A cloud-edge collaborative device linkage early warning system for expressways

Abstract

The application discloses a highway-oriented cloud-edge cooperative equipment linkage early warning system and relates to the technical field of highway active safety management and control. The application discloses a highway-oriented cloud-edge cooperative equipment linkage early warning system and relates to the technical field of highway active safety management and control. The system comprises a physical event anchoring module arranged in a roadside edge node, which is used for generating an absolute timestamp carrying a space coordinate at a moment of sensing an abnormal event, and generating a dynamic urgency factor based on the correlation between the aging time length of the absolute timestamp and the relative distance between the space coordinate and a downstream oncoming vehicle. The highway-oriented cloud-edge cooperative equipment linkage early warning system forms a stable priority inversion inhibition relationship between the delivery delay of the emergency braking message and the regular heartbeat message, inhibits the cyclic influence caused by the mutual coupling of MAC backoff and TCP retransmission at two levels of the physical layer and the transmission layer, and guarantees that the key early warning signaling can still obtain a determined priority transmission service under extreme concurrent load.

Description

Technical Field

[0001] The present invention relates to the technical field of expressway active safety management and control, and specifically relates to a cloud-edge collaborative device linkage early warning system for expressways. Background Art

[0002] At present, early warning systems for expressway active safety management and control are gradually evolving from centralized cloud-end decision-making to a cloud-edge collaborative architecture of "cloud end + roadside edge node RSU + on-board terminal OBU". Under this architecture, abnormal events captured by roadside sensing devices need to be preprocessed by edge nodes and then reported to the cloud-end decision-making center, and early warning instructions are issued in linkage with downstream variable message signs and incoming vehicle terminals. Existing solutions are generally based on the standard TCP / IP protocol stack or lightweight protocols such as MQTT / CoAP, use 4G / 5G air interfaces to carry data transmission, and rely on the IEEE 1588 Precision Time Protocol PTP or Network Time Protocol NTP to perform multi-node clock synchronization, so as to realize time alignment and fusion of multi-source sensing data.

[0003] However, existing systems have prominent hidden dangers of protocol layer failure under extreme high-concurrency traffic flow scenarios. First, at the time synchronization level, when sudden congestion or a chain accident occurs on a local section of an expressway, hundreds of edge nodes and on-board terminals almost simultaneously trigger alarm messages, forming instantaneous MAC layer channel conflicts. Subject to the binary exponential backoff algorithm in CSMA / CA Carrier Sense Multiple Access, a large number of emergency message sending ends are forced to introduce random backoff windows to avoid collisions, resulting in that the actual sending time is severely lagged behind the occurrence time of the physical event. More severely, after this non-deterministic transmission jitter is superimposed with the NTP / PTP synchronization error, the message of the remote node far away from the event location may arrive at the cloud-end fusion unit earlier than the message of the proximal node due to lighter network congestion, resulting in a logical sequence inversion of "remote false alarm covering proximal real alarm", which directly undermines the basic premise of spatio-temporal association of multi-source data.

[0004] Second, at the traffic priority scheduling level, existing systems only rely on the DiffServ Differentiated Services model to perform simple DSCP priority marking at the IP layer, and lack a dynamic mapping mechanism for the spatial urgency and timeliness sensitivity of early warning messages. When the edge gateway accumulates a large number of early warning loads of different risk levels within a window, the Retransmission Timeout RTO mechanism of the underlying TCP and the MAC layer backoff mechanism are coupled with each other, which easily induces the emergency braking message of high risk level to be repeatedly lost and retransmitted, resulting in that its actual delivery delay is longer than that of the regular heartbeat message, that is, a priority inversion phenomenon is formed. This chronic defect directly leads to the failure of the cloud-edge collaborative decision engine to obtain an effective on-site situation snapshot within the critical window, and the linkage early warning instruction is issued with delay or falsely triggered, which seriously weakens the reliability of the system under extreme harsh working conditions.

[0005] The essence of the above problems lies in the fact that the general transmission protocol does not fully consider the dual constraints of tight spatial location coupling and strong time synchronization dependence in the highway scenario. There is an urgent need to design a new cooperative transmission and signaling scheduling mechanism to ensure the deterministic response capability of the early warning system. Summary of the Invention

[0006] The purpose of this invention is to provide a cloud-edge collaborative device linkage early warning system for highways, in order to solve the problems mentioned in the background art.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a cloud-edge collaborative device linkage early warning system for highways, comprising roadside edge nodes, an edge aggregation gateway, and a cloud-based decision engine, and further comprising:

[0008] The physical event anchoring module set in the roadside edge node is used to generate an absolute timestamp carrying spatial coordinates at the moment when an abnormal event is detected and physically triggered, and to generate a dynamic urgency factor based on the correlation between the aging time of the absolute timestamp and the relative distance of the spatial coordinates from the downstream vehicle.

[0009] An urgency-aware scheduler set within the edge aggregation gateway is used to extract the dynamic urgency factor from the pre-warning message to be sent, and to inversely constrain the MAC layer backoff parameters of the edge aggregation gateway based on the magnitude of the dynamic urgency factor.

[0010] The spatiotemporal ordered buffer pool set in the edge aggregation gateway is used to receive the early warning message sent by the MAC layer, and logically rearrange the arrival order according to the absolute timestamp and spatial coordinates in the early warning message before sending it to the cloud decision engine.

[0011] The cloud-based decision engine is used to receive the rearranged early warning message, calculate the spatiotemporal correlation confidence, and feed back the backpressure signaling to the spatiotemporal ordered buffer based on the spatiotemporal correlation confidence to adjust the sliding window step size of the logical rearrangement.

[0012] Preferably, the physical event anchoring module generates the absolute timestamp in the following way:

[0013] In response to the physical quantity change signal output by the roadside sensing device, the current time is directly locked by the hardware-level clock source as the absolute timestamp, and the absolute timestamp is bound to the spatial coordinates as an event fingerprint anchor frame. The event fingerprint anchor frame is independent of the subsequent IP layer encapsulation process.

[0014] Preferably, the dynamic urgency factor is characterized as follows:

[0015] A nonlinear urgency metric that increases with the aging duration of the absolute timestamp and decreases with the relative distance between the spatial coordinates and the downstream vehicle.

[0016] Preferably, the reverse constraint on the MAC layer backoff parameters specifically includes:

[0017] When the dynamic urgency factor is greater than the preset emergency intervention threshold, the urgency-aware scheduler forcibly compresses the minimum backoff window of the MAC layer of the corresponding early warning message to one-quarter of the standard protocol value and simultaneously shortens the inter-frame interval; when the dynamic urgency factor is less than or equal to the emergency intervention threshold, the urgency-aware scheduler allows the corresponding early warning message to perform channel access according to the conventional binary exponential backoff logic.

[0018] Preferably, the logical rearrangement specifically includes:

[0019] The spatiotemporal ordered buffer pool initially sorts the incoming early warning messages in ascending order based on the spatial coordinates and distance from the event source. It then uses the timestamp difference between adjacent messages in the initial sorting result as a correction factor and inputs it back into the adaptive sliding time window for secondary sorting, so that the delivery order of remote messages is proactively delayed compared to near-end messages.

