A method and system for constructing a cross-road OD matrix and scheduling traffic prediction based on daily key double-factor HMAC license plate hashing

Abstract

The application relates to the technical field of traffic control systems and discloses a cross-intersection OD matrix construction and traffic prediction scheduling integrated method and system based on daily key double-factor HMAC license plate hash, which comprises the following steps: an edge processing unit calculates a license plate hash value by using a daily random key and overwrites memory plaintext data in real time; a central platform matches the hash value in an ordered set with a 900s survival period, generates an anonymous origin-destination matrix statistical table, fuses the statistical table and a real-time traffic input prediction model, and outputs predicted entering traffic; and a multi-agent controller adjusts the weight of a signal timing gain in a reward function according to the deviation value of the predicted entering traffic and measured traffic, and adjusts a phase switching instruction. The application realizes accurate path association on the premise of ensuring that identity information cannot be restored by using a key rotation and time limit matching mechanism, and solves the conflict between data security and scheduling stability by cooperating with a deviation feedback adjustment mechanism.

Description

Technical Field

[0001] This invention relates to an integrated method and system for constructing cross-intersection OD matrix and predicting and scheduling traffic flow based on daily key dual-factor HMAC license plate hashing, belonging to the field of traffic control system technology. Background Technology

[0002] In the field of intelligent transportation, using data acquisition equipment to obtain vehicle operation characteristics to guide traffic light adjustment is a standard way to improve road network traffic efficiency. Common solutions use video recognition, geomagnetic induction, and other means to obtain cross-sectional traffic flow information, providing a data foundation for control decisions. With the increasing demand for refined road network scheduling, cross-intersection origin-destination correlation features have become a key element in predicting traffic flow trends. However, when processing such data, the collection and retention of original license plate features face constraints on data privacy protection compliance. Traditional methods use one-way hash algorithms to anonymize vehicle features, but this approach has inherent limitations in balancing data utilization and privacy security.

[0003] Introducing collection point location parameters during hash calculation to enhance anonymity can lead to inconsistent summaries generated for the same vehicle at different intersections, causing cross-intersection matching mechanisms to fail. While globally fixed summary parameters support origin-destination association, they cannot prevent the risk of tracing long-term trajectories through historical summary information. Analysis reveals the following main shortcomings of existing technologies: 1. Anonymous feature extraction logic struggles to balance spatial correlation and identity isolation; 2. Summarized data possesses long-term traceability, making automatic trajectory information cut-off impossible; 3. Signal scheduling schemes are primarily based on real-time queuing observation, lacking feedforward adjustment capabilities tailored to flow evolution trends.

[0004] Therefore, the technical problem to be solved by this invention is how to extract irreversible license plate summaries that meet privacy compliance requirements, ensure the consistency of cross-intersection features within a single day cycle to support the construction of the origin-destination matrix, achieve physical isolation of historical trajectories, and use flow distribution features to drive the feedforward adjustment of traffic lights. Summary of the Invention

[0005] To address the problems mentioned in the background section, the technical solution of this invention is as follows:

[0006] A method for integrating cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing includes the following steps:

[0007] Step S101: The edge processing unit receives the plaintext of the license plate collected by the road sensing device, reads the daily updated random key generated by the central platform, uses the daily updated random key as the hash key to process the plaintext of the license plate, and calculates the license plate hash value.

[0008] In step S102, after calculating the license plate hash value, the edge processing unit physically overwrites the plaintext of the license plate in memory, and the residence period of the plaintext of the license plate in memory is less than 1 second.

[0009] In step S103, the edge processing unit encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform;

[0010] In step S104, the central platform maintains an ordered set with a preset storage period of 900s in the memory database. By comparing anonymous tracking messages with the same license plate hash value, it extracts the vehicle travel time between different intersection numbers and generates an anonymous origin-destination matrix statistics table.

[0011] In step S105, the central platform merges the anonymous origin-destination matrix statistics table with real-time traffic flow data, inputs it into the gating loop unit prediction model, and predicts the predicted inbound traffic flow of the target intersection within a preset future time period.

