Intelligent temperature control system for flight simulator cabin

Abstract

The invention belongs to the technical field of flight simulation training equipment, particularly relates to an intelligent temperature control system for a flight simulator cabin, and solves the problem that the requirements of flight simulator temperature control on multi-dimensional physiological perception fusion, multi-target collaborative optimization and rapid personalized adaptation and active safety protection of new pilots cannot be met in the prior art. According to the method, a multi-modal sensing layer processes multi-channel skin physiological data, physical environment data and pilot electrocardiosignals, a human body heat balance equation for blood flow correction is constructed, and physiological feature vectors including predicted core body temperature, predicted average votes and predicted dissatisfaction percentage are generated; the decision-making layer generates a temperature control instruction and a confidence evaluation value through a meta-learning model, a reward function is in a four-dimensional dynamic weighting form, and the weight is adjusted through an IF-THEN rule according to the state; and the dynamic execution layer adopts a PID closed loop to adjust the partition temperature and triggers response in a grading manner. Accurate and personalized temperature control is achieved, and the training quality and safety are guaranteed.

Description

Technical Field

[0001] This invention belongs to the technical field of flight simulation training equipment, and specifically relates to an intelligent temperature control system for a flight simulator cockpit. Background Technology

[0002] As a key piece of equipment for training pilots' core skills, the cabin temperature environment of flight simulators plays a decisive role in long-term, high-intensity simulated training—directly affecting pilots' physiological homeostasis, cognitive load levels, and operational performance. Therefore, building an efficient and comfortable temperature control system adapted to flight training scenarios has become a core requirement for ensuring training quality and pilots' physiological safety.

[0003] In existing technologies, mainstream solutions (such as those disclosed in Chinese patent CN119674344A) generally employ coarse-grained control logic based on fixed thresholds. These systems rely on monitoring data from a single or limited number of environmental sensors in the cockpit (such as temperature and humidity sensors), triggering the air conditioning system to start or stop when the values ​​exceed preset thresholds. Their core flaw lies in the limitation of their decision-making dimensions: they completely ignore individual physiological differences among pilots and dynamic physiological changes under different training tasks, failing to achieve the precise goal of "on-demand control." In practical applications, problems such as response lag and energy waste due to over-control often occur.

[0004] Furthermore, although some in-vehicle intelligent temperature control solutions (such as CN119396221A) incorporate intelligent algorithms such as reinforcement learning, their model architecture and control strategies are difficult to directly transfer due to the fundamental differences in application scenarios (daily driving vs. high-intensity flight training). Additionally, these solutions suffer from three inherent drawbacks: Reward function design is out of touch with actual needs: Reinforcement learning reward functions often rely on energy consumption E and temperature fluctuations. For a single optimization objective (such as The optimization system did not incorporate the PMV / PPD (predicted average votes / predicted percentage of dissatisfaction) parameter, which is considered the "gold standard" for thermal comfort, resulting in a significant deviation between the system's control targets and people's actual thermal sensations. Insufficient personalized adaptive capability: When facing new users (new pilots), traditional reinforcement learning models need to be trained from scratch or retrained on large-scale data, with a cold start cycle of more than 2 hours, which cannot meet the needs of rapid personnel rotation in flight training scenarios. Lack of safety protection mechanisms: Existing systems are mostly passive response controls, lacking predictive assessment and graded intervention mechanisms for pilots' physiological limits (such as excessively high core body temperature), posing potential health risks in high-intensity training scenarios; In summary, current technology cannot meet the needs of flight simulators for multi-dimensional physiological perception fusion, multi-target collaborative optimization, and rapid personalized adaptation. There is an urgent need to develop an intelligent temperature control system for flight training scenarios. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, namely, the inability of existing technologies to meet the needs of flight simulator temperature control for multi-dimensional physiological perception fusion, multi-objective collaborative optimization, and rapid personalized adaptation and active safety protection for new pilots, the first aspect of this invention proposes an intelligent temperature control system for a flight simulator cockpit, comprising: The multimodal sensing layer is configured to: acquire multi-channel skin physiological data and perform mean processing to obtain average skin physiological data; acquire physical environment data and calculate heat dissipation based on the average skin physiological data; acquire pilot electrocardiogram signals, perform preprocessing and feature extraction to obtain feature parameters; and normalize the feature parameters to obtain blood flow correction coefficients. Based on the heat dissipation and the blood flow correction coefficient, a blood flow-corrected human body heat balance equation is constructed, thereby generating a physiological feature vector; the physiological feature vector includes predicted core body temperature, predicted average votes, and predicted dissatisfaction percentage. The decision-making layer is configured as follows: based on the physiological feature vector and training task load instructions, a meta-learning model is used to generate temperature control instructions and confidence evaluation values; wherein, the reward function of the meta-learning model is a multi-objective dynamic weighted reward function, including four-dimensional objectives: comfort, energy consumption, stability, and response speed; the weights of the reward function are dynamically adjusted according to the real-time status through the IF-THEN rule; the real-time status includes: predicted core body temperature, predicted percentage of dissatisfaction, training task load level, and safety margin; The dynamic execution layer is configured to: based on the temperature control command and the average skin physiological data, use PID closed-loop adjustment to regulate the operating parameters of the airflow control equipment in each temperature control zone of the flight simulator cabin, thereby regulating the temperature of each zone; Based on the predicted core body temperature and the real-time numerical range of the predicted unsatisfactory percentage, a progressive temperature control response is triggered in stages, including local adjustment, composite adjustment, and global intervention alarm.