[0020] Preferably, the cloud-based decision engine calculates the spatiotemporal correlation confidence level in the following way:

[0021] Based on the difference between the arrival timestamp of the rearranged early warning message and its carried absolute timestamp, and combined with the continuity of the spatial coordinates, a spatiotemporal correlation confidence level characterizing the spatiotemporal consistency of multi-source data is generated.

[0022] Preferably, the edge aggregation gateway further includes a dynamic retransmission timeout compensator, which is used to read the dynamic urgency factor carried by the message to be retransmitted, linearly reduce the TCP retransmission timeout base according to the dynamic urgency factor, and directly inject the reduced retransmission message into the absolute head position of the sending queue of the urgency-aware scheduler.

[0023] Preferably, the dynamic urgency factor is embedded in the custom extended header of the warning message to replace the static DSCP priority mark of the IP layer. The dynamic urgency factor is updated in real time in the transmission path as the absolute timestamp ages.

[0024] Preferably, the hardware-level clock source is connected to the GPS / BeiDou timing module to lock the absolute timestamp when the physical quantity change signal is triggered, and the accuracy of the absolute timestamp is not affected by the synchronization error of the network time protocol or the precise time protocol.

[0025] Preferably, when the cloud decision engine determines that the spatiotemporal correlation confidence is less than a preset qualified threshold, it sends the backpressure signaling to the spatiotemporal ordered buffer pool. The backpressure signaling includes a sliding window step size reduction instruction. The spatiotemporal ordered buffer pool responds to the backpressure signaling by reducing the current window width and triggers the re-execution of the logical rearrangement of the messages remaining in the window.

[0026] This invention provides a cloud-edge collaborative device linkage early warning system for highways. It has the following beneficial effects:

[0027] By constructing a physical event anchor timestamp generation mechanism independent of network time protocols and precision time protocols, the original time value carried by each early warning message is locked by hardware at the moment of physical perception and remains immutable throughout the entire subsequent transmission path. This mechanism eliminates the contamination of the event occurrence time by non-deterministic transmission jitter and network synchronization errors in general transmission protocols. Based on this, a spatiotemporally ordered buffer pool uses the original coordinate values ​​in the event fingerprint anchor frame as the initial sorting basis and uses the timestamp difference between adjacent messages as a correction factor to adjust the sliding window step size in reverse. This ensures that the arrival order is not affected by differences in the actual network transmission path, but is strictly rearranged according to the spatial logic of the event occurrence location. The cloud-based decision engine further calculates the spatiotemporal correlation confidence based on the rearranged message sequence and, when the confidence does not meet the requirements, feeds back to the buffer pool through a backpressure signaling closed loop, triggering multiple iterative rearrangements until the confidence meets the standard, thus forming a complete timing guarantee link of "physical anchoring - spatial rearrangement - confidence verification - backpressure correction". This mechanism ensures that remote messages, regardless of the reason why they arrive before near-end messages, can be proactively delayed in delivery within the buffer pool. This effectively eliminates the risk of the spatiotemporal correlation of multi-source sensing data being compromised and ensures that the message sequence received by the cloud-edge collaborative decision engine is consistent with the actual occurrence logic of physical events.

[0028] The dynamic urgency factor quantifies the spatial urgency and temporal aging of warning messages into a unified metric, and uses this factor as the sole basis for the reverse constraint of the MAC layer backoff parameters. This ensures that high-urgency messages receive a compressed minimum backoff window and a shortened inter-frame interval during channel contention, thus gaining higher channel access priority than regular messages at the protocol layer level. Simultaneously, the dynamic retransmission timeout compensator, for TCP retransmission scenarios, linearly reduces the retransmission timeout base based on the current value of the dynamic urgency factor and directly injects the retransmitted message into the absolute head position of the sending queue. This prevents the retransmission waiting time of high-risk messages from exponentially expanding with the number of retransmissions when encountering packet loss and retransmission; instead, it remains at a low level due to the aging compensation effect of the urgency factor. This mechanism eliminates the direct dependency on the static differentiated service label of the IP layer, so that priority scheduling no longer depends on simple labels in the upper-layer protocol stack, but is related to the real-time evolution of the threat level of events in a highway scenario. The urgency factor ages and updates synchronously with the queuing and propagation time of each hop in the message transmission path, ensuring that it accurately reflects the current urgency level at every moment before reaching the cloud. The entire scheduling mechanism establishes a stable priority inversion suppression relationship between the delivery delay of emergency braking messages and regular heartbeat messages. It suppresses the cyclical impact caused by the coupling between MAC backoff and TCP retransmission at both the physical and transport layers, ensuring that critical early warning signaling can still receive deterministic priority transmission service under extreme concurrent loads. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the module interaction of the cloud-edge collaborative device linkage early warning system for highways according to the present invention.

[0030] Figure 2 This is the dynamic urgency factor generation and flow diagram of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example: Please refer to Figure 1 and Figure 2 This invention provides a cloud-edge collaborative device linkage early warning system for highways:

[0033] The physical event anchoring module is deployed within the sensing and processing unit of each roadside edge node, and is directly connected to the data output interface of the roadside sensing device. When the sensing device outputs a sudden change signal in a physical quantity representing an abnormal event, such as a jump in radar echo intensity or a pixel frame difference in a visual image exceeding a judgment threshold, the physical event anchoring module immediately latches the current time value from the hardware-level clock source at the moment the change occurs. Simultaneously, it reads the positioning coordinates bound to the current edge node and encapsulates the time value and coordinate values ​​together into an event fingerprint anchoring frame. This anchoring frame maintains its initial time value and coordinate values ​​throughout all subsequent transmission and processing flows, and remains unchanged regardless of IP layer or MAC layer re-encapsulation, retransmission, or queue scheduling operations.

[0034] The physical event anchoring module further extracts the time value and coordinate value from the anchoring frame. Based on the time value, it calculates the time length from the occurrence of the abnormal event to the current scheduling time, and based on the coordinate value, it calculates the spatial interval length between the location of the abnormal event and the current location of the downstream vehicle. Then, it converts this time length and spatial interval length into a dynamic urgency factor according to a preset mapping relationship. This mapping relationship is configured as follows: aging time intervals are divided into several levels according to fixed time intervals, with each level assigned an increasing contribution value; relative distance intervals are divided into several levels according to fixed distance intervals, with each level assigned a decreasing contribution value. The number of time intervals is equal to the number of distance intervals. The two contribution values ​​are combined using an equal-weighted addition rule. This mapping relationship ensures that the longer the time length or the shorter the spatial interval, the higher the output dynamic urgency factor value. This dynamic urgency factor is embedded in the custom protocol header of the early warning message and transmitted downstream along with the early warning message.

