Remote first-aid collaborative platform construction method and system based on AI decision

Abstract

The invention provides a remote first-aid cooperation platform construction method and system based on AI decision making. Belongs to the technical field of intelligent first-aid. The method comprises the following steps: selecting a communication frequency band; satellite communication data, sensor data and medical database information are mapped to a unified space-time coordinate system, a dynamic rescue scene model is constructed based on a knowledge graph, and key information is projected in real time through an AR interface; and a rescue scheme is generated and optimized in real time, and training courses are personalized. The switching interruption time is compressed from a millisecond level to a microsecond level through a pre-switching and conflict avoidance mechanism; data classification and distribution based on machine learning can maximize the transmission reliability of key data.

Description

Technical Field

[0001] The present invention proposes a method and system for constructing a remote emergency collaborative platform based on AI decision-making, which belongs to the field of intelligent emergency technology. Background Art

[0002] With the increasing frequency of global maritime activities and the frequent occurrence of maritime safety incidents, the demand for remote emergency coordination platforms has become increasingly urgent. However, traditional emergency response systems face numerous challenges, particularly in complex and volatile marine environments. These challenges include unstable communications, delayed information, and irrational allocation of rescue resources. Furthermore, existing systems often utilize independent coordinate systems (such as the WGS-84 geographic coordinate system and the ship's local coordinate system), requiring frequent manual conversions during data fusion. Furthermore, due to the dynamic displacement of the ship (such as roll and pitch), position mapping errors can reach several meters, making them unable to meet the needs of high-precision rescue operations. Summary of the Invention

[0003] The present invention provides a method and system for constructing a remote emergency collaborative platform based on AI decision-making to solve the problems mentioned in the above background technology: The present invention proposes a method for constructing a remote emergency collaborative platform based on AI decision-making, the method comprising: S1. Select the communication frequency band; S2, mapping satellite communication data, sensor data, and medical database information into a unified spatiotemporal coordinate system, building a dynamic rescue scenario model based on the knowledge graph, and projecting key information in real time through the AR interface; S3. Generate and optimize rescue plans and personalized training courses in real time.

[0004] The system for constructing a remote emergency collaborative platform based on AI decision-making proposed in the present invention includes a memory, a processor, and a computer program stored in and runnable on the memory. The processor executes the program to implement any of the methods for constructing a remote emergency collaborative platform based on AI decision-making described above.

[0005] Beneficial effects of the invention: The technical solution of the invention breaks through the traditional fixed threshold mode and can realize intelligent switching decisions that are adaptive to the meteorological environment; through pre-switching and conflict avoidance mechanisms, the switching interruption time can be compressed from milliseconds to microseconds; data classification and diversion based on machine learning can maximize the transmission reliability of key data; through differentiated protocol design and seamless switching of primary and backup links, the communication success rate can be increased from 90% to 97%; the dual-link load balancing efficiency is improved by 40%, and the bandwidth utilization rate reaches 92%; under extreme environments (wind speed 20m / s), the communication success rate is still ≥95%. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1This is a step diagram of the method of the present invention. DETAILED DESCRIPTION

[0007] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0008] One embodiment of the present invention, as Figure 1 As shown, a method for constructing a remote emergency collaborative platform based on AI decision-making includes: S1. Adopts multi-band adaptive switching technology, combined with a channel state prediction algorithm (based on LSTM neural network), to dynamically select the optimal communication frequency band, ensuring a communication success rate of ≥95% in complex marine environments. It also integrates multiple sensors (temperature and humidity, GPS, and vital signs monitoring) with a lightweight AR headset, and performs local data pre-processing through edge computing. S2, mapping satellite communication data, sensor data, and medical database information into a unified spatiotemporal coordinate system, building a dynamic rescue scenario model based on the knowledge graph, and projecting key information (such as casualty location, vital sign trends, and 3D navigation paths) in real time through an AR interface; S3. With the shortest rescue time, lowest resource consumption and highest medical priority as the objective function, the deep reinforcement learning (DRL) algorithm is used to generate rescue plans in real time; through the online learning module, the decision weight is dynamically adjusted by combining real-time environmental data (such as sudden weather changes, equipment failures) and expert feedback; a distributed medical expert database is built, and cross-institutional medical data is securely shared based on federated learning technology. The AI engine automatically matches the best expert and pushes diagnosis and treatment recommendations to the AR terminal; a high-fidelity maritime emergency simulation environment is built based on the physics engine and AR technology, and AI automatically generates multi-dimensional emergency plans (including extreme weather and multiple casualties scenarios); through computer Vision (CV) identifies rescuer operations in real time and compares them with the standard process knowledge base. If deviations are detected, AR warnings and correction instructions are triggered; reinforcement learning is used to simulate different first aid scenarios, dynamically generate personalized training courses, and monitor trainees' stress responses through brain-computer interfaces (BCIs) to optimize training programs; satellite communication links adopt a layered transmission strategy, with key instructions (such as first aid operation instructions) responded to in real time through edge nodes, and non-critical data (such as historical cases) uploaded asynchronously to the cloud for analysis; when the satellite signal is interrupted, the system automatically switches to mesh network communication mode and starts the local AI reasoning module to generate emergency decisions based on historical data.

[0009] The working principle and effects of the above technical solution are as follows: Through the multi-band adaptive switching technology and the channel state prediction algorithm based on the LSTM neural network, it is possible to ensure the selection of the optimal communication band in a complex marine environment, greatly improving the communication success rate. Especially in the case of unstable or limited satellite signals, the real-time performance and reliability of the platform are ensured, helping the first aid mission to proceed smoothly; integrating temperature and humidity, GPS, and vital sign monitoring sensors, and combining edge computing for local data preprocessing, it is possible to obtain and analyze the rescue environment and the condition of the wounded in real time. At the same time, through the lightweight AR headset device, key information (such as the location of the wounded, the trend of vital signs, the navigation path, etc.) can be projected onto the AR interface in real time for first aid personnel, greatly improving the first aid efficiency and decision-making quality; based on the deep reinforcement learning (DRL) algorithm, the system can generate the optimal rescue plan in real time. By combining the online learning module with expert feedback, the rescue plan is made more flexible, and the decision-making weight can be dynamically adjusted to cope with environmental changes and emergencies (such as meteorological changes, equipment failures, etc.), further improving the first aid efficiency; by constructing a distributed medical expert library and using federated learning technology for secure sharing of medical data, the AI engine can automatically match the most suitable experts and push diagnosis and treatment suggestions through the AR terminal to ensure that patients receive timely and accurate medical guidance. At the same time, cross-institutional data sharing and expert collaboration enhance the coverage and efficiency of the overall medical resources of the rescue system; by combining the physics engine with AR technology, a high-fidelity maritime first aid simulation environment is constructed, enabling first aid personnel to train and practice in multi-dimensional emergency scenarios, such as complex scenarios of extreme weather and multiple wounded. This simulation environment can not only help rescue personnel improve their actual combat capabilities but also generate targeted emergency plans in real scenarios; based on reinforcement learning technology, personalized training courses can be generated according to different first aid scenarios, and the stress response of the trainees can be monitored through the brain-computer interface (BCI) to further optimize the training plan. Such a training system can not only improve the capabilities of first aid personnel but also better adapt to different emergency scenarios and provide personalized emergency responses; adopting a hierarchical transmission strategy, the satellite communication link can effectively distinguish between critical instructions and non-critical data, ensuring that critical instructions are transmitted in a timely manner under real-time response, while non-critical data is asynchronously uploaded to the cloud for subsequent analysis. In addition, the system also has the function of automatically switching to the mesh network. In the case of satellite signal interruption, it can generate emergency decisions through the local AI inference module to ensure the continuous operation of the system.

