Intelligent road system for car-road cooperation safety warning

Abstract

The application discloses an intelligent road system for vehicle-road cooperation safety warning and relates to the technical field of vehicle-road cooperation warning, and comprises a communication evaluation module, an optimal communication path is input into a communication mapping model, a space matching layer is used for fusing road geometry and communication topology, a state mapping layer is used for marking signal strength and stud node state, and a communication quality heat map is generated; a risk decision module is used for evaluating the risk level of each region based on the communication quality heat map and outputting priority transmission parameters; and an information transmission module is used for starting an audible and visual alarm according to the priority transmission parameters, acquiring an execution state report, and feeding back the execution state report to a command center. Through the construction of the communication mapping model and the dynamic generation of the priority transmission parameters, the application improves the precision of communication state evaluation, realizes efficient cooperative processing of multi-source alarm information and intelligent allocation of priorities.

Description

Technical Field

[0001] This invention relates to the field of vehicle-road cooperative alarm technology, and in particular to an intelligent road system for vehicle-road cooperative safety alarm. Background Technology

[0002] Road traffic safety is a global concern, and as the country with the world's longest highway network, China places extreme importance on road traffic safety. Currently, when a road accident occurs, information is typically transmitted through various alarm systems or by placing warning triangles at designated safety distances to prevent secondary accidents. In recent years, vehicle-road cooperative methods based on 5G, edge computing, and artificial intelligence have further optimized real-time data fusion and decision-making capabilities, such as dynamic traffic signal control and collision warning systems.

[0003] However, existing vehicle-road cooperative safety alarm systems have two shortcomings: First, traditional risk level assessment models struggle to effectively integrate vehicle dynamic data, road environment parameters, and historical accident characteristics, resulting in limited risk prediction accuracy and difficulty in meeting the real-time safety assessment needs of complex traffic scenarios. Second, existing alarm strategies largely rely on fixed thresholds or simple priority settings, lacking dynamic coupling analysis of key safety indicators such as risk propagation paths and vehicle trajectories. This hinders the collaborative processing of multi-source alarm information and the intelligent generation of priority strategies, thus affecting the relevance of warning information and traffic management efficiency. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an intelligent road system for vehicle-road cooperative safety alarms to solve the problems of insufficient risk prediction accuracy and poor dynamic adaptability of alarm strategies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] This invention provides an intelligent road system for vehicle-road cooperative safety alarm, comprising,

[0008] The signal generation module receives and analyzes danger signals emitted by vehicle lights and manually triggered devices in real time via visible light communication, and simultaneously collects surrounding environmental data to generate an initial alarm signal.

[0009] The path planning module triggers road spike nodes based on the initial alarm signal, obtains the road spike network status, filters nodes based on the road spike network status, and generates the optimal communication path.

[0010] The communication evaluation module inputs the optimal communication path into the communication mapping model, the spatial matching layer fuses road geometry and communication topology, and the state mapping layer marks signal strength and road stud node status to generate a communication quality heatmap.

[0011] The risk decision-making module assesses the risk level of each area based on the communication quality heatmap and outputs priority transmission parameters.

[0012] The information transmission module activates audible and visual alarms based on priority transmission parameters, obtains execution status reports, and simultaneously sends the execution status reports back to the command center.

[0013] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the hazard signal includes trigger event type, accident classification identifier, precise location data, accident severity level, and timestamp information;

[0014] The surrounding environment data includes road stud node communication status parameters, road geometric feature data, real-time traffic flow data, and ambient light intensity parameters.

[0015] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for generating the initial alarm signal are as follows:

[0016] By integrating hazard signals with surrounding environmental data, dynamic risk characteristic data is obtained, and the risk of the dynamic risk characteristic data is quantified to generate an initial alarm signal.

[0017] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for obtaining the road stud network status are as follows:

[0018] The initial alarm signal triggers the road spike node through a fixed-frequency light pulse and performs adjacency detection to establish a road spike network connection link;

[0019] The signal strength and connection quality of each spike node in the spike network connection link are detected, the effective response range and distribution are statistically analyzed, and the spike network status is output.

[0020] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for generating the optimal communication path are as follows:

[0021] Based on the network status of the road spikes, valid candidate road spike nodes are selected, and communication quality is evaluated for the valid candidate road spike nodes to generate the road spike node transfer probability. Dynamic path optimization is then performed on the road spike node transfer probability to obtain the optimal communication path.

[0022] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for generating the communication quality heatmap are as follows:

[0023] A multimodal feature fusion method is used to integrate cross-domain information and extract hierarchical features from the spatial geometry layer and the communication state layer, thereby constructing a communication mapping model;

[0024] The optimal communication path is input into the communication mapping model, and the spatial matching layer uses a geometric topology association algorithm to fuse the road geometry and communication connection relationship to establish a spatial location correspondence.

[0025] The state mapping layer uses signal feature extraction methods to mark the signal strength and connection quality parameters of each spike node and obtain the link state parameters.

[0026] By integrating spatial location correspondences and link status parameters, a communication quality heatmap is generated.

[0027] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for assessing the risk level of each area are as follows:

[0028] The communication quality heatmap is divided into multiple sub-region units, and the communication quality parameters within each sub-region unit are obtained.

