A wireless intercom safety emergency message priority forwarding method based on artificial intelligence

Abstract

The application discloses a wireless intercom safety emergency message priority forwarding method based on artificial intelligence, and belongs to the technical field of wireless intercom communication. The method acquires voice, text, position, event type, terminal state, wireless link state and regional emergency strategy data of a wireless intercom terminal, generates a multi-modal risk feature, a link reliability vector and a terminal reachability vector, obtains a message emergency degree, a safety risk level and a model confidence through an edge risk identification model, and generates a dynamic priority to perform hierarchical queue scheduling accordingly. For a high-risk emergency message, the high-risk emergency message is split into an emergency control frame and a content carrying frame, an emergency control frame is preferentially forwarded based on an emergency reachability graph by selecting a main path and a backup path, and path recalculation and abnormal feedback are performed according to an acknowledgement message ACK returned by a receiving end and a tracking token, so as to reduce a first-arrival delay of the high-risk message and improve reliable delivery capability under a complex wireless link.

Description

Technical Field

[0001] This invention relates to the field of wireless intercom communication technology, specifically to a method for prioritizing the forwarding of wireless intercom security emergency messages based on artificial intelligence. Background Technology

[0002] Wireless intercom communication is commonly used in fire rescue, mining operations, rail transit, park security, disaster relief, and large-scale event support due to its convenient deployment, fast call response, and strong group call capability. In these situations, on-site personnel typically report information such as voice, short text, location, alarm status, or event type through intercom terminals. In cases of trapped personnel, fire spread, equipment failure, communication interruption, or emergency evacuation, whether relevant messages can be forwarded to the command center, regional gateway, or relevant rescue terminals in a timely manner often directly affects the efficiency of on-site response.

[0003] In existing wireless intercom systems, message forwarding is generally prioritized according to preset service type, user role, or manual dispatch instructions. For example, alarm messages, command messages, or messages from specific personnel are set to higher priority and sent first when the channel is congested. Some systems also combine link quality, link load, or path hop count to select forwarding paths to improve the likelihood of message delivery.

[0004] However, in actual emergency communication, the urgency of a message cannot always be accurately reflected by the service type or the sender's role. For example, voice messages sent by ordinary terminals may contain high-risk content such as distress calls, evacuation, explosions, and collapses. The forwarding strategy for the same type of alarm message may also be different in different areas, different link states, and different terminal reach states. If we still rely solely on fixed priorities or manual marking, it is easy to encounter problems such as delayed identification of high-risk messages and untimely priority adjustment.

[0005] Furthermore, existing forwarding controls typically focus more on the service priority of the message itself, or select paths solely based on link load, rarely combining message content risk, terminal location, regional emergency strategies, target terminal reachability, and wireless link status for unified judgment. In wireless intercom scenarios, factors such as RSSI / RSRP, SINR, channel occupancy, packet loss rate, queue backlog length, terminal's most recent online time, PTT talk time, terminal battery level, and the status of nearby relay nodes all affect the actual delivery effectiveness of emergency messages. If these factors are not involved in priority generation and path selection, messages with high priority may still be sent to unstable links or unreachable terminals.

[0006] When links are congested or narrowband resources are limited, the transmission method of high-risk messages also needs further improvement. If the complete voice, text, or attachment content is directly copied and forwarded through multiple paths, it will significantly increase channel occupancy and even aggravate local congestion. If only a single path is used for transmission, critical emergency information may not arrive in time when the link is interrupted, the target terminal is intermittently online, or the relay node is under high load. Therefore, it is necessary to distinguish between critical control information and complete message content, so that critical content such as risk level, location, event type, and remaining valid time can arrive first, and then the complete content is sent based on the confirmation result.

[0007] In addition, existing systems primarily use records of acknowledgment (ACK) timeouts, packet loss, path loops, or forwarding failures for post-event queries or log analysis, and rarely use them in real time to correct path edge weights, adjust queue resources, or trigger backup path retransmissions. In complex wireless intercom networks, if abnormal feedback cannot participate in subsequent forwarding control in a timely manner, the system may repeatedly retransmit on the same abnormal link, affecting the reliable delivery of high-risk emergency messages.

[0008] Therefore, there is an urgent need to propose an AI-based method for prioritizing the forwarding of emergency security messages in wireless intercoms. This method would enable AI identification results to be used not only to determine whether a message is urgent, but also to be further transformed into queue scheduling parameters, frame structure control parameters, routing parameters, ACK retransmission parameters, and anomaly tracking feedback parameters in wireless intercom communication. This would improve the priority transmission capability and reliable delivery capability of high-risk emergency messages under complex wireless link conditions. Summary of the Invention

[0009] The purpose of this invention is to provide an artificial intelligence-based method for prioritizing the forwarding of emergency security messages in wireless intercom communication. This method addresses the technical problems in existing wireless intercom communication where message risk identification results are disconnected from wireless link scheduling, frame structure control, and routing control. This results in long initial delivery delays, large amounts of repeated transmissions, and insufficient forwarding reliability of high-risk emergency messages when there is link congestion, intermittent online status of the target terminal, or local link anomalies.

[0010] To address the aforementioned technical problems, this invention provides a method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence, comprising the following steps: When a wireless intercom terminal generates a message to be forwarded, it acquires message data, terminal status data, wireless link status data, and area emergency strategy data. The message data, terminal status data, wireless link status data, and regional emergency strategy data are preprocessed to obtain risk characteristics, link reliability data, terminal reachability data, target terminal reachability probability, link congestion level, and remaining message validity time. By inputting risk characteristics, link reliability data, and terminal reachability data into the edge risk identification model, the message urgency, security risk level, and model confidence level are obtained. Based on message urgency, security risk level, model confidence, target terminal reachability probability, link congestion level, and regional emergency strategy, generate emergency control frame priority, content carrying frame priority, reachability trigger flag, and main path retransmission limit, and schedule hierarchical queues. Messages with a security risk level higher than the high-risk level are split into emergency control frames and content carrier frames. An emergency reachability map is constructed based on the wireless link, and the primary path and backup path are determined. Emergency control frames are sent first through the primary path and backup path, while the transmission of content carrier frames is temporarily suspended. After receiving an emergency control frame confirmation message from any path, the content carrier frame is sent according to the path status that returned the confirmation message. If an emergency control frame confirmation message is not received within the preset time, the abnormal link is identified based on the tracking token of the emergency control frame, its edge weight in the emergency reachability graph is corrected, a path that does not contain the abnormal link is re-identified, and the emergency control frame is resent.

