Tunnel partition illumination dynamic optimal control method and system based on traffic flow space-time distribution prediction

Abstract

The invention relates to a tunnel partition illumination dynamic optimal control method and system based on traffic flow space-time distribution prediction, and belongs to the technical field of tunnel illumination energy-saving operation. The method aims to solve the problems of black hole effect caused by response lag, trailing illumination caused by rough partition and lack of pre-judgment ability in the existing control technology. According to the technical scheme, data are collected through a multi-source traffic perception layer, and advanced prediction is carried out by fusing a macroscopic and microscopic model and deep learning through a traffic flow space-time trajectory prediction layer; the dynamic partition optimization decision-making layer maps a prediction result into a virtual light packet and solves an optimal dimming instruction meeting illuminance, uniformity and vehicle speed correlation type brightness change rate constraints through model prediction control, and finally the optimal dimming instruction is output through the smooth execution and closed-loop correction layer. According to the system, the black hole effect can be fundamentally eliminated, the safety is improved, fine energy-saving control over the lamp moving along with a vehicle is achieved, the visual comfort is guaranteed, and the system has high adaptability and robustness for multiple traffic scenes.

Description

Technical Field

[0001] This invention belongs to the field of tunnel lighting energy-saving operation technology, and relates to a dynamic optimal control method and system for tunnel zone lighting based on traffic flow spatiotemporal distribution prediction. Background Technology

[0002] Highway tunnels are special road environments that are closed or semi-closed, with extremely limited internal lighting. Artificial lighting is required both day and night, necessitating the continuous operation of tunnel lighting systems. Statistics show that lighting energy consumption accounts for 50% to 70% of the total energy consumption of highway tunnel operations, making it the largest single source of energy consumption in the tunnel's electromechanical systems. Therefore, achieving refined energy-saving control of tunnel lighting while ensuring a safe driving environment remains a core technical challenge in this field.

[0003] Currently, highway tunnel lighting control technology can be mainly divided into the following two categories: The first type is time-series control and external brightness linkage control. This method adjusts the brightness levels of all lamps in the tunnel according to a preset time schedule or the measurement values ​​of the external environment brightness sensor. Its advantages are simple implementation and low equipment cost. However, this control method has the following obvious defects: (1) It cannot sense the real-time traffic flow status in the tunnel, resulting in serious "ineffective lighting" waste during low traffic flow periods; (2) The zoning granularity is coarse, usually dividing the tunnel into only 3 to 4 lighting zones: entrance section, middle section, and exit section. Differentiated dimming cannot be carried out in each zone according to the traffic flow distribution; (3) The lighting strategy lags behind the actual needs and cannot cope with special traffic scenarios such as holidays and construction control, resulting in poor adaptability.

[0004] The second type is vehicle detection trigger control. This method deploys vehicle detectors at several cross sections inside the tunnel. When a vehicle is detected entering a certain zone, the corresponding lighting circuit is triggered to increase the brightness; the delay decreases after the vehicle leaves. Compared with pure timing control, it is an improvement, but still has the following fundamental defects: (1) Response lag, the "black hole effect" is difficult to eliminate. The system triggers the lighting action only after the sensor detects the vehicle, and there is a delay in the process of perception, transmission, response and lighting. During this period, the driver has entered an area that is not fully illuminated; (2) The granularity of zone control is coarse, and the "trailing lighting" problem is prominent. After the vehicle passes, there is a delay before the lights can be turned off, resulting in areas where there are no vehicles maintaining a high brightness for a long time; (3) It is difficult to take into account both sparse and dense traffic flow. The fixed triggering logic cannot adaptively distinguish different traffic flow density scenarios; (4) It lacks forward-looking prediction ability. The system is completely driven by events that have already occurred and does not have the ability to predict the future spatiotemporal distribution of traffic flow.

[0005] In summary, existing tunnel lighting control technology suffers from three major technical bottlenecks: slow response, coarse zoning granularity, and lack of predictive ability for future traffic flow spatiotemporal distribution. These bottlenecks make it difficult to balance energy efficiency with driving visual safety. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a method and system for dynamic optimal control of tunnel zone lighting based on traffic flow spatiotemporal distribution prediction.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A dynamic optimal control method for tunnel zone lighting based on traffic flow spatiotemporal distribution prediction includes the following steps: The multi-source traffic sensing steps involve deploying detection devices in front of the tunnel entrance and inside the tunnel to collect and process traffic information in real time, forming a unified traffic state vector. The traffic flow spatiotemporal trajectory prediction step, based on the traffic state vector, uses a hierarchical hybrid prediction architecture to predict the spatiotemporal distribution of traffic flow in the entire tunnel within the future prediction time domain, and outputs a traffic flow spatiotemporal distribution prediction matrix; The dynamic zoning optimization decision-making step maps the traffic flow spatiotemporal distribution prediction matrix to the lighting demand weights of each lighting zone, and constructs a model predictive control (MPC) optimization model. With the goal of minimizing total energy consumption and dimming smoothness, and under the conditions of satisfying illuminance constraints, brightness uniformity constraints, and vehicle speed-related brightness change rate constraints, the optimal dimming command sequence for each lighting circuit in the future prediction time domain is solved. The smooth execution and closed-loop correction steps involve performing time-domain smoothing on the optimal dimming command sequence to obtain the final dimming command, which is then sent to the lighting controller. Simultaneously, the measured vehicle position inside the tunnel is compared with the predicted position, and the prediction error is compensated by a rolling time-domain corrector.

