Cabin environment self-adaptive regulation and control method and system driven by vital signs of wounded

Abstract

The invention discloses a self-adaptive regulation and control method and system for an in-cabin environment driven by vital signs of a wounded person, and relates to the technical field of crossing of medical equipment and environment control. According to the self-adaptive regulation and control method and system for the in-cabin environment driven by the vital signs of the wounded person, a four-dimensional data acquisition system composed of physiology, environment, medical intervention and space-time scenes is built; the method comprises the following steps: synchronously acquiring multi-dimensional vital signs including brain electricity, skin electricity and other nerve and stress indexes, an in-cabin dynamic environment including detailed parameters such as airflow velocity and particulate matter concentration, external medical operation covering drug dosage and equipment working parameters, and transfer scene data including altitude change and bumping intensity; the limitation that only single environment or physiological data is acquired in the prior art is solved; when a craniocerebral injury wounded person is transported in an air transport mode, the system can be combined with the altitude change rate and the electroencephalogram alpha wave power, the blood oxygen lowering risk can be pre-judged 10 minutes in advance, and the problem that misjudgment is conducted only through current data traditionally is avoided.

Description

Technical Field

[0001] This invention relates to the field of medical equipment and environmental control technology, and in particular to a method and system for adaptive control of the cabin environment driven by the vital signs of the injured. Background Technology

[0002] In scenarios such as emergency rescue and patient transfer, and patient transfer during public health emergencies, sealed cabins are key carriers connecting on-site treatment and hospital treatment. The stability of their environmental parameters directly affects the stability of the patient's vital signs and the subsequent treatment effect. With the development of intelligent medical equipment, the existing cabin environment control technology has evolved from early manual adjustment to an automated mode of sensor plus cyclic control. However, there are still many technical defects in practical applications, which can be reflected one by one through typical application scenarios.

[0003] However, during the implementation of the above technical solution, at least the following technical problems were discovered:

[0004] Most existing cabin control systems only collect environmental data from a single dimension. A typical example is the control architecture of a certain vehicle-mounted emergency cabin, which only obtains data through temperature and humidity sensors and oxygen concentration sensors. The control logic relies on fixed thresholds: heating is activated when the temperature is below 22°C and cooling is activated when the temperature is above 26°C; the oxygen generator is activated when the oxygen concentration is below 21% and shut down when it is above 25%. It does not access the patient's vital signs monitoring data, nor does it collect data on scenarios such as bumps and altitude during transport. As a result, the control can only maintain the stability of environmental values ​​and cannot adapt to the dynamic physiological needs of the patient. For example, when the patient's fever reaches 38.5°C, the system still controls according to the fixed range, or when the vehicle climbs uphill and the altitude increases, causing the oxygen partial pressure to decrease, the system does not replenish oxygen in time, resulting in a drop in blood oxygen.

[0005] While some hyperbaric oxygen chamber control systems attempt to integrate multiple data sources, they employ static weight models where the weights of each parameter remain fixed and cannot be dynamically adjusted according to the scenario or injury. For example, when transporting traumatic brain injury patients by air, the increase in altitude reduces the partial pressure of oxygen, significantly enhancing the correlation between oxygen concentration and blood oxygen saturation. However, the model fails to correct this correlation or increase the weight of the oxygen concentration parameter, leading to the system misjudging environmental suitability when the patient's blood oxygen level drops, failing to replenish oxygen in time, and consequently causing transient latent hypoxia in the brain tissue. Furthermore, these models rely solely on current data for judgment, lacking the ability to predict trends and intervene in potential risks in advance, such as failing to predict that further increases in altitude will lead to a further decrease in blood oxygen.

[0006] Existing systems generally suffer from a one-size-fits-all approach to control parameters, applying a uniform safety range regardless of the patient's condition, such as controlling oxygen concentration between 40% and 60%. When transporting patients with chronic obstructive pulmonary disease, high oxygen concentrations can inhibit respiratory drive, leading to carbon dioxide retention. Conversely, when transporting patients with traumatic brain injury, this concentration is insufficient to meet the oxygen supply needs of brain tissue. Furthermore, the system lacks integration with external medical equipment. For instance, the system does not establish coordinated logic for adjusting oxygen concentrations after the infusion of vasoactive drugs to reduce microcirculation or adjusting temperature after the infusion of analgesics to decrease metabolic rate. This lack of coordination can easily lead to treatment side effects. Therefore, we propose a patient-driven adaptive control method and system for the cabin environment. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides a method and system for adaptive control of the cabin environment driven by the vital signs of the wounded, solving the five-dimensional deficiencies of existing technologies in terms of data dimension, fusion capability, individualized collaboration, reliable execution, and secure evidence storage.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] An adaptive control method for the cabin environment driven by the vital signs of wounded soldiers, the method comprising the following steps:

[0012] A four-tiered data collection system was established, comprising physiological data, environmental data, medical intervention data, and spatiotemporal scenario data. This system acquires multi-dimensional vital sign data of the injured, dynamic environmental data within the cabin, operational data of external medical equipment, and spatiotemporal scenario data of the transport. Multi-dimensional vital sign data includes 12-lead electrocardiogram data, non-invasive blood pressure data, blood oxygen saturation data, body temperature data, end-tidal carbon dioxide data, electroencephalogram (EEG) activity data, and conductance of skin activity data. Dynamic environmental data within the cabin includes temperature data, humidity data, oxygen concentration data, carbon dioxide concentration data, air pressure data, air particulate matter concentration data, and airflow velocity data. Operational data of external medical equipment includes drug dosage and type, infusion rate, and operating parameters of medical equipment with respiratory support capabilities, including inhaled oxygen concentration, tidal volume, and positive end-expiratory pressure. Spatiotemporal scenario data of the transport includes transport method, rate of altitude change, and intensity of bumps and vibrations; transport methods include vehicle transport, air transport, and ship transport.

[0013] A dynamic fusion model for four types of data was built based on an improved deep graph neural network model. Feature extraction and correlation analysis were performed on the collected data to generate a three-dimensional assessment result that includes an individual dynamic baseline, risk evolution trajectory, and intervention response prediction. The individual dynamic baseline is generated by dynamically comparing real-time EEG activity data with the patient's historical health records and is updated every 5 minutes to match the patient's current physiological state. The risk evolution trajectory is generated by fitting the rate of change and acceleration of change of vital signs data to reflect the trend of risk development. The intervention response prediction combines pharmacokinetic data, including drug half-life and drug target, to predict the duration of the impact of medical operations on vital signs.

[0014] Based on the three-dimensional assessment results and the dual-objective optimization logic of risk reduction and side effect control, an adaptive regulation strategy with functions of early intervention, dynamic correction, and side effect suppression is generated. The adaptive regulation strategy includes a precise oxygen concentration titration scheme, a pressure stepwise regulation scheme, a temperature, humidity and physiological state coordinated regulation scheme, and a carbon dioxide concentration cyclic control scheme. The pressure stepwise regulation scheme is only applicable to pressurized chambers. Each scheme embeds an environmental sensitivity coefficient, which is calculated through skin conductance data and reflects the patient's tolerance to environmental changes.

[0015] Through a distributed collaborative architecture, the in-cabin environmental control equipment, external medical equipment and remote medical center are executed in real time. During the collaborative execution, a dynamic priority and conflict arbitration mechanism is adopted. When multiple control needs conflict, the oxygen supply status of brain tissue is prioritized. The oxygen supply status of brain tissue is judged based on EEG activity data. At the same time, the urgency of the control needs is ranked according to the deterioration rate of the risk evolution trajectory to determine the priority of the control needs.