[0006] In some preferred embodiments, the pilot's electrocardiogram signal is acquired, preprocessed, and feature extracted to obtain feature parameters; the feature parameters are then normalized to obtain blood flow correction coefficients, the method of which is as follows: Obtain the pilot's electrocardiogram signal; The electrocardiogram (ECG) signal is preprocessed. The preprocessing includes using the Pan-Tompkins algorithm to detect the R-peak of the ECG signal and generating an RR interval sequence. Outliers in the RR interval sequence are removed and interpolated to obtain the preprocessed ECG signal. The preprocessed electrocardiogram signal features were extracted by short-time Fourier transform to obtain feature parameters, including low-frequency power, high-frequency power, and LF / HF ratio. Based on the extracted LF / HF ratio, the blood flow correction coefficient is obtained through a normalized sigmoid function. ; ; in, Let be the blood flow correction factor at time t. The minimum correction factor. To adjust the parameters, The real-time LF / HF ratio at time t. This is the LF / HF threshold.

[0007] In some preferred embodiments, the human body heat balance equation corrected by the blood flow correction coefficient is: ; in, To generate heat through metabolism, For convective heat dissipation, For radiative heat dissipation, For evaporative heat dissipation, This is the blood flow correction factor. The heat capacity of the human body The core body temperature This represents the rate of change of the human body's core temperature over time.

[0008] In some preferred embodiments, a blood flow-corrected human body heat balance equation is constructed based on the heat dissipation and the blood flow correction coefficient, thereby generating a physiological feature vector. The method is as follows: By integrating the human body heat balance equation, the predicted core body temperature can be obtained; Based on the predicted core body temperature, the physical environment data, and the average skin physiological data, the predicted average number of votes is obtained through a thermal comfort model, and then the predicted percentage of dissatisfaction is obtained.

[0009] In some preferred embodiments, the physical environment data includes cabin temperature, humidity, and airflow velocity; the heat dissipation includes convective heat dissipation, radiative heat dissipation, and evaporative heat dissipation. The multi-channel skin physiological data includes 12-channel skin temperature and 12-channel skin humidity; the average skin physiological data includes average skin temperature and average skin humidity.

[0010] In some preferred embodiments, the method for obtaining the meta-learning model is as follows: Define a meta-training dataset D_meta-train, where D_meta-train contains N tasks for different pilots. Each task is defined as a time-series data sequence {(S, A, R, S')} from a specific pilot during a training session; S is the state vector, A is the action vector, R is the reward value, and S' is the new state vector at the next moment after executing the action vector. A meta-learning model with a dual-network structure of teacher network and student network is constructed. Based on the meta-training dataset, knowledge from the teacher network is transferred to the student network through knowledge distillation and trained to obtain the meta-trained student network parameters θ. The teacher network is trained using historical data and its loss function includes physical constraint loss. The loss function of the student network includes KL divergence and hard label loss. The student network parameter θ is divided into general layer parameters, task layer parameters, and individual layer parameters by hierarchical parameter decoupling. When a new pilot joins, time-series data of the new pilot for a preset duration is collected as the fine-tuning dataset D_finetune; Load the meta-trained student network parameters θ, and perform finite gradient descent update with the D_finetune as input to obtain personalized student network parameters θ'; The student network updated using parameter θ' yields temperature control instructions and confidence assessment values.