[0035] The urgency-aware scheduler is deployed at the front end of the MAC layer transmission queue of the edge aggregation gateway. This scheduler continuously monitors the dynamic urgency factor value in the custom header of each pre-alarm message entering the transmission queue and compares this value with an internally preset emergency intervention threshold in real time. For pre-alarm messages with a dynamic urgency factor value higher than the threshold, the scheduler forcibly modifies the corresponding MAC layer minimum backoff window parameter to a proportion of the standard protocol value, while simultaneously shortening the inter-frame interval parameter to a proportion lower than the standard value, thus giving the message higher access priority than regular messages during channel contention. For pre-alarm messages with a dynamic urgency factor value not higher than the threshold, the scheduler allows them to access the channel according to the conventional binary exponential backoff logic. During this process, any pre-alarm message that triggers a retransmission due to channel conflict will have its current dynamic urgency factor value read again when it re-enters the transmission queue. If this value is still higher than the threshold, the scheduler repeats the parameter modification operations for the minimum backoff window and inter-frame interval until the message is successfully transmitted or the dynamic urgency factor naturally decays to below the threshold due to aging.

[0036] The spatiotemporally ordered buffer pool is deployed between the output of the edge aggregation gateway and the input of the cloud decision engine. Internally, this buffer pool maintains a storage queue with a sliding window structure. Warning messages, after being sent through the MAC layer, first enter this buffer pool. Upon receiving each warning message frame, the buffer pool extracts the original time value and original coordinate value from the event fingerprint anchor frame, without using the actual arrival order of the messages as the queue arrangement basis. The buffer pool initially sorts the messages in the queue according to their distance from the source of the abnormal event based on their original coordinate values, placing closer messages at the front of the queue. After the initial sorting, the buffer pool extracts the original time values ​​of two adjacent message frames, calculates the difference between them, and uses this difference as a correction factor input into the sliding window's window step size adjuster. The adjuster adjusts the window step size in the opposite direction based on the magnitude of this correction factor, causing the window to redetermine the message dequeue order based on the adjusted step size during the second queue traversal. Specifically, the dequeue order of far-end messages is actively shifted to the back of the queue for near-end messages. The buffer pool delivers messages to the cloud decision engine in the order of the secondary sorting.

[0037] After receiving the early warning message sequence delivered from the spatiotemporally ordered buffer, the cloud-based decision engine extracts the original time value carried by each frame within the current receiving window and compares it with the arrival time recorded locally by the engine. Simultaneously, it extracts the original coordinate values ​​of multiple adjacent frames for spatial trajectory continuity verification. Based on the above time comparison and spatial continuity verification results, the engine generates a confidence metric value characterizing the spatiotemporal consistency of multi-source data within the current receiving window. The cloud-based decision engine compares this confidence metric value with an internally preset pass / fail threshold. When the confidence metric value falls below this threshold, the engine sends a backpressure signaling message to the spatiotemporally ordered buffer, which includes a window step size reduction instruction. After receiving the backpressure signal, the buffer pool readjusts the width of the sliding window according to the reduced step size, and re-executes the above-mentioned initial sorting and correction factor back-substitution secondary sorting operations on all the delayed messages in the current window that have not yet been delivered. This process is repeated until the confidence metric value generated by the cloud decision engine for the received message sequence is not lower than the qualified judgment threshold. After that, the cloud decision engine outputs the message sequence to the backend linkage decision unit to generate the final warning instruction.

[0038] The physical event anchoring module is connected to the roadside sensing device via a high-speed data bus. The sensing device continuously outputs a processed sensing data stream to the module. The module contains an edge-triggered register, whose trigger pin is directly connected to the anomaly flag signal output pin of the sensing device. When the sensing device determines an anomaly after performing feature comparison on the raw radar echo or visual image, it simultaneously outputs a pulse-shaped physical quantity change signal to the edge-triggered register. Upon detecting the rising or falling edge of this pulse, the edge-triggered register immediately sends a latch request to the connected hardware-level clock source. The hardware-level clock source maintains a free-running counter independent of network time protocols and precise time protocols; this counter does not participate in any network-side time synchronization correction process. In response to the latch request, the hardware-level clock source freezes the counter value at the corresponding time value and outputs this time value to the binding register within the physical event anchoring module. Meanwhile, the physical event anchoring module reads the latest coordinate data from the positioning receivers deployed on the roadside edge nodes through a separate input channel. This coordinate data is then concatenated bit-wise with the aforementioned time value in a binding register to form an indivisible event fingerprint anchoring frame. After generation, this event fingerprint anchoring frame is written by the physical event anchoring module into a fixed-length field in its output buffer. The first half of this fixed-length field is the time value field, and the second half is the coordinate value field, separated by a separator identifier used to distinguish the boundary between the time value and the coordinate value during subsequent parsing. This output buffer is independent of the IP protocol stack's transmission buffer. After being written to the output buffer, the event fingerprint anchoring frame is transmitted directly to the link layer in its raw bitstream form for physical frame encapsulation without any encapsulation processing by the IP layer. During the link layer encapsulation process, the original time and coordinate values ​​of the event fingerprint anchor frame are placed in the payload header of the physical frame. This payload header retains its original bit content throughout subsequent MAC layer retransmissions, queue scheduling, and routing forwarding via the edge aggregation gateway. No intermediate node has the right to modify or overwrite the time and coordinate values ​​already written in the payload header. When a pre-alarm message is retransmitted multiple times due to channel conflicts, the time value in the event fingerprint anchor frame carried in each retransmission is always locked to the time of the first occurrence of the physical event, and does not change with the number of retransmissions or the accumulation of network latency. The physical event anchoring module also has an output acknowledgment loop. After the event fingerprint anchor frame is successfully read by the link layer, this loop sends a clear signal to the edge trigger register, restoring the edge trigger register to a ready-to-trigger state, preparing it to receive the input of the next physical quantity change signal.For multiple early warning messages generated after the same physical event is triggered, the physical event anchoring module assigns the same event fingerprint anchoring frame to the multiple early warning messages. The time value and coordinate value in the same event fingerprint anchoring frame are exactly the same, so that the subsequent spatiotemporal ordered buffer can identify the multiple early warning messages as a data set belonging to the same event source based on the same anchoring frame information.

[0039] After generating the event fingerprint anchoring frame, the physical event anchoring module outputs the anchoring frame to the urgency calculation submodule deployed within the same edge node. This urgency calculation submodule is internally configured with a clock counter read interface and a coordinate buffer. The clock counter read interface is used to obtain the current system time in real time from a hardware-level clock source, and the coordinate buffer is used to temporarily store the latest position coordinates of downstream vehicles obtained from the positioning receiver.

[0040] After parsing the original time value and event location coordinates from the event fingerprint anchor frame, the urgency calculation submodule performs the following two parallel parameter extraction operations:

[0041] In the first operation, the submodule performs a difference calculation between the current system time and the original time value to obtain the aging duration parameter, which continuously increases as the current system time progresses.

[0042] In the second operation, the submodule performs spatial distance calculations between the coordinates of the event occurrence location and the coordinates of the downstream vehicle location stored in the coordinate buffer to obtain a relative distance parameter. This relative distance parameter continuously decreases as the downstream vehicle moves towards the event occurrence location.

[0043] The urgency calculation submodule has a pre-set aging duration mapping table and a relative distance mapping table. The aging duration mapping table records the correspondence between multiple aging duration intervals and urgency contribution values, and the relative distance mapping table records the correspondence between multiple relative distance intervals and urgency contribution values.