[0012] In step S106, the multi-agent reinforcement learning controller calculates the signal timing gain coefficient based on the predicted incoming flow, and dynamically adjusts the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted incoming flow and the actual flow measured by the road sensing device, and adjusts the signal phase switching command according to the weighting ratio.

[0013] Preferably, step S101 includes: step S1011, the central platform generates a new random key at 00:00 every day based on the global time synchronization protocol and sends it to each intersection edge processing unit. At the same time, it sends an old key physical destruction instruction to each intersection edge processing unit to synchronously trigger the old key in the memory of each intersection edge processing unit to perform multiple overwrite physical deletion.

[0014] Preferably, step S104 includes: step S1041, whereby the central platform automatically and physically deletes license plate hash value entries that have not been matched successfully for more than 900 seconds from the ordered set of the memory database.

[0015] Preferably, in step S105, the input data of the gated loop unit prediction model includes the vehicle turning ratio and path transition probability, which reflect the spatiotemporal evolution trend of road network traffic, extracted from the anonymous origin-destination matrix statistics table. The vehicle turning ratio and path transition probability are obtained by performing dimensionality reduction and aggregation processing on the spatiotemporal feature dimensions of the anonymous origin-destination matrix statistics table.

[0016] Preferably, in step S106, the adjustment logic for the weighted ratio includes: step S1061, calculating the absolute error between the predicted incoming flow and the actual measured flow at the corresponding time; step S1062, when the absolute error continues to exceed a preset judgment threshold, progressively reducing the dynamic weight of the signal timing gain coefficient in the reward function. The formula for calculating the reward function is: ,in, The overall feedback reward value for a multi-agent reinforcement learning controller. The instant reward component is calculated based on the current queue length at the intersection. The forward-looking reward component is calculated based on the predicted inflow. The dynamic weights of the signal timing gain coefficients.

[0017] Preferably, the method also includes step S701, which is an overflow safety protection mechanism, to monitor the occupancy rate of the downstream exit lane of the intersection signal controller. When the occupancy rate of the downstream exit lane reaches 90% and the duration exceeds 30 seconds, the green light duration of the upstream inflow phase is cut off, and the corresponding demand weight of the associated intersection in the anonymous origin-destination matrix statistics table is reduced by introducing a penalty factor to correct the output instructions of the multi-agent reinforcement learning controller.

[0018] Preferably, in step S102, a random bit stream is used to cover the physical memory sector where the license plate plaintext is located, and the intersection edge processing unit blocks unauthorized access requests to the physical memory sector.

[0019] Preferably, in step S104, the process of generating the anonymous origin-destination matrix statistics table includes: based on the preset road segment length between intersection numbers, removing license plate hash value entries with an average driving speed lower than 3km / h.

[0020] Preferably, the multi-agent reinforcement learning controller generates a cross-intersection phase-coordinated scheduling scheme by synchronizing the spatiotemporal evolution trend of the anonymous origin-destination matrix statistics table among agents at adjacent intersections.

[0021] A cross-intersection OD matrix construction and traffic prediction scheduling integrated system based on daily key dual-factor HMAC license plate hashing includes:

[0022] The edge processing unit is connected to the road sensing device to receive the license plate plaintext and read the daily updated random key. It uses the daily updated random key as a hash key to calculate the license plate hash value. After the calculation is completed, it physically overwrites the license plate plaintext in memory and encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform.

[0023] The central platform, which communicates with the edge processing unit, is used to maintain an ordered set of anonymous tracking messages in an in-memory database. It extracts vehicle travel time by comparing the same license plate hash value and generates an anonymous origin-destination matrix statistics table. The anonymous origin-destination matrix statistics table is then fused with real-time traffic flow data and input into the gating loop unit prediction model to calculate the predicted inflow of the target intersection.