[0029] Risk levels are assigned to communication quality parameters within each sub-region unit.

[0030] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the output priority transmission parameter is implemented through the following steps.

[0031] Based on the regional risk distribution map, an optimized path to avoid high-risk areas is obtained, and priority transmission parameters are output in combination with the risk level.

[0032] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for activating the audible and visual alarm based on priority transmission parameters are as follows:

[0033] Parse the risk level identifier and alarm condition field of the priority transmission parameter, verify the validity of the parameter, and then activate the audible and visual alarm.

[0034] As a preferred embodiment of the intelligent road system for vehicle-road cooperative safety alarm described in this invention, the specific steps for obtaining the execution status report are as follows:

[0035] The system monitors the status of audible and visual alarms in real time, verifies the validity of the alarm status, generates an execution status report, and uploads the execution status report to the command center using the optimal communication path.

[0036] The beneficial effects of this invention are as follows: By constructing a communication mapping model, integrating road geometry and communication topology, marking signal strength and node status, a communication quality heatmap is generated, realizing a visual assessment of communication status, solving the problem of insufficient single-point detection accuracy, and providing spatial support for risk classification; by parsing priority parameters to activate audible and visual alarms, loading visible light configuration to transmit information and return status, a "perception-decision-execution-feedback" closed loop is formed, solving the problems of alarm response lag and status uncertainty, realizing accurate early warning and dynamic resource scheduling in high-risk areas, and improving the real-time performance and reliability of vehicle-road cooperative safety alarms. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of an intelligent road system used for vehicle-road cooperative safety alarms.

[0039] Figure 2 The flowchart for generating the initial alarm signal.

[0040] Figure 3 This is a flowchart showing the output of the road spike network status.

[0041] Figure 4 This is a flowchart of the audible and visual alarm and information feedback process. Detailed Implementation

[0042] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0043] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0045] Reference Figures 1-4 This is one embodiment of the present invention, which provides an intelligent road system for vehicle-road cooperative safety alarm, comprising the following steps:

[0046] The signal generation module receives and analyzes danger signals emitted by vehicle lights and manually triggered devices in real time via visible light communication, and simultaneously collects surrounding environmental data to generate an initial alarm signal.

[0047] Hazard signals include triggering event type, accident classification identifier, precise location data, accident severity level, and timestamp information;

[0048] It should be noted that when a vehicle accident occurs, the high-sensitivity photoelectric detector receives the light signals emitted by the vehicle's LED lights and the manual triggering device in real time. First, the received light signals are demodulated, and the triggering event type and accident classification identifier are parsed according to the pulse position modulation rules. Then, the precise positioning data encoded in the subsequent light pulse sequence is extracted, and the severity level of the accident is identified by decoding the changes in the pulse interval. Finally, the timestamp information is parsed to complete the collection and verification of all hazard signal fields.

[0049] It should also be noted that visible light communication is a wireless communication technology that uses the visible light band for data transmission. By modulating the brightness, flashing frequency, or pulse timing of LED lights, the data to be transmitted is encoded into an optical signal, which is then received and demodulated by a photodetector to restore the data, thus achieving high-speed and secure two-way communication while providing illumination.

[0050] Surrounding environmental data includes road stud node communication status parameters, road geometric feature data, real-time traffic flow data, and ambient light intensity parameters;

[0051] It should be noted that the photoelectric detector monitors the intensity and frequency changes of the received light signal to evaluate the communication status parameters of the road stud node; the inertial measurement unit detects geometric features such as road slope and curvature and records road geometric feature data; the infrared sensor senses the number and speed of passing vehicles to obtain real-time traffic flow data; and the ambient light sensor measures the light intensity parameters in real time.

[0052] Kalman filtering is used to fuse hazard signals with surrounding environmental data to obtain dynamic risk characteristic data, and risk quantification is performed through edge computing to generate an initial alarm signal;

[0053] It should be noted that, firstly, the hazard signal and surrounding environmental data are spatiotemporally aligned and noise suppressed using the Kalman filter algorithm to obtain a multi-source fusion dataset; then, the obtained multi-source fusion dataset is dynamically fused using real-time incremental correction; subsequently, dynamic risk characteristic data containing features such as real-time collision risk probability value, risk diffusion rate, and accident impact area range are obtained using real-time data correlation analysis; at the same time, each feature of the dynamic risk characteristic data is compared with a preset risk threshold one by one, and the risk level is defined based on the feature index exceeding the preset risk threshold, and the current impact range and risk level are encapsulated as an initial alarm signal.

[0054] It should also be noted that the preset risk threshold is determined by the distribution law of the characteristic indicators output by Kalman filter fusion, and the high percentile values ​​of the real-time collision risk probability value, risk diffusion rate and accident impact area range are used as the judgment threshold.

[0055] Dynamic fusion refers to continuously integrating the latest collected hazard signals and environmental data, and updating the existing fusion dataset through incremental algorithms;

[0056] Real-time data correlation analysis collects real-time data such as the relative speed and distance between vehicles and obstacles, the speed of obstacle movement, road surface friction state and direction of movement. Based on the correlation relationship that "high speed and close distance lead to high collision risk", it assesses the real-time collision risk probability value. According to the correlation rule that "fast movement and slippery road surface lead to fast spread", it calculates the risk spread rate. Based on the coverage of "direction of movement and duration", it delineates the accident impact area.

