair operators, the optimization objective is to achieve bus priority while minimizing the negative impact on background traffic flow. S203: Define constraints: decision variables The following two types of constraints must be satisfied simultaneously: 1) Basic physical constraints for signal control, used to ensure that traffic light timings comply with basic traffic engineering specifications, and to prevent safety hazards caused by extremely short green lights or queue overflow caused by extremely long red lights, i.e., phase duration boundary constraints: （5） in, To allow pedestrians to cross the street and vehicles to start, a minimum green light time of 10-15 seconds is adopted. This refers to the maximum green light time limit; 2) Hard constraints on bus priority in time and space: When a bus arrives at the stop line at an intersection, the target phase must be within a valid green light window. Assume the bus is expected to arrive at the [number missing] stop line. The time at each intersection is If the target phase is Then the following is required: （6） In formula (6): This indicates that buses will be in the future. One signal cycle, and The first The start and end times of the green light for cyclical goals are decision variables. The function; This is a safety buffer period to ensure that vehicles do not pass through at the edge of the green light's start / stop point, thus avoiding the risk of running a yellow light; S3: Based on the multi-intersection bus priority control optimization model, output the globally optimal traffic light phase timing scheme to the roadside controller for execution.

2. The method for real-time optimization of bus priority at multiple intersections based on vehicle-road-cloud collaboration as described in claim 1, characterized in that: The reverse initialization in S202 specifically involves reverse spatial reconstruction and population initialization: After obtaining the constraint boundary of the objective function, a hybrid initialization stage based on the reverse propagation of spatiotemporal constraints is entered. By decoupling the decision variables, the feasible interval of the key phase that satisfies the bus priority constraint is derived in reverse using a system of linear inequalities. The initial population generated by sampling within this interval ensures that 100% of the individuals fall within the feasible region.

3. The method for real-time optimization of bus priority at multiple intersections based on vehicle-road-cloud collaboration as described in claim 2, characterized in that: The hybrid initialization process based on spatiotemporal constraint backpropagation is as follows: 1) Non-target phase adaptive boundary contraction sampling based on kinematic prediction: First, the intersection... The phase set is divided into target phase and non-target phase set. The moment when the vehicle-road-cloud cooperative control system triggers the cooperative decision based on the distance threshold between the bus and the intersection is taken as the origin of the time coordinate. In order to ensure that the bus can enter the green light window of the target phase without deceleration, the total green light duration of all non-target phases experienced when switching from the current phase to the target phase must have a physical upper limit: Let the set of non-target phases that need to be crossed be . Phase index is Target phase The number of times the yellow light / all-red light switch was experienced was Under extreme conditions, when all non-target phases are taken as the lower bound of the minimum green light duration allowed by traffic engineering. At that time, the earliest activation time of the target phase : （7） In equation (7) For the first The remaining green light duration for the currently executing phase at each intersection is used to define the first... Free time budget for signal cycles at each intersection Under the condition of meeting green wave traffic requirements, the vehicle-road-cloud cooperative control system can allocate a maximum of extra green light time surplus to non-target phases. : （8） Using this time budget, a global search upper bound for non-target phases is established. Dynamic clipping is performed to derive an adaptive new upper bound after spatiotemporal constraint clipping. : （9） After boundary contraction is completed, the algorithm model randomly generates the green light duration for non-target phases within the new safety boundary constraints. : （10） Through non-target phase adaptive boundary contraction sampling based on kinematic prediction, the first The total duration of the non-target phases at each intersection is converted into a physically determined known constant. : （11） 2) Reverse derivation of the feasible interval of the critical phase: taking the intersection as an example. Target phase For example, let its decision variables be... Each phase ends with a fixed yellow / red light duration, recorded as follows: Assume that there are a total of If there are 1 phase, then the fixed total loss time within a single signal period is 1 / 2. According to the bus arrival time Remaining time with the current phase The relationship divides the timeline into different periodic intervals. Let's discuss: Case A ( If the bus catches the current green light and arrives within the current cycle, that is... And meets safety buffer requirements. If the timing is not adjusted for the next cycle, the bus can pass directly. The value of only needs to satisfy the basic boundary constraints, and the feasible interval is: （12） Case B ( ): Public transport in the future Each cycle passes, when When the distance is large, the bus will miss the current green light and will need to wait until the next... It passes through the target phase window of one signal cycle, where... Indicates the next cycle. This indicates the period after the next, and so on. In this context, the first... The moment the green light for the secondary target is activated. It consists of three parts: the remaining time of the current phase. ,forward Total green light duration for the next non-target phase ,forward Green light duration for the secondary target phase and Accumulated yellow light time over a period of time The details are as follows: 1) When At this time, the bus arrives in the next cycle: the start time of the target phase is no longer affected by the future. The influence is only affected by the remaining time of the current cycle and the non-target phase, therefore there is no upper bound constraint determined by the start time, and it is only limited by the physical upper bound: （13） 2) When hour: （14） No. The end of the green light for the secondary target This is the start time plus the duration of the current green light. : （15） According to the criteria for determining bus priority, the following must be met: （16） Will and Substituting into equation (13), we can obtain the following about The system of bilateral linear inequalities: The upper bound constraint is determined by the green light start time: ensuring that the green light is on when the bus arrives: （17） when Solving for: （18） The lower bound constraint is determined by the green light end time: ensuring that the green light has not yet ended when the bus arrives. （19） Solving for: （20） （21） In summary, based on the derived theoretical upper limit... and the lower bound of the theory For a given periodicity, Target phase Constructive feasible interval for: （22） in, and These are the derived lower bound constraint value and upper bound constraint value, respectively; 3) Adaptive construction generation: If the calculated lower bound constraint value is greater than the upper bound constraint value, then Algorithm traversal , To determine the maximum number of prediction periods, typically 2-3, calculate the set of all non-empty intervals. ;like Then, randomly select an interval and uniformly sample integers from it. Ensure the generated decision vector It enables buses to pass smoothly through a future green light window, and strictly conforms to the physical law of the periodic wear and tear of traffic lights.