[0011] Furthermore, the terminal status data includes the terminal identifier or its hash identifier, terminal role, most recent online time, most recent heartbeat time, PTT talk right status, terminal battery level, moving speed, historical acknowledgment success rate, and historical retransmission count; the wireless link status data includes received signal strength or reference signal received power, signal-to-interference-plus-noise ratio, bit error rate, packet loss rate, channel occupancy rate used to determine the degree of link congestion, queue backlog length, transmission delay, reachability of nearby relay nodes, relay node load, and gateway congestion status.

[0012] Furthermore, the risk features include semantic risk features, acoustic anomaly features, location fence features, and event type features; the semantic risk features are obtained from text or speech-to-text through keyword recognition, entity extraction, intent recognition, and risk classification; the acoustic anomaly features are obtained from speech data through acoustic analysis; and the location fence features are obtained by matching location information with regional emergency response strategies.

[0013] Furthermore, the edge risk identification model is constructed and trained in the following manner: voice or text data, location and regional policy data, wireless link status data, and terminal status data are respectively input into semantic, acoustic, location policy, link status, and terminal reachability coding branches and mapped to features of the same dimension; gating weights are generated and fused according to link congestion level, target terminal reachability probability, message remaining effective time, and regional policy parameters; the model is trained with historical emergency communication samples containing risk level labels and urgency labels, and the training loss includes risk level cross-entropy loss, urgency regression loss, and confidence calibration loss, so that the multi-task output layer outputs message urgency, security risk level, and model confidence.

[0014] Furthermore, when the model confidence level is lower than the first confidence level threshold, rule rollback processing is enabled. Temporary risk levels and temporary priorities are generated based on emergency keywords, event types, regional emergency strategies, terminal roles, remaining message validity time, and wireless link status data. Queue scheduling and path selection are then performed according to the temporary priorities.

[0015] Furthermore, the reachability probability of the target terminal is used to characterize the probability that the target terminal will successfully receive the message within the remaining valid time of the message; when the reachability probability of the target terminal is lower than the first reachability threshold and the security risk level is higher than the high-risk level, a reachability trigger flag is generated, and backup path forwarding is triggered according to the reachability trigger flag, while limiting the number of retransmissions on the primary path.

[0016] Furthermore, the hierarchical queue includes an emergency preemption queue, a high-priority emergency queue, and a normal queue; messages with a dynamic priority higher than the first priority threshold are added to the emergency preemption queue, messages with a dynamic priority no higher than the first priority threshold but higher than the second priority threshold are added to the high-priority emergency queue, and messages with a dynamic priority no higher than the second priority threshold are added to the normal queue; when there are messages in the emergency preemption queue, the sending window of the normal queue is paused or reduced.

[0017] Furthermore, the emergency control frame includes a message identifier or message fingerprint, a location grid, an event code, a security risk level, a message urgency, a dynamic priority, a remaining message validity time, and a tracking token; the content carrying frame includes a complete voice segment, speech-to-text, original text, or an extended event description, and includes a message identifier or message fingerprint corresponding to the emergency control frame.

[0018] Furthermore, the nodes of the emergency reachability map include wireless intercom terminals, gateway nodes, base station nodes, relay nodes, edge nodes, and command and dispatch nodes; the edge weights of the emergency reachability map are determined based on link congestion, transmission delay, packet loss rate, node reachability probability, regional policy penalty value, node credibility, and node remaining power; among the paths that satisfy the constraints of remaining message validity time and security risk level, the path with the lowest path cost is selected as the main path, and the path with the lowest path cost and a link overlap with the main path below a preset overlap threshold is selected as the backup path.

[0019] Furthermore, when each hop node receives and forwards the emergency control frame, it associates the current hop reception time, current hop forwarding time, current hop queuing delay, current hop link status, next hop node, and current hop forwarding result with the tracking token. When the emergency control frame confirmation timeout occurs, the abnormal link is determined based on the last successfully forwarded node and the next hop node, and the edge weight correction amount of the abnormal link is generated based on the confirmation timeout duration, the number of consecutive packet losses, and the current hop link status information.

[0020] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention uses semantic risk, acoustic anomaly, location fence, event type, wireless link status and terminal reachability together to generate basic message priority, and determines the priority generation method through model confidence gating and rule fallback mechanism, so that the identification result of high-risk emergency message can be directly converted into queue scheduling parameters in wireless intercom communication, thereby improving the reliability of priority forwarding of high-risk emergency message; 2. When the link is congested or narrowband resources are limited, the present invention splits high-risk emergency messages into emergency control frames and content-bearing frames, and prioritizes sending emergency control frames with smaller data volumes, so that key emergency information such as location, event encoding, risk level, TTL and tracking token can reach the target terminal, regional gateway or command and dispatch node before the complete message content, thereby reducing the first-delivery delay of high-risk messages. 3. This invention constructs an emergency reachability map and incorporates link congestion, transmission delay, packet loss rate, node reachability probability, regional policy penalty value, node credibility and remaining power into the path cost calculation. After determining the main path, a backup path with a link overlap of less than a threshold is selected, thereby improving the reliability of path selection in scenarios where the target terminal is intermittently online, the link quality fluctuates or the local link is abnormal. 4. Based on abnormal results such as ACK timeout, continuous packet loss, path loop, sudden drop in target terminal reachability probability, and regional policy conflict, this invention corrects the abnormal link edge weight, target terminal reachability probability calculation parameters, and queue transmission resources. When the manual scheduling confirmation result is inconsistent with the risk level output by the model, a risk identification model correction sample is formed, thereby forming a closed-loop control between risk identification, queue scheduling, path selection, and abnormal retransmission. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of a wireless intercom security emergency message priority forwarding method based on artificial intelligence according to the present invention; Figure 2 This is a schematic diagram of the multimodal risk identification and dynamic priority generation process in this invention; Figure 3 This is a schematic diagram illustrating the splitting and priority forwarding of emergency control frames and content-bearing frames in this invention. Detailed Implementation

[0022] 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.