[0057] Kalman filtering is a recursive estimation algorithm. First, it predicts the current state based on the estimated value of the previous moment (e.g., using the vehicle speed of the previous second to estimate the current position). Then, it compares the newly collected sensor data with the predicted value and dynamically assigns weights according to the size of the difference and the reliability of the sensor data itself (data with small measurement error and high stability is more reliable, while data with large error or easy interference is less reliable). Finally, it fuses the data to generate the current optimal state estimate and updates the uncertainty range of the next round of prediction.

[0058] Edge computing is a distributed computing paradigm that deploys computing resources at the network edge, close to the data source, to achieve localized data processing and real-time analysis. Its operation involves three key steps: First, data collected by terminal devices undergoes preprocessing at edge nodes, including data cleaning, format conversion, and feature extraction, to reduce data redundancy. Second, lightweight computing units built into the edge nodes execute pre-defined data analysis algorithms, such as anomaly detection, pattern recognition, and real-time decision-making, directly extracting key information from the data and generating preliminary processing results. Finally, edge nodes, according to a pre-defined priority strategy, upload high-value information or results of urgent events to the cloud while discarding low-value data, thereby reducing data transmission latency and bandwidth consumption, achieving efficient and low-power real-time response.

[0059] The edge computing unit consists of a data processing unit, a storage unit, and a communication unit. The data processing unit performs real-time computing tasks, the storage unit temporarily stores collected data and intermediate computing results, and the communication unit is responsible for data transmission between local devices and with external devices.

[0060] The path planning module triggers the road spike network based on the initial alarm signal, obtains the road spike network status, filters nodes based on the road spike network status, and generates the optimal communication path.

[0061] The initial alarm signal triggers the surrounding road spike nodes through a fixed-frequency light pulse and initiates the communication protocol. The adjacent road spike nodes are simultaneously activated and perform adjacency detection to establish a road spike network connection link.

[0062] It should be noted that the initial alarm signal first triggers the built-in optical signal transmitter of the road stud node, generating a periodic light pulse sequence at a preset frequency, which is then used as a communication trigger signal to propagate to surrounding road stud nodes. Adjacent road stud nodes monitor changes in ambient light intensity in real time through photodetectors. When a light pulse signal matching the preset frequency is detected, the internal communication protocol parsing program is immediately activated to complete signal feature extraction and protocol matching verification. After successful verification, adjacent road stud nodes are activated, and point-to-point connections are established sequentially, forming a chain communication network centered on the road stud node that generates the periodic light pulse sequence, ensuring that alarm information is rapidly and orderly disseminated to vehicles and facilities behind along the road.

[0063] It should also be noted that the parameters of the fixed-frequency optical pulse are determined by the PPM encoding protocol, and the frequency range is 1Hz~2Hz;

[0064] The communication protocol is a standardized rule system for information exchange between road spike nodes. It includes optical pulse coding methods, timing synchronization mechanisms, data format definitions, and error detection methods (such as duplicate verification) to ensure that different road spike nodes can accurately identify, parse, and respond to trigger signals, thereby achieving reliable transmission of alarm information and link synchronization.

[0065] The signal strength and connection quality of each spike node in the spike network are detected by optical signal interaction, the effective response range and distribution are statistically analyzed, and the spike network status data is output.

[0066] It should be noted that each road spike node transmits uncoded raw light pulses to adjacent road spike nodes according to a preset fixed period, and records the transmission time simultaneously. After the adjacent road spike node captures the light pulse through a photoelectric sensor, it immediately replies with a response signal at the same light frequency, while simultaneously detecting the received light intensity and signal duration. The transmitting end obtains the communication delay based on the pulse round-trip time difference, evaluates the signal attenuation degree based on the received light intensity (the light intensity attenuation threshold is set to the range of 10% to 90% of the initial transmission intensity), and establishes a connection quality database in each direction through multiple consecutive tests. It continuously tracks the changes in the number of effective response nodes, the range of signal intensity fluctuations, and the distribution pattern of delay time, and finally generates road spike network status data.

[0067] It should also be noted that optical signal interaction is a physical communication method in which data is transmitted between road spike nodes through light-emitting and photosensitive elements. The transmitting road spike node converts the electrical signal into a sequence of light pulses of a specific frequency and intensity, and emits it into the air through an optical lens. The receiving road spike node uses a photodetector to capture the light pulses, and decodes the original information after photoelectric conversion, filtering and amplification, so as to realize the state synchronization, data transmission and connection quality assessment between nodes.

[0068] The preset fixed period duration is 10 milliseconds ± 0.5 milliseconds, and the value is determined by the hardware timer precision and communication protocol requirements.

[0069] Based on the road spike network state data, a belief propagation algorithm is used to screen valid candidate road spike nodes, and a dynamic programming algorithm is used to construct a road spike node pool.