[0011] In some preferred embodiments, the finite gradient descent update employs dynamic parameter freezing and selective fine-tuning; The dynamic parameter freezing and selective fine-tuning are as follows: the parameter freezing ratio is calculated based on the similarity between the network parameters of new pilots and historical pilots. The parameter freezing ratio is combined with the training time to select and unfreeze parameters of different dimensions in stages, and perform finite gradient descent updates to obtain personalized model parameters.

[0012] In some preferred embodiments, the student network parameters θ after loading meta-training are updated by finite gradient descent with D_finetune as input to obtain personalized student network parameters θ'. The update also includes feature space alignment and transfer optimization: minimizing the feature distribution difference between new and old pilots by maximizing the mean difference, and dynamically adjusting the feature weights of the personalized layer with a progressive fusion function, thereby correcting the parameter gradient updated by finite gradient descent.

[0013] In some preferred embodiments, the formula for calculating the parameter freezing ratio is: ; in, The parameter is the freeze ratio. This is a natural exponential function used to map similarity to a freeze ratio that follows an exponential decay law. Here, represents the similarity calculation function, indicating the initial model parameters for the new pilot. Cluster centers of parameters from historical pilot models Similarity measure This is a freeze adjustment parameter used to control the degree of influence of similarity on the freeze ratio; The parameter freezing ratio is combined with the training duration to select and unfreeze parameters of different dimensions in stages, including: in the initial stage, only the individual layer parameters are unfrozen; in the adaptation stage, the individual layer parameters are unfrozen and some task layer parameters are unfrozen; in the stable stage, all parameters are unfrozen and key parameters are updated using elastic weight regularization.

[0014] In some preferred embodiments, the IF-THEN rule includes: a safety criticality rule, an energy efficiency priority rule, and a steady-state optimization rule.

[0015] The beneficial effects of this invention are: 1) Addressing the limitations of traditional reliance on single environmental sensors, this technology integrates 12-channel skin sensing with blood flow correction, expanding the sensing dimension to three dimensions: "physiology + environment + hemodynamics". By combining HRV characteristics and the thermal balance equation, the core body temperature prediction error is reduced to within ±0.2℃, accurately capturing the impact of peripheral vasoconstriction on heat dissipation and providing comprehensive physiological-environmental data support for regulation. 2) Incorporate PMV / PPD thermal comfort parameters into the core of the reward function and dynamically adjust the weights through the "IF-THEN" rule: increase the comfort weight by 50% during the safety critical period, increase the energy consumption weight by 30% during high energy consumption periods, increase the PMV control accuracy by 30%, and reduce the overall energy consumption by 28%, thereby achieving a scenario-based balance between comfort and energy saving and avoiding a disconnect between the goal and the experience. 3) By adopting hierarchical parameter decoupling and meta-learning adaptation, and reducing redundant training by 70% through dynamic parameter freezing, and based on 10-minute fine-tuning of data and feature distribution alignment, the convergence time of the new pilot model is reduced from 2 hours to within 10 minutes, meeting the rapid rotation requirements of "immediate replacement and adaptation" in flight training. 4) Construct a three-level response mechanism: local ventilation is activated when the core body temperature is >37.3℃, additional synergistic cooling is added when PMV is >1.0, and a global alarm is issued when PMV is >2.0. The response time for heat discomfort is reduced from 30 seconds to 8 seconds, and the risk of heat stress is reduced by more than 60%, forming a proactive safety closed loop of "prediction-intervention-alarm" to ensure physiological safety. Attached Figure Description

[0016] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart of the calculation of the human body heat balance equation with blood flow correction according to the present invention.