[0044] The interval division of the aging duration mapping table is based on the historical average communication delay upper bound of the highway segment. The delay upper bound is divided into a set number of intervals, and the contribution values ​​corresponding to each interval are arranged in an arithmetic progression. The interval division of the relative distance mapping table is based on the safe following distance under the design speed of the highway segment. The safe distance is divided into a set number of intervals, and the contribution values ​​corresponding to each interval are arranged in an arithmetic progression. The number of intervals in the two mapping tables is equal to ensure the dimensional balance of the two contribution components when added together.

[0045] The urgency calculation submodule extracts the urgency contribution value corresponding to the interval into which the calculated aging duration parameter falls as the first contribution component, and extracts the urgency contribution value corresponding to the interval into which the calculated relative distance parameter falls as the second contribution component. The first and second contribution components are then combined according to an additive rule to obtain the base value of the dynamic urgency factor. In the aforementioned aging duration mapping table, a higher aging duration interval corresponds to a higher urgency contribution value; in the aforementioned relative distance mapping table, a lower relative distance interval corresponds to a higher urgency contribution value. The urgency calculation submodule also includes a rate of change detector, which continuously monitors the instantaneous change rate of the aging duration parameter and the instantaneous change rate of the relative distance parameter.

[0046] When the instantaneous rate of change of the aging duration parameter exceeds a preset rate threshold, which is set based on the upper confidence boundary of instantaneous flow change in historical traffic flow statistics of the road segment, the rate of change detector outputs an additional acceleration increment to the adder. This acceleration increment is a fixed proportion of the current base value. When the instantaneous rate of change of the relative distance parameter exceeds another preset rate threshold, which is calculated based on the maximum braking deceleration of the vehicle at the design speed of the road segment, the rate of change detector outputs another acceleration increment to the adder. This other acceleration increment is another fixed proportion of the current base value. This acceleration increment is also superimposed on the base value of the dynamic urgency factor, causing the dynamic urgency factor to generate a non-linear additional increase when the relative distance shortens rapidly. The rate of change detector is equipped with a hold timer. This hold timer starts timing after any acceleration increment is triggered. Before the timer reaches the predetermined hold duration, even if the instantaneous rate of change falls below the rate threshold, the rate of change detector continues to output the acceleration increment to avoid frequent jumps in the dynamic urgency factor due to parameter fluctuations in a short period of time.

[0047] The urgency calculation submodule outputs the dynamic urgency factor, after the aforementioned rate of change compensation, to the message encapsulation unit. The message encapsulation unit then writes this dynamic urgency factor into a fixed field in the pre-alarm message's custom header and forwards the pre-alarm message to the edge aggregation gateway. When the same event fingerprint anchor frame corresponds to multiple pre-alarm messages, the urgency calculation submodule uses the same current system time as a reference within the generation time window of these multiple pre-alarm messages, ensuring that the dynamic urgency factor carried by these multiple pre-alarm messages remains consistent within the generation time window.

[0048] The urgency-aware scheduler is deployed within the MAC layer transmission queue management unit of the edge aggregation gateway. This scheduler is connected to both the header read port and the tail write port of the transmission queue. Internally, the scheduler contains an emergency intervention threshold register. During system initialization, a fixed threshold is written to this register by the upper-layer management plane, and this fixed threshold remains unchanged during system operation. The scheduler reads the dynamic urgency factor value from the custom header of newly arriving pre-alarm messages at the tail of the transmission queue and compares this value with the threshold stored in the emergency intervention threshold register. The comparison result determines the scheduler's control path selection for the subsequent transmission of this pre-alarm message.

[0049] When the comparison result shows that the dynamic urgency factor value is greater than the emergency intervention threshold, the scheduler marks the warning message as a high-urgency message and triggers a MAC layer parameter reconfiguration operation for the message. In this reconfiguration operation, the scheduler issues a first write command to the backoff parameter register of the MAC layer protocol processor via the control bus. This first write command rewrites the minimum backoff window base stored in the backoff parameter register to one-quarter of the default minimum backoff window value specified by the IEEE 802.11 standard protocol. Simultaneously, the scheduler issues a second write command to the inter-frame interval parameter register of the MAC layer protocol processor via the same control bus. This second write command rewrites the current value of the inter-frame interval parameter register to a value smaller than the default inter-frame interval specified by the standard distributed coordination function, specifically one-third of the default inter-frame interval value. After these rewrites, when the warning message enters a channel contention state, the rewritten minimum backoff window base is used as the starting backoff window for the binary exponential backoff algorithm, and the rewritten inter-frame interval is used as the waiting time for the channel to become idle. The scheduler also sets a locking flag for the warning message. This locking flag prevents the rewritten backoff window base and inter-frame interval from being restored by the MAC layer protocol processor's automatic retransmission counter in subsequent retransmission attempts. The locking is only released when the message is successfully sent or actively discarded by the transmission queue management unit.

[0050] When the comparison result shows that the dynamic urgency factor is less than or equal to the emergency intervention threshold, the scheduler marks the warning message as a regular message and does not perform any write operations on the backoff parameter register and inter-frame interval parameter register of the MAC layer protocol processor. The warning message performs channel access in full accordance with the default minimum backoff window and default inter-frame interval specified by the protocol standard.

[0051] The scheduler also includes an anti-starvation timer, which is associated with the locking flag of the aforementioned high-urgency messages. The anti-starvation timer starts timing simultaneously with the locking flag being set. After the preset protection duration is reached, even if the message has not been successfully transmitted, the anti-starvation timer automatically releases the locking flag, allowing the message to revert to normal backoff parameters in subsequent retransmissions. This ensures that other lower-urgency messages in the same transmission queue have a chance to access the channel. The protection duration of the anti-starvation timer is dynamically determined by the scheduler based on the current depth of the transmission queue; the greater the current depth of the transmission queue, the shorter the protection duration.

[0052] For high-urgency messages that are retransmitted due to channel collisions or lack of acknowledgment at the receiver, the scheduler reads the dynamic urgency factor value of the message again when it re-enters the head of the transmission queue. If this value decays to less than or equal to the emergency intervention threshold due to aging during the message's retention period, the scheduler cancels the message's marking and parameter reconfiguration, allowing it to perform subsequent retransmission attempts according to normal backoff logic. If the value is still greater than the emergency intervention threshold at the time of retransmission, the scheduler keeps the lock mark valid and repeats the above-mentioned operation of rewriting the backoff window base and inter-frame interval before each retransmission attempt, ensuring that each retransmission attempt is performed under compressed backoff parameters.

[0053] The scheduler's comparison results and parameter rewriting actions are all recorded to the local storage medium through the internal log bus of the edge aggregation gateway. The record contains the timestamp of each comparison, the value of the dynamic urgency factor, the category of the comparison result, and the rewritten backoff window base and inter-frame interval values. This record is available for the upper management plane to retrieve during system debugging or fault backtracking.