[0070] It should be noted that, based on the road spike network status data, the comprehensive confidence score of each road spike node is first obtained through the confidence propagation algorithm. Road spike nodes with a comprehensive confidence score above a preset confidence threshold are selected as valid candidate road spike nodes. Then, the road spike network is transformed into a graph structure of road spike nodes and connection relationships using a dynamic programming algorithm. The cost of each path is determined according to the communication quality index (considering both signal strength and delay). Starting from the road spike node that generates the periodic light pulse sequence, the optimal path cost and forwarding order to reach the adjacent road spike node are compared and recorded layer by layer. The road spike node sequence with the lowest cost and the highest connection quality is selected by backtracking the path, and a road spike node pool is generated.

[0071] It should also be noted that the confidence propagation algorithm is a method for evaluating the reliability of network nodes through information transmission. It simulates the process of messages being transmitted back and forth between adjacent network nodes. Each network node collects feedback information from its neighbors and then calculates a confidence score by weighting the real-time status data of the network nodes with the feedback information. The confidence score is continuously updated and adjusted as it receives the latest feedback from its neighbors. After multiple rounds of information exchange, each network node can obtain a relatively accurate confidence value, thereby selecting the most reliable connection path in the network.

[0072] The preset reliability threshold is defined based on the basic reliability requirements of the road spike node communication function, and the value range is usually the middle to high end of the scoring interval;

[0073] Dynamic programming is a mathematical method for finding optimal solutions in stages. It breaks down complex problems into several interrelated subproblems and avoids redundant calculations by recording the optimal solution of each subproblem. In road spike networks, dynamic programming transforms the path selection process from the alarm source to all adjacent road spike nodes into a multi-stage decision problem. First, it determines the next road spike node that is closest to the alarm source. Then, it plans the transmission order of subsequent road spike nodes step by step. Each step comprehensively considers indicators such as signal strength and communication delay to calculate the overall cost. Finally, it backtracks to find a complete transmission path with the lowest overall cost (i.e., high reliability and low latency).

[0074] Based on the road spike node pool, the artificial slime mold algorithm is used to dynamically calculate the road spike node transfer probability, and dynamic path optimization is performed on the road spike node transfer probability to obtain the optimal communication path.

[0075] The expression for calculating the transition probability of a road spike node is:

[0076] ;

[0077] Among them, P ij This represents the probability that the current spike node i will move to its neighbor node j. The pheromone concentration is represented from the spike node i to its neighbor node j, and α represents the influencing factor of the pheromone concentration, with a value range of 0.5≤α≤3; Let β represent the heuristic factor from spike node i to neighbor node j, and let β represent the influence factor of the heuristic factor, with a value range of 0.5≤β≤2.5. The pheromone concentration is represented from the spike node i to the candidate node k, and α represents the influence factor of the pheromone concentration, with a value range of 0.5≤α≤3. d represents the heuristic factor from spike node i to candidate node k, β represents the influence factor of the heuristic factor, and its value ranges from 0.5 to 2.5; ik d represents the physical distance from spike node i to candidate node k.ij This represents the physical distance from spike node i to its neighbor node j. γ represents the average distance of all possible paths in the spike node pool, γ represents the distance sensitivity coefficient, and its value ranges from 0.1 to 1.5; N(i) represents the set of neighboring nodes of the current spike node i, k represents the summation index, and e represents the exponential function.

[0078] It should be noted that, firstly, a static road spike node pool is constructed using an ant colony algorithm. During initialization, multiple virtual ants start synchronously from the alarm source road spike node and select a communication path based on a comprehensive evaluation index of signal strength and communication latency. The higher the signal strength and the lower the communication latency, the greater the probability of the communication path being selected. During their movement, the virtual ants release pheromones along the communication path to mark high-quality communication paths. After multiple iterations, road spike nodes with pheromone concentrations reaching a preset threshold (defined based on historical network performance statistics and engineering experience, with the value range set in the high percentile interval of pheromone concentration distribution) are selected to form the initial road spike node pool. Then, a dynamic path optimization is achieved using an artificial slime mold algorithm, which monitors the signal quality and communication status of each node in real time. When a decrease in signal strength or an increase in communication latency is detected at a node, the probability of the road spike node being selected is reduced, while the transfer priority of neighboring road spike nodes with good status is increased. Through continuous iterations of evaluation and adjustment of the path selection strategy, an optimal communication path that can adapt to changes in the current network status is finally determined.

[0079] It should also be noted that the ant colony algorithm is a metaheuristic optimization algorithm that simulates the foraging behavior of ant colonies in nature. It explores paths in parallel in the solution space through multiple virtual ants, dynamically releases pheromones to mark high-quality solutions based on the quality of the paths, and uses a positive feedback mechanism to make better paths have a higher selection probability. After multiple generations of iteration, pheromones accumulate on the optimal path to form positive reinforcement, and finally guide the colony to converge to the global optimal or near-optimal solution.

[0080] The artificial slime mold algorithm is a dynamic path optimization algorithm that dynamically adjusts the search strategy to balance global exploration and local exploitation capabilities by simulating the pulsating contraction, expansion and morphological reconstruction behavior of slime molds in the process of searching for food.