[0010] In one embodiment of the present invention, S1 includes: S11. Construct a channel quality prediction model based on the LSTM neural network, input historical channel parameters (such as signal-to-noise ratio, bit error rate, Doppler shift) and real-time marine meteorological data (wind speed, wave height), and output the stability scores of each frequency band within the next 5 seconds; S12. Based on the priority weight function (communication bandwidth × stability score / handover delay), when the current frequency band score is lower than the threshold, trigger a millisecond-level frequency band handover; perform data transmission based on the dual-link parallel transmission mode, with the primary link transmitting first aid instructions and the secondary link transmitting compressed sensor data, so that the communication success rate ≥ 95%.

[0011] S13. Embed a micro temperature and humidity sensor (accuracy ±0.5°C), a multi-mode GPS module (supporting Beidou / Galileo dual-frequency positioning), and a millimeter-wave radar vital sign monitor (detecting heart rate and respiratory rate with an error ≤ 3%) in the AR headset; deploy a lightweight YOLOv5 model on the AR terminal to filter sensor noise data in real time and extract key features (such as the offset of the casualty's position and abnormal vital sign markers), and increase the data compression rate to 70%; S14. Dynamically adjust the sensor sampling rate according to the network bandwidth (for example, from 100Hz to 30Hz), and prioritize ensuring the complete transmission of high-priority data (such as electrocardiogram signals).

[0012] The working principle and effects of the above technical solution are as follows: Through the channel quality prediction model based on the LSTM neural network, historical channel parameters and marine meteorological data can be obtained in real time, and the frequency band stability score within the next 5 seconds can be accurately predicted. This provides a more scientific basis for frequency band switching, ensuring the stability and reliability of the first aid communication link in a changing environment; by introducing a priority weight function, when the current frequency band stability is lower than the threshold, a millisecond-level frequency band switching is triggered, which can quickly respond to changes in the marine environment, avoid communication interruption, and at the same time improve the success rate of data transmission based on the dual-link parallel transmission mode, ensure the priority transmission of first aid instructions, and reduce transmission delay; by embedding a variety of high-precision sensors in the AR headset, such as temperature and humidity sensors, GPS modules, and millimeter-wave radar vital sign monitors, the environmental conditions and vital signs of the wounded can be monitored in real time. The sensor accuracy and multi-mode positioning ability (supporting Beidou / Galileo dual-frequency positioning) ensure the accuracy of data and the precision of positioning, improving the decision-making quality during first aid; by deploying the lightweight YOLOv5 model on the AR terminal, the noise in the sensor data can be effectively filtered, and key features such as the offset of the wounded's position and abnormal marks of vital signs can be extracted in real time. This data processing method not only improves the data quality but also significantly reduces the redundancy of data transmission, with the compression rate increased to 70%, helping to reduce bandwidth occupancy; by dynamically adjusting the sampling rate of the sensors according to the change of network bandwidth, the complete transmission of important high-priority data (such as electrocardiogram signals) can be prioritized. This strategy can be flexibly adjusted according to the bandwidth situation to ensure that key data during first aid is not lost, thereby improving the execution efficiency and accuracy of first aid instructions; by dynamically prioritizing the complete transmission of high-priority data, it can ensure that in the case of limited bandwidth, first aid personnel can obtain the most critical information, such as changes in the vital signs of the wounded. This ensures the timeliness and accuracy of the medical decision-making process, helping to improve the success rate of first aid; by transmitting first aid instructions through the main link and compressed sensor data through the backup link, the stability of the data is effectively guaranteed. Even if one link fails, the backup link can still continue to work, ensuring high communication reliability, and increasing the success rate of data transmission to ≥95%, significantly improving the success rate of first aid tasks; through the dual-link and frequency band switching mechanism, the system can effectively cope with the interference and changes of communication signals in complex marine environments, ensuring the smooth transmission of first aid data and instructions, especially in harsh weather and sea environments with strong signal interference.

[0013] In one embodiment of the present invention, the S12 includes: Define a weight function, and the weight function is obtained through the following formula:

[0014] Wherein, 、 And Indicates the dynamic coefficient (the initial value is , , ); Based on the Qlearning algorithm, the online optimization coefficients are collected every 10 seconds, including link performance data (such as packet loss rate and delay jitter). The weight parameters are updated with the goal of maximizing the communication success rate. An initial threshold is set, which is a stability score of 70 points (out of 100). A sliding window (window size = 30 seconds) is used to collect historical score distribution statistics. If the score standard deviation is greater than 15, the threshold is dynamically lowered to 60 points to reduce the switching frequency. A weather-related factor is introduced. When the wind speed is greater than 15m / s or the wave height is greater than 3m, the threshold is automatically increased by 10 points to prioritize maintaining the current link stability. Before the LSTM predicts that the current frequency band score is about to fall below the threshold, the system establishes a physical layer connection (PHY layer handshake) for the target frequency band 5ms in advance and buffers 500ms of communication data. Software-defined radio (SDR) technology is used to support frequency band switching time of ≤ 2ms, ensuring uninterrupted transmission of emergency instructions. Based on a multi-device coordination strategy, when multiple rescue terminals trigger a handover simultaneously, a priority Token Ring mechanism is used to allocate handover timing to avoid channel congestion. A collision detection module monitors the signal-to-noise ratio (SNR). If the SNR drops by more than 3dB after a handover, the system rolls back to the original frequency band and triggers a backup plan. The main link (emergency command) uses the UDP protocol + forward error correction (FEC) encoding, sets an ultra-short frame structure (50ms per frame), and has the highest priority (DSCP=EF). The backup link (sensor data) is based on the QUIC protocol, which improves data compression by more than 80% (based on the LZ4 algorithm) through multiplexing and header compression. Based on the sensitivity classification of data content, ECG signals and GPS coordinates are marked as critical data (mandatory main link transmission), while temperature and humidity are marked as non-critical data (dynamically allocated according to link load); A random forest classifier is used to input data type, link load rate, and environmental threat level into the traffic diversion decision model, and output the optimal diversion path (decision delay ≤ 1ms). An asynchronous verification mechanism is deployed between the primary and backup links, comparing key data hash values (SHA3) every 10 seconds. Any inconsistency triggers retransmission on the primary link. Duplicate data packets are automatically filtered based on timestamp and content similarity (cosine similarity > 90%). When the continuous packet loss rate of the primary link exceeds 5%, the backup link is automatically enabled to take over the transmission of emergency commands and start bandwidth preemption mode (forcibly releasing non-critical data bandwidth). After the performance of the primary link stabilizes, the data flow is gradually migrated back (migration rate = 50Mbps / second) to avoid instantaneous congestion.