[0008] Furthermore, the traffic state vector is ,in, for t Flow rate at any given time, in units of vehicles per hour. for t Spatial average density over time, in units of vehicles per kilometer. for t Average speed over time, in kilometers per hour. for t Average headway at any given time, in seconds. for t Vehicle type classification vector at any given time.

[0009] Furthermore, the traffic flow spatiotemporal trajectory prediction step specifically includes: Traffic condition determination steps, based on real-time traffic density Distance between the front and rear of the vehicle With preset threshold , Based on the comparison results, the micro-trajectory tracking mode or the macro-density wave prediction mode is adaptively switched. In the micro-trajectory tracking step, a state estimator is established for each vehicle in micro-trajectory tracking mode. An improved Kalman filter is used to predict the position and speed of a single vehicle within the tunnel. This improved Kalman filter introduces an adaptive noise covariance adjustment mechanism, dynamically adjusting the process noise covariance matrix Q based on the variance of speed changes within adjacent control cycles. The adjustment rule is as follows: ,in, Forgetting factor, The Kalman gain matrix is... For information vectors; The macroscopic density wave prediction step involves using a macroscopic density wave predictor based on a traffic flow dynamics model to predict traffic density waves within the tunnel. The corrected velocity-density relationship of the model is as follows: ,in, For free flow velocity, For blocking density, For shape parameters, This is a correction factor for the tunnel's enclosed environment. The auxiliary prediction and fusion step introduces a deep learning auxiliary predictor based on the Transformer architecture to learn from historical traffic flow time series data, generate auxiliary prediction results, and then weights and fuses the output of the micro trajectory tracking step or macro density wave prediction step with the auxiliary prediction results to generate the traffic flow spatiotemporal distribution prediction matrix.

[0010] Furthermore, the constraint on the vehicle speed-related brightness change rate is as follows: ,in, For the first j The average brightness value of the road surface corresponding to the loop. The maximum permissible rate of change of brightness function related to vehicle speed is defined as follows: ,in, Let be the predicted average velocity inside the tunnel at time t. The maximum design brightness value, This is the low-speed baseline rate of change coefficient. The velocity coupling attenuation coefficient is... This represents the free-flow velocity.

[0011] Furthermore, the time-domain smoothing process employs an exponentially weighted moving average method, as shown in the formula: ,in, The solution obtained for the dynamic partitioning optimization decision step is the first... j Optimal dimming ratio for the circuit This is the final dimming command after smoothing. This is the smoothing coefficient.

[0012] A dynamic optimal control system for tunnel zone lighting based on traffic flow spatiotemporal distribution prediction includes: The multi-source traffic perception layer includes radar detection units and high-definition cameras deployed in front of the tunnel entrance, as well as tunnel cross-section detectors deployed in multiple sections inside the tunnel, used to collect and fuse raw traffic information to generate traffic state vectors. The traffic flow spatiotemporal trajectory prediction layer is connected to the multi-source traffic perception layer. It is used to receive the traffic state vector and output the traffic flow spatiotemporal distribution prediction matrix in the future prediction time domain using a hierarchical hybrid prediction architecture. The dynamic partitioning optimization decision layer is connected to the traffic flow spatiotemporal trajectory prediction layer. It is used to map the traffic flow spatiotemporal distribution prediction matrix into lighting demand weights, and generate the optimal dimming command sequence for each lighting circuit by solving the model prediction control MPC optimization model. The smooth execution and closed-loop correction layer is connected to the dynamic partition optimization decision layer, the traffic flow spatiotemporal trajectory prediction layer, and the lighting controller. It is used to smooth the optimal dimming command sequence and issue it, and at the same time, to perform rolling time-domain compensation for the prediction error based on the measured data of the tunnel cross-section detector.

[0013] Furthermore, the traffic flow spatiotemporal trajectory prediction layer includes a traffic state discriminator, a micro-trajectory tracking module, a macro-density wave prediction module, a deep learning-assisted predictor, and a weighted fusion processor. The traffic state discriminator selects to activate either the micro-trajectory tracking module or the macro-density wave prediction module based on real-time traffic density and headway. The weighted fusion processor performs weighted fusion of the output of the micro-trajectory tracking module or the macro-density wave prediction module with the output of the deep learning-assisted predictor.

[0014] Furthermore, the dynamic zoning optimization decision layer includes a virtual light packet mapper and a model predictive control (MPC) optimization solver. The virtual light packet mapper is used to convert the traffic flow spatiotemporal distribution prediction matrix into dynamic lighting requirements for each lighting zone. The model predictive control (MPC) optimization solver is used to solve a quadratic programming problem that includes total energy consumption, dimming smoothing, illuminance lower limit constraint, luminance uniformity constraint, and vehicle speed-related luminance change rate constraint.