[0024] The multi-agent reinforcement learning controller, connected to the central platform, is used to determine the signal timing gain coefficient based on the predicted inbound traffic flow, and dynamically adjust the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted inbound traffic flow and the measured traffic flow, so as to issue phase switching commands to the intersection signal controller.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] 1. By adopting a hash mechanism that combines plaintext license plates with periodically rotating random keys, and excluding spatial features of collection points in the hash calculation process, the anonymous feature stream has global consistency within a preset time window. This ensures that the original identity information cannot be restored, and achieves accurate association of cross-intersection paths. Furthermore, the continuity of historical trajectories is automatically cut off through key iteration, thus meeting the balance requirements of data security and business utilization in traffic control systems.

[0027] 2. The intersection sensing device quickly extracts features in memory and immediately overwrites the original cache. In conjunction with the central platform, it uses an ordered set structure in memory to asynchronously match anonymous message streams. It automatically clears invalid entries through preset time-sensoring logic, reducing the occupation of persistent storage resources by high cardinality feature data and compressing the processing cycle from multi-point sensing to the generation of the origin-destination matrix, thus providing low-latency data support for backend traffic evolution analysis.

[0028] 3. The trend component of the traffic forecast output is incorporated into the scheduling logic feedback loop in the form of a dynamic weight term, driving the controller to adjust the signal phase in advance according to the expected arrival volume. This feedforward adjustment mechanism based on path distribution characteristics makes the control behavior take precedence over physical queuing, and starts peak shaving processing in the early stage of trunk traffic fluctuation, improving the overall traffic efficiency of the road network and the foresight of signal scheduling. Attached Figure Description

[0029] Figure 1 This is a flowchart of the integrated method of the present invention;

[0030] Figure 2 This is a schematic diagram of the system operation logic and feedback control of the present invention.

[0031] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0033] A method for integrating cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing includes the following steps:

[0034] Step S101: The edge processing unit receives the plaintext of the license plate collected by the road sensing device, reads the daily updated random key generated by the central platform, uses the daily updated random key as the hash key to process the plaintext of the license plate, and calculates the license plate hash value.

[0035] In step S102, after calculating the license plate hash value, the edge processing unit physically overwrites the plaintext of the license plate in memory, and the residence period of the plaintext of the license plate in memory is less than 1 second.

[0036] In step S103, the edge processing unit encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform;

[0037] In step S104, the central platform maintains an ordered set with a preset storage period of 900s in the memory database. By comparing anonymous tracking messages with the same license plate hash value, it extracts the vehicle travel time between different intersection numbers and generates an anonymous origin-destination matrix statistics table.

[0038] In step S105, the central platform merges the anonymous origin-destination matrix statistics table with real-time traffic flow data, inputs it into the gating loop unit prediction model, and predicts the predicted inbound traffic flow of the target intersection within a preset future time period.

[0039] In step S106, the multi-agent reinforcement learning controller calculates the signal timing gain coefficient based on the predicted incoming flow, and dynamically adjusts the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted incoming flow and the actual flow measured by the road sensing device, and adjusts the signal phase switching command according to the weighting ratio.

[0040] Preferably, step S101 includes: step S1011, the central platform generates a new random key at 00:00 every day based on the global time synchronization protocol and sends it to each intersection edge processing unit. At the same time, it sends an old key physical destruction instruction to each intersection edge processing unit to synchronously trigger the old key in the memory of each intersection edge processing unit to perform multiple overwrite physical deletion.

[0041] Preferably, step S104 includes: step S1041, whereby the central platform automatically and physically deletes license plate hash value entries that have not been matched successfully for more than 900 seconds from the ordered set of the memory database.

[0042] Preferably, in step S105, the input data of the gated loop unit prediction model includes the vehicle turning ratio and path transition probability, which reflect the spatiotemporal evolution trend of road network traffic, extracted from the anonymous origin-destination matrix statistics table. The vehicle turning ratio and path transition probability are obtained by performing dimensionality reduction and aggregation processing on the spatiotemporal feature dimensions of the anonymous origin-destination matrix statistics table.