[0017] Figure 2 This is an architectural diagram of an intelligent temperature control system for a flight simulator cockpit according to the present invention. Detailed Implementation

[0018] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] To more clearly explain the intelligent temperature control system for a flight simulator cockpit according to the present invention, the following description is provided in conjunction with... Figures 1 to 2 The steps in the embodiments of the present invention will be described in detail below.

[0021] This invention proposes an intelligent temperature control system for a flight simulator cockpit, see [link to relevant documentation]. Figure 1 The system includes: a multimodal sensing layer, see [link to documentation]. Figure 2 The configuration is as follows: acquire multi-channel skin physiological data and perform mean processing to obtain average skin physiological data; In this embodiment, the multi-channel skin physiological data includes 12-channel skin temperature and 12-channel skin humidity (a 12-channel flexible sensor array is installed in key areas such as the torso and limbs of the pilot's seat and clothing to collect skin temperature and humidity data in real time at a sampling rate of 100Hz); and the average skin physiological data, including average skin temperature, is obtained through averaging. Average skin humidity This provides a physiological basis for differentiated temperature control; Obtain physical environment data and calculate heat dissipation by combining the average skin physiological data; In this embodiment, the physical environment data includes the cabin temperature. ,humidity airflow speed The heat dissipation is precisely quantified to achieve the heat dissipation state, including convective heat dissipation. Radiative heat dissipation Evaporative heat dissipation ; wherein, the The calculation formula is: ,in, The convective heat transfer coefficient is... A The body surface area; The calculation formula is: ,in, For emission rate, Boltzmann's constant; The The calculation formula is: ; in, The evaporative heat transfer coefficient is... , These are the partial pressures of water vapor on the skin surface and inside the cabin, respectively. The pilot's electrocardiogram signal is acquired, preprocessed, and feature extracted to obtain feature parameters; the feature parameters are then normalized to obtain the blood flow correction coefficient. In this embodiment, the method for acquiring the pilot's electrocardiogram signal, preprocessing it, extracting features to obtain feature parameters, and normalizing these feature parameters to obtain the blood flow correction coefficient is as follows: Pilot electrocardiogram (ECG) signals can be acquired using an ECG / PPG sensor. The electrocardiogram (ECG) signal is preprocessed. The preprocessing includes using the Pan-Tompkins algorithm to detect the R-peak of the ECG signal and generating an RR interval sequence. Outliers in the RR interval sequence are removed and interpolated to obtain the preprocessed ECG signal. Feature parameters were obtained by extracting features from the preprocessed electrocardiogram signal using short-time Fourier transform (STFT). Including low-frequency power High-frequency power and LF / HF ratio The expression for the feature parameters is: ; The dynamic increase in the LF / HF ratio is considered a core physiological parameter of peripheral vasoconstriction and decreased heat dissipation capacity. Based on the extracted LF / HF ratio, the blood flow correction coefficient is obtained through a normalized sigmoid function. It captures the peripheral vasoconstriction effect caused by sympathetic nerve excitation, overcoming the deficiency of traditional heat balance models that ignore the influence