[0015] Furthermore, the smooth execution and closed-loop correction layer includes a temporal smoothing module, an error evaluation module, a rolling temporal corrector, and a safety redundancy protection module. The temporal smoothing module is used to smooth the dimming command. The error evaluation module is used to calculate the deviation between the predicted and measured values ​​of the vehicle position. The rolling temporal corrector is used to feed the deviation back to the traffic flow spatiotemporal trajectory prediction layer to correct the prediction. The safety redundancy protection module triggers a forced brightening command when the deviation exceeds a safety threshold.

[0016] The beneficial effects of this invention are as follows: (1) Through a feedforward control mechanism based on the prediction of traffic flow spatiotemporal distribution, the system can proactively and proactively increase the lighting brightness of the target lighting area before the vehicle actually arrives, so that the lighting action is completed before the vehicle arrives. This mechanism avoids the safety hazards caused by the inherent delays in signal transmission and lamp response of traditional sensor-triggered control, and ensures the driver's visual safety in each section of the tunnel.

[0017] (2) By constructing a dynamically moving virtual "light package" and improving the lighting control precision to the level of a single lamp group or circuit, the high-brightness lighting area can accurately follow the vehicle's position. Areas where the vehicle has not arrived or has left maintain the minimum basic lighting or are turned off, thereby eliminating the phenomena of "ineffective lighting" and "trailing lighting" to the greatest extent. Theoretical calculations show that this method can achieve significant energy-saving effects during low traffic periods such as late at night.

[0018] (3) By innovatively incorporating vehicle speed-related brightness change rate constraints into the optimization model and using a time-domain smoothing strategy, the system can dynamically limit the rate of brightness change based on real-time vehicle speed, ensuring that changes in the light environment are always within the physiological adaptation threshold of the driver's eyes. This effectively prevents discomfort such as flicker and glare caused by sudden changes in brightness, and provides a smooth, continuous, and comfortable driving light environment while pursuing energy conservation.

[0019] (4) Through a hierarchical prediction architecture that combines macroscopic density wave prediction with microscopic single-vehicle trajectory tracking, the system can automatically switch to the optimal prediction mode according to the real-time traffic conditions (sparse flow or congested flow), solving the problem of poor adaptability of a single control strategy in extreme scenarios. At the same time, relying on the closed-loop error correction and safety redundancy protection mechanism in the rolling time domain, even in abnormal situations such as sudden vehicle speed change or prediction deviation, the system can quickly correct and forcibly ensure the lighting safety of critical areas, enhancing the reliability and fault tolerance of the entire control system.

[0020] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 Flowchart of the dynamic optimal control system for tunnel zone lighting; Figure 2 Constructing and optimizing the decision-level logic diagram for dynamic virtual optical packets; Figure 3 This is a diagram illustrating the working mechanism of the rolling time-domain closed-loop correction layer. Detailed Implementation

[0022] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0023] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0024] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0025] Figure 1 This is a flowchart of the dynamic optimal control system for tunnel zone lighting. The diagram illustrates the complete workflow of the intelligent tunnel lighting system from initialization to closed-loop control: After system startup, the multi-source sensing layer collects and fuses traffic data through entrance radar, cameras, and tunnel cross-section detectors; the traffic state discriminator selects macro and micro modes based on real-time density and vehicle type information, activating Kalman filtering for micro-level single-vehicle tracking under sparse traffic flow, and employing prediction adjustments based on congestion conditions under congested conditions; the prediction results enter the Transformer deep learning and auxiliary prediction stage to form a spatiotemporal distribution matrix, which is then processed by virtual light... The package mapping is converted into lighting requirements; then the MPC solver performs optimal dimming solution under the goal of minimizing energy consumption. If the constraint check is not met, the solution is iterated again. If it is met, time-domain smoothing is performed. While performing dimming, the system compares the measured position with the predicted position through the error evaluation module. If the deviation exceeds the safety threshold, the safety redundancy protection is triggered to force high brightness. Otherwise, the rolling time-domain correction stage is entered to reset the initial state of the prediction model. Finally, it is determined whether there are still vehicles in the tunnel to decide whether to continue the control loop or enter standby mode, forming a complete intelligent control system that combines feedforward prediction and closed-loop compensation.