[0023] Example 1: like Figure 1 As shown, this embodiment provides a method for prioritizing the forwarding of emergency security messages in wireless intercom systems based on artificial intelligence. This method is applicable to emergency communication scenarios involving wireless intercoms; the wireless intercom terminal can be a digital intercom terminal, a private network intercom terminal, a public-private hybrid intercom terminal, a handheld intercom terminal, a vehicle-mounted intercom terminal, or an emergency communication terminal with intercom functionality; the wireless intercom communication link can be a narrowband private network link, a broadband private network link, a trunking communication link, a relay forwarding link, a base station forwarding link, or a short message link. The method includes the following steps: S1: Acquire multimodal message data, terminal status data, wireless link status data, and regional emergency strategy data. When a wireless intercom terminal generates a message to be forwarded, it acquires the multimodal message data corresponding to that message; the multimodal message data may include voice data, text data, location information, and event type information.

[0024] Among them, voice data can be real-time voice streams or voice segments obtained by dividing by time windows; text data can be terminal input text, scheduling short text, speech-to-text text, or event description text; location information can be GPS positioning information, Beidou positioning information, base station positioning information, Bluetooth positioning information, UWB positioning information, or regional grid positioning information; event type information is the event type selected by the user on the terminal, or the event type identified based on voice or text content.

[0025] While acquiring multimodal message data, terminal status data is also acquired. The terminal status data includes the terminal ID or its hash identifier, terminal role, terminal's most recent online time, terminal's most recent heartbeat time, PTT talk rights status, terminal battery level, terminal movement speed, historical ACK success rate, and historical retransmission count.

[0026] Terminal roles can include ordinary members, on-site operators, rescuers, commanders, inspectors, security personnel, or temporarily authorized personnel; PTT talk rights status is used to characterize whether the terminal is occupying talk rights, whether there is a preemption request, or whether it is in a waiting talk rights state; the terminal's most recent heartbeat time and most recent online time are used to determine whether the terminal has the risk of intermittent online or disconnection; the terminal roles are only used to determine the preemption level of wireless communication resources, queue scheduling parameters, and routing forwarding constraints, and are not used for identity evaluation, credit evaluation, or management punishment of terminal users that are unrelated to emergency communication.

[0027] Furthermore, wireless link status data is acquired; the wireless link status data includes RSSI / RSRP, SINR, bit error rate, packet loss rate, channel occupancy rate, current link queue length, end-to-end transmission delay, per-hop forwarding delay, reachability of nearby relay nodes, relay node load, and gateway congestion status.

[0028] Among them, RSSI / RSRP and SINR represent the signal quality of the wireless link, channel occupancy rate represents the current busyness of wireless resources, packet loss rate and bit error rate represent the reliability of data transmission, queue backlog length represents the node buffer occupancy and data forwarding load, and the reachability of nearby relay nodes represents whether there are alternative forwarding links.

[0029] Furthermore, regional emergency strategy data is obtained; regional emergency strategy data can be determined based on the current location of the terminal, the location of the target terminal, the target area of ​​the message, or the current emergency task area; regional emergency strategy data includes regional risk level, regional broadcast range, regional TTL requirement, regional redundant forwarding level, regional relay priority, regional trusted node set, regional prohibited forwarding node set, regional congestion threshold, and regional emergency preemption ratio.

[0030] For example, when the source of the message is located in a fire area, a deep mine area, a tunnel area, a sealed-off area, a base station failure area, a temporary command area, or a densely populated area, the regional emergency strategy data corresponding to that area is obtained so as to subsequently affect priority, path edge weight, and forwarding redundancy.

[0031] The voice data, text data, location information, and terminal status data are collected and processed within the scope necessary for legal authorization or the execution of emergency communication tasks, and are acquired according to the minimum data range required to achieve message risk identification and wireless forwarding control; before being used for model training, anomaly analysis, or forwarding tracking, the terminal ID, personnel identification, and message identification are hashed or de-identified.

[0032] Training samples do not use personal attributes such as gender, age, ethnicity, and religious belief, which are not directly related to the technical control of wireless communication transmission, as risk level labels or priority labels; regional emergency strategies are only used to adjust the transmission queue, frame structure, path edge weights, and retransmission methods.

[0033] S2: Generate multimodal risk features, link reliability vectors, and terminal reachability vectors, and obtain risk identification results. like Figure 2 As shown, the data obtained by S1 is preprocessed to generate multimodal risk features, link reliability vectors, and terminal reachability vectors.

[0034] Multimodal risk features include semantic risk features, acoustic anomaly features, location fence features, and event type features.

[0035] Semantic risk features are obtained by keyword recognition, entity extraction, intent recognition, and risk classification of text data or speech-to-text data. Semantic risk features may include emergency keywords, dangerous behavior words, disaster type words, casualty words, location entities, equipment failure entities, distress call intent, evacuation intent, loss of contact intent, fire intent, explosion intent, collapse intent, or leakage intent.

[0036] Acoustic anomaly features are obtained through acoustic analysis of speech data; acoustic anomaly features may include speech energy, fundamental frequency, speech rate, volume abrupt changes, background noise intensity, spectral abrupt changes, distress call features, scream features, explosion sound features, impact sound features, Mel spectrum features, and MFCC features.

[0037] Location fence features are obtained by matching terminal location information with preset area policies; for example, if the terminal location falls into a high-risk work area, fire area, deep mine area, tunnel area, sealed-off area, weak communication coverage area or temporary command area, the corresponding location fence features are generated.

[0038] Event type characteristics are derived from event type information. Event types can include ordinary phone calls, inspection reports, equipment malfunctions, personnel distress calls, fire alarms, medical rescue, public security incidents, natural disasters, hazardous chemical leaks, communication interruptions, command and dispatch, and evacuation orders.

[0039] The link reliability vector L is used to characterize the communication reliability of the current forwarding link; the link reliability vector L can be expressed as: L={rssi,sinr,occ,loss,delay,queue,relay,ack}; where rssi represents the received signal strength or the received power of the reference signal, sinr represents the signal-to-interference-plus-noise ratio, occ represents the channel occupancy rate, loss represents the packet loss rate, delay represents the transmission delay, queue represents the queue backlog length of the current node, relay represents the reachability of nearby relay nodes, and ack represents the historical ACK success rate.

[0040] The terminal reachability vector A is used to characterize the reachability status of the target terminal or related terminals within the target area; the terminal reachability vector A can be represented as: A={hb,online,speed,battery,ptt,neighbor,ackhistory}; where hb represents the most recent heartbeat interval, online represents the most recent online time, speed represents the terminal's moving speed, battery represents the terminal's battery level, ptt represents the PTT talk time status, neighbor represents the response status of neighboring nodes, and ackhistory represents the historical message delivery status.