[0016] A cyclical optimization mechanism of real-time monitoring, effect feedback, and model iteration is established. The matching degree between environmental control results and vital signs is calculated every 2 minutes, and the qualified threshold for this matching degree is set at 90%. If the matching degree is lower than 80% or the risk level of vital signs is upgraded, an emergency control and remote consultation dual response mode is immediately triggered. The data collection interval is shortened to one-quarter of the original frequency, and real-time data and control logs are synchronized to the remote medical center through 5G network and edge computing technology to receive correction instructions from professional medical personnel.

[0017] The preferred improvements to the deep graph neural network model include:

[0018] Breaking through the limitations of traditional static graph neural networks, this paper constructs a dual dynamic update logic for parameter importance weights and parameter correlation strength. Each parameter in the four data categories is treated as an independent analysis object, and the weight of each parameter is adjusted in real time according to data importance. For example, when assessing the risk of shock, the weight of the blood pressure parameter is increased to 75%, while the weight of the skin conductance parameter is decreased to 10%. The correlation strength between parameters is calculated by fusing statistical correlation coefficients and information difference. For example, the correlation strength between oxygen concentration and blood oxygen saturation is adjusted with altitude; for every 1000 meters increase in altitude, the correlation strength increases by 20%.

[0019] A new spatiotemporal scene feature integration layer is added, which converts spatiotemporal data such as transportation mode, altitude change rate, and bump vibration intensity into quantifiable feature information. The feature information corresponding to vehicle transportation is vibration frequency of 5 Hz to 10 Hz and acceleration of 0.5g to 1g. The feature information corresponding to air transportation is altitude change rate of 500 meters to 1000 meters per hour and air pressure change rate of 0.05 ATA to 0.1 ATA per hour. Then, the attention mechanism is used to integrate these feature information into the data fusion process, avoiding the traditional model from ignoring the impact of spatiotemporal scene on physiological state.

[0020] Introducing risk probability distribution analysis into the results output stage not only outputs the current risk level, but also the probability percentage of each risk level in the next 10 minutes, providing a quantitative basis for early intervention of control strategies, and improving the accuracy of risk prediction to over 92%, while the accuracy of traditional models does not exceed 75%.

[0021] Preferred methods for precise titration of oxygen concentration include:

[0022] We developed a customized oxygen metabolism analysis logic for patients with different injuries. By combining alpha wave power, which reflects oxygen supply to brain tissue, and end-tidal carbon dioxide partial pressure, which reflects lung ventilation efficiency, we determined the safe range of oxygen concentration for each patient. For patients with chronic obstructive pulmonary disease, the safe range of oxygen concentration was set at 55% to 65% to avoid carbon dioxide retention. For patients with traumatic brain injury, the safe range of oxygen concentration was set at 45% to 55% to balance oxygen supply to brain tissue and the risk of oxygen toxicity.

[0023] The system employs a dual-regulation cycle logic: The first level of regulation targets blood oxygen saturation and dynamically sets the adjustment range each time based on the environmental sensitivity coefficient. When the patient has a high tolerance to environmental changes, the oxygen concentration is reduced by 8% to 10% each time, and when the tolerance is low, it is reduced by 3% to 4% each time. The second level of regulation targets the alpha wave power of the brain. When the alpha wave power is 20% lower than the baseline level, the first level of regulation is immediately suspended, and the oxygen concentration is increased by 2% to 3% to prevent latent hypoxia in the brain tissue.

[0024] A logic for co-correcting drug and oxygen concentration is introduced. When an external medical device injects a vasoactive drug, the target oxygen concentration is adjusted according to the duration of drug action, which is the drug half-life plus 2 hours. This increases the target oxygen concentration by 5% to 8%, offsetting the drug's impact on microcirculation oxygen supply and avoiding the problem of traditional titration methods neglecting the correlation between medical procedures and oxygen concentration.

[0025] The preferred design of the pressure step-by-step control scheme includes:

[0026] Breaking away from the traditional stepped decompression model with fixed pressure platforms and fixed dwell times, this paper constructs a three-dimensional pressure regulation parameter calculation logic based on the rate of risk change, spatiotemporal scenarios, and individual tolerance. First, a base decompression rate is set according to the rate of deterioration of the risk evolution trajectory; for example, when the shock risk is moderate, the base decompression rate is set to 0.06 ATA per minute. Then, adjustments are made based on the intensity of bumps and vibrations during transport; for each increase in bump and vibration intensity, the decompression rate decreases by 0.01 ATA per minute. Simultaneously, adjustments are made based on the patient's body mass index (BMI); when the BMI is ≥30, the decompression rate is further reduced by 0.02 ATA per minute.

[0027] The system establishes a three-tiered decompression system: an early warning system, a buffer system, and an emergency system. The early warning system is triggered by the end-tidal carbon dioxide partial pressure. When the end-tidal carbon dioxide partial pressure is ≥45 mmHg, a transition buffer platform is set up three pressure gradients in advance. The pressure value of this platform is the average of the pressure values ​​of two adjacent gradients, and the duration of the buffer is twice that of the original pressure platform. The emergency system is triggered by the proportion of delta waves in the brain activity. When the proportion of delta waves is ≥30%, decompression is immediately paused and the current pressure is maintained. Decompression is resumed only when the proportion of delta waves drops below 20%, in order to avoid intracranial pressure fluctuations during the decompression process.

[0028] Develop a decompression effect simulation logic that, based on historical decompression logs and current four types of data, simulates the impact curve of pressure changes on heart rate and blood pressure within the next 5 minutes. If the simulation curve shows a blood pressure fluctuation of ≥15 mmHg, automatically adjust the decompression speed and pressure platform parameters to control the actual blood pressure fluctuation within 10 mmHg, while the blood pressure fluctuation of traditional methods is usually ≥20 mmHg.

[0029] Preferably, the specific architecture of the distributed collaborative architecture includes:

[0030] At the hardware level, a three-channel redundancy design is adopted, consisting of a low-latency communication channel, a high-speed data transmission channel, and a real-time computing channel. The low-latency communication channel is used for communication between devices within the cabin, with communication latency controlled within 30 milliseconds. The high-speed data transmission channel is used for data transmission with external medical devices, with a transmission speed of no less than 50 megabytes per second and communication latency controlled within 100 milliseconds. The real-time computing channel is equipped with dedicated computing components, responsible for real-time data processing and execution of collaborative logic, to avoid equipment collaboration interruption due to the failure of a single channel.

[0031] At the protocol level, a custom medical and environmental coordination protocol is defined, which converts medical operation instructions into correction parameters for environmental control. For example, after the injection of sedative drugs, a correction instruction to increase the target temperature by 1°C to 1.5°C is automatically sent to the temperature control device, and a correction instruction to reduce the decompression rate by 0.01 ATA per minute is sent to the pressure control device, thus avoiding the problem that traditional protocols cannot coordinate medical operations and environmental control.

[0032] At the application level, a dynamic priority arbitration logic is integrated to construct a three-dimensional priority evaluation standard based on risk urgency, physiological impact, and operational complexity. When oxygen concentration regulation and temperature regulation conflict, the risk urgency of oxygen concentration regulation is 9, and the risk urgency of temperature regulation is 6, so oxygen concentration regulation is executed first. If oxygen concentration regulation requires suspending medical equipment parameter coordination, the operational complexity of medical equipment parameter coordination is 8, so the backup oxygen supplementation equipment is activated, and the operational complexity of the backup oxygen supplementation equipment is 3, to ensure the efficiency of regulation execution.

[0033] Preferably, the specific implementation methods of the spatiotemporal scene feature integration layer include:

[0034] Establish a correspondence between transfer scenarios and feature information, and convert different transfer methods, altitude change rates, and bump vibration intensity into quantifiable feature information. For example, the feature information corresponding to vehicle transfer is a vibration frequency of 5 Hz to 10 Hz and an acceleration of 0.5g to 1g, while the feature information corresponding to air transfer is an altitude change rate of 500 meters to 1000 meters per hour and an air pressure change rate of 0.05 ATA to 0.1 ATA per hour.