[0026] Figure 2 A logic diagram for constructing and optimizing the decision layer of the dynamic virtual light packet is presented. The dynamic partitioning optimization decision layer takes the traffic flow spatiotemporal distribution prediction matrix as input. First, the virtual light packet mapping module C1 constructs a high-brightness demand zone, i.e., the virtual light packet, that moves in real-time with the traffic flow position based on the probability of vehicle presence at each time and position in the prediction matrix. The leading edge C1a of the virtual light packet gradually brightens a few seconds in advance according to the human eye's dark adaptation curve, ensuring that the driver is in a fully illuminated environment before the vehicle arrives, completely eliminating the black hole effect. The trailing edge C1b of the virtual light packet quickly decreases to the basic lighting level after the vehicle passes, effectively eliminating trailing illumination. Based on this, the model predictive control solver C2, starting from the current moment, establishes an optimization problem in the prediction time domain with the objective function of minimizing comprehensive lighting energy consumption. It simultaneously applies three types of constraints: first, illuminance constraint, meaning the illuminance of each zone must not be lower than the standard illuminance value specified in the highway tunnel lighting design code; second, brightness uniformity constraint, meaning the brightness difference between adjacent lighting circuits must not exceed the upper limit of the design code; and third, speed-related brightness change rate constraint, meaning that at any control moment, the rate of change of lighting brightness must meet the human eye adaptation threshold constraint related to the current predicted vehicle speed. The higher the vehicle speed, the smaller the allowable rate of change of brightness, to prevent flickering and glare hazards at high speeds. Solver C2 outputs the optimal dimming command sequence for each lighting circuit in the next control cycle. After smoothing by the time-domain smoothing module C3, it is sent to the LED driver power supply C4 of each circuit via DALI or a 0-10V analog signal interface to drive the corresponding lamp group to perform dimming actions.

[0027] Figure 3 This diagram illustrates the working mechanism of the rolling time-domain closed-loop correction layer. The time-domain smoothing module C3 performs a moving average process on the loop-by-loop dimming command sequence output by the MPC solver, limiting the brightness jumps between adjacent control cycles to within the range of human eye adaptation. The final dimming command is then sent to the LED driver power supply D2 of each zone via the DALI bus D1, driving the corresponding lamp group D3 to execute continuous and smooth dimming actions. Simultaneously, the detectors A4 at each section inside the tunnel continuously upload the measured vehicle position information to the error evaluation module D4. D4 compares the measured positions with the predicted positions output by the prediction layer section by section, calculates the position deviation vector, and injects it into the rolling corrector D5 of the prediction layer. D5 resets the initial state of the prediction model every control cycle, using the latest measured data as an anchor point to re-roll and predict the spatiotemporal distribution of traffic flow within subsequent time windows, thereby eliminating prediction errors within the closed-loop correction process of each control cycle. When the deviation exceeds the safety threshold, the system triggers the safety redundancy protection mechanism D6, which automatically and forcibly increases the lighting brightness of the zone where the abnormal vehicle is located to the highest level and maintains it until the vehicle leaves the tunnel or the deviation returns to the normal range, ensuring the safety baseline of driving lighting under any extreme conditions.

[0028] This invention proposes a dynamic optimal control system for tunnel zone lighting based on traffic flow spatiotemporal distribution prediction. Its overall architecture consists of four functional layers connected in series: a multi-source traffic perception layer, a traffic flow spatiotemporal trajectory prediction layer, a dynamic zone optimization decision layer, and a smooth execution and closed-loop correction layer, from bottom to top. These layers interact via a standardized real-time data bus. Control commands are sent unidirectionally from top to bottom, while sensor-measured feedback data is continuously uploaded from bottom to top. A rolling time-domain corrector injects feedback deviations into the prediction layer in real time, forming a composite control loop combining feedforward and closed-loop control. The overall control logic of the system follows a periodic iteration of four stages: prediction, optimization, execution, and correction. Within each control cycle, the system completes a full state estimation, trajectory prediction, optimal dimming solution, and error compensation to ensure that the lighting control maintains optimal matching with the actual traffic flow spatiotemporal distribution at any given time.

[0029] Compared to existing timing control and traditional loop-triggered control, this system achieves fundamental breakthroughs in three dimensions: response mechanism, control accuracy, and predictive capability. In terms of response mechanism, this system replaces the passive response triggered by sensors with predictive feedforward, completing the lighting action ahead of the vehicle's arrival time, completely eliminating the black hole effect caused by the inherent delay introduced by the sensor-triggered response link. Regarding control accuracy, this system replaces physically fixed partitions with algorithm-level virtual light packet dynamic mapping, raising the granularity of lighting control from the traditional whole-segment or coarse-grained loop level to the single-lamp group level, achieving precise spatiotemporal tracking of vehicle positions within high-brightness lighting areas. In terms of predictive capability, this system integrates deep learning timing prediction models and traffic flow dynamics models to predict the spatiotemporal distribution of traffic flow throughout the tunnel within a future time window, enabling the control system to proactively perceive and respond to future traffic conditions.

[0030] 1. Multi-source traffic perception layer The multi-source traffic perception layer is the data foundation for the entire system to achieve accurate prediction and control. It is responsible for the real-time collection and fusion processing of raw traffic information from the tunnel entrance and various cross-sections inside the tunnel. The perception layer deploys forward radar detection units and high-definition cameras 100 to 200 meters in front of the tunnel entrance to detect the initial speed, vehicle type, and headway of approaching vehicles in real time. Inside the tunnel, a set of cross-section detectors is deployed every 50 to 100 meters along the direction of traffic to continuously collect vehicle speed, traffic density, and cross-section flow data, which are then uploaded to the data fusion center via an industrial Ethernet bus.