[0041] The target terminal reachability probability Pr is calculated based on the terminal reachability vector A. The target terminal reachability probability Pr is used to characterize the probability that the target terminal will successfully receive the message within the remaining TTL of the message.

[0042] In one implementation, hb, online, speed, battery, ptt, neighbor, and ackhistory are normalized to the interval [0,1]. Larger values ​​for hb, online, and speed indicate lower reachability, while larger values ​​for battery, neighbor, and ackhistory indicate higher reachability. ptt indicates whether the target terminal is in a receiveable state. The reachability probability Pr of the target terminal is calculated using the following formula: ;in, For the Sigmoid function, , , , , , and For normalization parameters, to The non-negative weights are obtained through training on historical delivery samples or preset according to regional emergency strategies. When Pr is less than the first reachable threshold, the message queue priority is not directly increased, but at least one of the following actions is triggered: backup path, regional broadcast, relay assistance, or limiting the number of retransmissions.

[0043] The edge risk identification model includes a semantic coding branch, an acoustic coding branch, a location policy coding branch, a link state coding branch, a terminal reachability coding branch, a gating fusion layer, and a multi-task output layer.

[0044] The semantic coding branch receives speech-transcribed text and terminal input text, and outputs a semantic risk feature vector; the acoustic coding branch receives Mel spectrum, MFCC, speech energy, and background noise features extracted according to a preset time window, and outputs an acoustic anomaly feature vector; the location policy coding branch receives location grid, regional risk level, and regional policy identifier, and outputs a location policy feature vector; the link state coding branch receives normalized RSSI / RSRP, SINR, channel occupancy rate, packet loss rate, transmission delay, and queue backlog length, and outputs a link state feature vector; the terminal reachability coding branch receives the most recent heartbeat interval, most recent online time, terminal battery level, mobile speed, PTT talk right status, and historical ACK success rate, and outputs a terminal reachability feature vector.

[0045] The feature vectors output by each coding branch are mapped to the same dimension through a fully connected layer and then input into a gated fusion layer. The gated fusion layer determines the fusion weight of each feature vector based on the link congestion and the reachability probability Pr of the target terminal. The multi-task output layer outputs the message urgency U, security risk level R, and model confidence C based on the fused feature vectors.

[0046] In one implementation, the semantic coding branch, acoustic coding branch, location policy coding branch, link state coding branch, and terminal reachability coding branch each output a feature vector. , , , and And mapped to the same dimension d through a fully connected layer. The gated fusion layer determines the link congestion level. The probability of reaching the target terminal (Pr), the remaining time to read the message (TTL), and the regional policy parameters. Generate gate vector ,in , Including corresponding , , , and Fusion weights , , , and And satisfy ; Fusion features The multi-task output layer includes an urgency output header, a risk level output header, and a confidence output header, which output the message urgency U, security risk level R, and model confidence C based on the fusion feature h, respectively.

[0047] When training the edge risk identification model, the training samples include corresponding voice or text data, location and regional policy data, wireless link status data, terminal status data, risk level labels and urgency labels; wherein, the risk level labels are generated based on the confirmed emergency event type, on-site handling records or scheduling confirmation results; the urgency labels are generated based on the event handling time limit, message remaining TTL and confirmed message processing order.

[0048] The training samples are input into each encoding branch, and after fusion by the gated fusion layer, the predicted risk level and urgency values ​​are obtained respectively. The training loss is then calculated according to the following formula: ;in, The cross-entropy loss is the difference between the predicted risk level and the risk level label. The regression loss is the difference between the predicted urgency values ​​and the urgency labels. To calibrate the loss for model confidence, , and The weights are non-negative loss weights.

[0049] During training, the training samples are divided into training, validation, and test sets. The training set is used to update model parameters, the validation set is used to determine the first confidence threshold, the first reachability threshold, and the model's stopping condition, and the test set is used to evaluate the model's performance after training. If the number of high-risk samples is lower than the number of ordinary samples, at least one of the following methods—class weighting, resampling, or hard example augmentation—is applied to high-risk samples to ensure that the recall rate of high-risk messages meets the preset requirements. Model training uses a mini-batch iterative approach, and the optimizer can be a stochastic gradient descent or adaptive moment estimation optimizer. The learning rate, batch size, and number of training epochs are determined based on the convergence of the validation set loss.

[0050] in, Expected calibration error loss or Brier fractional loss can be used to constrain the consistency between the model output confidence and the actual recognition accuracy on the validation set. Training samples are derived from historical emergency communication records, dispatch confirmation records, manually annotated voice or text messages, and corresponding link status records. The risk level labels for high-risk samples are determined by on-site handling records or dispatch confirmation results. The model is deployed on wireless intercom gateways, edge nodes, or command and dispatch servers. The inference latency of a single message does not exceed a preset inference latency threshold, which can be between 1 and 3 seconds.

[0051] The model parameters are updated via backpropagation based on the training loss until the training loss on the validation set meets the convergence condition. The model parameters and the first confidence threshold are determined based on the high-risk message recall rate, risk level identification accuracy, false positive rate, confidence calibration error, and single-message inference latency on the validation set. and a first reachable threshold; wherein, the first confidence threshold The first reachability threshold is used to determine whether to use the model output result. The single message inference latency is used to determine whether the model meets the real-time processing requirements of the wireless intercom edge node or command and dispatch server.

[0052] The message urgency level U is a value between 0 and 1; the security risk level R is a discrete level among ordinary, attention, urgent, high risk, and extremely high risk, and is mapped to a normalized risk value before participating in dynamic priority calculation. The values ​​for ordinary, attention, emergency, high risk, and extremely high risk are 0, 0.25, 0.5, 0.75, and 1, respectively. The model confidence score C is used to characterize the credibility of the risk identification model for the current identification results.

[0053] The trained edge risk identification model outputs message urgency U, security risk level R, and model confidence C; where U and R are used to generate the basic message priority, and C is used to determine whether to adopt the model output; when C is not lower than the first confidence threshold... When U and R are used, the model output is used; when C is lower than 1, the model output is used. When this occurs, a rule rollback mechanism is activated to generate temporary risk levels and temporary priorities based on emergency keywords, event types, regional policies, terminal roles, message TTL, and link status. Among these, message urgency U and security risk level R are used to generate dynamic priorities, model confidence C is used to select model output or rule rollback processing, and target terminal reachability probability Pr is used to generate reachability trigger flags and affect the retransmission limit of backup paths, regional broadcasts, or primary paths.