[0015] The working principle and effect of the above technical solution are as follows: By dynamically adjusting the weight function (including bandwidth, stability score, and switching delay) and using the Q-learning algorithm for online optimization, the weight parameters can be updated in real time based on link performance data (such as packet loss rate and delay jitter), thereby continuously maximizing the communication success rate and ensuring link stability and efficient transmission. Meteorological factors (such as wind speed and wave height) are introduced to automatically adjust the threshold to ensure priority link stability in extreme weather conditions, reduce the number of switches, and avoid communication interruptions. This approach enhances the system's environmental adaptability, enabling it to provide stable communication services in various environments. When the LSTM predicts that the current frequency band score may be lower than the threshold, a physical layer connection is established in advance, and software-defined radio (SDR) technology is used to ensure that the frequency band switching time is less than 2ms, thereby minimizing switching delay. For critical data such as emergency instructions, communication interruptions can be effectively avoided, ensuring real-time transmission. When multiple devices are switching simultaneously, a token ring mechanism is used to allocate switching timing to avoid channel congestion. Signal-to-noise ratio and collision detection mechanisms ensure signal quality after switching, and rollback to the original frequency band is performed when necessary to ensure communication reliability. The primary link uses the UDP protocol with forward error correction (FEC) encoding and an ultra-short frame structure to improve the transmission efficiency and priority of emergency instructions. The backup link uses the QUIC protocol and uses data compression and multiplexing to improve bandwidth efficiency and maximize data transmission rates. Dynamic link allocation is used for non-critical data, ensuring that critical data is prioritized. A random forest classifier is used to make accurate data flow diversion decisions and select the optimal path based on factors such as link load and environmental threat level. Decision latency is kept below 1ms, ensuring rapid data processing and transmission. An asynchronous checksum mechanism and hash value comparison ensure data consistency between the primary and backup links, triggering retransmission in the event of data anomalies. Automatically filtering duplicate packets through timestamp and content similarity detection effectively improves data transmission efficiency and accuracy. When packet loss on the primary link is high, it automatically switches to the backup link and activates bandwidth preemption mode to ensure the transmission of emergency instructions is not affected. The system also has the ability to gradually re-migrate data streams to avoid sudden congestion and improve the intelligent and rational use of bandwidth.

[0016] In one embodiment of the present invention, the S2 includes: S21. Adopt the UTM coordinate system conversion algorithm to unify GPS data, satellite remote sensing images (resolution ≤ 1m), and medical databases (casualty medical history, allergy information) into the same spatio-temporal framework; S22. Construct a rescue knowledge graph based on the Neo4j graph database. The nodes include casualties, rescue equipment, and medical resources. The edge attributes include spatio-temporal relationships (such as "Casualty A is 15 meters northeast of lifeboat B") and medical dependency relationships (such as "Medicine C needs to be used after equipment D arrives"). Parse medical guidelines through natural language processing (NLP) to generate structured operation instructions (such as "The compression depth for cardiopulmonary resuscitation is 5 - 6 cm") and bind them to AR interface elements; S23. Based on the path planning algorithm combined with ocean current field data, plan an obstacle avoidance path in real time, and project dynamic guiding arrows (accuracy ±0.5m) on the AR interface through light field rendering technology; S24. Adopt a time series prediction model (Prophet algorithm) to generate a future 10 - minute trend curve based on historical vital sign data and overlay it on the AR annotation box of the casualty. Integrate the LeapMotion sensor to support rescue personnel to retrieve the AR menu through gestures (such as making a fist to confirm, waving to turn the page), with a response delay ≤ 50ms.

[0017] The working principle and effects of the above technical solution are as follows: Through the UTM coordinate system conversion algorithm, GPS data, satellite remote sensing images, and medical database data are successfully integrated into the same spatio-temporal framework, providing a high-precision and real-time rescue environment map. Such data integration not only improves the visualization of information but also provides accurate data support for real-time decision-making, enabling rescue personnel to make rapid responses based on geographical and medical data; The rescue knowledge graph constructed based on the Neo4j graph database combines spatio-temporal relationships and medical dependency relationships, effectively integrating various types of rescue information. In the form of a graph, rescue personnel can intuitively understand the distribution of the wounded and medical resources and make accurate decisions based on the relationships in the graph; NLP parses medical guidelines and generates structured operation instructions, further reducing errors in manual interpretation and ensuring the standardization and accuracy of rescue operations; The path planning algorithm combined with ocean current field data can automatically generate the optimal obstacle avoidance path according to real-time environmental changes. This intelligent path planning not only improves rescue efficiency but also avoids obstacles in complex environments, ensuring the safety of rescue personnel; The dynamic guiding arrows projected on the AR interface, through light field rendering technology, provide real-time navigation with an accuracy of ±0.5 meters, reducing the operation burden on rescue personnel; Using the time series prediction model (Prophet algorithm), it can accurately predict the vital sign trends of the wounded in the next 10 minutes, providing forward-looking support for medical decision-making. This function is particularly crucial for severely wounded patients, being able to early warn of potential life threats and ensure timely treatment. In addition, the superimposed trend curves can help rescue personnel more intuitively understand the condition of the wounded and enhance the accuracy of decision-making; By combining the LeapMotion sensor and gesture control technology, the interaction efficiency between rescue personnel and the AR system is greatly improved. Through simple gesture operations (such as making a fist to confirm or waving to turn the page), rescue personnel can quickly retrieve the required operation menu. The gesture control response delay is less than 50ms, ensuring the real-time and efficient operation of the system. Especially in emergency situations, it can quickly respond to the needs of rescue personnel, reducing the time delay of traditional operation methods; This technical solution, through the combination of diversified data sources (GPS, remote sensing images, medical data) and various algorithms (graph database, path planning, vital sign prediction, etc.), provides comprehensive data support and intelligent optimization for rescue decision-making. With the help of the AR interface, NLP, and gesture control technology, the efficiency and accuracy in the rescue process are improved, enabling rescue personnel to quickly and accurately execute tasks under extreme conditions, greatly enhancing the success rate of rescue work.

[0018] In one embodiment of the present invention, the S21 includes: S211. Define a standardized data interface protocol to support seamless access to GPS modules (NMEA0183 format), satellite remote sensing data (GeoTIFF format, with metadata), and medical databases (HL7 FHIR standard); through a data verification module, use cyclic redundancy check (CRC32) and hash value comparison (SHA256) to verify the transmitted data, with an error rate ≤ 0.01%; for the scenario of GPS signal loss, adopt the Kalman filter algorithm to fuse inertial navigation (IMU) data and predict the instantaneous position of the vessel (error ≤ 2m); for low-resolution satellite images (>1m), apply super-resolution reconstruction technology (ESRGAN model) to enhance the satellite images; enhance to the target resolution, with PSNR ≥ 30dB. S212. Dynamically select the UTM zone number (Zone160) according to the real-time position (longitude range) of the rescue vessel and calculate the in-zone planar coordinates (Easting / Northing); fuse the astronomical tide table and real-time buoy data to correct the elevation datum deviation (accuracy ±0.1m); construct a drift prediction model based on long short-term memory network (LSTM), input the vessel speed, sea current vector, and wind speed, and output the coordinate offset (Δx, Δy) in the next 10 seconds, and compensate it to the UTM conversion result. S213. Based on the model online update mechanism, collect the latest environmental data every 5 minutes and fine-tune the network parameters through transfer learning to keep the prediction error ≤ 0.5m; add four-dimensional spatio-temporal tags (longitude, latitude, elevation, UTC timestamp) to each frame of data and use spatio-temporal Rtree indexing to accelerate the query (response time ≤ 10ms). S214. Based on the preset medical data spatialization rules, convert the geographical location description (such as "the third cabin on the deck") in the casualty's medical history into UTM coordinates and map it in combination with the ship's 3D model (error ≤ 0.3m). S215. Through the in-memory data lake based on Apache Arrow, perform columnar storage and parallel processing on GPS trajectories, satellite image slices, and medical texts; adopt spatio-temporal association rule mining (Apriori algorithm) to automatically discover the implicit relationships between data (such as "the conflict risk between the casualty's allergy history and the nearby drug inventory"). S216. Deploy a lightweight UTM conversion library (implemented in C++, with memory occupancy ≤ 10MB) at the edge node to support a single-frame data processing delay ≤ 5ms; and through the incremental synchronization protocol, transmit the coordinate difference data (ΔEasting / ΔNorthing), reducing the bandwidth occupancy by 60%. S217. Use the Lamport logical clock algorithm to synchronize the timestamps of multi-terminal data. When coordinate conflicts are detected (for example, the positioning deviation between two devices > 3m), trigger the blockchain-based consensus mechanism and select the data of the majority of nodes as the benchmark.