[0054] The sliding window structure maintained internally by the spatiotemporally ordered buffer pool is defined by a window start pointer, a window end pointer, and a step register. The storage area between the window start pointer and the window end pointer constitutes the currently active window. The step value stored in the step register determines the address offset by which the window end pointer moves forward after each queue traversal. After receiving the pre-alarm message from the MAC layer, the buffer pool writes each frame of the message into the tail of the storage queue in the order of arrival. Simultaneously, it extracts the original coordinate values ​​from the event fingerprint anchor frame of each frame of the message and performs a difference operation between these original coordinate values ​​and the coordinates of the event occurrence location to obtain the distance metric value for each frame of the message. After completing the above enqueue operation, the buffer pool triggers the initial sorting process. In the initial sorting process, the buffer pool traverses all messages within the currently active window, using the distance metric value corresponding to each frame of the message as the sorting key value, and rearranges the physical storage locations of the messages within the window in ascending order of the key value, so that the message with the smallest distance metric value is placed at the front of the window, and the message with the largest distance metric value is placed at the back of the window, completing the initial ascending sort.

[0055] After the initial sorting, the buffer pool extracts the original time values ​​of two adjacent frames in the sorting result and calculates the difference between the original time value of the later frame and the original time value of the earlier frame, using this difference as the first correction factor. The buffer pool then extracts the difference between the distance metrics of the two adjacent frames and uses this distance difference as the second correction factor. The buffer pool combines the first and second correction factors into a correction vector, which is input to the step size adjuster of the adaptive sliding time window. The step size adjuster internally has a non-linear mapping table that records the correspondence between multiple correction vector intervals and step size adjustment amounts. Based on the interval into which the input correction vector falls, the step size adjuster reads the corresponding step size adjustment amount from the mapping table and outputs this adjustment amount to the step size register. The step size register updates the current step size value based on this adjustment amount, ensuring that the updated step size value differs from the previous step size value. In the above mapping table, the larger the value of the first correction factor, the smaller the corresponding step size adjustment will result in the updated step size value; the larger the value of the second correction factor, the larger the corresponding step size adjustment will result in the updated step size value. The two factors have opposite effects on the step size value.

[0056] After the step size register is updated, the buffer pool triggers a secondary sorting process using the updated step size value. In this secondary sorting process, the window end pointer moves forward from the window start pointer position according to the updated step size value. Each time the pointer moves by the address offset corresponding to the step size value, a packet at the current address passed by the window end pointer is selected sequentially and arranged in the temporary output queue in the order of selection. When the window end pointer moves beyond the original range of the window start and end pointers, the secondary sorting process ends, and the packet order in the temporary output queue is the delivery order after the secondary sorting. In this secondary sorting process, near-end packets with smaller distance metrics but relatively lagging original time values ​​are prioritized because they were already at the front of the window during the initial sorting, and are selected first as the window end pointer moves. Conversely, far-end packets with larger distance metrics but relatively earlier original time values ​​are delayed because they were at the back of the window during the initial sorting, and are selected later as the window end pointer moves. This results in a shift in the delivery order of far-end packets relative to their position in the initial sorting result.

[0057] After completing the secondary sorting, the buffer pool reads the first message in the temporary output queue from the storage queue and delivers it to the cloud decision engine. Then, the remaining messages in the temporary output queue are sequentially refilled into the storage queue to await the next delivery trigger. If the buffer pool receives a pre-alarm message for a newly enqueued message while waiting for delivery, it does not immediately interrupt the currently completed secondary sorting. Instead, it temporarily stores the newly enqueued message in the tail isolation area of ​​the storage queue. Once all messages in the current temporary output queue have been delivered, the new message in the isolation area is merged into the main storage queue, and the initial sorting process is retried. The buffer pool also has a sorting counter that records the positional offset direction and amount of each message frame relative to the initial sorting result after secondary sorting. This record, bound to the message's event fingerprint anchor frame, is delivered to the cloud decision engine along with the message. This allows the cloud decision engine to refer to the message's historical sorting adjustment records within the buffer pool when performing spatial trajectory continuity verification.

[0058] The cloud-based decision engine internally deploys a spatiotemporal correlation confidence calculation unit. The inputs of this unit are connected to the message receiving buffer and the system clock interface. The message receiving buffer stores the complete sequence of early warning messages within the current receiving window, delivered from the spatiotemporally ordered buffer pool. The system clock interface provides the calculation unit with the engine's local current reference time. The calculation unit extracts the original time value and original coordinate value from each message frame according to the order of messages in the receiving window. It then performs a difference operation between the original time value and the engine's local current reference time to obtain the transmission delay observation value for that message frame. The calculation unit compares the transmission delay observation value of each message frame with the transmission delay observation values ​​of previous messages from the same event source stored in its historical buffer, generating a delay consistency score for that message frame. This score characterizes the degree of deviation of the current message from historical messages from the same event source in terms of transmission time.

[0059] The calculation unit simultaneously extracts the original coordinate values ​​of two adjacent frames of messages from the current receiving window, performs spatial span calculations on these two frames, and obtains the spatial displacement between adjacent messages. The calculation unit then converts this spatial displacement with the downstream vehicle speed information carried by the previous frame of messages within the window to obtain a theoretical displacement reference value, and uses the deviation between the spatial displacement and the theoretical displacement reference value as the displacement deviation. The calculation unit sequentially performs the above calculations of spatial displacement and displacement deviation for all adjacent message pairs within the current receiving window, and assigns a spatial continuity score to each pair of adjacent messages based on the deviation range into which the displacement deviation falls. This spatial continuity score is higher when the displacement deviation is in a low deviation range than when it is in a high deviation range. The calculation unit performs a weighted addition operation on the delay consistency score corresponding to the same frame of message and its spatial continuity score as the next message to obtain the individual confidence sub-score of the frame of message. The weight coefficients of the delay consistency score and the spatial continuity score are pre-configured during system initialization based on the average traffic speed of the road segment and the typical round-trip delay experience value of the wireless communication link. The sum of the two weight coefficients is equal to a fixed constant. The arithmetic mean of the individual confidence sub-scores of all messages in the current receiving window is performed to obtain the spatiotemporal correlation confidence.

[0060] The calculation unit also includes an outlier remover. Before calculating the arithmetic mean, this outlier remover performs outlier detection on the individual confidence sub-scores of each frame within the window. The outlier remover calculates the median and absolute median difference of all individual confidence sub-scores within the window. Messages whose individual confidence sub-score differs from the median by more than three times the absolute median difference are identified as outliers and temporarily removed from the dataset to be averaged, thus not participating in the arithmetic mean calculation. The message sequence number of the removed outlier is recorded in the anomaly log and output to the backend processing unit along with the confidence score calculation results.

[0061] If multiple consecutive receiving windows fail to reach the pass / fail threshold, the calculation unit also includes a trend analyzer. This trend analyzer records the trajectory of recent confidence calculations and fits the slope of this trajectory. When the slope is negative and its absolute value exceeds a preset trend threshold, the trend analyzer outputs an acceleration trigger signal to the backpressure signaling generator. This acceleration trigger signal causes the backpressure signaling generator to issue backpressure signaling with doubled reduction, thereby accelerating the adjustment speed of the window step size by the spatiotemporally ordered buffer. After each round of confidence calculation, the calculation unit copies the original time value and original coordinate value of each frame in the current receiving window to the storage location corresponding to the event source identifier in the history buffer, for use in the differential comparison and spatial continuity verification benchmark update in the next round of calculation.