[0054] S3: Generate dynamic priorities and perform hierarchical queue scheduling Based on the message urgency U, security risk level R, model confidence C, target terminal reachability probability Pr obtained from S2, as well as message remaining TTL, queue waiting time, regional policy weight, and link congestion parameters, a dynamic priority P is generated to control the message sending queue, sending window, and emergency preemption resources.

[0055] An optional dynamic priority calculation method includes message base priority. Emergency control frame priority and content carrying frame priority .

[0056] When the model confidence level C is not lower than the first confidence threshold hour, When C is lower than When generating, a rule rollback mechanism is used. For messages with a security risk level higher than the high-risk level, the emergency control frame priority is increased. Content-carrying frame priority .

[0057] in, R represents the normalized risk value obtained by mapping the security risk level R; G represents the regional policy gain, which is obtained by normalizing the regional risk level, regional TTL requirement, regional redundant forwarding level, regional relay priority, regional emergency preemption ratio and regional prohibition forwarding constraint, and takes a value of 0 to 1. This indicates the role weight corresponding to the terminal role. Ordinary members, on-site workers, rescuers, and commanders can be set to 0.2, 0.5, 0.8, and 1, respectively. Inspectors, security personnel, or temporarily authorized personnel are set according to their authorization level.

[0058] in, Indicates the degree of link congestion. The data volume factor representing the content-carrying frame. and The weights are configurable; therefore, when the link is congested, the preemption priority of high-risk emergency control frames will not be reduced due to congestion, while the immediate transmission priority of ordinary messages or large-capacity content-bearing frames will be reduced.

[0059] The time-to-live (TTL) urgency level (T) can be calculated based on the remaining validity time of the message: ;in, Indicates the remaining validity period of the message. Indicates the initial validity period of the message.

[0060] Queue waiting for aging factors It can be calculated based on the waiting time after a message enters the queue: ;in, This indicates the time the message has been waiting. This indicates the preset maximum waiting time.

[0061] The reachability probability Pr of the target terminal is not directly used as a linear bonus to increase the priority of the queue. When Pr is lower than the first reachability threshold and the message security risk level is higher than the high-risk level, the system generates a reachability trigger flag ReachFlag and triggers alternative paths, area broadcasts, relay-assisted forwarding, or limits the number of retransmissions on the main path according to ReachFlag. This avoids the continuous occupation of wireless transmission resources when the target terminal is unreachable for a long time.

[0062] Link congestion regulation item It may be determined according to the channel occupancy rate, queue backlog length and relay node load; when the link congestion degree is high, the system may reduce the immediate forwarding priority of non-emergency messages or large-capacity content frames, while maintaining or increasing the forwarding priority of high-risk emergency control frames.

[0063] Messages are divided into different levels according to the dynamic priority P; when P≥P1, the message is determined as an extremely critical emergency message; when P2≤P<P1, the message is determined as a high-risk emergency message; when P3≤P<P2, the message is determined as a normal emergency message; when P<P3, the message is determined as a normal service message; wherein P1, P2, and P3 are configurable thresholds, and can be dynamically adjusted according to regional emergency policies and network congestion status.

[0064] After the message classification is completed, hierarchical queue scheduling is performed on the messages according to the dynamic priority P; the hierarchical queues at least include an emergency preemption queue, a high-priority emergency queue and a normal queue; the extremely critical emergency messages enter the emergency preemption queue, the high-risk emergency messages enter the high-priority emergency queue, and the normal emergency messages and normal service messages enter the corresponding normal queues.

[0065] When there are to-be-forwarded messages in the emergency preemption queue, the transmission window of the normal queue is suspended or reduced, and radio resources are released for transmitting emergency control frames; when there are to-be-forwarded messages in the high-priority emergency queue and the channel occupancy rate is higher than the first congestion threshold, the emergency control frames of the high-risk emergency messages are preferentially transmitted, and the corresponding content bearing frames or normal service messages are delayed to be transmitted.

[0066] When the message security risk level is higher than the high-risk level, the dynamic priority P is further used to determine whether the message performs separate forwarding of the emergency control frame and the content bearing frame, the active / standby path transmission mode of the emergency control frame, and the delayed transmission condition of the content bearing frame, so that the dynamic priority directly controls the subsequent frame structure and radio link forwarding process.

[0067] To avoid normal services from not being transmitted for a long time, this step sets a queue waiting aging factor W; after the waiting time of a normal message exceeds a preset threshold, its dynamic priority may be appropriately increased, but the increased priority does not exceed the minimum protection priority of high-risk emergency messages; thus, while ensuring the priority forwarding of high-risk emergency messages, the long-term starvation of normal services is avoided.

[0068] S4: Split the high-risk emergency message into an emergency control frame and a content bearing frame As Figure 3 shown, when a message is determined to be an extremely critical emergency message or a high-risk emergency message, the message is split into an emergency control frame and a content bearing frame.

[0069] Emergency control frames are used to quickly transmit critical emergency information in situations of link congestion or limited narrowband resources. The data size of an emergency control frame is smaller than that of a content carrier frame, and it can be sent using low bitrate encoding, short frame format, strong error correction encoding, or repetitive encoding.

[0070] An emergency control frame includes at least some of the following fields: FrameType (frame type); MsgID (message identifier); MsgHash (message fingerprint); TerminalHash (terminal ID hash); LocationGrid (location grid); EventCode (event encoding); RiskLevel (security risk level); Urgency (message urgency); Priority (dynamic priority); Confidence (model confidence); TTL (remaining message validity time); RouteConstraint (routing constraint); RegionPolicyID (region policy identifier); TraceToken (trace token); and Signature (signature digest).

[0071] The content carrier frame is used to carry the complete message content; the content carrier frame includes complete voice segments, speech-to-text, original text, images, attachments, sensor data or extended event descriptions, and includes the MsgID or MsgHash corresponding to the emergency control frame.

[0072] When the channel occupancy rate is greater than the first congestion threshold and the message security risk level is higher than the high-risk level, the emergency control frame is sent to the main path and the backup path respectively, and the transmission of the content carrier frame is temporarily suspended, so that the main path and the backup path only transmit emergency control frames with a data volume smaller than the content carrier frame.

[0073] When either the primary path or the backup path returns an emergency control frame ACK, the transmission path for the content carrier frame is selected based on the link quality, target terminal reachability probability, and message remaining TTL of the path that returned the ACK, and the content carrier frame is forwarded through the selected transmission path.