[0019] The working principle and effects of the above technical solution are as follows: By defining a standardized data interface protocol, seamless access to multiple data sources (such as GPS modules, satellite remote sensing data, medical databases, etc.) is supported, realizing the unification and standardization of data sources and effectively solving the compatibility problems of different data formats and sources. In addition, the data verification module is used to accurately verify the transmitted data to ensure the accuracy and stability of data transmission. The Kalman filter algorithm is used in combination with inertial navigation (IMU) data fusion, which can effectively compensate for the influence of GPS signal loss and accurately predict the instantaneous position of the ship, with the error remaining within 2 meters. At the same time, a drift prediction model is constructed by combining factors such as ocean currents and wind speeds to further improve the accuracy of ship position prediction. For low-resolution satellite images, super-resolution reconstruction technology is applied, which can effectively improve the image resolution to reach the target resolution, and the peak signal-to-noise ratio (PSNR) reaches more than 30dB, improving the usability and clarity of image data and further supporting accurate environmental analysis and decision-making. Through technologies such as long short-term memory network (LSTM), the system can update environmental data in real time and fine-tune the network model to keep the prediction error below 0.5 meters, ensuring the precise positioning of the ship in a complex environment. At the same time, using the spatio-temporal Rtree index to accelerate queries can respond in milliseconds, greatly improving the real-time performance of the system. By converting the geographical location information in the medical history of the wounded into UTM coordinates and accurately mapping them in combination with the 3D model of the ship, it can effectively help the medical team understand the specific location of the wounded. In addition, the spatio-temporal association rule mining technology is used to discover the implicit relationships between data (such as conflicts between the allergic history of the wounded and drug inventory), providing an important basis for medical decision-making. The in-memory data lake based on Apache Arrow can efficiently perform columnar storage and parallel processing of data, helping to improve the processing capacity and storage efficiency of the system. At the same time, the incremental synchronization protocol is used to transmit the coordinate difference data, reducing the bandwidth occupancy and optimizing the network resources of the system. Through the Lamport logical clock algorithm and the blockchain-based consensus mechanism, the system can achieve multi-terminal data timestamp synchronization, ensuring the consistency of positioning data of multiple devices. When coordinate conflicts are found, the consensus mechanism can be used to select the data of the majority of nodes as the benchmark, thus avoiding the influence of incorrect data on decision-making.

[0020] In one embodiment of the present invention, the S215 includes: Integrate GPS trajectory data, high-resolution satellite image slices, medical text records, and ship environmental monitoring data (such as temperature, humidity, air quality, etc.) using in-memory data lake technology based on Apache Arrow; the data is managed in a columnar storage format; Preprocess the data and, based on the Apriori algorithm, mine the implicit relationships between medical events (such as changes in the health status of the wounded) and environmental factors (such as weather conditions, geographical locations); for example, analyze the probability of the deterioration of the wounded's condition under specific weather conditions, or identify the correlation between specific areas inside the ship and the frequent occurrence of allergic reactions in the wounded; Use geographic information systems (GIS) and statistical software to conduct spatial autocorrelation analysis and hotspot analysis to identify hotspots of medical needs or potential risk areas; combine historical rescue data to build a prediction model to evaluate future rescue demand trends; Based on the mined spatio-temporal association rules and statistical analysis results, develop an intelligent decision support system. Based on the intelligent decision support system, monitor environmental changes, the status of the wounded, and the position of the ship in real time, automatically evaluate the urgency and priority of rescue operations, and send warning signals to the command center; Use genetic algorithms in combination with ship navigation path planning to dynamically adjust the allocation of rescue resources to ensure that medical resources can be delivered to the location of the wounded in the shortest possible time.

[0021] The working principle and effects of the above technical solution are as follows: By using the in-memory data lake technology based on Apache Arrow, GPS trajectory data, high-resolution satellite image slices, medical text records, and ship environmental monitoring data are integrated on a unified platform, enabling efficient columnar storage and processing. This data integration method can significantly improve the speed of data access and processing efficiency, ensuring that different types of data (such as environmental monitoring data and medical records) can be interoperated in real time in the same system, thus providing a solid data foundation for subsequent analysis; By applying the Apriori algorithm to mine the potential correlation between medical events (such as changes in the health status of the wounded) and environmental factors (such as weather conditions, geographical locations), it can help the medical team predict risks under specific conditions. For example, analyzing the possibility of the deterioration of the wounded's condition in a specific weather, or the frequent occurrence of allergic reactions in certain ship areas, can take preventive measures in advance or carry out personalized medical interventions to improve the accuracy and timeliness of medical responses; Using geographic information system (GIS) and statistical software for spatial autocorrelation analysis and hotspot analysis can effectively identify hotspots or potential risk areas of medical needs inside and around the ship. Combining historical rescue data, the constructed prediction model can accurately evaluate the future trend of rescue needs, helping the command center take proactive measures when problems are detected in advance and avoiding slow responses in case of emergencies; Based on the results of spatio-temporal association rule mining and statistical analysis, the intelligent decision support system can monitor environmental changes, the status of the wounded, and the position of the ship in real time, and automatically evaluate the urgency and priority of rescue operations. Through this real-time monitoring, the system can issue timely warning signals for potential medical emergencies and automatically adjust response strategies to ensure the real-time and accuracy of decisions; Based on the genetic algorithm combined with ship navigation path planning technology, the system can dynamically adjust the allocation of rescue resources. By calculating the optimal path and resource scheduling, it ensures that medical resources can be accurately delivered to the location of the wounded in the shortest time. Different from the traditional static scheduling method, dynamic adjustment can cope with complex and changeable environments and emergencies, thus greatly improving the rescue efficiency and success rate; Through the real-time fusion and collaborative processing of multi-dimensional data, this technical solution not only improves the system processing speed, but also enhances the system's prediction ability and response ability. Through the accurate analysis and prediction of large-scale data, the system can anticipate various potential risks in advance, make accurate decisions, and greatly improve the overall efficiency in multiple aspects such as medical rescue, ship management, and environmental monitoring.