[0043] Preferably, in step S106, the adjustment logic for the weighted ratio includes: step S1061, calculating the absolute error between the predicted incoming flow and the actual measured flow at the corresponding time; step S1062, when the absolute error continues to exceed a preset judgment threshold, progressively reducing the dynamic weight of the signal timing gain coefficient in the reward function. The formula for calculating the reward function is: ,in, The overall feedback reward value for a multi-agent reinforcement learning controller. The instant reward component is calculated based on the current queue length at the intersection. The forward-looking reward component is calculated based on the predicted inflow. The dynamic weights of the signal timing gain coefficients.

[0044] Preferably, the method also includes step S701, which is an overflow safety protection mechanism, to monitor the occupancy rate of the downstream exit lane of the intersection signal controller. When the occupancy rate of the downstream exit lane reaches 90% and the duration exceeds 30 seconds, the green light duration of the upstream inflow phase is cut off, and the corresponding demand weight of the associated intersection in the anonymous origin-destination matrix statistics table is reduced by introducing a penalty factor to correct the output instructions of the multi-agent reinforcement learning controller.

[0045] Preferably, in step S102, a random bit stream is used to cover the physical memory sector where the license plate plaintext is located, and the intersection edge processing unit blocks unauthorized access requests to the physical memory sector.

[0046] Preferably, in step S104, the process of generating the anonymous origin-destination matrix statistics table includes: based on the preset road segment length between intersection numbers, removing license plate hash value entries with an average driving speed lower than 3km / h.

[0047] Preferably, the multi-agent reinforcement learning controller generates a cross-intersection phase-coordinated scheduling scheme by synchronizing the spatiotemporal evolution trend of the anonymous origin-destination matrix statistics table among agents at adjacent intersections.

[0048] A cross-intersection OD matrix construction and traffic prediction scheduling integrated system based on daily key dual-factor HMAC license plate hashing includes:

[0049] The edge processing unit is connected to the road sensing device to receive the license plate plaintext and read the daily updated random key. It uses the daily updated random key as a hash key to calculate the license plate hash value. After the calculation is completed, it physically overwrites the license plate plaintext in memory and encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform.

[0050] The central platform, which communicates with the edge processing unit, is used to maintain an ordered set of anonymous tracking messages in an in-memory database. It extracts vehicle travel time by comparing the same license plate hash value and generates an anonymous origin-destination matrix statistics table. The anonymous origin-destination matrix statistics table is then fused with real-time traffic flow data and input into the gating loop unit prediction model to calculate the predicted inflow of the target intersection.

[0051] The multi-agent reinforcement learning controller, connected to the central platform, is used to determine the signal timing gain coefficient based on the predicted inbound traffic flow, and dynamically adjust the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted inbound traffic flow and the measured traffic flow, so as to issue phase switching commands to the intersection signal controller.

[0052] Example 1: This example combines Figures 1 to 2 This paper describes an integrated method and system for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing. Figure 1 As shown, firstly, step S101 is executed, where the edge processing unit receives the plaintext license plate data collected by the road surface sensing device, reads the daily updated random key generated by the central platform, and uses this daily updated random key as the hash key to process the plaintext license plate data, calculating the license plate hash value. Then, in step S102, after calculating the license plate hash value, the edge processing unit physically overwrites the plaintext license plate data in memory, and the plaintext license plate data resides in memory for less than 1 second. Next, step S103 is executed, where the edge processing unit encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform. In step S104, the central platform maintains an ordered set in its memory database with a preset storage period of 900 seconds. By comparing anonymous tracking messages with the same license plate hash value, the travel time of vehicles between different intersection numbers is extracted and an anonymous origin-destination matrix statistical table is generated. Then, in step S105, the central platform merges the anonymous origin-destination matrix statistical table with real-time traffic flow data and inputs it into the gating loop unit prediction model to predict the predicted inbound traffic flow of the target intersection within a preset future time period. Finally, in step S106, the multi-agent reinforcement learning controller calculates the signal timing gain coefficient based on the predicted inbound traffic flow and dynamically adjusts the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted inbound traffic flow and the actual traffic flow measured by the road surface sensing device. The signal phase switching command is adjusted according to the weighting ratio.