[0081] The communication evaluation module inputs the optimal communication path into the communication mapping model, the spatial matching layer fuses road geometry and communication topology, and the state mapping layer marks signal strength and road stud node status to generate a communication quality heatmap.

[0082] A multimodal feature fusion method is used to integrate cross-domain information and extract hierarchical features from the spatial geometry layer and the communication state layer, thereby constructing a communication mapping model;

[0083] It should be noted that in the PyTorch deep learning framework, the spatial feature extraction network is invoked via the `nn.Module` command, and a three-level progressive spatial unit is constructed for the spatial feature extraction network, including coarse-grained units, medium-grained units, and fine-grained units. Different spatial analysis operators are configured for each spatial unit. The coarse-grained unit uses graph convolution to capture the global topological relationships of roads and outputs an initial spatial feature map. The medium-grained unit uses an attention mechanism to obtain the structural importance scores of each region in the initial spatial feature map, enhancing the feature weights of high-importance regions and suppressing redundant information in low-importance regions, thereby refining the initial spatial feature map and generating regional structural feature maps. The fine-grained unit uses dynamic edge weight adjustment to capture local detail changes, further optimizing the intermediate spatial feature map and outputting a complete spatial feature map, thus completing the process. The spatial geometry layer is constructed by calling the temporal state evolution network (TDN) through the `torchdiffeq.odeint` parameter and embedding a state transition solver into the TDN using the `ODESolver` class. Simultaneously, the time step parameter and convergence threshold are set for the state transition solver: a time step of 0.01 ensures temporal continuity, and a convergence threshold of 1e-5 ensures state stability. The TDN receives temporal communication data (such as channel quality and signal strength sequences) and simulates state changes over time using a dynamic function parameterized by its internal neural network. The `torchdiffeq.odeint` function calls `ODESolver` to progressively acquire state evolution at 0.01-second time steps. Finally, the states at each time step are merged to generate a communication feature map, completing the construction of the communication state layer.

[0084] A multimodal feature fusion method is used to integrate cross-domain information and extract hierarchical features from the spatial geometry layer and the communication state layer. Specifically, firstly, the bimodal feature concatenation unit is called using the torch.cat parameter to concatenate the complete spatial feature map output by the spatial geometry layer with the communication feature map output by the communication state layer along the channel dimension. Then, a multi-scale attention mechanism is embedded using the CrossModalFusion class, and initial parameters are set for the attention mechanism simultaneously: the spatial feature weight is set to 0.6 and the communication feature weight is set to 0.4. Finally, the communication mapping model is constructed.

[0085] Next, the constructed communication mapping model is trained. Specifically, the multimodal feature dataset is first divided into a sample set, a training set, and a validation set in a 7:1:2 ratio. On the sample set, features are normalized using StandardScaler, and data distribution is balanced using stratified sampling to generate standardized training samples. On the training set, the AdamW optimizer is used for backpropagation on the standardized training samples, and a focus loss function is applied for gradient optimization to obtain optimized communication mapping model parameters. On the validation set, forward inference is performed on the optimized communication mapping model parameters to obtain the validation loss. When the validation loss reaches the convergence threshold, training is stopped, and the trained communication mapping model is output using the `torch.jit.script` parameter.

[0086] It should also be noted that the multimodal feature fusion method is a collaborative processing technology for multi-source heterogeneous feature information. Its core lies in the deep fusion of feature information from different sources and with different forms of expression through specific information association and complementarity mechanisms.

[0087] The optimal communication path is input into the communication mapping model, and the spatial matching layer uses a geometric topology association algorithm to fuse the road geometry and communication connection relationship to establish a spatial location correspondence.

[0088] It should be noted that, firstly, the spatial coordinate sequence of the optimal communication path and the road topology data of the road geometry layer are extracted and unified under the same geographic coordinate system; then, based on the geometric topology association algorithm, by analyzing the similarity of spatial locations and the compatibility of topological structures, bidirectional matching of communication path nodes and road geometry nodes is performed to establish a preliminary spatial-communication correspondence; finally, for erroneous matching caused by coordinate errors, the matching results are adjusted through topology verification, and the final output is a spatial location correspondence table containing one-to-one binding of road location identifiers and communication path nodes.

[0089] It should also be noted that the geometric topology association algorithm is a computational method for establishing the relationship between spatial geometric information and topological structure. Its core lies in deeply integrating and associating the two by analyzing the geometric features and topological attributes of spatial elements.

[0090] The spatial coordinate sequence contains geometric information such as the latitude and longitude and direction angle of the path nodes, while the road topology data of the road geometry layer contains geometric features such as the coordinates of road segment nodes, lane connection relationships, and the location of traffic signs.

[0091] The state mapping layer uses signal feature extraction methods to mark the signal strength and connection quality parameters of each spike node and obtain the link state parameters.

[0092] It should be noted that, firstly, the raw signal data is collected in real time through the radio frequency unit of the communication node, covering basic indicators such as received RSSI (Signal Strength Indicator), SNR (Signal-to-Noise Ratio), end-to-end transmission delay, and packet loss rate. Then, the collected raw signal data is smoothed by a sliding window. By obtaining the mean and standard deviation of the raw signal data within the window, instantaneous noise interference is eliminated, and a stable signal characteristic sequence is obtained. Next, the mean RSSI within the statistical window is used to characterize the overall signal strength, and the connection quality is quantified by combining the mean SNR, mean delay, and packet loss rate. Finally, the signal strength value and connection quality parameters are bound one by one to the unique identifier of the corresponding road spike node (such as node ID) to generate link status parameters, which dynamically reflect the real-time communication status of each node.