[0022] In one embodiment of the present invention, the S3 includes: S31. With the goal of the shortest rescue time, the lowest resource consumption, and the highest medical priority as the objective function, use the deep reinforcement learning (DRL) algorithm to generate rescue plans in real time; through the online learning module, combine real-time environmental data (such as meteorological mutations, equipment failures) with expert feedback to dynamically adjust decision weights; build a distributed medical expert library, and based on federated learning technology, securely share cross-institutional medical data. The AI engine automatically matches the optimal expert and pushes diagnosis and treatment suggestions to the AR terminal; S32. Build a high-fidelity maritime first aid simulation environment based on the physics engine and AR technology. The AI automatically generates multi-dimensional emergency plans (including extreme weather, multi-casualty scenarios); through computer vision (CV), real-time identify the operations of rescue personnel, compare with the standard process knowledge base, and if deviations are detected, trigger AR warnings and correction guidelines; S33. Use reinforcement learning to simulate different first aid scenarios, dynamically generate personalized training courses, and monitor the stress response of trainees through the brain-computer interface (BCI) to optimize the training plan.

[0023] S34. The satellite communication link adopts a hierarchical transmission strategy. Key instructions (such as first aid operation guidelines) are responded to in real time by edge nodes, and non-critical data (such as historical cases) are asynchronously uploaded to the cloud for analysis; when the satellite signal is interrupted, the system automatically switches to the mesh network communication mode and starts the local AI inference module to generate emergency decisions based on historical data.

[0024] The working principle and effects of the above technical solution are as follows: The deep reinforcement learning (DRL) algorithm can continuously learn and dynamically adjust the rescue strategy by processing environmental changes, casualty status, and equipment failure information in real time. The objective function (shortest rescue time, lowest resource consumption, and highest medical priority) ensures optimized decision-making in emergencies, enabling the maximum utilization of each resource; the online learning module combines real-time environmental data and expert feedback to continuously optimize the decision-making model, enhancing the system's ability to handle unknown emergencies. For example, sudden weather changes, equipment failures, etc. can be considered during the decision-making process, thereby dynamically adjusting the rescue plan to ensure the effectiveness of medical rescue; the federated learning technology ensures the secure sharing of data from different medical institutions without revealing privacy, thus building a distributed medical expert database. The AI engine can automatically match the optimal expert and push diagnosis and treatment suggestions to provide real-time remote medical support. This not only improves the quality of remote medical services but also accelerates the medical decision-making process for the injured; the expert feedback mechanism enables the system to not only rely on algorithms for decision-making but also be updated and optimized in real time, ensuring continuous improvement and the best effect of the medical rescue plan; the high-fidelity simulation environment combining AR technology and the physics engine allows rescue personnel to conduct actual operation training in a virtual environment. By using computer vision to real-time identify the operations of rescue personnel and compare them with the standard operation procedures, deviations in operations can be detected in a timely manner, and correction guidance can be provided through the AR interface, thus avoiding medical risks caused by operation errors; simulations of extreme weather and multi-casualty scenarios help training personnel to handle medical rescue in complex situations and enhance their emergency response capabilities during actual rescue; reinforcement learning is applied to the first aid scenario simulation to dynamically generate personalized training courses according to different training needs and the situations of trainees. In this way, the training can better meet the needs of the trainees and enable them to achieve the most effective improvement; the application of the brain-computer interface (BCI) enables the monitoring of the stress responses of trainees during the training process, and then optimizes the training plan. This not only helps to improve the training effect but also ensures the reaction speed and accuracy of trainees in real emergencies; through the hierarchical transmission strategy, the system can still ensure the real-time transmission of critical instructions (such as first aid operation guidelines) even when the satellite communication link is interrupted, guaranteeing the smooth progress of the rescue work. The edge nodes respond quickly and generate emergency decisions through the local AI inference module, ensuring the effectiveness of rescue decisions even in an environment with unstable signals; the automatic switching of the Mesh network mode ensures uninterrupted communication, maximizing the reliability and emergency response capabilities of the system; by dynamically adjusting the rescue strategy, the system can not only implement the rescue in the shortest time but also effectively reduce resource consumption.The deep reinforcement learning algorithm optimizes resource allocation and personnel deployment, enabling each resource (such as medical staff and equipment) to be utilized promptly and effectively where it is most needed, ensuring maximum resource utilization; the personalized rescue plan targets the needs of different wounded, avoiding the inefficiency of a general rescue plan and further improving the overall efficiency of the rescue work; by integrating multiple technical means, the system achieves highly intelligent emergency response and decision support. From real-time environmental changes, the status of the wounded to the detection of equipment failures, it can automatically respond and quickly make emergency decisions through an accurate inference model, providing immediate guidance for rescue personnel; the disaster prediction and early warning mechanism combines historical data and real-time data, enabling rescue decisions to be based not only on the current situation but also on predicting possible future risks, deploying resources in advance to prevent the situation from deteriorating.

[0025] In one embodiment of the present invention, S31 includes: S311. Define the state space, including the classification of the vital signs of the wounded (using the MEWS score), the resource distribution matrix (lifeboat, drug stock), and the environmental threat index (wind speed, sea current speed); set triple weighted rewards (reward for shortened rescue time + penalty for resource consumption + medical priority score), and optimize the parameters of the policy network through Monte Carlo Tree Search (MCTS); S312. When a meteorological mutation is detected (such as a 20% sudden increase in wind speed), trigger the retraining of the policy network, retain the historical policy weights based on transfer learning, and adjust the time threshold ≤ 300 ms; S313. Locally deploy a lightweight AI model (such as DenseNet121) in each medical institution, aggregate the model parameters through the Federated Averaging algorithm (FedAvg), share the feature extraction layer but isolate sensitive data; calculate the matching degree between the symptoms of the wounded and the expert's area of expertise based on cosine similarity, and preferentially push the expert video stream with a matching degree ≥ 85% to the AR terminal; S314. Integrate the opinions of multiple experts using Bayesian inference to generate a joint diagnosis and treatment plan with a confidence level ≥ 90%, and highlight the key operation steps through the AR interface.