[0053] like Figure 2As shown, after receiving the plaintext license plate and the daily updated random key, the system first processes the plaintext license plate and calculates the license plate hash value. Then, it performs a physical overwrite operation on the plaintext license plate and encapsulates and sends an anonymous tracking message, thus entering the maintenance ordered set stage. If the license plate hash value entries in the set fail to match within 900 seconds, the system triggers automatic physical deletion logic. If the match is successful, the system extracts the vehicle travel time to generate an anonymous origin-destination matrix statistical table and integrates it with real-time traffic flow data to calculate the predicted inbound flow. When the absolute error between the predicted inbound flow and the actual flow measured by the road surface sensing equipment continuously exceeds the preset judgment threshold, the system triggers a program to adjust the signal timing gain coefficient, thereby adjusting the signal phase switching command and feeding it back to the front-end plaintext license plate processing stage to form a closed loop. At the same time, the system independently monitors the downstream exit lane occupancy rate. When the downstream exit lane occupancy rate reaches 90% and lasts for more than 30 seconds, the system will take measures to cut off the green light duration of the inbound phase and reduce the demand weight of the corresponding associated intersection. This adjustment result directly affects the maintenance ordered set to adjust the data processing priority and prevent congestion from spreading.

[0054] Example 2: First, in the data privacy processing stage, the two factors of this invention refer to the plaintext license plate data as the identity factor and the daily-updated random key as the time factor. The edge processing unit uses the plaintext license plate data as the message input of the HMAC algorithm and the daily-updated random key as the key input. Through two-factor coupling calculation, the global uniqueness of the hash digest within the same natural day is ensured. At the same time, physical isolation of cross-day trajectories is achieved by using key iteration. To ensure memory security, the edge processing unit delineates the storage area as a restricted area through the kernel-mode direct memory access (DMA) protection mechanism, and the hardware-level memory management unit (MMU) intercepts unauthorized access requests. Specifically, the central... The platform distributes daily-updated random keys through an encrypted tunnel with SSL / TLS two-way authentication, and works with a heartbeat detection mechanism to ensure that the edge processing units at each intersection complete key synchronization on time at 00:00 every day. The HMAC algorithm preferably adopts the HMAC-SHA256 standard, in which the plaintext of the license plate is used as the message body and the daily-updated random key is used as the encryption factor. To further enhance the ability to resist rainbow table attacks, the system also introduces a dynamic salt value in the hash calculation process. This salt value is composed of the intersection number and the hourly timestamp, thus ensuring that even on the same day, the digests of the same vehicle at different times have time-sensitive characteristics, but maintain matchability within a 900-second logical window.

[0055] Within 1 second of the summary generation, a pseudo-random bitstream is continuously written three times to the physical address via the underlying driver. The repeated toggling of physical voltage levels eliminates the residual hysteresis effect of the storage medium. In the data cleaning and transformation stage, the central platform removes abnormal entries with an average driving speed below 3 km / h based on the road segment length to eliminate interference from non-traffic-related stops such as temporary loading and unloading at the roadside on statistical accuracy. Subsequently, the system aggregates discrete matching records into a multi-dimensional flow direction matrix in 900-second steps, and concatenates the turning distribution component with the cross-sectional flow sequence into a fixed-dimensional feature tensor input gated recurrent unit (GRU) model. Traffic flow inertia features are extracted from the model's hidden layer state, achieving dimensionality reduction from discrete individual data to continuous traffic flow trends. This invention introduces daily updates... The random key and 900s lifespan are not simply cryptographic applications, but rather a decoupling of statistical stationarity and identity uniqueness in traffic flow theory. The Gated Recurrent Unit (GRU) prediction model adopts a three-layer cascaded structure: the input layer contains 256 neurons, the hidden layer uses Tanh as the activation function to capture nonlinear fluctuations, and the output layer normalizes the predicted traffic through the Sigmoid function. Its core innovation lies in constructing a privacy-aware dynamic feedback loop: the daily key ensures the determinism of cross-intersection path associations within the same day, while the physical isolation of the next day's key cuts off the traceability of long-term trajectories. This design solves the industry problem of the incompatibility between personal privacy protection and fine-grained traffic scheduling in large-scale urban road networks.