of blood flow; ; in, Let be the blood flow correction factor at time t, preferably [0.6, 1.0]. The minimum correction factor is preferably 0.6. To adjust the parameters, 5.0 is preferred. The real-time LF / HF ratio at time t. The LF / HF threshold is preferably 2.5; Based on the heat dissipation and the blood flow correction coefficient, a blood flow-corrected human body heat balance equation is constructed, thereby generating a physiological feature vector; the physiological feature vector Including predicting core body temperature Predicted average vote count Predicting the percentage of dissatisfaction The physiological feature vector expression is: This enables precise mapping from physiological signals to thermal comfort states; The human body heat balance equation corrected by the blood flow correction coefficient is as follows: ; in, Heat generated by metabolism (unit: W). Heat dissipation via convection (in W). Heat dissipation by radiation (unit: W). Evaporative heat dissipation (unit: W). This is the blood flow correction factor. Human body heat capacity (unit: J / K) The core body temperature (unit: °C) The rate of change of core body temperature over time; The original heat transfer equation of the blood flow-corrected human body thermal equilibrium model is: ,in Where is the thermal diffusivity, Metabolic heat production density, For human tissue density, For specific heat capacity, The core and skin temperature difference coefficient, Factors affecting blood flow; Based on the heat dissipation and the blood flow correction coefficient, a blood flow-corrected human body heat balance equation is constructed, and then a physiological feature vector is generated. The method is as follows: By integrating the human body heat balance equation, the predicted core body temperature can be obtained; The formula for integral operation is: ; in, The predicted core body temperature at time t. Let t be the initial core body temperature, and t be the integration time variable, with a lower limit of 0 and an upper limit of t. For a moment The blood flow correction factor has a value range of [0.6, 1.0]. Between the lower limit 0 and the upper limit t; Based on the predicted core body temperature The physical environment data (cabin interior temperature) ,humidity airflow speed The average skin physiological data ( The predicted average number of votes was obtained using the Fanger thermal comfort model. This leads to the predicted percentage of dissatisfaction. ; specific: ; ; Where W represents the external work done, the value of which is estimated by the joystick pressure sensor. For normal flight: W ≈ 0 (very little joystick force); for aerobatic / emergency flight: W = 15-25 W / m² (work done by high G-force countermeasures); for ejection: W = 30-50 W / m² (work done by instantaneous impact). The decision layer is configured as follows: based on the physiological feature vector and training task load instructions, a meta-learning model is used to generate temperature control instructions and confidence evaluation values. The reward function of the meta-learning model is a multi-objective dynamic weighted reward function, which includes four dimensions of objectives: comfort, energy consumption, stability, and response speed. The weights of the reward function are dynamically adjusted according to the real-time status using the IF-THEN rule; the real-time status includes: predicted core body temperature, predicted percentage of unsatisfactory results, training task load level, and safety margin. In this embodiment, the multi-objective dynamic weighted reward function is: ; in, Total reward function value, The system state vector at time t. : Control action vector at time t , , , : Dynamic weighting coefficient (summation = 1); Constraints: ; in: ; ; ;