[0026] The working principle and effects of the above technical solutions are as follows: The state space definition and triple weighted reward mechanism ensure the comprehensiveness and flexibility of the decision-making process. Through weighted evaluation in multiple dimensions (such as the vital signs of the wounded, resource distribution, environmental threats, etc.), the deep reinforcement learning model can make more accurate decisions in a dynamic environment. Especially when dealing with complex rescue tasks (such as multiple wounded and complex environmental changes), this weighted reward mechanism helps to balance the optimization of different goals, such as shortening the rescue time, reducing resource waste, and enhancing medical priorities; The Monte Carlo Tree Search (MCTS) optimized policy network can efficiently explore the optimal decision-making path in a changing environment, avoiding local optimal solutions that may occur in traditional algorithms, thus achieving a global optimal strategy. This enables the rescue task to still achieve the best results under time and resource constraints; The meteorological mutation triggers the retraining mechanism. When sudden meteorological changes such as a sharp increase in wind speed occur, the system can quickly retrain the policy network to ensure accurate decision-making under extreme weather conditions. The introduction of transfer learning enables the network to avoid training from scratch by retaining historical policy weights, significantly reducing the training time (time threshold ≤ 300ms), ensuring real-time performance and the continuity of decision-making; This mechanism ensures that the system can cope with a rapidly changing environment, improves the system's emergency response ability, and avoids a decline in rescue efficiency due to sudden environmental changes; Federated learning (FedAvg) enables multiple medical institutions to achieve knowledge sharing and model collaboration while ensuring data privacy and security. By aggregating the model parameters and feature extraction layers of different institutions, the system can improve the accuracy and effectiveness of diagnosis and treatment decisions on the basis of multi-party collaboration; By calculating the cosine similarity between the symptoms of the wounded and the areas of expertise of experts, the system can intelligently push the expert resources with the highest matching degree to ensure that each wounded person can receive the most suitable expert remote support. This intelligent matching mechanism greatly improves the efficiency of telemedicine and also makes the allocation of expert resources more efficient; The Bayesian inference method is used to integrate the opinions of multiple experts and generate a joint diagnosis and treatment plan with a confidence level ≥ 90%. This can not only improve the accuracy of diagnosis and treatment decisions but also avoid misdiagnosis or biases that may exist in a single expert, enhancing the reliability during the diagnosis and treatment process; The highlighting function of the AR interface enables rescue personnel to quickly master important medical operations by intuitively displaying key operation steps, thus reducing operation errors in a high-pressure environment. This real-time feedback mechanism significantly improves the first aid efficiency and ensures medical intervention at critical moments; The comprehensive evaluation of the medical priority score enables the wounded with more critical vital signs to be treated first in case of resource shortages, avoiding a significant reduction in rescue effects due to unreasonable resource allocation.The resource consumption penalty in the triple weighted reward can also guide the system to reduce resource waste and ensure that the use of each resource achieves the best benefit. The optimization of resource scheduling and allocation automates the management of the deployment of lifeboats, medicines, and personnel by the system, avoiding errors that may occur in traditional manual scheduling and ensuring the efficient use of resources during the rescue process. The combination of the AR terminal and the real-time feedback mechanism provides more intuitive and practical operation guidance for rescue personnel. Through the instant feedback and operation prompts provided by the AR interface, it can help rescue personnel reduce anxiety during the rescue process and improve the accuracy of decision-making and operation. The intelligent operation prompts automatically adjust the operation process and priorities by analyzing the casualty situation and environmental changes in real time, and display the key operation steps through AR to ensure that no details are missed during the rescue process. The combination of the lightweight AI model (such as DenseNet121) and edge computing enables the system to not only operate efficiently under resource constraints but also reduce the dependence on the central server, improving the stability and response speed of the overall system. Especially in extreme environments (such as maritime rescue), edge computing can ensure that rescue operations are not interrupted. The intelligent optimization of resource and task scheduling greatly reduces manual intervention, improves the degree of automation, saves time, and enhances efficiency. Through the dynamic environment perception and adjustment mechanism, the system can quickly identify the current state and adjust the rescue strategy and medical plan according to real-time data. This rapid response mechanism ensures that even in extremely urgent and complex situations, the system can quickly respond, maximizing the survival probability of the casualties.

[0027] In one embodiment of the present invention, the S32 includes: S321. Construct digital twins of scenarios such as ship capsizing and fires based on the Unity3D engine, with a fluid dynamics simulation accuracy reaching the CFD level (mesh size ≤ 1 cm³); use the generative adversarial network (GAN) to synthesize rare disaster data (such as a compound event of oil spill + casualty poisoning) to enhance the scenario adaptation ability of the AI model; S322. Combine the historical case database with the current environmental parameters to output a three-level emergency plan including a material scheduling plan, an escape route, and a remote consultation process (response time ≤ 2 seconds); S323. Use the OpenPose algorithm to capture the key points of the rescue personnel's bones in real time and perform dynamic comparison with the standard operation templates (such as the CPR compression angle and the position of the tourniquet); S324. When a deviation is detected (such as a compression depth of less than 4 cm), the AR interface synchronously displays a red flashing warning box, plays a voice prompt (decibel ≥ 80 dB), and gives a vibration feedback (frequency 10 Hz); generate a dynamic correction strategy based on reinforcement learning (such as "rescue by shifting 15 cm to the left"), and guide the operation correction through the superposition of AR arrows and text.

[0028] The working principle and effectiveness of the above technical solution are as follows: By constructing disaster scenarios such as ship capsizing and fire based on the Unity3D engine and combining them with CFD-level fluid dynamics simulation, the fluid and heat transfer processes at the disaster site can be extremely accurately reproduced. The accuracy of the grid size of ≤1cm³ ensures the accuracy of the simulation, providing rescue personnel with a reliable virtual training environment to help them better understand the disaster situation, predict possible risk areas, and prepare. Using GAN to synthesize complex and rare disaster scenarios (such as a combination of an oil spill and poisoning), not only enhances the diversity of the AI model's training data, but also improves the model's generalization ability. By generating diverse scenarios and data, the AI system can cope with various sudden disasters, ensuring more efficient emergency response and decision support. By combining historical case libraries with real-time environmental data (such as weather and sea conditions), the AI model can output diversified and personalized emergency plans. This dynamic emergency response plan, based on big data analysis, can quickly generate rescue plans covering material dispatch, escape routes, remote consultations, and other content within a response time of ≤2 seconds, effectively improving the response speed and accuracy of emergency decision-making. The system can generate up-to-date emergency response plans based on the specific characteristics of each disaster, further ensuring that in complex and changing disaster environments, rescue teams can quickly adjust strategies based on system prompts to maximize the safety of the injured and resource utilization. By using the OpenPose algorithm to capture key points of the rescuer's skeleton in real time, every movement of the rescuer can be dynamically tracked to ensure that the operation complies with standard operating procedures (SOPs). In particular, in critical operations, such as the correctness of CPR compression angles and depths and tourniquet placement, deviations can be promptly identified and corrected by comparing with standard templates. This technology can closely integrate rescue processes with best practices to ensure that every rescuer's operations meet optimization standards. The real-time feedback mechanism improves operational accuracy and reduces human error through data-driven analytics, particularly in high-pressure, high-risk rescue environments, ensuring maximum safety and effectiveness. When a deviation is detected (e.g., CPR compression depth less than 4cm), the system not only alerts rescuers with a flashing red warning box, an audible prompt (≥80dB), and vibration feedback (frequency 10Hz), but also uses augmented reality to display dynamic correction strategies, such as "move 15cm left to begin rescue." This multi-dimensional feedback mechanism not only enhances operator awareness but also improves the success rate of real-time corrections through visual, auditory, and tactile feedback. Using reinforcement learning, the system automatically generates the optimal correction strategy for potential deviations during the rescue process and displays it in real time through the AR interface. For example, if a rescuer's movements deviate, the system automatically adjusts and guides them toward the optimal rescue path.This intelligent correction and feedback system improves rescue efficiency and minimizes operational errors of rescue personnel. By combining multiple feedback methods such as AR, audio, and vibration, it ensures that rescue personnel can accurately understand and execute operation correction instructions even in extreme situations (such as emotional stress and complex environments). Through these precise feedbacks, the operation accuracy of rescue personnel can be effectively improved, and delays caused by operation errors can be reduced. The system can dynamically adjust strategies according to real-time feedback through reinforcement learning. This adaptive correction ability enables the rescue process to be continuously optimized, ensuring high rescue execution efficiency in changing environments and complex situations. The system can not only capture and analyze the actions of rescue personnel in real time but also integrate data from various sources such as the on-site environment, equipment, and casualties to achieve global collaboration. Through the integration and processing of multi-modal data, the system can comprehensively consider various factors such as environmental conditions, resource allocation, and casualty situations, optimize decisions in different scenarios, and significantly improve rescue efficiency. In disaster scenarios, combined with the remote consultation process, the system can push the real-time data of casualties to appropriate medical experts, provide accurate medical advice, and guide on-site rescue personnel to take necessary treatment measures. This collaborative operation between experts and on-site operators can greatly improve the rescue effect in disaster rescue.