[0074] When the channel occupancy rate is not greater than the first congestion threshold, an emergency control frame is sent through the main path, and a content bearer frame is sent after receiving the ACK of the emergency control frame.

[0075] Therefore, when the link is congested, the complete high-risk message is not copied and sent through the primary and backup paths to reduce the amount of data transmitted repeatedly and shorten the first-delivery delay of critical emergency information.

[0076] For example, when a firefighter's terminal sends a voice message containing "Someone is trapped, requesting assistance," the system identifies the voice message as a high-risk emergency message. It then generates an emergency control frame containing a location grid, event code, risk level, dynamic priority, and tracking token, and sends it to the command and dispatch node and nearby relay nodes first. After receiving the ACK corresponding to the emergency control frame, the system then sends the complete voice segment as the content carrier frame, so that the command terminal can obtain key emergency information in advance before the complete voice content arrives.

[0077] S5: Construct an emergency reachability map and select the primary and backup paths. Based on the available communication nodes and available wireless links in the current wireless intercom communication network, an emergency reachability graph G=(V,E) is constructed; where V represents the set of graph nodes, including at least some of the nodes among wireless intercom terminals, gateway nodes, base station nodes, relay nodes, edge nodes and command and dispatch nodes; E represents the set of edges, where each edge represents an available wireless link or available forwarding relationship between two graph nodes.

[0078] For each edge Calculate the corresponding edge weights Border rights It can be determined using the following formula: ;in, This indicates the congestion level of the link between node i and node j. Indicates link transmission delay; Indicates the packet loss rate of the link; This represents the reachability probability of the next hop node j or the target node; Indicates the regional policy penalty value; Indicates the security and trustworthiness of a node; This represents the penalty for the remaining power of the node; all the above parameters are normalized to the interval [0,1]. to The weights are configurable and can be preset or dynamically adjusted based on the message security risk level and regional emergency response strategy.

[0079] Regional strategy penalty value The penalty value is determined based on the regional emergency response strategy. For example, when a link is located in a base station failure area, the regional policy penalty value of that link is increased; when a node is an authorized command node in a temporary command area, its regional policy penalty value as a receiving node or relay node is reduced; when a node is in a security authentication abnormal state, its regional policy penalty value is increased; when a relay node is located near a high-risk area and has a high relay priority, its path cost is reduced.

[0080] For each candidate path Calculate the total path cost : ;in, This represents the sum of the weights of all edges along the path; Indicates the number of hops in the path; This indicates the penalty for exceeding the message TTL constraint; This indicates a penalty for a mismatch between the path forwarding capability and the message security risk level. Indicates penalties for conflicts between the route and the regional emergency response strategy; to Configurable weights.

[0081] Determining the main path Then, candidate backup paths are calculated. Link overlap with the main path: ;in, Let be the set of edges for candidate alternative paths. The set of edges for the main path; only if If the overlap is less than the preset overlap threshold, the candidate path is determined as a backup path candidate; if there are multiple backup path candidates, the path with the lowest total cost is selected as the backup path.

[0082] Among the candidate paths that meet the constraints of message TTL and security risk level, the path with the lowest total path cost is selected as the primary path, and the candidate path with a link overlap of less than a preset threshold with the primary path is selected as the backup path. The primary path and backup path are first used to transmit emergency control frames. The transmission path of the content-bearing frame is not directly determined according to the initial total path cost, but is determined from the primary path and backup path after receiving the emergency control frame ACK, based on the link quality of the returned ACK, the reachability probability of the target terminal, and the remaining TTL of the message.

[0083] For critical emergency messages, emergency control frames can be sent simultaneously via the primary and backup paths. For high-risk emergency messages, emergency control frames can be sent via the primary path first, and then switched to the backup path when the primary path ACK times out or the link quality deteriorates. For ordinary emergency messages or ordinary business messages, single-path forwarding or delay-tolerant forwarding can be used.

[0084] Therefore, this step does not select the optimal path in general, but rather uses the emergency reachability graph to combine wireless link quality, target terminal reachability, regional emergency strategy, security and trustworthiness, and backup path overlap for path selection, making the path selection result more suitable for emergency communication scenarios. The edge weights and total path cost of the emergency reachability graph are used to generate wireless link forwarding control instructions to determine the next-hop node, primary path, and backup path of emergency control frames and content-bearing frames.

[0085] S6: Perform priority forwarding and form an anomaly feedback loop based on ACK and tracking token. Message forwarding is performed based on the dynamic priority obtained in S3, the emergency control frame and content carrier frame obtained in S4, and the primary path and backup path obtained in S5.

[0086] Each high-risk emergency message generates a tracking token. Tracking tokens are generated as follows: ;in, Represents the terminal ID hash; Represents a message fingerprint; Indicates the level of safety risk; Indicates dynamic priority; Represents a timestamp; Indicates the path identifier; Represents a random number; This represents a hash operation.

[0087] and The tracking record is obtained by a one-way hash operation from the terminal identifier and the message identifier respectively. The tracking record does not directly store the identity information of the terminal user. The tracking token and the forwarding status of each hop are only accessible to authorized wireless intercom gateways, edge nodes or command and dispatch nodes, and are deleted or anonymized after the preset tracking retention period expires.

[0088] When receiving and forwarding emergency control frames or content bearer frames, each hop node associates its current hop forwarding status with a tracking token. Each hop forwarding status includes the current hop node identifier, current hop reception time, current hop forwarding time, current hop queuing delay, current hop RSSI / RSRP, current hop SINR, current hop packet loss rate, current hop queue length, current hop forwarding result, next hop node identifier, and signature digest.

[0089] Each hop node can update the tracking summary as follows: ;in, This represents the summary of the k-th hop tracing. This indicates a summary of the previous hop's tracking. Indicates the identifier of the k-th hop node; This indicates the time information for the k-th hop; Indicates the state of the k-th hop link; Indicates the queuing delay for the k-th hop; This represents the forwarding result of the k-th hop; This represents the signature digest for the k-th hop.

[0090] When the target terminal, regional gateway, or command and dispatch node receives an emergency control frame, it returns an ACK. If no emergency control frame ACK is received within a preset time, it is determined that an emergency control frame ACK timeout has occurred, and the path, abnormal link, and corresponding ACK timeout time are determined based on the tracking token.