[0062] The dynamic retransmission timeout compensator is deployed on the TCP / IP protocol stack processing path of the edge aggregation gateway, located between the TCP retransmission timer management unit and the MAC layer sending queue management unit. The input of the compensator is connected to the output of the TCP retransmission timer management unit, and the output is connected to the sending queue entry of the urgency-aware scheduler. When the TCP retransmission timer management unit determines that a pre-alarm message has not received an acknowledgment from the receiver after the acknowledgment timer has expired, the message is marked as a retransmission message and sent to the input buffer of the compensator, instead of being directly sent to the MAC layer sending queue.

[0063] The compensator reads the custom header of the retransmitted packet from the input buffer and extracts the dynamic urgency factor value carried within. Internally, the compensator has a base value register that stores the standard retransmission timeout base corresponding to the current TCP connection. This standard retransmission timeout base is determined by the TCP protocol stack's slow start algorithm and round-trip time estimation algorithm during connection establishment. The compensator also has an attenuation coefficient table, which records the attenuation coefficients corresponding to multiple dynamic urgency factor intervals. The compensator compares the extracted dynamic urgency factor value with the intervals in the attenuation coefficient table and reads the attenuation coefficient corresponding to the interval in which the dynamic urgency factor value falls. The compensator then multiplies the standard retransmission timeout base stored in the base value register with the read attenuation coefficient to obtain the reduced retransmission timeout value. In the aforementioned attenuation coefficient table, the higher the dynamic urgency factor value, the lower the corresponding attenuation coefficient value, resulting in a proportional contraction of the reduced retransmission timeout value relative to the standard retransmission timeout base.

[0064] After calculating the reduced retransmission timeout value, the compensator writes this value back to the initial timer register of the TCP retransmission timer management unit, replacing the original standard retransmission timeout base. This ensures that the reduced initial timer value is used for timeout counting in subsequent retransmission attempts. Simultaneously, the compensator generates a control command containing the reduced retransmission timeout value and the packet's target position identifier in the transmission queue. The compensator sends this control command along with the retransmission packet to the urgency-aware scheduler's transmission queue entry, forcibly specifying the head position of the transmission queue in the target position identifier of the control command. Upon receiving the control command, the urgency-aware scheduler writes the retransmission packet directly to the front of the transmission queue, placing it before all regular packets already waiting to be transmitted and other packets awaiting retransmission. When the channel becomes idle, the MAC layer prioritizes processing this retransmission packet.

[0065] The compensator also incorporates an attenuation amplitude limiter, which is connected in series between the base register and the multiplication unit. This limiter constrains the read attenuation coefficient, forcing it up to the pre-stored lower threshold before sending it to the multiplication unit when the coefficient falls below it. This prevents unnecessary network congestion caused by excessively short retransmission timeout values ​​after reduction. The lower threshold of the attenuation amplitude limiter is dynamically adjusted by the compensator based on the current transmission queue depth; a higher transmission queue depth results in a higher lower threshold, maintaining the basic transmission order of other packets in the transmission queue.

[0066] If a pre-alarm message fails to transmit successfully in multiple retransmission attempts and the dynamic urgency factor continues to decay due to aging, the compensator re-executes the complete operation of reading the dynamic urgency factor, looking up the attenuation coefficient, performing multiplication, and writing the initial value of the timer each time a new retransmission is triggered. If the dynamic urgency factor value extracted in a retransmission attempt has decayed to the lowest interval in the attenuation coefficient table, corresponding to an attenuation coefficient of one, the compensator will not perform any reduction operation on subsequent retransmissions of that message, will resume using the standard retransmission timeout base for timing, and will release the message from the head of the sending queue to the tail of the queue, allowing it to queue for transmission in the normal order. The compensator records the dynamic urgency factor value, the reduced retransmission timeout value, and the actual number of retransmissions for each retransmission message in each retransmission attempt in the local retransmission log. This log can be accessed by the cloud operation and maintenance platform through the remote management interface of the edge aggregation gateway.

[0067] During the protocol stack encapsulation process of the early warning message at the edge aggregation gateway, the message encapsulation unit inserts a custom extended header between the IP header and the upper-layer protocol payload. This custom extended header consists of a header identifier field, a header length field, a urgency field, and reserved fields in that order. The header identifier field is filled with a fixed type encoding value to enable the receiver's parser to recognize the presence of the extended header; the header length field contains the total byte length of the custom extended header; the urgency field occupies a fixed bit width in the custom extended header and is used to write the current value of the dynamic urgency factor; the reserved fields are reserved for subsequent protocol extensions. When constructing the custom extended header, the message encapsulation unit does not modify the original Differentiated Services code point field in the IP header; that is, the Differentiated Services code point field in the IP header remains at zero or its default value and does not carry any marking information related to the warning priority. This ensures that the priority scheduling of the early warning message is entirely handled independently by the urgency field in the custom extended header, rather than relying on the IP layer's Differentiated Services model.

[0068] Custom extension headers are forwarded along with the message throughout the entire transmission path of each pre-warning message. When a pre-warning message passes through an intermediate edge node or edge aggregation gateway on the path, the forwarding processing unit in the intermediate node, after parsing the IP header and before delivering the payload to the upper-layer protocol, first reads the header identifier field in the custom extension header. After confirming the presence of the extension header by comparing the type encoding value, the forwarding processing unit then extracts the current value of the dynamic urgency factor stored in the urgency field. The forwarding processing unit compares this current value with the current time provided by the node's local hardware clock source, calculates the path propagation time of the dynamic urgency factor since it was sent from the previous hop node, and adds this propagation time to the original aging time component of the dynamic urgency factor to obtain the updated aging time parameter. Based on the updated aging time parameter, the forwarding processing unit recalculates the updated value of the dynamic urgency factor according to the same mapping rule as the urgency calculation submodule, writes the updated value back to the urgency field of the custom extension header of the same pre-warning message, overwriting the original factor value, and then continues to forward the message to downstream nodes. During the recalculation process described above, the spatial coordinate components of the dynamic urgency factor remain unchanged, while only its aging duration component increases with the actual time consumed on the transmission path. This ensures that the dynamic urgency factor is refreshed to reflect the true urgency of the message at the current transmission moment each time the message passes through a hop node.

[0069] The custom extended header also includes an update counter field, which is initialized to zero during packet encapsulation. Each time the forwarding processing unit performs an overwrite operation on the urgency field, it increments the current value of the update counter field to record the number of hops the packet has been updated on the transmission path. The forwarding processing unit has an internal maximum update threshold. Before each write operation, the forwarding processing unit reads the current value of the update counter field. If the current value has reached the maximum update threshold, the forwarding processing unit will not perform the overwrite operation on the urgency field and will simply forward the packet as is, preventing the dynamic urgency factor from being over-updated during long-path transmission, which could lead to value overflow or flipping.