[0029] In one embodiment of the present invention, S33 includes: S331. Dynamically adjust the difficulty of the simulation scenario according to the historical performance of the trainee (for example, the primary scenario only has a single casualty, and the advanced scenario includes an explosion + multiple language casualties); adopt the Curriculum Learning strategy to gradually increase the scenario complexity and interference factors (such as communication interruption, equipment failure) to improve the stress resistance of the trainee. S332. Construct an evaluation matrix including operation speed (seconds / step), decision accuracy rate (%), and psychological stability (heart rate variability), and output a personalized improvement report; monitor the activity of the prefrontal cortex of the trainee through a non-invasive EEG headset (sampling rate 256Hz) to identify states of anxiety (enhanced β waves) or distraction (abrupt change in α waves). S333. When it is detected that the stress level exceeds the threshold (such as heart rate > 120 bpm), automatically reduce the scenario complexity or insert a virtual coach's voice for comfort; generate a special stress resistance training course based on the trainee's stress pattern through time series clustering analysis (such as night rescue simulation + sudden noise interference).

[0030] The working principle and effects of the above technical solution are as follows: By dynamically adjusting the difficulty of the simulation scenario according to the trainee's historical performance (for example, from a single casualty to multiple complex scenarios), it can effectively ensure that each trainee conducts training within their ability range. Adopting the Curriculum Learning strategy, gradually increasing the complexity of the scenario and interference factors enables trainees to improve their skills in gradually increasing challenges, avoiding anxiety or frustration caused by overly complex scenarios at the beginning. This gradually increasing strategy helps trainees stay calm under pressure, improve their stress resistance, and promote continuous progress in skills; By constructing an evaluation matrix including multiple dimensions such as operation speed, decision-making accuracy, and psychological stability, not only the execution ability of trainees is evaluated from a technical level, but also their psychological stability is evaluated by monitoring physiological signals such as heart rate variability, providing more comprehensive feedback on training effects. This comprehensive evaluation method can deeply understand the performance of trainees during training, and by generating personalized improvement reports, help them identify and optimize their own weaknesses, promoting the formulation of targeted training; Using a non-invasive EEG headset to real-time monitor the activity of the prefrontal cortex of trainees can accurately identify their psychological states, such as anxiety (enhanced beta waves) or distraction (abrupt alpha wave changes). This electroencephalogram monitoring technology has high precision in psychological state evaluation, can timely capture the emotional changes of trainees, avoid trainees making wrong decisions or operations under the condition of anxiety or excessive stress, and provide effective physiological feedback to assist their emotion management; When the system detects that the physiological stress level of the trainee exceeds a predetermined threshold (for example, heart rate > 120 bpm), it can automatically adjust the scenario complexity, reduce the burden on the trainee, or comfort the trainee through the voice of a virtual coach to help them quickly regain calm. This automated stress regulation mechanism ensures that trainees can receive effective support when facing pressure, avoiding excessive emotional load from interfering with their operations; Based on time series clustering analysis, the system can real-time identify the stress patterns of trainees, identify the characteristic manifestations of their stress (such as heart rate changes, electroencephalogram fluctuations, etc.), and design targeted stress resistance training courses accordingly, such as simulating noise interference in night rescue and other scenarios. Through personalized stress management training, it can help trainees effectively adapt to different stress patterns and enhance their stability and adaptability in a stressful environment; This technical solution not only trains the stress resistance of trainees by adjusting the scenario complexity and interference factors, but also combines psychological and physiological monitoring and feedback mechanisms to ensure that trainees can effectively cope with challenges in a high-pressure environment and improve their ability to respond to emergencies.Especially in some high-pressure and high-risk training scenarios (such as night rescue, post-explosion environment, etc.), trainees can gradually cultivate strong coping abilities and situational adaptation abilities through training with gradually increasing complexity and dynamic adjustment support; during the training process, the voice comfort and guidance provided by the virtual coach not only relieve the trainees' anxiety, but also can adjust the training content in real time according to the trainees' physiological feedback. This adaptive ability and feedback mechanism of the virtual coach provide continuous emotional support for the trainees, ensuring that they always maintain a high performance in the simulated environment; by accumulating the trainees' training data (including physiological data, operation speed, decision-making accuracy, etc.), the system can perform big data analysis based on these data to further optimize the training content and scenario design. This data-driven training mode can not only be customized according to the personalized needs of each trainee, but also continuously adjust and improve the training plan to improve the accuracy and effectiveness of the training effect; according to the trainees' feedback data, the system can intelligently recommend advanced training courses suitable for their current level, and gradually guide the trainees to improve their skills. During the personalized training process, trainees can not only receive guidance at the technical level, but also exercise their psychological qualities, ensuring that they can perform tasks calmly and accurately when facing actual disasters or emergencies.

[0031] In one embodiment of the present invention, the S34 includes: S341. Define critical instructions (such as hemostasis operation guidelines) as QoSLevel1 (transmission delay ≤ 100 ms), and non-critical data (such as training logs) as QoSLevel3 (delay tolerance ≥ 5 s); use the LRU algorithm to dynamically manage the local cache, retain the high-priority data in the recent 10 minutes, and ensure that it can be quickly called when the network is disconnected; S342. Deploy the Spark streaming processing engine to mine association rules for historical cases (such as the response rate of drug A to symptom B), and the cycle for updating the medical knowledge base ≤ 1 hour; S343. When the satellite link is interrupted, automatically activate the LoRa module between rescue devices to form a Mesh network with a maximum hop count ≤ 5, and expand the communication radius to 1 km; S354. Embed the pruned MobileNetV3 model (parameter quantity < 1MB) in the edge device to support the classification of the wounded (light / medium / severe) and the generation of basic solutions in the offline state; after the network is restored, automatically upload the local decision log to the cloud, and protect sensitive information through differential privacy technology to ensure the consistency of the global model.

[0032] The working principle and effects of the above technical solution are as follows: By setting different QoS levels for critical instructions (such as hemostasis operation guidelines) and non-critical data (such as training logs), it ensures that in a high-pressure emergency environment, important operations can be transmitted with extremely low latency (≤100 ms), while non-critical data can tolerate longer latency (≥5 s) when communication resources are limited. This hierarchical management optimizes the allocation of network resources and guarantees the priority handling of emergency tasks; The LRU (Least Recently Used) algorithm is adopted to dynamically manage the local cache, retaining high-priority data in the most recent 10 minutes to ensure that the most important data can be quickly retrieved under limited resources even when the network is interrupted. This mechanism ensures the efficient operation of the system when the network is unstable or interrupted, avoiding decision-making delays caused by data loss; The Spark streaming processing engine is deployed to conduct real-time association rule mining on historical cases, such as the response rate analysis of drug A and symptom B, which can provide intelligent support for decision-making based on big data. By regularly updating the medical knowledge base (≤1 hour), it ensures that medical operations and decisions are always based on the latest medical information and case data, enhancing the scientific nature and accuracy of decision-making; This data analysis-based strategy can rapidly generate targeted medical plans at the disaster rescue site, and with the update of the medical knowledge base, it ensures the continuous optimization of emergency response measures for different situations; When the satellite link is disconnected, the system can automatically activate the LoRa module and form a Mesh network with a maximum hop count ≤5, extending the communication radius to 1 km. This provides stable communication guarantee for remote areas or disaster sites. Even when the main network is interrupted, communication between devices can be maintained to ensure the transmission of instructions and data sharing; This communication method can continue in the case of network instability or interruption, reducing communication breaks and ensuring that rescue operations are not affected, especially in places without traditional communication infrastructure; The pruned MobileNetV3 model (parameter quantity <1MB) is embedded in the edge device, which can classify the wounded (mild / moderate / severe) and generate basic plans when the device is in an offline state. Through edge computing, the system can directly classify the wounded and generate preliminary medical plans on-site without relying on cloud resources, greatly improving the rescue efficiency, especially in cases of poor communication or no network connection; This design not only solves the processing problem in case of network instability but also can automatically upload the local decision log to the cloud after the network is restored and protect sensitive information through differential privacy technology. This privacy protection mechanism ensures that user data leakage can be prevented during the data sharing process, maintaining data security and compliance; Even when the edge device performs offline operations, after the network is restored, the system can automatically upload the decision log to ensure consistency between the cloud and local devices.This design avoids the inconsistency of decision-making models among different devices, improving the reliability and accuracy of the overall system. Through differential privacy technology, the system can ensure the protection of personal sensitive information, such as the identities of the wounded and medical records, when uploading logs and data, preventing data leakage. This is crucial for the privacy protection of medical data, especially in cross-regional and cross-organizational data sharing, which can minimize the risk of privacy leakage to the greatest extent. Based on the classification results of the wounded and real-time data analysis, the system can quickly generate personalized medical plans to help rescue personnel make the most appropriate treatment decisions in the shortest time. Whether it is minor or serious injuries, corresponding treatment plans can be automatically generated by the edge devices and guide on-site operations to optimize the rescue efficiency.