[0035] The influence weight of different scenarios is introduced. Based on historical data, the influence of different scenarios on vital signs is statistically analyzed. For example, the influence weight of altitude change on blood oxygen saturation during air transport is set to 0.3, and the influence weight of vehicle bumps on heart rate is set to 0.2. This weight is updated in real time during the transport process. When the rate of altitude change increases, the corresponding influence weight increases accordingly.

[0036] By fusing scene feature information with physiological feature information through temporal fusion logic, the lagged correlation between scene changes and physiological state changes is captured. For example, the impact of altitude change on blood oxygen saturation reaches its peak 3 minutes later. This makes the contribution of scene factors to the evaluation results quantifiable, with the error controlled within 5%, avoiding the problem of traditional models ignoring spatiotemporal scenes.

[0037] An adaptive control system for the cabin environment driven by the vital signs of the wounded is provided. The control system includes four types of data acquisition units, a depth map neural fusion unit, an individual dynamic baseline modeling unit, a risk and side effect optimization control engine, a distributed intelligent collaboration unit, an edge and cloud collaborative learning unit, a blockchain secure storage unit, as well as cabin environment control equipment and external medical equipment.

[0038] The four types of data acquisition units include various physiological sensors, dynamic environment sensors, medical device data interfaces, and spatiotemporal scene acquisition components. The various physiological sensors include electrodes for measuring electroencephalogram (EEG) activity and sensors for measuring skin conductance. The dynamic environment sensors include sensors for measuring airflow velocity and sensors for measuring vibration intensity. The medical device data interface can read the operating parameters of medical devices with respiratory support functions and infusion pumps in real time. The spatiotemporal scene acquisition components include positioning devices and altitude sensors. All data are acquired at a frequency of no less than 100 times per second to ensure data synchronization.

[0039] The deep graph neural fusion unit deploys an improved deep graph neural network model, which outputs a three-dimensional assessment result that includes risk evolution trajectory and intervention response prediction.

[0040] The individual dynamic baseline modeling unit generates an individual dynamic baseline based on EEG activity data and the patient's historical health records, and generates an update report every 5 minutes.

[0041] The risk and side effect optimization and control engine, based on the three-dimensional evaluation results, executes oxygen concentration titration logic and pressure control logic to generate an adaptive control strategy with early intervention capabilities.

[0042] The distributed intelligent collaborative unit adopts a three-channel redundant architecture and a medical and environmental collaborative protocol to achieve low-latency collaboration between devices.

[0043] The edge and cloud collaborative learning unit adopts a simplified model training mode at the edge and a full model optimization mode in the cloud. The simplified model with no more than 500,000 parameters is deployed at the edge, and the data processing latency is controlled within 300 milliseconds. The full model with no less than 20 million parameters is deployed in the cloud. The model parameters are updated monthly through joint training with data from multiple edge devices, and the updated key parameters are then distributed to the edge to ensure that the evaluation accuracy of the edge model is maintained above 90% in the long term.

[0044] The blockchain secure evidence storage unit adopts a multi-node verification blockchain architecture to encrypt and store four types of data, adaptive control strategies, and control execution results. Each data block must be verified by at least three nodes before it can be stored, ensuring that the data is tamper-proof and traceable.

[0045] The preferred specific logic of the risk and side effect optimization and control engine includes:

[0046] A dual-objective calculation logic for regulatory effects and side effects was constructed. The extent to which the risk of vital signs of the wounded was reduced was used as the regulatory effect objective, with a weight of 0.6. The degree to which environmental regulation interfered with medical procedures was used as the side effect objective, with a weight of 0.4. The optimal regulatory parameters that balance the two objectives were found through a dual-objective optimization algorithm. For example, when adjusting oxygen concentration, a suitable oxygen concentration value was determined between reducing the risk of oxygen toxicity and avoiding affecting the drug's effect. Reducing the risk of oxygen toxicity corresponds to the regulatory effect, while avoiding affecting the drug's effect corresponds to side effect control.

[0047] An intervention response prediction correction parameter was introduced, and the peak period of drug action was calculated based on the drug half-life in the operation data of external medical devices. The peak period of drug action is 1 to 2 hours after drug injection. During this period, the environmental regulation amplitude was reduced by 40% to 50% to avoid physiological fluctuations caused by the superposition of environmental changes and drug action.

[0048] Set up scenario adaptability verification logic to verify the spatiotemporal scenario adaptability of the generated adaptive control strategy. For example, in the air transport scenario, automatically verify whether the pressure control speed matches the altitude change rate, with the error controlled within 0.02 ATA per minute. If they do not match, re-optimize the parameters to ensure that the strategy is effective in different transport scenarios.

[0049] Preferred improvements to the edge and cloud collaborative learning unit include:

[0050] A hierarchical parameter update logic is adopted, where the edge devices only upload the feature parameters of the intermediate layers of the model, without uploading the original data; the cloud performs full model training based on the feature parameters of multiple edge devices, and only sends out the updated parameters of the key layers of the model after training. The key layers of the model include the parameter weight layer of the deep graph neural network model. This logic reduces the amount of data transmission by more than 80%, avoiding the problem of insufficient network bandwidth in transportation scenarios.

[0051] By introducing scenario transfer learning, the cloud trains scenario-specific sub-models for different transfer scenarios, including vehicle transfer, air transfer, and ship transfer; the edge device calls the corresponding scenario-specific sub-models for evaluation based on real-time spatiotemporal scenario data, which reduces the model's adaptation time in new scenarios from the traditional 2 hours to 10 minutes.

[0052] The built-in model reliability monitoring logic determines the model reliability by calculating the deviation between the model prediction results and the actual vital signs data. The acceptable threshold for the deviation value is ≤8%. When the deviation value exceeds the acceptable threshold three times in a row, the edge model is automatically reset and the cloud parameters are updated urgently to ensure the long-term reliability of the model.

[0053] Preferably, the specific design of the blockchain secure evidence storage unit includes:

[0054] Multiple data verification nodes are set up, including the verification node of the in-cabin system itself, the verification node of external medical equipment, and the verification node of the telemedicine center.

[0055] Each data block requires verification by at least three nodes before it can be stored on the blockchain to prevent a single node from tampering with the data; each data block contains a timestamp of the data generation, the identity signature of the data generating device, and verification information from the telemedicine center to ensure that the data is traceable and cannot be tampered with.

[0056] The system employs a balance logic between authorized querying and privacy protection, using privacy protection technology to ensure that authorized medical personnel can only query data fragments related to the treatment of the injured. These data fragments include control strategies for specific time periods, and complete vital sign data cannot be obtained, thus balancing the needs of data sharing and privacy protection.

[0057] (III) Beneficial Effects

[0058] 1. Establish a four-dimensional data acquisition system comprising physiological, environmental, medical intervention, and spatiotemporal scenarios. This system simultaneously acquires multi-dimensional vital signs including neurological and stress indicators such as EEG and skin conductance; dynamic cabin environment with detailed parameters such as airflow velocity and particulate matter concentration; external medical operations covering drug dosage and equipment operating parameters; and transport scenario data including altitude changes and turbulence intensity. This overcomes the limitations of existing technologies that only collect single environmental or physiological data. During air transport of traumatic brain injury patients, the system can combine the rate of altitude change with EEG alpha wave power to predict blood oxygen saturation 10 minutes in advance. First, it avoids the problem of misjudgment caused by relying solely on current data in traditional methods, and the accuracy of risk prediction is significantly improved. Second, by improving the deep graph neural network model, the parameter weights and data correlation strength are dynamically updated. That is, the weights of key parameters are flexibly adjusted when different injuries are assessed, and the data correlation is corrected as the scenario changes. At the same time, a special control plan is formulated for different injuries and linked with drug metabolism data. This avoids the adaptation bias caused by the static model with fixed parameters in the existing technology, and solves the problem of the disconnect between environmental control and medical operation, so that individualized control meets the physiological needs of the injured.