[0070] For multiple early warning messages originating from the same edge aggregation gateway that belong to the same event source and carry the same event fingerprint anchor frame, the message encapsulation unit independently constructs a custom extended header for each frame, ensuring that the initial write value of the urgency field in the custom extended header of each frame remains consistent. However, due to differences in the number of routing hops and queuing time of each hop in the transmission path, the timing and magnitude of the update operation performed by intermediate nodes on the dynamic urgency factor of each frame differ. This results in different final values ​​in the urgency field of each frame when it reaches the cloud decision engine. These differences indirectly reflect the differences in transmission paths and queuing delays experienced by each frame, providing a reference signal for the cloud decision engine when calculating confidence. After the early warning message finally reaches the cloud decision engine, the engine's parser reads the final version of the dynamic urgency factor value and update counter value from the custom extended header, and uses the final version factor value in the time delay consistency scoring stage of the spatiotemporal correlation confidence calculation. The final version factor value replaces the direct difference between the message arrival time and the original time value as the basis for representing the degree of time delay deviation, so that the time delay consistency score also covers the path hop count factor.

[0071] The hardware-level clock source within the physical event anchoring module consists of a free-running counter driven by a high-precision crystal oscillator and a timing interface unit. The timing interface unit connects to a Global Navigation Satellite System (GNSS) receiver module deployed on the roadside edge node via a physical serial bus. This GNSS receiver module is compatible with both GPS and BeiDou Navigation Satellite System dual-mode signal reception. The timing interface unit reads Coordinated Universal Time (UTC) reference values ​​from the GNSS receiver module at fixed time intervals and writes these reference values ​​into the phase calibration register of the free-running counter. During normal operation, the free-running counter relies entirely on the output pulses of its internal high-precision crystal oscillator for incrementing, without receiving any clock correction commands from the network side via network time protocols or precise time protocols, and does not participate in any clock servo loop adjustment process of the network time synchronization protocol stack. The Coordinated Universal Time (UTC) reference value output by the Global Navigation Satellite System (GNSS) receiver module is only used to set the initial count value of the free-running counter during the system startup phase, and to constrain the cumulative drift of the free-running counter at fixed intervals during long-term system operation. This boundary constraint forces the deviation between the current count value of the free-running counter and the UTC reference value to be compressed into a fixed window through hard calibration. The width of this fixed window is jointly determined by the frequency stability of the crystal oscillator and the calibration period of the time synchronization interface unit, and the width of this fixed window is smaller than the time synchronization error range that the network time protocol can achieve under the same network conditions.

[0072] When the edge-triggered register in the physical event anchoring module detects a sudden change in a physical quantity, it immediately outputs a latch enable signal to the hardware-level clock source. Within the same hardware clock cycle that receives the latch enable signal, the hardware-level clock source freezes the instantaneous count value of the free-running counter to the output latch. This freezing operation does not involve any software interrupt handling or protocol stack scheduling; the time span from the arrival of the latch enable signal to the count value fully stabilizing in the output latch is determined solely by the propagation delay of the hardware gate circuitry. The count value in the output latch is then read and written to the time value field of the event fingerprint anchoring frame.

[0073] The count value of the aforementioned free-running counter is unaffected by the Coordinated Universal Time (UTC) reference value update operation output by the Global Navigation Satellite System (GNSS) receiver module during the process of receiving the latch enable signal and freezing the edge-triggered register. The hard calibration operation of the timing interface unit and the latch operation of the edge-triggered register are driven by independent clock domains at the hardware level. Data exchange between the two clock domains is achieved through an asynchronous first-in-first-out (FIFO) buffer, preventing glitches or transient disturbances caused by the timing calibration signal at the moment of latching to the output of the free-running counter. The count value of the free-running counter continues to increment when no latch enable signal is received. Even if the interval between two adjacent latch operations spans multiple timing calibration cycles, the count value obtained from each latch uniquely corresponds to the instantaneous state of the free-running counter at that latching moment, and does not backtrack or jump due to hard calibration adjustments of the count value during the timing calibration operation.

[0074] The hardware-level clock source also features a verification readback path. After each event fingerprint anchor frame is read, this path performs a hardware-level bit comparison between the frozen count value stored in the output latch and the value written to the time value field. If the comparison matches, a latch completion confirmation signal is sent to the edge-triggered register. If the comparison fails, the hardware-level clock source automatically re-latches the count value in the output latch and rewrites it to the time value field, while simultaneously recording this re-latch event in the error counter of the verification readback path. The error counter's count value is provided to the outside world through the status register of the physical event anchoring module, allowing maintenance personnel to read the count value via a handheld terminal during on-site inspections to assess the operational health of the hardware-level clock source. The processing of physical quantity change signals between each edge node is performed independently. Within a single sensing input channel of the same edge node, the edge trigger register is only allowed to return to the ready-to-trigger state after receiving a latch completion confirmation signal. Before receiving the confirmation signal, the edge trigger register shields all subsequent physical quantity change signals to avoid triggering new latch operations when the clock source is in a frozen output or verification readback state, which would lead to data confusion.

[0075] After calculating the spatiotemporal correlation confidence score, the cloud-based decision engine sends the confidence score result to a threshold comparator inside the engine. This threshold comparator pre-stores a pass / fail threshold value, which is configured and stored in non-volatile memory based on the actual communication environment of the highway segment during system deployment. The threshold comparator compares the confidence score value with the pass / fail threshold value using binary values ​​and outputs the comparison result to the trigger input of the backpressure signaling generator.

[0076] When the comparison result indicates that the confidence level is less than the pass / fail threshold, the backpressure signaling generator is triggered. The backpressure signaling generator first reads the step size value from the sliding window step size register of the current spatiotemporally ordered buffer pool, and simultaneously reads the number of outlier samples recorded by the outlier sample remover during this confidence level calculation. Based on the proportion of outlier samples to the total number of messages in the current received window, the backpressure signaling generator determines a step size reduction ratio coefficient, which is positively correlated with the proportion of outlier samples. The backpressure signaling generator multiplies the current step size value by this step size reduction ratio coefficient to obtain the target reduction amount, and encodes this target reduction amount as a window step size reduction instruction, writing it into the message body field of the backpressure signaling. This message body field also includes an instruction type identifier and a sequence number field. The instruction type identifier is filled with a type code value representing the step size reduction operation, and the sequence number field is filled with an auto-incrementing sequence number maintained locally by the cloud decision engine to uniquely identify each round of backpressure signaling. After the backpressure signaling generator encapsulates the complete backpressure signaling into a control message, it sends it to the control interface of the spatiotemporally ordered buffer pool through a dedicated control channel between the cloud decision engine and the edge aggregation gateway.

[0077] The spatiotemporally ordered buffer pool's control interface continuously monitors control messages on the dedicated control channel. Upon receiving a backpressure signaling message with a matching type-coded value, the buffer pool first parses the sequence number field in the message body and compares this sequence number with the sequence number of the last backpressure operation recorded internally by the buffer pool. If the current sequence number is less than or equal to the previously recorded sequence number, the buffer pool discards the backpressure signaling message and does not execute it, to prevent delayed old backpressure signaling messages from the network from causing unexpected rollback effects on the current window state. If the current sequence number is greater than the previously recorded sequence number, the buffer pool writes the window step size reduction instruction from the message body into the update port of the step size register. The step size register subtracts the target reduction amount from the current step size value according to the reduction instruction and stores the new step size value in the register. At the same time, the sequence number is written into the last execution record and the record is updated.