[0022] As a comfort reward, To predict the average number of votes for the target, The current actual predicted average number of votes (i.e., the predicted average number of votes obtained through the thermal comfort model). ), As an energy consumption reward, The total power consumption of the system is The rated power of the system, As a reward for temperature stability, Standard deviation, The recent cabin temperature, Rewards for response speed The temperature after the system responds, that is, the temperature reached in the cabin after the control action is executed. The optimal temperature represents the ideal cabin temperature. The IF-THEN rules include: safety criticality rules, energy efficiency priority rules, and steady-state optimization rules; intelligent adaptation of optimization objectives is achieved through precise matching of scenario features, and dynamic weight vectors. W ( st ) as system state st The function is not a fixed constant, but rather adaptively adjusted based on the training task type, the pilot's real-time physiological state, and external environmental parameters. The specific rule logic is as follows: Safety Criticality Rule: When the system identifies a safety criticality state (such as detecting high-pressure scenarios through HRV characteristic parameters, sudden changes in HRV signals, and other physiological warning signals), it automatically triggers the "comfort weight priority enhancement" mechanism, increasing the comfort weight. W 1. Dynamically increase by 50% to prioritize pilot physiological safety and operational stability; Energy efficiency priority rule: When in low-load cruise phases or high-energy-consumption periods (such as peak grid electricity consumption), the "energy-saving target enhancement" strategy is activated, increasing the weight of energy consumption. W 3. Improvement by 30% guides the system to adopt energy efficiency optimization control strategies while meeting basic comfort requirements; Steady-state optimization rules: During the steady-state phase of regular training, stability weights are dynamically balanced based on cabin environment stability indicators and response speed requirements. W 2. Weighting based on response speed W The 4:1 ratio ensures that the temperature control system achieves the optimal balance between accuracy and dynamic response; This structured rule system, through clear scenario triggering conditions and quantitative adjustment strategies, not only ensures the scientific nature of weight adjustment but also enables flexible switching of optimization targets in different scenarios, effectively improving the rationality and robustness of system control. The method for obtaining the meta-learning model is as follows: The meta-training phase specifically includes task construction, where a "task" is defined from the time-series data of a specific pilot during a training session, and the time-series data of a single task for a specific object is defined as a meta-learning data unit. This includes a state vector S (State), such as multimodal perception layer data; an action vector A (Action), such as temperature control commands; a reward value R (Reward); and a new state vector S' at the next moment after executing the action vector; and a meta-training dataset. The mission includes N different pilots; A meta-learning model with a dual-network structure of teacher and student networks is constructed. Based on the meta-training dataset, knowledge from the teacher network is transferred to the student network through knowledge distillation and trained to obtain the meta-trained student network parameters θ. The teacher network is trained using historical data (i.e., state vector S), and its loss function includes physical constraint loss, with the output being the temperature control command prediction distribution and confidence level. The student network's loss function includes KL divergence and hard label loss, achieving lightweight deconstruction and addressing the problem of insufficient model generalization ability. Among them, the loss function of the teacher network for: ; in, For physical constraint loss, For comfort loss function, For the first Comfort labels / target values ​​for each sample. For the teacher network to the first Input Samples The predicted value of the temperature control output. These are the weighting coefficients for the physical constraint loss; Loss function of student network for: ; in, These are the weighting coefficients. This is the raw output of the student network to input x. This is the raw output of the teacher network to input x. Temperature parameter controls the smoothness of the soft label. KL divergence measures the difference between the distributions of student and teacher outputs. Hard label loss; Hierarchical parameter decoupling is used to decouple student network parameters Divided into general layer parameters Task layer parameters and personality layer parameters These are used to model the universal laws of human thermophysiology (such as basal metabolic rate and heat dissipation mechanisms), the characteristics of different training tasks (such as takeoff and landing, aerobatic maneuvers, and emergency situations), and individual differences among pilots (such as physical condition, preferences, and adaptability); the specific expressions are as follows: ; New pilot adaptation phase: When a new pilot joins, collect the time series data (S, A, R, S') of the new pilot for a preset duration (first 10 minutes) as the fine-tuning dataset D_finetune; Load the meta-trained student network parameters θ, and perform finite gradient descent update with D_finetune as input to obtain personalized student network parameters θ'; The finite gradient descent update employs dynamic parameter freezing and selective fine-tuning. The dynamic parameter freezing and selective fine-tuning are as follows: the parameter freezing ratio is calculated based on the similarity between the model parameters of the new pilot and the historical pilot. The parameter freezing ratio is combined with the training time to select and unfreeze parameters of different dimensions in stages to obtain personalized model parameters. The formula for calculating the freezing ratio of the parameter is: ; in, The parameter is the freeze ratio. This is a natural exponential function used to map similarity to a freeze ratio that follows an exponential decay law. Here, represents the similarity calculation function, indicating the initial model parameters for the new pilot. Cluster centers of parameters from historical pilot models Similarity measure This is a freeze adjustment parameter used to control the degree of influence of similarity on the freeze ratio.