[0059] 2. A three-channel redundant architecture is adopted, consisting of a low-latency communication channel, a high-speed data transmission channel, and a real-time computing channel. This is coupled with a custom medical and environmental collaboration protocol and dynamic priority arbitration logic to ensure stable collaboration between in-cabin environmental equipment, external medical devices, and remote medical centers. Even in the event of a single-channel failure, communication can be maintained through the backup channel. This addresses the issues of easy interruption and delayed response in existing single-channel technologies for high-risk control needs, ensuring real-time control and reliability. Furthermore, by establishing a monitoring and feedback mechanism and a monthly model iteration optimization mechanism, strategies can be adjusted in real-time based on changes in the patient's physiological state, maintaining a high level of assessment accuracy over the long term. Simultaneously, a multi-node verified blockchain architecture is used to encrypt and store control data, including timestamps, device signatures, and medical verification information. Privacy protection technology allows authorized queries to retrieve only necessary data, thus balancing data sharing and privacy protection requirements. Attached Figure Description

[0060] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0061] Figure 1 This is a diagram illustrating the overall control architecture of an embodiment of the present invention.

[0062] Figure 2 This is a system flowchart of an embodiment of the present invention. Detailed Implementation

[0063] This application provides a method and system for adaptive control of the cabin environment driven by the vital signs of injured personnel. It addresses the five-dimensional deficiencies in existing technologies regarding data dimensions, fusion capabilities, individualized collaboration, reliable execution, and secure evidence storage. It establishes a four-dimensional data acquisition system comprising physiological, environmental, medical intervention, and spatiotemporal scenarios. This system simultaneously acquires multi-dimensional vital signs including neurological and stress indicators such as EEG and skin conductance; the dynamic cabin environment with detailed parameters such as airflow velocity and particulate matter concentration; external medical operations covering drug dosage and equipment operating parameters; and transport scenario data including altitude changes and turbulence intensity. When transporting traumatic brain injury patients by air, the system can combine the rate of altitude change with EEG alpha wave power to predict the risk of blood oxygen saturation 10 minutes in advance, avoiding the misjudgment problems of traditional methods relying solely on current data, thus significantly improving the accuracy of risk prediction.

[0064] In casualty transport scenarios, the compatibility between the environmental parameters of the sealed chamber and the vital signs of the casualty directly determines the safety of the transport. Existing technologies suffer from shortcomings such as single data dimensions, static models, and lack of individualized control, making it difficult to meet the needs of complex injuries. This technical solution focuses on physiological adaptation, dynamic response, collaborative reliability, and safety traceability as its core objectives. Through a two-tiered implementation architecture of methods and systems, it enables precise control of the entire process from data collection to effect cycle. The following sections provide a detailed analysis of the operation steps, parameter settings, core formulas, and equipment selection for each technical solution from the two dimensions of method implementation and system implementation, ensuring that each step is reproducible and verifiable, and covering all the contents described in the claims.

[0065] Example: The technical solution in this application aims to address the five-dimensional deficiencies in existing technologies, namely, data dimension, fusion capability, personalized collaboration, reliable execution, and secure evidence storage. The overall approach is as follows:

[0066] To address the problems existing in the prior art, this invention provides a patient vital signs-driven adaptive control system for the cabin environment, the specific details of which are as follows:

[0067] 1. A four-dimensional data acquisition system collects physiological data, environmental data, medical intervention data, and spatiotemporal scene data, addressing the issue of missing data dimensions in existing technologies. Details are as follows:

[0068] 1.1 Multidimensional vital sign data collection:

[0069] Multiple types of sensors were used to capture the physiological state of the injured, balancing signal stability and patient comfort. Specifically, electroencephalogram (EEG) data was acquired using four silver chloride EEG electrodes deployed at the FP1 and FP2 points in the forehead and the T3 and T4 points in the temporal lobe. Medical conductive gel was used for fixation to prevent poor contact due to vibration. Interference was removed through 50Hz power frequency filtering and 0.5 to 30Hz bandpass filtering. Alpha wave power and delta wave percentage were extracted in real time. Alpha wave power reflects brain tissue oxygen supply, with a normal range of 50 to 70 μV, while delta wave percentage reflects brain hypoxia, with a normal range of ≤20%.

[0070] The circulation and oxygenation data were recorded using a wrist-worn non-invasive blood pressure module and a finger clip pulse oximeter, recording systolic blood pressure, diastolic blood pressure, mean arterial pressure and blood oxygen saturation, and aligned by timestamps to form a circulation and oxygenation related dataset; while the metabolism and ventilation data were acquired through a body surface-attached temperature sensor and a nasal cannula end-tidal carbon dioxide sensor, the former deployed in the armpit or neck, and the latter used end-tidal carbon dioxide partial pressure to determine lung ventilation efficiency.

[0071] Stress data is collected by a wrist-worn skin conductance sensor, reflecting the stress level. The normal range is 0.2 to 1.0 μS, and the higher the value, the lower the tolerance. All physiological sensors are aggregated to the data acquisition box via a bus. Timestamp synchronization technology is used to avoid data misalignment. The cable is fixed inside the injured person's clothing to prevent it from being pulled off.

[0072] 1.2 In-cabin dynamic environment data acquisition:

[0073] A three-point distribution and multi-parameter synchronization scheme is adopted to ensure data coverage of different areas inside the cabin. Temperature and humidity are collected by three SHT35 sensors deployed at three points: the head, torso, and feet of the cabin, with a spacing of at least 1.5m. The average value of the three points is taken every 10 seconds as the overall temperature and humidity inside the cabin. The average value is calculated using the logic T. avg = (T1+T2+T3) / 3, H avg = (H1+H2+H3) / 3, where T1, T2, and T3 are the temperatures of the head, torso, and feet, respectively, in °C, and H1, H2, and H3 are the humidity of the corresponding areas, in %RH;

[0074] Gas parameters are collected by an NDIR sensor, which is deployed in the middle of the chamber, 1.2m above the ground, at the same breathing height as the injured person, to avoid local concentration deviations caused by proximity to the air outlet; an absolute pressure sensor dedicated to the pressurization chamber records changes in pressure inside the chamber in real time.

[0075] Airflow status is collected by an airflow velocity sensor, which is deployed below the air outlet in the middle of the cabin to ensure that the airflow velocity is ≥0.5m / s to avoid local air stagnation. These environmental sensors need to be calibrated regularly. Temperature and humidity are compared monthly with a standard constant temperature and humidity chamber, and gas parameters are calibrated quarterly with standard gas, i.e., using O2 with a concentration of 50%.

[0076] 1.3 Medical Intervention and Spatiotemporal Data Acquisition:

[0077] Medical intervention data is read through an interface from devices including ventilators, infusion pumps, and portable monitors. It supports mainstream medical device protocols such as HL7 and DICOM, and obtains drug injection and ventilator parameters. Drug injection includes name, dosage, and rate; ventilator parameters include inhaled oxygen concentration (FiO2), tidal volume (Vt), and positive end-expiratory pressure (PEEP). When an operation is triggered, the operation command and timestamp are immediately sent to the acquisition system to ensure synchronization with physiological data.