4. The real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration as described in claim 1, characterized in that: The repair operator in S202 is specifically based on the standard simulated binary crossover SBX operator, which introduces a phased adaptive control of the distribution index, and coordinates the SBX crossover operator and the mutation operator based on deterministic spatiotemporal constraints.

5. The real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration as described in claim 4, characterized in that: The control methods for the SBX crossover operator are as follows: 1) The principle of the SBX operator is as follows: For the selected parent Simulate binary crossover to generate offspring For each decision variable First, generate random numbers. Calculate the expansion factor : （23） Offspring generation formula: （24） 2) Phased adaptive parameter adjustment: The process is divided into three stages—early, middle, and late—based on the number of evolutionary generations, and the distribution index is dynamically adjusted. ,because The variance of the offspring distribution is inversely proportional to the variance of the offspring distribution, gradually transitioning from a wide distribution in the early stages to a narrow distribution in the later stages, using the variation with the number of generations. Distribution index of change : （25） in, The maximum number of generations, the smaller It will produce offspring that are distantly related to their parents, and larger offspring. This will produce offspring that are similar to their parents; 3) Adaptive crossover probability: The crossover probability is dynamically calculated based on the fitness value of the parent individual relative to the average fitness of the population. To protect superior individuals and accelerate the elimination of low-fit individuals, we assume... For the maximum fitness of the population, The average fitness of the population. The larger fitness value among the two parent generations to be crossovered: （26） in, and For adaptive parameters that decrease with evolutionary stage, if the fitness value of the two parents to be crossbred is larger... When the fitness of a species is greater than the average fitness of the population, a higher crossover probability is adopted; otherwise, a lower crossover probability is adopted.

6. The method for real-time optimization of bus priority at multiple intersections based on vehicle-road-cloud collaboration as described in claim 4, characterized in that: The method for regulating the mutation operator employs evolutionary state-aware adaptive mutation, as detailed below: 1) Evolutionary state perception mutation probability: Construct a dynamic mutation probability model based on population fitness distribution and evolutionary process, assuming the... The first generation of the population The fitness of each individual is The average fitness of the population is Maximum fitness is Define evolutionary process factors ,in For the maximum number of iterations, the th The probability of variation for each individual Defined as: （27） in, These are the upper and lower bounds of the basic mutation rate and the radical mutation rate, respectively. The damping coefficient decreases with each generation of evolution and is used to reduce the perturbation frequency in the later stages of convergence, thus protecting the obtained superior gene patterns. 2) Multi-strategy hybrid perturbation model: based on evolutionary process factors Dynamic switching of mutation operators Mathematical form: （28） This represents the mutated gene value. The gene values ​​before mutation are used; this mechanism initially works through a uniform distribution. Perform large-scale jumps, and later utilize variance. Gaussian distribution with algebraic decay A small neighborhood search is performed to achieve adaptive adjustment of the search step size.

7. The real-time optimization method for bus priority at multiple intersections based on vehicle-road-cloud collaboration as described in claim 4, characterized in that: It also includes a deterministic repair operator based on boundary projection, which maps out-of-bounds individuals back to the nearest constraint boundary to alleviate the uncertainty caused by random retries, as detailed below: Under strong constraints, intermediate solutions generated by mutation and crossover operators are considered. May violate minimum / maximum green light duration constraints Instead of using a random retry strategy, a repair operator is adopted. Project it to the nearest constraint boundary using the following piecewise function: （29） in, This is the final solution after processing by the repair operator. If an individual falls within the feasible region, the operator does not intervene and preserves the diversity characteristics of genetic evolution; if an individual falls to the left of the feasible region, the operator forcibly pulls it back to the lower boundary. If an individual falls to the right of the feasible region, the operator forcibly pulls it back to the upper boundary. .

Images