[0033] An embodiment of the present invention is a system for constructing a remote first-aid collaboration platform based on AI decision-making, which is characterized by including a memory, a processor, and a computer program stored on the memory and executable on the memory. The processor executes the program to implement the method for constructing a remote first-aid collaboration platform based on AI decision-making as described in any one of claims 1-9.

[0034] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A method for constructing a remote first aid collaboration platform based on AI decision-making, characterized in that, The method includes: S1. Select a communication frequency band; S2. Map satellite communication data, sensor data, and medical database information to a unified spatio-temporal coordinate system, construct a dynamic rescue scenario model based on a knowledge graph, and project key information in real time through an AR interface; S3. Generate and optimize rescue plans and personalized training courses in real time.

2. The method for constructing a remote first aid collaboration platform based on AI decision-making according to claim 1, wherein The S1 includes: S11. Input historical channel parameters and real-time ocean meteorological data, and output the stability scores of each frequency band within the next 5 seconds; S12. When the current frequency band score is lower than the threshold, trigger a millisecond-level frequency band switch; and perform data transmission; S13. Deploy a lightweight YOLOv5 model on the AR terminal to filter sensor noise data in real time and extract key features; S14. Dynamically adjust the sensor sampling rate according to the network bandwidth.

3. The method for constructing a remote first aid collaboration platform based on AI decision-making according to claim 1, characterized in that, The S2 includes: S21. Adopt a UTM coordinate system conversion algorithm to unify GPS data, satellite remote sensing images, and the medical database into the same spatio-temporal framework; S22. Construct a rescue knowledge graph based on the Neo4j graph database; parse medical guidelines through natural language processing (NLP) to generate structured operation instructions and bind them to the AR interface elements; S23. Based on a path planning algorithm combined with ocean current field data, plan an obstacle avoidance path in real time and project dynamic guiding arrows on the AR interface through light field rendering technology; S24. Adopt a time series prediction model to generate a future 10-minute trend curve based on historical vital sign data and superimpose it on the AR annotation box of the wounded.

4. The method for constructing a remote first aid collaboration platform based on AI decision-making according to claim 3, wherein, The S21 includes: Define a standardized data interface protocol, and through a data verification module, use cyclic redundancy check and hash value comparison to verify the transmitted data; adopt a Kalman filter algorithm to fuse inertial navigation data and predict the instantaneous position of the ship; for low-resolution satellite images, apply super-resolution reconstruction technology to process the satellite images; Dynamically select the UTM zone number according to the real-time position of the rescue ship, and calculate the in-zone plane coordinates; fuse the astronomical tide table and real-time buoy data to correct the elevation datum deviation; construct a drift prediction model based on a long short-term memory network, input the ship speed, sea current vector, and wind speed, and output the coordinate offset within the next 10 seconds to compensate the UTM conversion result; Based on the model online update mechanism, collect the latest environmental data every 5 minutes and fine-tune the network parameters through transfer learning; add four-dimensional spatio-temporal tags to each frame of data and use a spatio-temporal Rtree index to accelerate the query; Based on the preset medical data spatialization rules, convert the geographical location description in the wounded's medical history into UTM coordinates and map them in combination with the ship 3D model; Through an in-memory data lake based on Apache Arrow, perform columnar storage and parallel processing on GPS trajectories, satellite image slices, and medical texts; adopt spatio-temporal association rule mining to automatically discover the implicit relationships between data; Deploy a lightweight UTM conversion library on the edge node to support a single-frame data processing delay ≤ 5ms; and transmit the coordinate difference data through an incremental synchronization protocol; Adopt the Lamport logical clock algorithm to synchronize the timestamps of multi-terminal data; when coordinate conflicts are detected, trigger the blockchain-based consensus mechanism and select the data of the majority of nodes as the benchmark.

5. The method for constructing a remote first-aid collaboration platform based on AI decision-making according to claim 1, wherein The S3 includes: S31. Generate a rescue plan in real time and dynamically adjust the decision weights; S32. Conduct intelligent first aid process management and compliance monitoring; S33. Train reinforcement learning and optimize the brain-computer interface; S34. Conduct edge-cloud collaboration.

6. The method for constructing a remote first-aid collaboration platform based on AI decision-making according to claim 5, characterized in that, The S31 includes: S311. Define the state space and set the reward function; S312. When a meteorological mutation is detected, trigger the retraining of the policy network, retain the historical policy weights based on transfer learning, and adjust the time threshold ≤ 300 ms; S313. Aggregate model parameters, share the feature extraction layer and isolate sensitive data; perform expert mechanism matching; S314. Conduct diagnosis and treatment advice fusion.

7. The method for constructing a remote first aid collaboration platform based on AI decision-making according to claim 5, characterized in that, The S32 includes: S321. Conduct high-fidelity physical modeling and conduct generalization training in extreme scenarios; S322. Automatically generate an emergency plan; S323. Capture the key points of the rescue personnel's bones and conduct dynamic comparison with the standard operation template; S324. Trigger multi-modal alerts and correct the guidance in real time.

8. The method for constructing a remote first aid collaboration platform based on AI decision-making according to claim 5, characterized in that, The S33 includes: S331. Conduct scene complexity grading; S332. Conduct multi-dimensional ability assessment and collect neural signals; S333. Adjust the dynamic difficulty.

9. The method for constructing a remote first aid collaboration platform based on AI decision-making according to claim 5, wherein, The S34 includes: S341. Divide the data priority and set the edge node caching policy; S342. Conduct cloud asynchronous analysis; S343. Build a multi-hop ad hoc network; S354. Deploy a lightweight inference model and conduct decision backtracking and synchronization.

10. A system for constructing a remote first-aid collaboration platform based on AI decision-making, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the memory. The processor executes the program to implement the method for constructing a remote first aid collaboration platform based on AI decision-making as described in any one of claims 1-9.

Images