[0078] Spatiotemporal scene data is collected by a small module integrating a GPS module, an altitude sensor, and a vibration sensor. It is deployed on the top of the cabin to avoid obstruction. The transportation method is determined by the GPS movement trajectory. When transported by vehicle, the movement is a zigzag line. When transported by air, the movement is a straight line with a sudden increase in altitude. The rate of altitude change and the intensity of bumps and vibrations are recorded simultaneously and aligned with the timestamps of physiological data to form a scene-physiological correlation dataset. The medical device interface needs to complete compatibility testing with Mindray SV300 ventilator and Smith Fusion2 infusion pump in advance. The spatiotemporal module needs to complete GPS positioning initialization before transportation to avoid loss of initial signal.

[0079] 2. Improved deep graph neural network fusion evaluation: The improved deep graph neural network can perform dynamic data association and risk prediction, overcoming the static limitations of traditional models. Details are as follows:

[0080] 2.1 Dual Dynamic Update Logic:

[0081] The system dynamically adjusts data weights and correlation strength based on assessment goals and scenario parameters, mainly in two steps. The first step involves parameter importance weighting, which is based on the sensitivity of the injury to physiological indicators. Initial weights are determined through expert scoring and historical data verification, and are dynamically adjusted according to the assessment goals. For example, when assessing the risk of traumatic brain injury, the weight of EEG parameters increases from a baseline of 50% to 70%, and the weight of blood pressure parameters increases from 40% to 60%. When assessing the risk of respiratory failure, the weight of end-tidal carbon dioxide parameters increases from 30% to 65%, and the weight of blood oxygen saturation increases from 50% to 60%. A mapping table between assessment goals and weights is established, and the system automatically calls the corresponding weights based on the patient's injury, with verification every 5 minutes.

[0082] The second step is to determine the correlation strength of the parameters by fusing the Pearson correlation coefficient and mutual information entropy. The correlation coefficient represents a linear correlation, while the mutual information entropy represents a non-linear correlation. R0 total =0.6×R Pearson +0.4×(1-H / H max ), where R total R represents the final association strength, ranging from 0 to 1. Pearson The Pearson correlation coefficient reflects a linear relationship, ranging from -1 to 1, and its absolute value is used in the calculation. H is the mutual information entropy between the two parameters, reflecting a non-linear relationship, measured in bits. max To determine the maximum mutual information entropy for this type of parameter combination, based on historical data statistics, such as the Hi of oxygen concentration and blood oxygen saturation. max =1.0 bit; For example, in a sea-level scenario, the baseline correlation strength between oxygen concentration and blood oxygen saturation is 0.3, and for every 1000 meters increase in altitude, R... Pearson As the concentration increases from 0.7 to 0.9, H decreases from 0.5 to 0.3. Substituting this into the formula R... total=0.6×0.9+0.4×(1-0.3 / 1.0)=0.54+0.28=0.82. The correlation strength is calculated every minute. When the change exceeds 20%, it is marked as a high-sensitivity parameter and given priority in subsequent regulation. The weight mapping table is trained and validated using data from 100+ injured patients. The correlation strength calculation is pre-set with H. max That is, the data of 100+ injured persons includes 50 cases of traumatic brain injury, 30 cases of COPD, and 20 normal controls.

[0083] 2.2 Spatiotemporal Scene Feature Integration Layer:

[0084] Unstructured scene data is transformed into quantifiable feature vectors and incorporated into the fusion model. First, a table mapping scene types to feature vectors is established: vehicle-mounted transport corresponds to vibration frequencies of 5 to 10 Hz, accelerations of 0.3 to 0.5 g, and altitude change rates ≤ 100 m / h; air transport corresponds to vibration frequencies of 2 to 5 Hz, accelerations of 0.1 to 0.3 g, altitude change rates of 500 to 1000 m / h, and air pressure change rates of 0.05 to 0.1 ATA / h; ship transport corresponds to vibration frequencies of 1 to 3 Hz, accelerations of 0.2 to 0.4 g, and altitude change rates of 0.05 to 0.1 ATA / h. The rate of change in pressure is 0; the transportation method is automatically identified through GPS trajectory and altitude data, and the corresponding feature vector is called; and the scene influence weight is based on historical data statistics. The influence weight of air transport altitude change on blood oxygen saturation is 0.3, that is, for every 1000-meter change in altitude, the fluctuation of blood oxygen saturation by 3% is directly caused by altitude; while the influence weight of vehicle bumps on heart rate is 0.2, that is, for every 0.1g increase in bump intensity, the heart rate increases by 2 beats / minute, and it is updated in real time with scene parameters. When the rate of change in altitude increases from 500 meters / hour to 800 meters, the influence weight increases from 0.3 to 0.4.

[0085] The gated recurrent unit (GRU) captures the hysteresis correlation between scene and physiological data, and the correlation strength is calculated. Where C is the correlation coefficient between scene and physiology, ranging from 0 to 1, n is the number of data samples, and S i Let P be the scene feature vector at time i. i+τ Let be the physiological feature vector at time i+τ, where τ is the lag time, such as 3 minutes, and Sim(·) is the cosine similarity function. For example, the effect of altitude change on blood oxygen saturation reaches its peak 3 minutes later. At this time, C=0.4 means that 40% of the blood oxygen change is caused by altitude. The model is trained with 100+ transport data to ensure that the contribution error of scene factors to the evaluation results is ≤5%. The scene feature vector and the physiological feature vector need to be standardized to the range of 0 to 1. The model is deployed in a lightweight manner on the edge (Raspberry Pi 4B) with a parameter scale of ≤500,000.

[0086] 2.3 Generation of 3D Evaluation Results:

[0087] Based on an improved DGNN, the output includes individual dynamic baselines, risk evolution trajectories, and intervention response predictions. Details are as follows:

[0088] An individual dynamic baseline, serving as a dynamic reference value to closely reflect the current physiological state of the injured person, is established by fusing real-time physiological data and historical health record data using a Kalman filter algorithm. t =0.7×X avg +0.3×B t-1 Among them, B t The individual's dynamic baseline at time t, such as heart rate and blood pressure, X avg B is the average of real-time physiological data over the last 5 minutes. t-1 This represents the baseline value at time t-1 (the first 5 minutes); for example, a traumatic brain injury patient's historical baseline heart rate is 70 beats / min, and their current heart rate is 75 beats / min, which is the 5-minute average. Substituting this into the formula yields B. t =0.7×75+0.3×70=52.5+21=73.5≈74 times / minute, updated every 5 minutes, excluding temporary interference, such as a brief increase in blood pressure caused by turbulence;

[0089] Risk evolution trajectory, predicting risk trends through the rate and acceleration of change in physiological data, v=(X t -X t-Δt ) / Δt,a=(v t -v t-Δt ) / Δt, where v is the rate of change of a physiological indicator, such as the rate of change of heart rate, in beats per minute. 2 'a' represents the changing acceleration, in units of accelerations per minute. 3 X t Let X be the physiological index value at time t. t-Δt v is the index value before Δt. t v is the velocity at time t. t-Δt This is the rate before Δt; for example, if the heart rate increases from 74 beats / min to 79 beats / min, that is, within 5 minutes, the rate v = (79-74) / 5 = 1 beat / min. 2 The rate increases to 85 beats per minute in the next 5 minutes, with a rate v = (85-79) / 5 = 1.2 beats per minute. 2 The acceleration a = (1.2 - 1) / 5 = 0.04 times / minute 3 The acceleration is positive and the speed is >0.8 cycles / min. 2 The time markers indicate a worsening risk trend, such as brain hypoxia progressing from mild to moderate.

[0090] Intervention response prediction involves forecasting the duration and magnitude of the impact of medical procedures on physiological indicators, based on pharmacokinetic data and the formula for calculating the duration of impact, T. impact =T 1 / 2 +2, where T impact Duration of drug effect, in hours, T1 / 2 This refers to the drug's half-life, expressed in hours; for example, if norepinephrine with a half-life of 2 hours is injected, substituting this value gives T. impact =2+2=4 hours, the impact is a 10% to 15% reduction in microcirculation oxygen supply, which needs to be offset by a 5% to 8% increase in oxygen concentration. Output the table of intervention type, impact duration and physiological fluctuation range.