[0031] The data fusion center performs time and spatial alignment processing on multi-source heterogeneous sensor data, and uses a system error calibration method based on least squares estimation to eliminate inherent biases between sensors, forming a traffic state vector in a unified format and feeding it into the prediction layer in real time. The fused traffic state vector at a single moment is defined as follows:

[0032] In the formula: The cross-sectional flow rate at time t is expressed in vehicles per hour. Let be the spatial average density at time t, in units of vehicles / km; The time-averaged velocity at time t is expressed in kilometers per hour. The average headway at time t is expressed in seconds. Let t be the vehicle type classification vector, which includes three components: passenger cars, medium-sized trucks, and large trucks.

[0033] The aforementioned state vector is pushed to the prediction layer with a fixed sampling period Δt, providing continuous and high-quality data input for subsequent trajectory prediction and optimization decisions.

[0034] 2. Spatiotemporal trajectory prediction layer for traffic flow The traffic flow spatiotemporal trajectory prediction layer is the core algorithm module of this system. It adopts a hierarchical hybrid prediction architecture that combines macroscopic traffic flow density wave prediction with microscopic single-vehicle trajectory tracking, and the prediction mode is adaptively switched by the traffic state discriminator according to the real-time traffic density.

[0035] (1) Traffic status identification and mode switching Traffic condition discriminator uses the current cross-sectional average density k ( t The primary criterion is the headway. h ( t As an auxiliary criterion, the adaptive switching between macro and micro prediction modes is achieved according to the following discrimination rules: when and At that time, the micro-trajectory tracking mode is activated; when or At that time, activate the macroscopic density wave prediction mode. In the formula: The preset density switching threshold; The preset headway switching threshold is used; both thresholds are calibrated based on the actual traffic data of the specific tunnel, reflecting the engineering adaptation principle of one plan for one tunnel.

[0036] (2) Microscopic trajectory tracking mode: Improved Kalman filter In the micro-trajectory tracking mode, the system establishes an independent state estimator for each vehicle and uses an improved Kalman filter to accurately predict the time-by-time position and velocity of a single vehicle within the tunnel. For the... i For a vehicle, whose state vector is defined as a combination of position and velocity, the discrete-time state equation and observation equation are as follows: Equations of state:

[0037] Observation equation:

[0038] In the formula: ,in For the first i The car is t The longitudinal position of the time relative to the tunnel entrance. The velocity at the corresponding moment; [0,1] represents the state transition matrix; To control the input matrix; For the first i Estimated longitudinal acceleration of the vehicle; H=[1, 0] is the observation matrix, and the tunnel cross-section detector only directly observes the longitudinal position of the vehicle; For process noise, To observe noise.

[0039] This invention introduces an adaptive noise covariance adjustment mechanism based on standard Kalman filtering. It dynamically adjusts the process noise covariance matrix Q according to the variance estimate of the speed change within adjacent control cycles to adapt to the frequent speed changes of vehicles in tunnels. The adaptive adjustment rules are as follows:

[0040] In the formula: This is the forgetting factor, with a value ranging from 0.95 to 0.99; The Kalman gain matrix; This is the information vector. This mechanism enables the system to quickly increase the process noise weight under sudden changes in operating conditions such as sudden braking and acceleration of the vehicle, thereby speeding up the filter's response to abnormal trajectories.

[0041] For the i-th vehicle, the system outputs a predicted trajectory sequence for the next N steps at the current time t based on the Kalman prediction equation:

[0042] In the formula: In order to be in t Time for the first i The car is Predicted value of time and location; This is an estimate of the acceleration. N To predict the number of time-domain steps, corresponding to the prediction duration. N ·Δ t .

[0043] (3) Macroscopic density wave prediction model A macroscopic density wave predictor for a traffic flow hydrodynamic model predicts the propagation direction and velocity of congestion waves within a tunnel. The core governing equation of the model is the conservation equation for traffic flow density:

[0044] In the formula: For position x Place t Traffic density at any given time; The corresponding cross-sectional flow rates; the two are related through a basic diagram. q = k · V ( k Establish contact, V ( k ) is the velocity-density relationship function.

[0045] This invention uses the Greenshields velocity-density relationship as the basic graphical model and introduces correction coefficients to address the specific characteristics of driver behavior in a closed tunnel environment. The corrected velocity-density relationship is:

[0046] In the formula: The free-flow velocity is determined based on the tunnel design speed limit. Blocking density; β The shape parameter reflects the nonlinear characteristics of the velocity-density curve; This is a correction factor for the enclosed environment of a tunnel, used to characterize the driver's proactive speed reduction behavior caused by psychological pressure in an enclosed tunnel. The specific value is determined based on the tunnel length and clearance height.

[0047] The model equations are discretized numerically using the Godunov scheme. The measured density of the detectors at each tunnel section is used as the boundary condition. The density distribution of each subdivided zone at future times is solved in the prediction time domain. The density distribution matrix is ​​then converted into the lighting demand weights of each lighting zone, which are used as the input to the optimization decision layer.