[0078] After the step size register is updated, the spatiotemporal ordered buffer pool identifies all pending messages in the current storage queue that have not yet been delivered to the cloud decision engine. The buffer pool marks these pending messages as a set to be rearranged and re-executes the logical rearrangement operation with the updated step size value. During the rearrangement process, the buffer pool sequentially places the pending messages into a new temporary output queue according to the new delivery order generated after the re-execution of the logical rearrangement operation. After the buffer pool completes the rearrangement, it reads the messages from the temporary output queue frame by frame and delivers them to the cloud decision engine again. Upon receiving the re-delivered message sequence, the cloud decision engine re-executes the spatiotemporal correlation confidence calculation process and re-sends the recalculated confidence value to the threshold comparator for comparison with the pass / fail judgment threshold value. If the recalculated confidence level is still less than the pass / fail threshold, the cloud decision engine triggers the backpressure signaling generator to issue a new round of backpressure signaling. The spatiotemporal ordered buffer responds to the new round of backpressure signaling by reducing the step size again and performing logical rearrangement and redelivery on the delayed messages. This cycle repeats until the confidence level recalculated by the cloud decision engine is not lower than the pass / fail threshold.

[0079] The backpressure signaling generator has an internal iteration counter that records the number of times backpressure signaling is continuously sent to the same event source message sequence. The iteration counter is connected in series with the trigger input of the backpressure signaling generator. Before each trigger, the iteration counter reads its current count value and compares it with a pre-stored maximum iteration threshold. If the current count value has reached the maximum iteration threshold, the backpressure signaling generator is disabled and no further backpressure signaling is sent. Simultaneously, the iteration counter outputs an iteration overflow flag to the alarm management unit of the cloud decision engine. The alarm management unit converts this overflow flag into an event log and stores it in the cloud database. If the current count value has not reached the maximum iteration threshold, the backpressure signaling generator triggers normally and increments the current count value by one, writing it to the update input of the iteration counter. The maximum iteration threshold is less than or equal to three to avoid control-level oscillations caused by continuous and repeated backpressure signaling interactions between the cloud and the edge.

[0080] After the confidence level is not lower than the pass / fail threshold, the cloud decision engine outputs the final confirmed message sequence within the current receiving window, after the aforementioned backpressure closed-loop adjustment, to the backend linkage decision unit. The linkage decision unit generates a warning instruction for downstream variable message signs and vehicle terminals based on the event fingerprint anchor frame and dynamic urgency factor in the final confirmed message sequence, and sends this warning instruction to the corresponding execution device via the downlink. Upon detecting that the cloud decision engine has output the final confirmed message sequence, the spatiotemporal ordered buffer restores the step size value in the current step size register to the default step size value set during system initialization, clears the iteration counter, and returns to the initial state awaiting the next batch of warning message sequences. Simultaneously, it clears the storage space occupied by delivered messages in the current storage queue to release buffer capacity for subsequent newly enqueued warning messages.

[0081] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A cloud-edge collaborative device linkage early warning system for highways, comprising roadside edge nodes, edge aggregation gateways, and a cloud decision engine, characterized in that, Also includes: The physical event anchoring module set in the roadside edge node is used to generate an absolute timestamp carrying spatial coordinates at the moment when an abnormal event is detected and physically triggered, and to generate a dynamic urgency factor based on the correlation between the aging time of the absolute timestamp and the relative distance of the spatial coordinates from the downstream vehicle. An urgency-aware scheduler set within the edge aggregation gateway is used to extract the dynamic urgency factor from the pre-warning message to be sent, and to inversely constrain the MAC layer backoff parameters of the edge aggregation gateway based on the magnitude of the dynamic urgency factor. The spatiotemporal ordered buffer pool set in the edge aggregation gateway is used to receive the early warning message sent by the MAC layer, and logically rearrange the arrival order according to the absolute timestamp and spatial coordinates in the early warning message before sending it to the cloud decision engine. The cloud-based decision engine is used to receive the rearranged early warning message, calculate the spatiotemporal correlation confidence, and feed back the backpressure signaling to the spatiotemporal ordered buffer based on the spatiotemporal correlation confidence to adjust the sliding window step size of the logical rearrangement.

2. The cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The physical event anchoring module generates the absolute timestamp in the following way: In response to the physical quantity change signal output by the roadside sensing device, the current time is directly locked by the hardware-level clock source as the absolute timestamp, and the absolute timestamp is bound to the spatial coordinates as an event fingerprint anchor frame. The event fingerprint anchor frame is independent of the subsequent IP layer encapsulation process.

3. The cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The dynamic urgency factor is characterized as follows: A nonlinear urgency metric that increases with the aging duration of the absolute timestamp and decreases with the relative distance between the spatial coordinates and the downstream vehicle.

4. The cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The reverse constraint on the MAC layer backoff parameters specifically includes: When the dynamic urgency factor is greater than the preset emergency intervention threshold, the urgency-aware scheduler forcibly compresses the minimum backoff window of the MAC layer of the corresponding early warning message to one-quarter of the standard protocol value, and simultaneously shortens the inter-frame interval. When the dynamic urgency factor is less than or equal to the emergency intervention threshold, the normal retreat logic is executed.

5. A cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The logical rearrangement specifically refers to: The spatiotemporal ordered buffer pool initially sorts the incoming early warning messages in ascending order based on the spatial coordinates and distance from the event source. It then uses the timestamp difference between adjacent messages in the initial sorting result as a correction factor and inputs it back into the adaptive sliding time window for secondary sorting, so that the delivery order of remote messages is proactively delayed compared to near-end messages.

6. A cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The cloud-based decision engine calculates the spatiotemporal correlation confidence score in the following way: Based on the difference between the arrival timestamp of the rearranged early warning message and its carried absolute timestamp, and combined with the continuity of the spatial coordinates, a spatiotemporal correlation confidence level characterizing the spatiotemporal consistency of multi-source data is generated.

7. A cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The edge aggregation gateway also includes a dynamic retransmission timeout compensator, which reads the dynamic urgency factor carried by the message to be retransmitted, linearly reduces the TCP retransmission timeout base according to the dynamic urgency factor, and directly injects the reduced retransmission message into the absolute head position of the sending queue of the urgency-aware scheduler.

8. A cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, The pre-warning message's custom extended header embeds the dynamic urgency factor, replacing the static DSCP priority flag at the IP layer. The dynamic urgency factor is updated in real time along the transmission path as the absolute timestamp ages.

9. A cloud-edge collaborative device linkage early warning system for highways according to claim 2, characterized in that, The hardware-level clock source is connected to the GPS / BeiDou timing module and is used to lock the absolute timestamp when the physical quantity change signal is triggered. The accuracy of the absolute timestamp is not affected by the synchronization error of the network time protocol or the precise time protocol.

10. A cloud-edge collaborative device linkage early warning system for highways according to claim 1, characterized in that, When the cloud-based decision engine determines that the spatiotemporal correlation confidence is less than a preset qualified threshold, it sends the anti-pressure signaling to the spatiotemporal ordered buffer pool. The anti-pressure signaling includes a sliding window step size reduction instruction. The spatiotemporally ordered buffer pool responds to the backpressure signaling by reducing the current window width and triggers the re-execution of the logical rearrangement of messages remaining in the window.

Images