[0091] 3. Individualized synergistic regulation strategy: This strategy optimizes two objectives based on the three-dimensional evaluation results. Details are as follows:

[0092] 3.1 Precise titration of oxygen concentration: Individualized safety ranges are established for different injuries, and precise titration is performed through dual circulation, combined with dynamic correction by drug intervention.

[0093] Individualized safety ranges are determined by combining EEG alpha wave power and end-tidal carbon dioxide partial pressure. For COPD patients, the respiratory center's sensitivity to carbon dioxide is reduced, and high oxygen concentrations inhibit respiratory drive. The safety range is set at 28% to 35%, which is 70% of the traditional uniform range of 40% to 60%, with end-tidal carbon dioxide controlled at ≤45 mmHg. For traumatic brain injury patients, ensuring oxygen supply to brain tissue requires a safety range of 45% to 55%, with EEG alpha wave power maintained at ≥45 μV. This is verified through blood gas analysis every 30 minutes, ensuring arterial oxygen partial pressure is maintained at 80 to 100 mmHg (COPD) and 90 to 110 mmHg (traumatic brain injury).

[0094] The dual-regulation cycle comprises an inner and an outer loop. The inner loop regulates blood oxygen saturation, while the outer loop regulates alpha wave power. The inner loop targets blood oxygen saturation of 92% to 94% for COPD patients and 95% to 97% for traumatic brain injury patients. The regulation amplitude is dynamically calculated based on the environmental sensitivity coefficient k, where k is used to calculate skin conductance activity, ranging from 0.1 to 1.0. ΔFiO2 = 8% to 5% × (k - 0.1) / 0.9. When k = 0.1, the tolerance is highest, with ΔFiO2 = 8% to 5% × 0 = 8%; when k = 1.0... For the lowest tolerance, ΔFiO2 = 8% to 5% × 1 = 3%, consistent with the logic of lowering by 8% to 10% when tolerance is high and by 3% to 4% when tolerance is low. After adjustment, monitor for 30 to 60 minutes to ensure stable blood oxygenation, i.e., 60 minutes for COPD and 45 minutes for traumatic brain injury. The outer loop is triggered when the alpha wave power is 20% lower than the baseline. For example, if the baseline is 50 μV and the current is 40 μV, immediately pause the inner loop adjustment and increase the oxygen concentration by 2% to 3%, such as from 50% to 52%. Wait until it recovers to above 90% of the baseline, such as 45 μV, before restarting the inner loop.

[0095] Drug and oxygen concentration are synergistically corrected. For drugs that affect oxygen supply, when injecting norepinephrine, the drug's duration of action (T) is considered. impactThe formula for increasing the target oxygen concentration value is as follows: =5%+3%×(T) impact -2) / 2, when T impact =2 hours, which is a short half-life drug. =5%; when T impact =4 hours, which is a long half-life drug. =5%+3%×1=8%, consistent with the adjustment logic of 5% to 8%; when injecting sedative drugs, increase by 3% to 5%, while reducing the adjustment range from 8% to 5%; oxygen concentration is adjusted through an oxygen concentrator and flow valve, such as a turtle oxygen concentrator with a flow rate of 0 to 10 L / min, and setting an emergency pause threshold, such as pausing immediately when blood oxygen is 2% below the target lower limit.

[0096] 3.2 Pressure is adjusted in a stepped manner, and a three-dimensional parameter model of risk, scenario and individual is constructed. The pressure reduction safety is ensured through three-level nodes and effect simulation.

[0097] For three-dimensional decompression parameter calculation, first set the basic decompression rate v. base The risk of shock was 0.07 ATA / min for mild, 0.06 ATA / min for moderate, and 0.05 ATA / min for severe. After scenario and individual adjustments, the scenario adjustment was that for every 0.1g increase in vibration intensity A, the speed decreased by 0.01 ATA / min; the individual adjustment was that for a BMI ≥ 30, the speed decreased by an additional 0.02 ATA / min. For example, moderate risk of shock (v) base =0.06ATA / min + vibration A = 0.3g + BMI = 28, no correction, substituting gives v final =0.06-0.01×0.3 / 0.1-0=0.06-0.03=0.03ATA / minute;

[0098] The three-tiered decompression system includes early warning nodes, emergency nodes, and normal nodes; details are as follows:

[0099] The warning node is the end-tidal carbon dioxide partial pressure. Triggered when pressure is ≥45mmHg, a transition buffer plateau is set 3 pressure gradients in advance, with plateau pressure P buffer =(P current +P next ) / 2, where P current For the current pressure, such as 0.18 ATA, P next For the next gradient pressure, such as 0.15 ATA, substitute it to get P. buffer =(0.18+0.15) / 2=0.165ATA, duration of stay = original platform duration × 2 (originally 5 minutes, now 10 minutes); emergency node in EEG delta wave proportion R δTriggered when ≥30%, immediately pause decompression, maintain current pressure, and simultaneously increase oxygen concentration by 3% to 5%, waiting for R... δ Reduce pressure to below 20%, then continue decompression at 50% of the original speed; normal nodes should press V when there are no warnings / emergency triggers. final Decompression: hold for 3 minutes for every 0.02 ATA decrease.

[0100] Stress reduction effect simulation: Based on a stress and physiological prediction model trained from historical stress reduction logs, the blood pressure fluctuation formula is simulated as ΔBP. pred =k BP ×ΔP× , where ΔBP pred To simulate blood pressure fluctuations, the unit is mmHg, kJ. BP The correlation coefficient between blood pressure and stress is based on historical data. ΔP represents the pressure change over 5 minutes, in ATA, and Δt represents the rehearsal time. For example, ΔP = 0.015 ATA, which gives ΔBP. pred =0.8×0.015× The pressure reading is approximately 0.054 mmHg. In actual scenarios, it needs to be corrected based on heart rate and BMI. If the blood pressure fluctuation during the simulation is ≥15 mmHg, the decompression rate will be automatically reduced by 0.005 ATA / minute, and the simulation will be repeated until the fluctuation is ≤10 mmHg. Pressure regulation is achieved through a pressurization pump and a safety valve. The three-level nodes are linked to the cabin's audible and visual alarms. The yellow light flashes at the warning node, and the red light flashes at the emergency node.

[0101] Temperature, humidity, and physiological state are synergistically regulated, along with carbon dioxide control. Temperature regulation is based on the depth of sleep: a delta wave percentage (≥20%) indicates deep sleep, 15% to 20% indicates light sleep, and <15% indicates wakefulness. During deep sleep, the temperature is increased by 1°C to 1.5°C (e.g., from 24°C to 25.5°C); during light sleep, it is increased by 0.5°C to 1°C; and during wakefulness, it is maintained at 23°C to 24°C. Humidity regulation is based on skin conductance activity (≥0.8 μS), indicating a stress state. During stress, humidity is increased by 5% to 10% (e.g., from 50% to 60%); during non-stress, it is maintained at 45% to 55%. Carbon dioxide is controlled at a specific concentration. When the concentration is ≥0.5%, start the ventilation device, such as a centrifugal fan with an air volume of 100m³. 3 / h, ventilation duration formula T vent =10×( -0.4), where T vent Ventilation time is measured in minutes. Substituting the current carbon dioxide concentration into the equation, we get T. vent =10×(0.6-0.4)=20 minutes; When the concentration is ≥0.8%, the sodium lime adsorbent is activated simultaneously. The particle diameter is 3 to 5 mm, and the loading amount is 500 g. The adsorbent failure is determined by the rate of concentration change. <0.1% / minute, for example, it took 5 minutes to decrease from 0.8% to 0.75%. =0.01% / minute, after failure, extend the ventilation time by 50%, such as 20 to 30 minutes.