[0048] (4) Transformer deep learning-assisted predictor Building upon the two physical model-driven predictions mentioned above, this invention introduces a deep learning-assisted predictor based on the Transformer architecture. This predictor learns long-range temporal dependencies in historical traffic flow time-series data, providing statistical learning-assisted prediction of global traffic conditions within future time windows. The Transformer predictor uses past... M The traffic state sequence at each time step is taken as input, and the future traffic state sequence is output. N Step-by-step state prediction sequence:

[0049] In the formula: For the past M Step through the historical state sequence; The parameters for the Transformer network are obtained through offline training using historical tunnel traffic data. M This represents the length of the historical window. The Transformer's multi-head self-attention mechanism enables the model to automatically capture periodic patterns such as morning and evening rush hours and holidays, compensating for the insufficient prediction accuracy of physical models in unconventional traffic scenarios.

[0050] Finally, the weighted fusion processor performs a weighted fusion of the prediction results from the microscopic or macroscopic physical model and the prediction results assisted by deep learning, outputting the final spatiotemporal distribution prediction matrix of traffic flow for optimization decision-making:

[0051]

[0052] In the formula: Predict density for physical models; Predict density for deep learning models; and The weights are time-varying, with the physical model as the main component in the short prediction time domain, and the weights of the deep learning model gradually increasing as τ increases in the long prediction time domain, thus balancing short-term accuracy and long-term stability.

[0053] 3. Dynamic partitioning optimizes the decision-making level. (1) Virtual optical packet mapping mechanism Virtual light packet mapping module for predicting density matrix As input, a highlighted area, or virtual light packet, is constructed at the algorithm level that moves in real time with the position of the traffic flow. The virtual light packet... t The spatial range is determined by the following formula:

[0054] In the formula: L The total length of the tunnel; The minimum density threshold for triggering high-brightness lighting; The safe advance lighting distance in single-vehicle mode is calculated based on vehicle speed and the human eye's dark adaptation time. ,in The human eye's dark adaptation time constant.

[0055] The leading edge of the virtual light packet performs a slow brightening operation, with the brightening rate determined based on the human eye's dark adaptation curve; the trailing edge performs a rapid brightening operation after the vehicle passes, with the brightening rate taking the maximum value within the allowable range within the vehicle speed-related constraint framework, in order to maximize the elimination of trailing lighting energy consumption.

[0056] (2) Model predictive control optimization framework The core of the dynamic partitioning optimization decision layer is the MPC optimal control solver, which uses the virtual light packet mapping result as a reference for lighting demand in the prediction time domain. N An optimization problem is established within the step, with the objective of minimizing the overall lighting energy consumption. This problem is applied to the entire tunnel section. J The lighting circuit is set as follows: j Each loop in t The dimming ratio at any given time is Then the objective function is:

[0057] In the formula: For the first j The rated power of each circuit; To predict the τ-th step in the time domain j The dimming ratio of the loop; μ is the penalty weight for dimming changes, used to suppress frequent and large fluctuations in dimming commands. The first term of the objective function constrains the minimization of total energy consumption, and the second term constrains the smoothness of the dimming process. The above objective function is subject to the following three types of constraints: The first category is the lower limit constraint of illuminance, which ensures that the illuminance level of each lighting zone is not lower than the standard value specified in the highway tunnel lighting design code:

[0058] In the formula: For the first j Illuminance values ​​for each zone at prediction step τ; adjacent loops j 'For partitions j The illuminance contribution coefficient is pre-calibrated by tunnel lighting simulation; To determine the standard illuminance requirement value dynamically based on traffic flow prediction density, the upper limit of the driving area illuminance specified in the standard is taken within the high-brightness area of ​​the virtual light packet, and the lower limit of the basic lighting illuminance is taken outside the light packet.

[0059] The second type is brightness uniformity constraint, which ensures that the brightness transition between adjacent lighting circuits meets visual comfort requirements:

[0060] In the formula: For the first j The average brightness value of the road surface corresponding to the circuit; The maximum brightness difference between adjacent sections is determined according to the tunnel lighting design specifications.

[0061] The third type is the vehicle speed-related brightness change rate constraint, which is one of the core innovative constraints of this invention. This constraint dynamically limits the rate of change of lighting brightness based on real-time prediction of vehicle speed, ensuring from a physiological perspective that changes in the light environment never exceed the adaptive capacity of the driver's eyes.

[0062]

[0063] In the formula: for t Predicted average speed inside the tunnel at any given time; The maximum permissible rate of change of brightness is a function related to vehicle speed; This is the coefficient of the low-speed baseline rate of change. This is the velocity coupling attenuation coefficient; This represents the maximum design brightness value. This constraint indicates that the higher the vehicle speed, the longer the road surface is scanned in the human eye's field of vision per unit time, and the more sensitive the perception of sudden changes in brightness is, thus allowing for a smaller rate of brightness change; the lower the vehicle speed, the more relaxed the allowable rate of brightness change can be, thereby achieving a more agile energy-saving dimming response under low-speed conditions.