[0093] It should also be noted that the signal feature extraction method is a process of extracting key parameters reflecting link quality from the raw signal data collected from communication nodes. The core is to remove noise interference through algorithmic processing and extract feature values ​​that can effectively characterize the communication status. In specific operations, raw signal data including received signal strength, signal-to-noise ratio, end-to-end delay, and packet loss rate are first obtained from the radio frequency unit. Then, instantaneous fluctuations are eliminated through preprocessing methods such as sliding window smoothing. Based on statistical rules (such as mean and standard deviation), signal strength (characterized by the mean RSSI value) and connection quality (combined with the mean SNR value reflecting anti-interference capability, the mean delay value reflecting transmission efficiency, and the packet loss rate reflecting reliability) are obtained. Finally, quantifiable node signal strength values ​​and connection quality parameters are generated, providing a direct basis for dynamically evaluating the node communication status.

[0094] By integrating spatial location correspondences and link state parameters using DS evidence theory, a communication quality heatmap is generated.

[0095] It should be noted that, firstly, the spatial location correspondences and link state parameters are associated with geographical locations to form a set of evidence bodies based on specific locations. Then, based on DS evidence theory, a Basic Probability Assignment (BPA) is defined for each evidence body, and the connection quality parameters are converted into initial confidence levels through normalization. Simultaneously, the initial confidence levels are corrected by considering the stability of the signal strength of the road spike nodes (such as the fluctuation coefficient calculated from the mean and standard deviation of RSSI), resulting in a preliminary BPA for each piece of evidence. Next, the Dempster synthesis rule is used to resolve conflicts and synthesize multiple evidence bodies in overlapping areas to obtain a comprehensive communication quality confidence value. Finally, the comprehensive communication quality confidence value is mapped to a geospatial grid and visualized through color gradients, generating a communication quality heatmap reflecting the correlation between spatial location and communication quality, visually displaying the distribution of communication reliability in different areas.

[0096] It should also be noted that the core of the Dempster evidence theory is to quantify the degree of trust in a proposition through the probability of confidence (BPA). The evidence space is divided into a set of mutually exclusive and exhaustive propositions, and each proposition is assigned a probability value (BPA) that reflects the degree of trust. The Dempster evidence theory achieves the fusion of multi-source evidence through the "Dempster synthesis rule"—when there is no conflict between the evidence, it is directly weighted and synthesized; when there is a conflict, the weight of conflicting evidence is reduced and the contribution of consistent evidence is increased, ultimately generating a comprehensive degree of trust, providing a more reliable basis for decision-making, and is widely used in information fusion, target identification, risk assessment and other fields.

[0097] The risk decision-making module, based on the communication quality heatmap, assesses the risk level of each area using the DQN algorithm and employs a dynamic path planning algorithm to output priority transmission parameters.

[0098] The communication quality heatmap is divided into multiple sub-regional units using spatial rasterization, and the communication quality parameters within each sub-regional unit are obtained through spatial statistical analysis.

[0099] It should be noted that, firstly, the basic size of the raster unit is set by comprehensively considering the level of detail of the spatial analysis (smaller raster for capturing local changes, larger raster for focusing on overall trends), the complexity of the regional space (smaller raster for densely built-up urban areas, larger raster for open plains), and the hardware processing capabilities (such as GPU memory limitations and real-time computing speed). Then, through geographic coordinate mapping, each pixel in the communication quality heatmap is assigned to the corresponding raster unit according to its latitude and longitude. The nearest neighbor interpolation method is used to process edge points to ensure that all raster units are closely arranged and do not overlap. After completing the raster segmentation, spatial statistical analysis is used to extract the communication quality parameters of each raster unit: the communication quality heatmap values ​​in each raster unit are aggregated and statistically analyzed to calculate the mean, standard deviation, maximum / minimum values, and other statistical quantities. At the same time, based on the spatial coordinates of the raster unit, the mean difference between adjacent raster units is obtained to quantify the quality gradient change. Finally, a sub-region communication quality parameter table containing "raster coordinates - mean - standard deviation - extreme values ​​- gradient" is output.

[0100] The DQN algorithm is used to classify the risk level of communication quality parameters in each sub-regional unit, forming a regional risk distribution map.

[0101] It should be noted that, firstly, the communication quality parameters of each sub-region unit are organized as state inputs, and risk level labels are marked in combination with historical data. Then, the DQN algorithm environment is initialized, and the state (communication quality parameter vector) of each sub-region unit is mapped to a high-dimensional state space that the DQN algorithm can process. The action space is defined as risk level classification, and reward rules are set (e.g., a +1 reward is given when the predicted level matches the actual label, and a -0.5 penalty is given when they do not match). Next, the state-action-reward interaction data is stored through an experience replay mechanism. A greedy algorithm is used to select the policy network, and it interacts with the environment (the task-related running environment that includes the communication quality parameter state and risk level label reward of the sub-region unit) to continuously try and optimize the policy network. Finally, the optimized policy network is used to predict the risk level of the communication quality parameters of all sub-region units, and the risk level prediction results are associated with the sub-region spatial coordinates to form a regional risk distribution map.