[0056] In the multi-agent cooperative scheduling and feedback control phase, adjacent intersection agents achieve coordinated guidance of cross-intersection environmental states by real-time synchronization of dynamic transition probability vector packets encapsulated with traffic growth gradients and arrival time offsets. The multi-agent reinforcement learning controller adopts a comprehensive feedback reward function based on negative exponential normalization. The portion of instant rewards According to the formula Mapping queue length, forward prediction of reward component According to the formula The predicted inflow is mapped, thus unifying physical quantities with different dimensions into dimensionless control feedback values. Dynamic weights are used to ensure the stability of the control closed loop. The adjustment follows the damping attenuation principle: when the absolute error between the predicted flow and the measured flow exceeds a threshold, the system gradually reduces the damping. The controller strategy is adjusted to a conservative approach based on measured queue lengths to prevent system oscillations caused by incorrect predictions; after the error subsides, the system is then slowly restored. In addition to the weighting of the downstream exit lanes, the system independently monitors the downstream exit lane occupancy rate and executes a safety cutoff logic. When the occupancy rate reaches the saturation flow critical point of 90% and remains there for 30 seconds (to filter out instantaneous low-speed interference), the green light duration of the upstream inflow phase is forcibly cut off, and the demand weight of the associated intersections is reduced, thus blocking the spread of congestion overflow from the closed-loop system level. Furthermore, this invention nonlinearly couples the 900s anonymous dwell period with the reward function weight of reinforcement learning (RL). When the prediction deviation of the OD matrix exceeds a preset threshold, the multi-agent reinforcement learning controller is not limited to adjusting the signal timing, but corrects the input gain of the prediction model by adjusting the data aggregation priority of the ordered set in the memory database. Specifically, the system temporarily shortens the anonymous data cleaning step size of high-deviation intersections and increases the sampling frequency of this flow direction feature in the GRU model weight update, thereby realizing the direct conversion of the cryptographic time factor into the adjustment parameter of traffic control. This deep coupling ensures that the system can sacrifice some historical prediction weights to obtain highly sensitive feedback control of real-time deviations when facing sudden traffic congestion.

[0057] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0058] 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 technical solutions of the present invention.

Claims

1. A method for integrating cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing, characterized in that, Includes the following steps: Step S101: The edge processing unit receives the plaintext of the license plate collected by the road sensing device, reads the daily updated random key generated by the central platform, uses the daily updated random key as the hash key to process the plaintext of the license plate, and calculates the license plate hash value. In step S102, after calculating the license plate hash value, the edge processing unit physically overwrites the plaintext of the license plate in memory, and the residence period of the plaintext of the license plate in memory is less than 1 second. In step S103, the edge processing unit encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform; In step S104, the central platform maintains an ordered set with a preset storage period of 900s in the memory database. By comparing anonymous tracking messages with the same license plate hash value, it extracts the vehicle travel time between different intersection numbers and generates an anonymous origin-destination matrix statistics table. In step S105, the central platform merges the anonymous origin-destination matrix statistics table with real-time traffic flow data, inputs it into the gating loop unit prediction model, and predicts the predicted inbound traffic flow of the target intersection within a preset future time period. In step S106, the multi-agent reinforcement learning controller calculates the signal timing gain coefficient based on the predicted incoming flow, and dynamically adjusts the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted incoming flow and the actual flow measured by the road sensing device, and adjusts the signal phase switching command according to the weighting ratio.

2. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, Step S101 includes: Step S1011, the central platform generates a new random key at 00:00 every day based on the global time synchronization protocol and sends it to each intersection edge processing unit. At the same time, it sends an old key physical destruction instruction to each intersection edge processing unit to synchronously trigger the old key in the memory of each intersection edge processing unit to perform multiple overwrite physical deletion.

3. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, Step S104 includes: Step S1041, the central platform automatically and physically deletes license plate hash value entries that have not been matched successfully for more than 900 seconds in the ordered set of the memory database.

4. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, In step S105, the input data of the gated loop unit prediction model includes the vehicle turning ratio and path transition probability, which reflect the spatiotemporal evolution trend of road network traffic, extracted from the anonymous origin-destination matrix statistics table. The vehicle turning ratio and path transition probability are obtained by performing dimensionality reduction and aggregation processing on the spatiotemporal feature dimensions of the anonymous origin-destination matrix statistics table.

5. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, In step S106, the adjustment logic for the weighted ratio includes: step S1061, calculating the absolute error between the predicted incoming flow and the actual flow at the corresponding time; step S1062, when the absolute error continues to exceed the preset judgment threshold, progressively reducing the dynamic weight of the signal timing gain coefficient in the reward function. The formula for calculating the reward function is: ,in, The overall feedback reward value for a multi-agent reinforcement learning controller. The instant reward component is calculated based on the current queue length at the intersection. The forward-looking reward component is calculated based on the predicted inflow. The dynamic weights of the signal timing gain coefficients.

6. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, It also includes step S701, which is an overflow safety protection mechanism, to monitor the downstream exit lane occupancy rate of the intersection signal controller. When the downstream exit lane occupancy rate reaches 90% and lasts for more than 30 seconds, the green light duration of the upstream inflow phase is cut off. The corresponding demand weight of the associated intersection in the anonymous origin-destination matrix statistics table is reduced by introducing a penalty factor to correct the output instructions of the multi-agent reinforcement learning controller.

7. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, In step S102, a random bit stream is used to cover the physical memory sector where the license plate plaintext is located, and the intersection edge processing unit blocks unauthorized access requests to that physical memory sector.

8. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, In step S104, the process of generating the anonymous origin-destination matrix statistics table includes: based on the preset road segment length between intersection numbers, removing license plate hash value entries with an average driving speed of less than 3km / h.

9. The integrated method for cross-intersection OD matrix construction and traffic prediction scheduling based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, The multi-agent reinforcement learning controller generates a phase-coordinated scheduling scheme across intersections by synchronizing the spatiotemporal evolution trend of the anonymous origin-destination matrix statistics table among agents at adjacent intersections.

10. A cross-intersection OD matrix construction and traffic prediction scheduling integrated system based on daily key dual-factor HMAC license plate hashing, used to implement the cross-intersection OD matrix construction and traffic prediction scheduling integrated method based on daily key dual-factor HMAC license plate hashing as described in claim 1, characterized in that, include: The edge processing unit is connected to the road sensing device to receive the license plate plaintext and read the daily updated random key. It uses the daily updated random key as a hash key to calculate the license plate hash value. After the calculation is completed, it physically overwrites the license plate plaintext in memory and encapsulates the license plate hash value and intersection number into an anonymous tracking message and sends it to the central platform. The central platform, which communicates with the edge processing unit, is used to maintain an ordered set of anonymous tracking messages in an in-memory database. It extracts vehicle travel time by comparing the same license plate hash value and generates an anonymous origin-destination matrix statistics table. The anonymous origin-destination matrix statistics table is then fused with real-time traffic flow data and input into the gating loop unit prediction model to calculate the predicted inflow of the target intersection. The multi-agent reinforcement learning controller, connected to the central platform, is used to determine the signal timing gain coefficient based on the predicted inbound traffic flow, and dynamically adjust the weighting ratio of the signal timing gain coefficient in the reward function according to the deviation between the predicted inbound traffic flow and the measured traffic flow, so as to issue phase switching commands to the intersection signal controller.

Images