[0064] Combining the objective function and the three types of constraints described above, the complete MPC optimization problem can be summarized into the following quadratic programming form:

[0065]

[0066]

[0067]

[0068] In the formula: To optimize the variable vector, it includes the dimming ratio of all loops in the prediction time domain; This is the coefficient matrix of the quadratic terms; This is the vector of linear term coefficients; and These are the inequality constraint coefficient matrix and the right-hand vector, respectively, covering the lower limit constraint of illuminance, uniformity constraint and luminance change rate constraint; and The constraints are equality constraints (such as a fixed boundary illumination condition). This quadratic programming problem can be solved efficiently within each control cycle using the efficient set method or the interior point method.

[0069] 4. Smooth Execution and Closed-Loop Correction Layer (1) Temporal smoothing The loop-by-loop optimal dimming command sequence output by the MPC solver is processed by a time-domain smoothing module using a moving average. This further constrains the brightness jumps between adjacent control cycles within the allowable range related to vehicle speed, ensuring that the driver experiences a continuous and smooth brightness gradient throughout the driving process. The smoothing process employs an exponentially weighted moving average method.

[0070] In the formula: The optimal dimming ratio for the j-th loop output by the MPC solver; This is the final dimming command after smoothing. This is a smoothing coefficient, with a value ranging from 0 to 1. The larger the value, the weaker the smoothing effect and the faster the response. A smaller value results in a stronger smoothing effect but a slower response. It's recommended to adjust the value dynamically in real-time based on vehicle speed: at higher speeds... Take a smaller value to enhance smoothness, especially at lower vehicle speeds. Choose a larger value to ensure response sensitivity.

[0071] (2) Rolling time-domain prediction error compensation The error assessment module compares the measured vehicle position data from the tunnel section detector with the predicted position output by the prediction layer section by section, and calculates the position deviation vector:

[0072] In the formula: for t The sensors inside the cave at any time are for the first i The actual measured location of the vehicle; For the prediction of the previous control cycle t Time position estimate; This represents the location prediction error.

[0073] The rolling time-domain corrector feeds back the position deviation vector to the prediction layer, resets the initial state of the prediction model using the latest measured data as the anchor point, and re-rolls the prediction within the new prediction time domain to achieve closed-loop error compensation for each control cycle. The corrected prediction trajectory is as follows:

[0074] In the formula: This is the corrected location prediction value; To correct for gain, its physical meaning is similar to that of Kalman gain, reflecting the degree of confidence in the measured bias; A larger value indicates that the system trusts the measured feedback more and has a larger correction range, making it suitable for scenarios where the sensor accuracy is high.

[0075] (3) Safety redundancy protection mechanism When position deviation When the absolute value exceeds the preset safety threshold, the safety redundancy protection mechanism is automatically triggered. The system immediately sets the distances before and after the abnormal vehicle's current measured position by d. safe All lighting circuits within the range are forced to increase to the highest dimming ratio and maintain the highest brightness until the vehicle exits the tunnel or the position deviation returns to within the safe threshold.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic optimal control method for tunnel zone lighting based on traffic flow spatiotemporal distribution prediction, characterized in that: Includes the following steps: The multi-source traffic sensing steps involve deploying detection devices in front of the tunnel entrance and inside the tunnel to collect and process traffic information in real time, forming a unified traffic state vector. The traffic flow spatiotemporal trajectory prediction step, based on the traffic state vector, uses a hierarchical hybrid prediction architecture to predict the spatiotemporal distribution of traffic flow in the entire tunnel within the future prediction time domain, and outputs a traffic flow spatiotemporal distribution prediction matrix; The dynamic zoning optimization decision-making step maps the traffic flow spatiotemporal distribution prediction matrix to the lighting demand weights of each lighting zone, and constructs a model predictive control (MPC) optimization model. With the goal of minimizing total energy consumption and dimming smoothness, and under the conditions of satisfying illuminance constraints, brightness uniformity constraints, and vehicle speed-related brightness change rate constraints, the optimal dimming command sequence for each lighting circuit in the future prediction time domain is solved. The smooth execution and closed-loop correction steps involve performing time-domain smoothing on the optimal dimming command sequence to obtain the final dimming command, which is then sent to the lighting controller. Simultaneously, the measured vehicle position inside the tunnel is compared with the predicted position, and the prediction error is compensated by a rolling time-domain corrector.

2. The method for dynamic optimal control of tunnel zone lighting based on traffic flow spatiotemporal distribution prediction according to claim 1, characterized in that: The traffic state vector is ,in, for t Flow rate at any given time, in units of vehicles per hour. for t Spatial average density over time, in units of vehicles per kilometer. for t Average speed over time, in kilometers per hour. for t Average headway at any given time, in seconds. for t Vehicle type classification vector at any given time.