[0102] Based on the regional risk distribution map, an optimal path to avoid high-risk areas is obtained through a dynamic path planning algorithm. Combining the risk level and the optimal path, a multi-objective decision-making method is used to output priority transmission parameters.

[0103] It should be noted that, firstly, the regional risk distribution map is used as a spatial constraint input into the dynamic path planning algorithm. Simultaneously, dynamic environmental factors such as real-time traffic flow and road connectivity are combined to construct a comprehensive optimization objective function that includes both risk avoidance and path efficiency objectives. Subsequently, the dynamic path planning algorithm is used for path search, guiding the path to avoid high-risk areas through a heuristic function while ensuring global optimality. After obtaining multiple candidate optimized paths, a multi-objective decision-making method is employed. This method combines the risk level of each sub-regional unit traversed by the candidate path with the coverage of the candidate path within each sub-regional unit, weighting and summing the path risk exposure and path efficiency indicators (total mileage, estimated travel time) as dual objectives to obtain the comprehensive utility value of each candidate optimized path. Finally, the candidate optimized path with the highest comprehensive utility value is selected as the final recommended path, and key transfer parameters, including the path start-end coordinate sequence, estimated travel time, and path risk exposure, are extracted to form priority transfer parameters.

[0104] It should also be noted that dynamic path planning algorithms are a class of methods that can obtain the optimal path in real time in a dynamically changing environment. The core is to continuously update and optimize path decisions by searching the state space and using heuristic guidance, while taking into account dynamic environmental factors.

[0105] The path start-end coordinate sequence comes from the spatial trajectory of the final recommended path, and the estimated travel time is obtained from the total path mileage and real-time traffic flow.

[0106] Multi-objective decision-making methods are a class of decision analysis techniques used to seek a balance or optimal solution among multiple conflicting objectives. The core of these methods is to transform multiple objective functions into a comparable unified evaluation system and use Pareto front analysis to identify all non-dominated solutions, thereby determining the optimal decision solution in complex trade-offs.

[0107] The information transmission module enables the road stud network to activate audible and visual alarms based on priority transmission parameters, and transmit alarm information to surrounding facilities via visible light communication, while simultaneously sending the execution status back to the command center.

[0108] The road spike network uses a data decoding algorithm to parse the risk level identifier and alarm condition field of the priority transmission parameters, and activates the audible and visual alarm after verifying the validity of the parameters.

[0109] It should be noted that, firstly, the risk level identifier and alarm condition field of the priority transmission parameters are parsed through a data decoding algorithm: the received priority transmission parameters are first checked for errors to ensure that the data transmission is error-free; then, according to the risk level binary code lookup table clearly defined in the communication protocol (00 represents low risk, 01 represents medium risk, 10 represents high risk, and 11 is a reserved value), the binary code of the risk level identifier is converted into the corresponding risk level value, and the code of the alarm condition field is converted into specific trigger parameters; then, the road stud network matches the sound and light alarm mode according to the risk level value, and finally activates the sound and light alarm device, which sends a visual warning signal matching the risk level to surrounding road users through a combination of changes in light color, flashing frequency, and sound intensity.

[0110] It should also be noted that the audible and visual alarm modes are predefined by road safety standards: High risk: the red LED light flashes 5 times per second, while the buzzer sounds at a high frequency of 80 decibels; Medium risk: the yellow LED light flashes 3 times per second, and the buzzer sounds at a medium frequency of 60 decibels; Low risk: the blue LED light flashes once per second, and the buzzer sounds at a low frequency of 40 decibels.

[0111] A priority scheduling algorithm is used to extract the visible light communication configuration field from the priority transmission parameters, and the configuration parameters are loaded through an embedded communication protocol parser. Then, visible light communication is used to transmit alarm information to surrounding facilities.

[0112] It should be noted that, firstly, a priority scheduling algorithm is used to filter out the highest priority communication request from the received priority transmission parameters, and the visible light communication configuration field containing key parameters such as modulation method, transmission rate, and wavelength range is accurately located and extracted. Subsequently, the embedded communication protocol parser parses the extracted visible light communication configuration field, converts the binary encoded configuration parameters into specific communication control commands, and verifies the validity of the parameters. Then, the hardware device is initialized according to the parsed configuration parameters (adjusting the modulation frequency of the LED light source and setting the emission angle of the optical lens), and the alarm information is converted into an optical signal encoding that conforms to the requirements of the communication protocol. Finally, the modulated optical signal is sent to the surrounding facilities through the visible light emitting unit, and the surrounding receiving devices capture the optical signal through photoelectric sensors and parse the alarm content according to the same protocol.

[0113] The road spike network monitors the status of audible and visual alarms in real time and generates an execution status report, which is then uploaded to the command center using the optimal communication path.