[0091] The ACK timeout, number of consecutive packet losses, and link status recorded by the tracking token of the abnormal link are converted into edge weight correction values, and the edge weight correction values ​​are added to the original edge weights of the abnormal link in the emergency reachability graph to recalculate the emergency reachability graph.

[0092] After recalculation, the candidate path with the lowest total cost and that does not contain the abnormal link is selected as the updated main path, and the emergency control frame is retransmitted; after receiving the retransmitted emergency control frame ACK, the content bearer frame is sent according to the path that returned the ACK.

[0093] At the same time, the proportion of sending resources in the emergency preemption queue is increased according to the message security risk level, and the sending window of the ordinary queue is reduced until an emergency control frame ACK is received or the remaining TTL of the message is lower than a preset threshold.

[0094] During the forwarding process, an anomaly feedback loop is triggered when any of the following abnormal conditions are detected: ACK timeout; continuous repeated packet loss; path loop; abnormally reduced priority; regional policy conflict; decreased node security trustworthiness; sudden drop in the reachability probability of the target terminal; relay node load exceeding the threshold; emergency control frame has arrived but the content-carrying frame has not arrived; a node in the path does not perform forwarding according to the message security risk level.

[0095] The anomaly feedback closed loop includes the following processing: recalculating the emergency reachability map; switching to backup paths; temporarily increasing the transmission resource ratio of the emergency preemption queue; pausing or reducing the transmission window of the normal queue; retransmitting emergency control frames; performing regional broadcasts to the target area; marking abnormal nodes as low-trust nodes; adjusting regional policy weights; adjusting dynamic priority thresholds; correcting the reachability probability of the target terminal; and using the abnormal results as model update samples.

[0096] When the confidence level C of the edge risk identification model output in S2 is lower than the first confidence threshold... Or the current computing resources are below the second resource threshold. When this occurs, rule rollback processing is enabled; the computing resources can be determined based on model inference latency, CPU utilization, memory utilization, current number of concurrent messages, node power consumption, or node temperature.

[0097] Rule rollback processing generates a temporary priority based on at least some of the following parameters: emergency keyword, event type, regional emergency strategy, terminal role, link congestion, target terminal reachability probability, message TTL, queue waiting time, and historical ACK success rate. and according to temporary priority Perform queue scheduling and path selection.

[0098] When N consecutive ACK results are inconsistent with the predicted path success rate or the predicted probability of reaching the target terminal (Pr), the link edge weight parameters, the target terminal reachability calculation parameters, or the regional policy weights are adjusted. ACK results are used to correct link reachability and path success rate, and are not directly used as the true value of the event risk level.

[0099] When the manual dispatch confirmation result or on-site handling record is inconsistent with the security risk level R output by the edge risk identification model, the semantic risk features, acoustic anomaly features, location fence features and event type features of the corresponding message are marked as correction samples for the risk identification model.

[0100] For example, when a message experiences consecutive ACK timeouts and backup path switching during forwarding, the system uses the corresponding link status and tracing results as path success rate correction samples; when the message is confirmed as a high-risk event by manual scheduling, the system uses its semantic risk features, acoustic anomaly features, and location fence features as correction samples for the risk identification model.

[0101] Therefore, ACK feedback, tracking tokens, and anomaly detection results are not used to simply form log records, but are converted into queue sending resource adjustment parameters, emergency reachability graph edge weight correction parameters, ACK timeout retransmission parameters, and backup path switching parameters, thus forming a closed-loop communication control process between multimodal risk identification, wireless link scheduling, routing selection, and anomaly retransmission.

[0102] Example 2: This embodiment uses a fire rescue scenario as an example to illustrate the specific application of the method of the present invention.

[0103] After entering the fire scene with a wireless intercom terminal, the firefighter sent a voice message via PTT saying "Someone is trapped, requesting assistance"; at the same time, the terminal was located in a high-risk fire area, and there were popping sounds and high-intensity noise in the background, and the wireless link was currently congested.

[0104] The system acquires the voice data, terminal location, terminal battery level, PTT talk time status, RSSI / RSRP, SINR, channel occupancy rate, queue backlog length, and regional emergency strategy data; it transcribes the voice data, extracts semantic risk features such as "trapped" and "requesting support," and extracts acoustic anomaly features such as popping sounds, sudden changes in voice energy, and background noise intensity; based on the matching results between the terminal location and the regional emergency strategy, it determines that the message comes from a high-risk fire area.

[0105] The edge risk identification model outputs message urgency, security risk level, and model confidence based on the aforementioned multimodal risk characteristics, link reliability data, and terminal reachability data. When the model confidence is not lower than the first confidence threshold, the system uses the model output to generate a dynamic priority. When the model confidence is lower than the first confidence threshold, the system activates a rule fallback mechanism to generate a temporary risk level and temporary priority based on emergency keywords, event type, regional emergency strategy, terminal role, and wireless link status.

[0106] When a message is determined to have a security risk level higher than the high-risk level, the system splits the message into an emergency control frame and a content carrier frame. The emergency control frame includes a message identifier or message fingerprint, location grid, event code, security risk level, message urgency, dynamic priority, remaining message validity time, and tracking token. The content carrier frame includes the complete voice segment and speech-to-text corresponding to the emergency control frame.

[0107] The system constructs an emergency reachability map based on the wireless link status, and determines the path cost of candidate paths based on link congestion, transmission delay, packet loss rate, node reachability probability, area policy penalty value, node trustworthiness, and node remaining power. Among the candidate paths that meet the constraints of message remaining validity time and security risk level, the path with the lower path cost is selected as the primary path, and the candidate path with a link overlap of less than a preset overlap threshold with the primary path is selected as the backup path.

[0108] When the channel occupancy rate exceeds the first congestion threshold, the system prioritizes sending emergency control frames through the primary and backup paths and temporarily suspends the transmission of content carrier frames. When either the primary or backup path returns an emergency control frame confirmation message, the system determines the transmission path for the content carrier frame based on the link quality, the probability of reaching the target terminal, and the remaining validity time of the message corresponding to the path that returned the confirmation message, and then sends the content carrier frame.

[0109] If an emergency control frame acknowledgment message is not received within a preset time, the system identifies the abnormal link based on the last successful forwarding node, next-hop node, current hop reception time, current hop forwarding time, current hop queuing delay, and current hop link status recorded in the tracking token. It then generates an edge weight correction for the abnormal link based on the acknowledgment timeout duration, the number of consecutive packet losses, and the current hop link status. The corrected emergency reachability graph is used to re-determine the path and retransmit the emergency control frame.