3. The method for dynamic optimal control of tunnel zone lighting based on traffic flow spatiotemporal distribution prediction according to claim 1 or 2, characterized in that: The traffic flow spatiotemporal trajectory prediction step specifically includes: Traffic condition determination steps, based on real-time traffic density Distance between the front and rear of the vehicle With preset threshold , Based on the comparison results, the micro-trajectory tracking mode or the macro-density wave prediction mode is adaptively switched. In the micro-trajectory tracking step, a state estimator is established for each vehicle in micro-trajectory tracking mode. An improved Kalman filter is used to predict the position and speed of a single vehicle within the tunnel. This improved Kalman filter introduces an adaptive noise covariance adjustment mechanism, dynamically adjusting the process noise covariance matrix Q based on the variance of speed changes within adjacent control cycles. The adjustment rule is as follows: ,in, Forgetting factor, The Kalman gain matrix is... For information vectors; The macroscopic density wave prediction step involves using a macroscopic density wave predictor based on a traffic flow dynamics model to predict traffic density waves within the tunnel. The corrected velocity-density relationship of the model is as follows: ,in, For free flow velocity, For blocking density, For shape parameters, This is a correction factor for the tunnel's enclosed environment. The auxiliary prediction and fusion step introduces a deep learning auxiliary predictor based on the Transformer architecture to learn from historical traffic flow time series data, generate auxiliary prediction results, and then weights and fuses the output of the micro trajectory tracking step or macro density wave prediction step with the auxiliary prediction results to generate the traffic flow spatiotemporal distribution prediction matrix.

4. The method for dynamic optimal control of tunnel zone lighting based on traffic flow spatiotemporal distribution prediction according to claim 1, characterized in that: The vehicle speed-related brightness change rate constraint is: ,in, For the first j The average brightness value of the road surface corresponding to the circuit. The maximum permissible rate of change of brightness function related to vehicle speed is defined as follows: ,in, Let be the predicted average velocity inside the tunnel at time t. The maximum design brightness value, This is the low-speed baseline rate of change coefficient. The velocity coupling attenuation coefficient is... This represents the free-flow velocity.

5. The method for dynamic optimal control of tunnel zone lighting based on traffic flow spatiotemporal distribution prediction according to claim 1, characterized in that: The time-domain smoothing process employs an exponentially weighted moving average method, with the following formula: ,in, The solution obtained for the dynamic partitioning optimization decision step is the first... j Optimal dimming ratio for the circuit This is the final dimming command after smoothing. This is the smoothing coefficient.

6. A dynamic optimal control system for tunnel zone lighting based on traffic flow spatiotemporal distribution prediction, characterized in that: include: The multi-source traffic perception layer includes radar detection units and high-definition cameras deployed in front of the tunnel entrance, as well as tunnel cross-section detectors deployed in multiple sections inside the tunnel, used to collect and fuse raw traffic information to generate traffic state vectors. The traffic flow spatiotemporal trajectory prediction layer is connected to the multi-source traffic perception layer. It is used to receive the traffic state vector and output the traffic flow spatiotemporal distribution prediction matrix in the future prediction time domain using a hierarchical hybrid prediction architecture. The dynamic partitioning optimization decision layer is connected to the traffic flow spatiotemporal trajectory prediction layer. It is used to map the traffic flow spatiotemporal distribution prediction matrix into lighting demand weights, and generate the optimal dimming command sequence for each lighting circuit by solving the model prediction control MPC optimization model. The smooth execution and closed-loop correction layer is connected to the dynamic partition optimization decision layer, the traffic flow spatiotemporal trajectory prediction layer, and the lighting controller. It is used to smooth the optimal dimming command sequence and issue it, and at the same time, to perform rolling time-domain compensation for the prediction error based on the measured data of the tunnel cross-section detector.

7. The tunnel zone lighting dynamic optimal control system based on traffic flow spatiotemporal distribution prediction according to claim 6, characterized in that: The traffic flow spatiotemporal trajectory prediction layer includes a traffic state discriminator, a micro-trajectory tracking module, a macro-density wave prediction module, a deep learning-assisted predictor, and a weighted fusion processor. The traffic state discriminator selects to activate either the micro-trajectory tracking module or the macro-density wave prediction module based on real-time traffic density and headway. The weighted fusion processor performs weighted fusion of the output of the micro-trajectory tracking module or the macro-density wave prediction module with the output of the deep learning-assisted predictor.

8. The tunnel zone lighting dynamic optimal control system based on traffic flow spatiotemporal distribution prediction according to claim 6 or 7, characterized in that: The dynamic zoning optimization decision layer includes a virtual light packet mapper and a model predictive control (MPC) optimization solver. The virtual light packet mapper is used to convert the traffic flow spatiotemporal distribution prediction matrix into dynamic lighting requirements for each lighting zone. The model predictive control (MPC) optimization solver is used to solve a quadratic programming problem that includes total energy consumption, dimming smoothing, illuminance lower limit constraint, luminance uniformity constraint, and vehicle speed-related luminance change rate constraint.

9. The tunnel zone lighting dynamic optimal control system based on traffic flow spatiotemporal distribution prediction according to claim 6, characterized in that: The smooth execution and closed-loop correction layer includes a temporal smoothing module, an error evaluation module, a rolling temporal corrector, and a safety redundancy protection module. The temporal smoothing module is used to smooth the dimming command. The error evaluation module is used to calculate the deviation between the predicted and measured values ​​of the vehicle position. The rolling temporal corrector is used to feed the deviation back to the traffic flow spatiotemporal trajectory prediction layer to correct the prediction. The safety redundancy protection module triggers a forced brightening command when the deviation exceeds a safety threshold.

Images