[0102] 4. Distributed intelligent collaborative execution: Distributed intelligent collaborative execution achieves reliable collaboration through three-channel redundancy and dynamic arbitration. Details are as follows:

[0103] 4.1 A three-channel redundant hardware architecture is adopted, featuring low-latency intra-cabin communication, high-speed medical interaction, and real-time computing fault detection channels to avoid single-channel failure. The low-latency communication channel connects to intra-cabin devices via a CAN bus controller, using the CANopen protocol. The data frame format is device ID (1 byte) + instruction type (1 byte) + parameter value (2 bytes) + checksum (1 byte), and low latency is ensured through bus load rate monitoring. The high-speed data transmission channel connects medical devices to the remote center via a 5G RedCap module, compatible with Bluetooth 5.0 backup channels, and automatically switches when the 5G signal is weak. The real-time computing channel deploys an FPGA chip, sending heartbeat frames every 10ms to detect faults. If there are three consecutive no responses, it switches to the backup channel within 10ms. In case of CAN bus failure, RS485 is enabled, fault logs are recorded, and medical device instructions are parsed and converted into environmental control instructions. The three channels need to undergo fault injection testing to ensure that they can still maintain coordination after a fault. The 5G module is pre-enabled with dedicated data traffic to avoid congestion.

[0104] 4.2 Medical and Environmental Collaboration Protocol: A custom protocol converts medical operation instructions into environmental control and correction parameters. The protocol frame structure is: Medical Device ID (1 byte) + Operation Type (1 byte) + Operation Parameter (2 bytes) + Environmental Correction Parameter (4 bytes) + Checksum (1 byte). For example, when a ventilator executes an FiO2 reduction from 60% to 55% (Operation Type 0x01, Parameter 0x0037), the environmental correction parameters are: oxygen concentration simultaneously reduced to 55% (0x0037) + pressure control rate maintained at 0.03 ATA / min (0x0003). Typical collaborative scenarios include: when administering sedatives, the temperature control device increases the target temperature by 1°C to 1.5°C, and the pressure control device reduces the decompression rate by 0.01 ATA / min; when the ventilator increases tidal volume, the oxygen concentration increases by 2% to 3%, and the ventilation device duration is extended by 10%. The protocol requires joint debugging with mainstream medical device manufacturers, and the correction parameters have passed clinical validation in 10+ sedative injection scenarios.

[0105] 4.3 Dynamic priority arbitration is used to construct a three-dimensional evaluation standard based on risk urgency, physiological impact, and operational complexity. Risk urgency is assessed as follows: oxygen concentration control 9 points, pressure control 8 points, temperature control 6 points, humidity control 5 points, and carbon dioxide control 7 points. Physiological impact is assessed as follows: oxygen concentration control 8 points, pressure control 7 points, and temperature control 5 points. Operational complexity is assessed as follows: oxygen concentration control 5 points, pressure control 8 points, and backup oxygen replenishment 3 points. The arbitration score is S = 0.4 × S. risk +0.3×S impact +0.3×(10-S complex ), where S is the total score, S risk To assess the urgency of the risk, S impact For physiological impact, S complex For operational complexity, 10-S complex For ease of operation; for example, oxygen concentration control S risk =9,S impact =8,S complex =5, substituting, we get S=0.4×9+0.3×8+0.3×(10-5)=3.6+2.4+1.5=7.5, temperature control S risk =6,S impact =5,S complex =4, score S = 0.4×6 + 0.3×5 + 0.3×6 = 2.4 + 1.5 + 1.8 = 5.7, oxygen concentration control should be prioritized; the ventilator needs to be paused while the backup oxygen supplementation device S is activated. complex =3, oxygen concentration control score S=0.4×9+0.3×8+0.3×7=3.6+2.4+2.1=8.1, which is still higher than temperature control; the arbitration result is displayed on the 14-inch touch screen in the cabin and synchronized to the remote center. The priority standard is reviewed by experts, including 3 emergency medicine experts and 2 control engineering experts. The arbitration logic is solidified in FPGA.

[0106] 5. A loop optimization mechanism ensures continuous strategy adaptation through real-time monitoring, performance feedback, and model iteration. The specific process is as follows:

[0107] Real-time monitoring and matching degree calculation: Environmental and physiological matching degree is calculated every 2 minutes, M=N s / N total ×100%, where M is the matching degree, in % and N is the percentage. s To stabilize physiological indicator numbers, N total This refers to the total number of physiological indicators, such as 7 items: heart rate, blood pressure, blood oxygen, body temperature, end-tidal carbon dioxide, EEG alpha waves, and skin conductance; for example, 6 out of the 7 indicators are stable N. s=6, substituting into the equation, we get M=6 / 7×100%≈85.7%; the stability indicators are defined as blood oxygen saturation fluctuation ≤2%, blood pressure fluctuation ≤10mmHg, heart rate fluctuation ≤5 beats / min, and EEG alpha wave power fluctuation ≤10μV.

[0108] In dual-response mode, when the matching degree is 80% to 90%, a strategy fine-tuning is triggered, such as reducing the oxygen concentration adjustment range from 8% to 5% and extending the monitoring time to 45 minutes; when the matching degree is <80% or the risk level is upgraded, an emergency response is triggered, the control end shortens the data acquisition interval to 1 / 4 of the original frequency, such as 1 minute to 15 seconds, the backup environment equipment is activated, and the consultation end synchronizes data to the remote center via 5G to receive expert correction instructions.

[0109] The model is iterated and optimized monthly. Based on the monthly transport and regulation log, the DGNN weight mapping table, regulation strategy parameters, and drug and physiological effect database are optimized. The accuracy of risk prediction is ensured through testing in 10+ new scenarios.

[0110] 6. Implementation: Key implementation points include: sensor deployment should avoid proximity to metal and heat sources to ensure data representativeness; regular calibration and recording of data stored on the blockchain, with temperature and humidity tested monthly and gas parameters quarterly; equipment compatibility testing covering major medical equipment manufacturers, and the reliability of backup channel switching verified through extreme scenario testing; the technical solution must undergo ethical approval and be validated in 100+ real transport scenarios, with model iteration combined with clinical feedback to avoid incompatibility caused by pure algorithm optimization; the system uses dual power supplies, and the pressurized chamber is equipped with overpressure and hypoxia protection.

[0111] Effectiveness verification was conducted using scenario parameters including air transport, pressurized hard cabin, traumatic brain injury patients, ventilators, and infusion pumps. Data acquisition: all four data types were collected synchronously without loss, with timestamp deviations ≤1ms, and average temperature and humidity values ​​T. avg =24℃, H avg =50%, oxygen concentration 60%.

[0112] Fusion assessment, individual dynamic baseline heart rate B t =0.7×75+0.3×70=73.5≈74 beats / min, blood pressure B t =0.7×130+0.3×120=127≈125mmHg; Risk evolution trajectory heart rate v=(75-70) / 5=1 beat / min 2 Acceleration a=0, no worsening trend; intervention response prediction mannitol effect duration T impact =6 hours, half-life 4 hours; strategy generation, oxygen concentration safe range 45% to 55%, adjustment range ΔFiO2 = 8% to 5% × (0.6 - 0.1) / 0.9 ≈ 5%; decompression rate v final=0.06-0.01×0.3 / 0.1=0.03ATA / minute; Temperature increased to 25℃, humidity 55%; Collaborative execution: 5G channel response delay 82ms, CAN bus transmission normal, arbitration score oxygen concentration control 7.5 points > temperature control 5.7 points, priority execution; Cyclic optimization: every 2 minutes monitor matching degree 85% to 92%, blood oxygen saturation 94% to 96%, blood pressure fluctuation ≤8mmHg, EEG alpha wave power 50 to 55μV, no emergency response trigger.