[0023] The parameter freezing ratio is combined with the training duration to select and unfreeze parameters of different dimensions in stages, including: the initial stage (0-5 minutes), where only the individual layer parameters are unfrozen. Adaptation phase (5-10 minutes): Unfreeze personality layer parameters and some task layer parameters. During the stable phase (>10 minutes), all parameters are unfrozen, and key parameter updates are constrained using elastic weight regularization to prevent catastrophic amnesia; the elastic weight regularization formula is: ; in, Let be the total loss function for elastic weight regularization. These are the diagonal elements of the Fisher information matrix, used to measure the importance of parameters. Current mission losses, Regularization weight coefficients The model is currently updating the first Parameter values, No. The optimal values ​​for each parameter after training for the old mission (historical pilots) is completed. This represents the squared difference between the old and new parameters; The student network parameters θ after loading the meta-training are updated by finite gradient descent with D_finetune as input to obtain personalized student network parameters θ'. It also includes feature space alignment and transfer optimization update: minimizing the feature distribution difference between new and old pilots by the maximum mean difference, and dynamically adjusting the feature weights of the personalized layer with the progressive fusion function, thereby correcting the parameter gradient updated by finite gradient descent to achieve personalized adaptation optimization. The method to minimize the difference in characteristic distribution between new and old pilots by maximizing the mean difference is as follows: The state vector S in the fine-tuning dataset D_finetune of the new pilot (i.e. the target pilot) is input into the shared feature extractor; The distribution difference between the extracted new pilot feature representation and the general feature representation space learned by the old pilot (in the meta-training phase) is calculated. The feature alignment loss function is applied to minimize the distribution difference so that the feature representation of the new pilot is effectively projected into the general feature representation space. The aligned target pilot feature representation is used to drive parameter updates for subsequent decision layers. The formula for minimizing the feature alignment loss is: ; in, For the maximum mean difference, The characteristic distribution of the source pilots (“old data” distribution). The characteristic distribution of the target pilots (“new data” distribution). The number of source pilot feature samples (sample size). The number of characteristic samples of the target pilot (sample size). This is a single feature sample from the source pilot. A single feature sample of the target pilot. For feature mapping function, Let be the mean vector of the source pilot features in high-dimensional space. Let be the mean vector of the target pilot's features in a high-dimensional space. Let H be the square norm in the reproducing kernel Hilbert space (RKHS, denoted as H); The progressive fusion function is: ; ; in, The final feature representation after fusion, The dynamic weighting coefficients of personalized features (varying with time t, with values ​​ranging from [0,1]) The general feature representation (basic features) obtained from meta-training Personalized characteristics of the target pilot (exclusive characteristics); As the adaptation time increases, the weight of individual layer features gradually increases; In this way, the problems of simplified reward function and long cold start cycle can be solved through multi-objective dynamic optimization and rapid personalized adaptation. The student network, updated using parameter θ', generates real-time control commands, which in turn update the learning model to obtain temperature control commands. : ; in, Set the temperature for the temperature control system. These are the fan speed control parameters. Select the ventilation mode; Confidence level assessment value: ; This value is calculated based on the decision uncertainty of the personalized model θ' for the new input state or the similarity of the distribution with the meta-training data; when Confidence_score is lower than the preset threshold, a data sampling update request is triggered or the control strategy is adjusted to a conservative one to provide personalized temperature control; The dynamic execution layer is configured to: based on the temperature control command and the average skin physiological data, use PID closed-loop to adjust the operating parameters of the airflow control equipment in each temperature control zone of the flight simulator cockpit, thereby controlling the temperature of each zone and achieving precise and differentiated air delivery to different parts of the pilot's body; The flight simulator cockpit has multiple independent temperature control zones, such as the head, torso, and legs; each zone's air outlet is equipped with independent airflow control devices, such as fans, valves, and air conditioners. Based on the real-time numerical range of the predicted core body temperature and the percentage of unsatisfactory predictions, a progressive temperature control response is triggered in stages, including local adjustment, composite adjustment, and global intervention alarm, to achieve a three-level active safety response. Level 1 response for local regulation (early warning intervention): Triggered when the predicted core body temperature exceeds 37.3℃, immediately initiating localized high-temperature strong air supply in the high-temperature area; Level 2 response of composite regulation (synergistic cooling): Based on the Level 1 response, if the PMV is greater than 1.0, the seat ventilation will be activated and the fresh air exchange rate will be increased. Level 3 response to global intervention alarm (forced cooling): If the predicted unsatisfactory percentage further deteriorates to greater than 2.0, global high-power cooling will be forcibly activated and an alarm will be sent to the monitoring station to ensure the absolute physiological safety of the pilot; In this way, through precise zone control and proactive safety protection, a full-chain safety protection system from early warning to mandatory intervention is formed, eliminating potential health risks and solving the problems of crude and passive response in traditional control.