[0110] Therefore, in scenarios with congested links, such as fire rescue, the system does not repeatedly send complete voice messages through primary and backup paths. Instead, it first sends emergency control frames with smaller data volumes, enabling the command center or regional gateway to obtain key emergency information such as location, event type, risk level, and tracking token. After confirming that the link is available, the complete voice content is then sent, thereby achieving hierarchical, framing, and closed-loop forwarding control of high-risk emergency messages.

[0111] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence, characterized in that, Includes the following steps: When a wireless intercom terminal generates a message to be forwarded, it acquires message data, terminal status data, wireless link status data, and area emergency strategy data. The message data, terminal status data, wireless link status data, and regional emergency strategy data are preprocessed to obtain risk characteristics, link reliability data, terminal reachability data, target terminal reachability probability, link congestion level, and remaining message validity time. By inputting risk characteristics, link reliability data, and terminal reachability data into the edge risk identification model, the message urgency, security risk level, and model confidence level are obtained. Based on message urgency, security risk level, model confidence, target terminal reachability probability, link congestion level, and regional emergency strategy, generate emergency control frame priority, content carrying frame priority, reachability trigger flag, and main path retransmission limit, and schedule hierarchical queues. Messages with a security risk level higher than the high-risk level are split into emergency control frames and content carrier frames. An emergency reachability map is constructed based on the wireless link, and the primary path and backup path are determined. Emergency control frames are sent first through the primary path and backup path, while the transmission of content carrier frames is temporarily suspended. After receiving an emergency control frame confirmation message from any path, the content carrier frame is sent according to the path status that returned the confirmation message. If an emergency control frame confirmation message is not received within the preset time, the abnormal link is identified based on the tracking token of the emergency control frame, its edge weight in the emergency reachability graph is corrected, a path that does not contain the abnormal link is re-identified, and the emergency control frame is resent.

2. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence, as described in claim 1, is characterized in that... The terminal status data includes the terminal identifier or its hash identifier, terminal role, most recent online time, most recent heartbeat time, PTT talk right status, terminal battery level, moving speed, historical acknowledgment success rate, and historical retransmission count; the wireless link status data includes received signal strength or reference signal received power, signal-to-interference-plus-noise ratio, bit error rate, packet loss rate, channel occupancy rate used to determine the degree of link congestion, queue backlog length, transmission delay, reachability of nearby relay nodes, relay node load, and gateway congestion status.

3. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence, as described in claim 1, is characterized in that... The risk features include semantic risk features, acoustic anomaly features, location fence features, and event type features; the semantic risk features are obtained from text or speech-to-text through keyword recognition, entity extraction, intent recognition, and risk classification; the acoustic anomaly features are obtained from speech data through acoustic analysis; and the location fence features are obtained by matching location information with regional emergency response strategies.

4. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence, as described in claim 1, is characterized in that... The edge risk identification model is constructed and trained in the following way: voice or text data, location and regional policy data, wireless link status data and terminal status data are respectively input into semantic, acoustic, location policy, link status and terminal reachability coding branches and mapped to same-dimensional features; Gating weights are generated and fused based on link congestion level, target terminal reachability probability, remaining message validity time, and regional policy parameters. The model is trained using historical emergency communication samples with risk level and urgency labels. The training loss includes risk level cross-entropy loss, urgency regression loss and confidence calibration loss, so that the multi-task output layer outputs message urgency, security risk level and model confidence.

5. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence according to claim 1, characterized in that, When the confidence level of the model is lower than the first confidence level threshold, rule rollback processing is enabled. A temporary risk level and temporary priority are generated based on emergency keywords, event type, regional emergency strategy, terminal role, remaining validity time of message and wireless link status data, and queue scheduling and path selection are performed according to the temporary priority.

6. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence according to claim 1, characterized in that, The reachability probability of the target terminal is used to characterize the probability that the target terminal will successfully receive the message within the remaining valid time of the message. When the reachability probability of the target terminal is lower than the first reachability threshold and the security risk level is higher than the high-risk level, a reachability trigger flag is generated, and backup path forwarding is triggered according to the reachability trigger flag, while limiting the number of retransmissions on the main path.

7. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence, as described in claim 1, is characterized in that... The hierarchical queues include an emergency preemption queue, a high-priority emergency queue, and a normal queue. Messages with a dynamic priority higher than the first priority threshold are added to the emergency preemption queue, messages with a dynamic priority no higher than the first priority threshold but higher than the second priority threshold are added to the high-priority emergency queue, and messages with a dynamic priority no higher than the second priority threshold are added to the normal queue. When there are messages in the emergency preemption queue, the sending window of the normal queue is paused or reduced.

8. The method for prioritizing the forwarding of emergency security messages in wireless intercoms based on artificial intelligence according to claim 1, characterized in that, The emergency control frame includes a message identifier or message fingerprint, a location grid, an event code, a security risk level, a message urgency, a dynamic priority, a remaining message validity period, and a tracking token; the content carrier frame includes a complete voice segment, a voice-to-text transcript, the original text, or an extended event description, and includes a message identifier or message fingerprint corresponding to the emergency control frame.

9. A method for prioritizing the forwarding of emergency security messages for wireless intercoms based on artificial intelligence, as described in claim 1, is characterized in that... The nodes of the emergency reachability map include wireless intercom terminals, gateway nodes, base station nodes, relay nodes, edge nodes, and command and dispatch nodes. The edge weights of the emergency reachability map are determined based on link congestion, transmission delay, packet loss rate, node reachability probability, regional policy penalty value, node credibility, and node remaining power. Among the paths that meet the constraints of remaining message validity time and security risk level, the path with the lowest path cost is selected as the primary path, and the path with the lowest path cost and a link overlap of less than a preset overlap threshold with the primary path is selected as the backup path.

10. A method for prioritizing the forwarding of emergency security messages for wireless intercoms based on artificial intelligence, as described in claim 1, is characterized in that... When each hop node receives and forwards the emergency control frame, it associates the current hop reception time, current hop forwarding time, current hop queuing delay, current hop link status, next hop node, and current hop forwarding result with the tracking token. When the emergency control frame confirmation timeout occurs, the abnormal link is determined based on the last successfully forwarded node and the next hop node, and the edge weight correction amount of the abnormal link is generated based on the confirmation timeout duration, the number of consecutive packet losses, and the status information of the current hop link.

Images