[0113] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for adaptive control of the cabin environment driven by the vital signs of wounded soldiers, characterized in that, This control method includes the following steps: Acquire multi-dimensional vital sign data of the injured, dynamic environmental data inside the cabin, operation data of external medical equipment, and spatiotemporal scenario data of the transfer; the operation data of medical equipment includes parameters of drug injection, infusion and respiratory support, and the spatiotemporal scenario data includes the transfer method, altitude changes and turbulence intensity. Based on an improved deep graph neural network model, a dynamic fusion model for four types of data is constructed. Features are extracted and correlation analysis is performed to generate a three-dimensional assessment result containing individual dynamic baselines, risk evolution trajectories, and intervention response predictions. The risk trajectory reflects risk trends, and the intervention predictions combine drug metabolism data to predict the duration of the impact of medical procedures. By combining the three-dimensional evaluation results with the dual-objective optimization logic, an adaptive control strategy with functions of early intervention, dynamic correction, and side effect suppression is generated. The strategy includes oxygen concentration titration, pressure stepwise regulation, temperature and humidity coordinated regulation, and carbon dioxide cycle control schemes, with each scheme embedding an environmental sensitivity coefficient. The system utilizes a distributed collaborative architecture to coordinate the execution of equipment within the cabin, medical devices, and remote centers. It also employs a dynamic priority and conflict arbitration mechanism to prioritize ensuring oxygen supply to brain tissue in the event of a conflict, and then proceed according to the rate of risk deterioration. The system calculates the matching degree between the regulation and vital signs. When the matching degree is lower than the threshold or the risk escalates, it triggers a dual-response mode, shortens the data collection interval, synchronizes the data to the remote center, and receives feedback correction instructions.

2. The method according to claim 1, characterized in that, The improvements to the improved deep graph neural network model include: Construct a dual dynamic update logic for parameter importance weights and parameter correlation strength; adjust parameter weights according to data importance, calculate parameter correlation strength by statistical correlation coefficients and information difference, and dynamically correct correlation strength as the environment changes; Transform the spatiotemporal data related to transportation into quantifiable feature information and incorporate it into the fusion process through an attention mechanism; The output stage incorporates risk probability distribution analysis, outputting the current risk level and the probability percentage of each level in the future.

3. The method according to claim 2, characterized in that, Transforming transport-related spatiotemporal data into quantifiable feature information and incorporating it into the fusion process through an attention mechanism, specifically including: Transform transportation methods, altitude changes, and bump intensity into quantifiable feature information to determine the range of feature parameters corresponding to different scenarios; The impact of different scenarios on vital signs is statistically analyzed based on historical data. The weights are updated in real time during the transfer process, and the corresponding weights are increased synchronously when the rate of environmental change accelerates. By integrating scene and physiological feature information through temporal fusion logic, the lagged correlation between scene changes and physiological state is captured, making the contribution of the scene to the evaluation results quantifiable.

4. The method according to claim 1, characterized in that, The specific procedures for oxygen concentration titration include: A unique oxygen metabolism analysis logic is constructed for different injuries, and the safe range of oxygen concentration is determined by combining the power of EEG alpha waves and the partial pressure of carbon dioxide at the end of expiration. It employs a dual-regulation cycle logic: the first level targets blood oxygenation and sets the regulation amplitude according to the environmental sensitivity coefficient; the second level targets EEG alpha wave power, pausing the first level of regulation and increasing oxygen concentration when alpha wave power is 20% below the baseline. When injecting vasoactive drugs, increase the target oxygen concentration by 5% to 8% according to the duration of drug action.

5. The method according to claim 1, characterized in that, The specific design of the pressure step control scheme includes: Construct a three-dimensional pressure regulation parameter calculation logic based on the risk change rate, spatiotemporal scenario, and individual tolerance; set a basic decompression speed according to the risk deterioration rate, and dynamically correct it by combining the turbulence intensity and the patient's body mass index. The system sets up three levels of decompression nodes: early warning, buffer, and emergency. The early warning node is triggered by the end-tidal carbon dioxide partial pressure and a transition buffer platform is set up. The emergency node is triggered by the proportion of delta waves on the EEG and decompression is suspended until the proportion of delta waves drops to a safe range. Based on historical logs and current data, the system simulates the impact of stress changes on heart rate and blood pressure. When the simulation shows that blood pressure fluctuations exceed the blood pressure threshold, the system automatically adjusts parameters to control the actual fluctuations within the preset blood pressure range.

6. The method according to claim 1, characterized in that, The specific architecture of the distributed collaborative architecture includes: The design employs a three-channel redundancy system: a low-latency channel for communication between in-cabin devices, a high-speed channel for data transmission from medical devices, and a real-time computing channel with dedicated components for processing data and collaborative logic. Medical and environmental collaboration involves converting medical operation instructions into environmental control and correction parameters to coordinate medical operations with environmental control. Integrate dynamic priority arbitration logic, construct a three-dimensional priority evaluation standard, and execute according to risk urgency in case of conflict. When it is necessary to suspend high-complexity operations, start a low-complexity backup plan.

7. A cabin environment adaptive control system driven by the vital signs of wounded soldiers, characterized in that, The control system includes four types of data acquisition units, a depth map neural fusion unit, an individual dynamic baseline modeling unit, a risk and side effect optimization and control engine, a distributed intelligent collaboration unit, an edge and cloud collaborative learning unit, a blockchain secure storage unit, and an environmental control device and external medical device as described in any one of claims 1-6. The four types of data acquisition units include multiple types of physiological sensors, dynamic environmental sensors, medical device data interfaces, and spatiotemporal scene acquisition components. The depth graph neural fusion unit deploys an improved depth graph neural network model and outputs 3D evaluation results; the individual dynamic baseline modeling unit generates an update report. The control engine executes strategies for oxygen concentration titration and pressure control logic generation. The distributed coordination unit adopts a three-channel redundant architecture and a coordination protocol for low-latency coordination. The edge and cloud units ensure evaluation accuracy by simplifying model training and optimizing the complete model; the blockchain unit uses a multi-node verification architecture for encrypted storage.

8. The system according to claim 7, characterized in that, The specific logic of the risk and side effect optimization and control engine includes: The optimal control parameters are found by using a dual-objective optimization algorithm, with the reduction in vital sign risk as the control effect target and the degree of interference of environmental control on medical operations as the side effect target. By introducing intervention response prediction correction parameters, the peak period of drug action is calculated based on the drug half-life, during which the environmental control intensity is reduced by 40% to 50%. The spatiotemporal adaptability of the control strategy is verified, and the matching degree between the pressure control speed and the altitude change rate is verified. When the error exceeds the limit, the parameters are re-optimized.

9. The system according to claim 7, characterized in that, Specific improvements to the edge and cloud collaborative learning unit include: The model's intermediate layer feature parameters are uploaded at the edge, and the cloud trains the complete model based on the parameters from multiple edge devices, then sends down the updated parameters for the key layers. By introducing scenario transfer learning, dedicated sub-models are trained in the cloud for different transit scenarios, and the corresponding sub-models are called for evaluation at the edge according to the real-time scenario. The system calculates the deviation between the predicted results and the actual vital signs, and triggers a model reset and emergency parameter update when the deviation exceeds the threshold three times consecutively.

10. The system according to claim 7, characterized in that, The specific design of the blockchain secure evidence storage unit includes: Set up multiple data verification nodes, including in-cabin systems, medical equipment and remote center nodes. Each data block must be verified by at least three nodes before it can be stored. Each data block contains a timestamp, device identity signature, and remote center verification information; Privacy protections ensure that authorized personnel can only access data snippets related to diagnosis and treatment.

Images