[0114] It should be noted that the flashing frequency and photoelectric intensity of the LED lights are detected in real time by the photoelectric sensor built into the road spike node, while the sound decibel value of the buzzer is collected by the sound sensor, and the alarm trigger duration is continuously recorded. The collected data such as the flashing frequency of the lights, the sound decibel value, and the trigger duration are compared with the preset standard range (defined based on the road spike equipment specifications and communication protocol requirements) to determine whether the current audible and visual alarm is working properly. The information such as the flashing frequency of the LED lights detected by the photoelectric sensor, the sound decibel value of the buzzer collected by the sound sensor, the continuously recorded alarm trigger duration, the judgment result, and the timestamp are integrated to generate a structured execution status report. Based on the optimal communication path, the execution status report is forwarded hop by hop to the road spike gateway node using short-range wireless communication. The road spike gateway node then uploads the execution status report to the command center through the cellular network, realizing real-time remote transmission and monitoring of the alarm execution status.

[0115] In summary, this invention addresses the problem of insufficient accuracy in single-point detection by constructing a communication mapping model that integrates road geometry and communication topology, marks signal strength and node status, and generates a communication quality heatmap to achieve visualized assessment of communication status. This provides spatial support for risk classification. Furthermore, by parsing priority parameters to activate audible and visual alarms, loading visible light configurations to transmit information and return status, a closed loop of "perception-decision-execution-feedback" is formed. This solves the problems of delayed alarm response and unknown status, enabling precise early warning and dynamic resource scheduling in high-risk areas, and improving the real-time performance and reliability of vehicle-road cooperative safety alarms.

[0116] 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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An intelligent road system for vehicle-road cooperative safety alarm, characterized in that: include, The signal generation module receives and analyzes danger signals emitted by vehicle lights and manually triggered devices in real time via visible light communication, and simultaneously collects surrounding environmental data to generate an initial alarm signal. The path planning module triggers road spike nodes based on the initial alarm signal, obtains the road spike network status, filters nodes based on the road spike network status, and generates the optimal communication path. The communication evaluation module inputs the optimal communication path into the communication mapping model, the spatial matching layer fuses road geometry and communication topology, and the state mapping layer marks signal strength and road stud node status to generate a communication quality heatmap. The risk decision-making module assesses the risk level of each area based on the communication quality heatmap and outputs priority transmission parameters. The information transmission module activates audible and visual alarms based on priority transmission parameters, obtains an execution status report, and simultaneously sends the execution status report back to the command center. The specific steps are as follows. The system monitors the status of audible and visual alarms in real time, verifies the validity of the alarm status, generates an execution status report, and uploads the execution status report to the command center using the optimal communication path.

2. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 1, characterized in that: The hazard signal includes the triggering event type, accident classification identifier, precise location data, accident severity level, and timestamp information; The surrounding environment data includes road stud node communication status parameters, road geometric feature data, real-time traffic flow data, and ambient light intensity parameters.

3. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 2, characterized in that: The specific steps for generating the initial alarm signal are as follows: By integrating hazard signals with surrounding environmental data, dynamic risk characteristic data is obtained, and the risk of the dynamic risk characteristic data is quantified to generate an initial alarm signal.

4. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 1, characterized in that: The specific steps for obtaining the road spike network status are as follows: The initial alarm signal triggers the road spike node through a fixed-frequency light pulse and performs adjacency detection to establish a road spike network connection link; The signal strength and connection quality of each spike node in the spike network connection link are detected, the effective response range and distribution are statistically analyzed, and the spike network status is output.

5. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 4, characterized in that: The specific steps for generating the optimal communication path are as follows: Based on the network status of the road spikes, valid candidate road spike nodes are selected, and communication quality is evaluated for the valid candidate road spike nodes to generate the road spike node transfer probability. Dynamic path optimization is then performed on the road spike node transfer probability to obtain the optimal communication path.

6. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 1, characterized in that: The specific steps for generating the communication quality heatmap are as follows: A multimodal feature fusion method is used to integrate cross-domain information and extract hierarchical features from the spatial geometry layer and the communication state layer, thereby constructing a communication mapping model; The optimal communication path is input into the communication mapping model, and the spatial matching layer uses a geometric topology association algorithm to fuse the road geometry and communication connection relationship to establish a spatial location correspondence. The state mapping layer uses signal feature extraction methods to mark the signal strength and connection quality parameters of each spike node and obtain the link state parameters. By integrating spatial location correspondences and link status parameters, a communication quality heatmap is generated.

7. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 6, characterized in that: The specific steps for assessing the risk level of each region are as follows. The communication quality heatmap is divided into multiple sub-region units, and the communication quality parameters within each sub-region unit are obtained. Risk levels are assigned to communication quality parameters within each sub-region unit.

8. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 1, characterized in that: The output priority transmission parameter refers to obtaining an optimized path to avoid high-risk areas based on the regional risk distribution map, and outputting priority transmission parameters in conjunction with the risk level.

9. The intelligent road system for vehicle-road cooperative safety alarm as described in claim 8, characterized in that: The specific steps for activating the audible and visual alarm based on priority parameters are as follows. Parse the risk level identifier and alarm condition field of the priority transmission parameter, verify the validity of the parameter, and then activate the audible and visual alarm.

Images