[0024] It should be noted that the intelligent temperature control system for a flight simulator cockpit provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be further decomposed or combined. For example, the modules in the above embodiments can be merged into one module, or further divided into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the various modules or steps and are not considered as an improper limitation of the present invention.

[0025] Those skilled in the art will recognize that the modules and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The programs corresponding to the software modules and method steps can be placed in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium known in the art. To clearly illustrate the interchangeability of electronic hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention.

[0026] The terms “first”, “second”, etc., are used to distinguish similar objects, not to describe or indicate a specific order or sequence.

[0027] The term "comprising" or any other similar term is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus / device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent in such process, method, article, or apparatus / device.

[0028] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An intelligent temperature control system for a flight simulator cockpit, characterized in that, The system includes: The multimodal sensing layer is configured to: acquire multi-channel skin physiological data and perform mean processing to obtain average skin physiological data; acquire physical environment data and calculate heat dissipation based on the average skin physiological data; acquire pilot electrocardiogram signals, perform preprocessing and feature extraction to obtain feature parameters; and normalize the feature parameters to obtain blood flow correction coefficients. Based on the heat dissipation and the blood flow correction coefficient, a blood flow-corrected human body heat balance equation is constructed, thereby generating a physiological feature vector; the physiological feature vector includes predicted core body temperature, predicted average votes, and predicted dissatisfaction percentage. The decision-making layer is configured as follows: based on the physiological feature vector and training task load instructions, a meta-learning model is used to generate temperature control instructions and confidence evaluation values; wherein, the reward function of the meta-learning model is a multi-objective dynamic weighted reward function, including four-dimensional objectives: comfort, energy consumption, stability, and response speed; the weights of the reward function are dynamically adjusted according to the real-time status through the IF-THEN rule; the real-time status includes: predicted core body temperature, predicted percentage of dissatisfaction, training task load level, and safety margin; The dynamic execution layer is configured to: based on the temperature control command and the average skin physiological data, use PID closed-loop adjustment to regulate the operating parameters of the airflow control equipment in each temperature control zone of the flight simulator cabin, thereby regulating the temperature of each zone; Based on the predicted core body temperature and the real-time numerical range of the predicted unsatisfactory percentage, a progressive temperature control response is triggered in stages, including local adjustment, composite adjustment, and global intervention alarm.

2. The intelligent temperature control system for a flight simulator cockpit according to claim 1, characterized in that, The pilot's electrocardiogram signal is acquired, preprocessed, and feature extracted to obtain feature parameters; the feature parameters are then normalized to obtain blood flow correction coefficients, the method of which is as follows: Obtain the pilot's electrocardiogram signal; The electrocardiogram (ECG) signal is preprocessed. The preprocessing includes using the Pan-Tompkins algorithm to detect the R peak of the ECG signal and generating an RR interval sequence. Outliers in the RR interval sequence are removed and interpolated to obtain the preprocessed ECG signal. The preprocessed electrocardiogram signal features were extracted by short-time Fourier transform to obtain feature parameters, including low-frequency power, high-frequency power, and LF / HF ratio. Based on the extracted LF / HF ratio, the blood flow correction coefficient is obtained through a normalized sigmoid function. ; ； in, Let be the blood flow correction factor at time t. The minimum correction factor, To adjust the parameters, The real-time LF / HF ratio at time t. This is the LF / HF threshold.

3. The intelligent temperature control system for a flight simulator cockpit according to claim 2, characterized in that, The human body heat balance equation corrected by the blood flow correction coefficient is as follows: ； in, To generate heat through metabolism, For convective heat dissipation, For radiative heat dissipation, For evaporative heat dissipation, Let be the blood flow correction factor at time t. The heat capacity of the human body The core body temperature This represents the rate of change of the human body's core temperature over time.

4. The intelligent temperature control system for a flight simulator cockpit according to claim 1, characterized in that, Based on the heat dissipation and the blood flow correction coefficient, a blood flow-corrected human body heat balance equation is constructed, and then a physiological feature vector is generated. The method is as follows: By integrating the human body heat balance equation, the predicted core body temperature can be obtained; Based on the predicted core body temperature, the physical environment data, and the average skin physiological data, the predicted average number of votes is obtained through a thermal comfort model, and then the predicted percentage of dissatisfaction is obtained.

5. The intelligent temperature control system for a flight simulator cockpit according to claim 1, characterized in that, The physical environment data includes cabin temperature, humidity, and airflow speed; the heat dissipation includes convective heat dissipation, radiative heat dissipation, and evaporative heat dissipation. The multi-channel skin physiological data includes 12-channel skin temperature and 12-channel skin humidity; the average skin physiological data includes average skin temperature and average skin humidity.

6. The intelligent temperature control system for a flight simulator cockpit according to claim 1, characterized in that, The method for obtaining the meta-learning model is as follows: Define a meta-training dataset D_meta-train, where D_meta-train contains N tasks for different pilots. Each task is defined as a time-series data sequence {(S, A, R, S')} from a specific pilot during a training session; S is the state vector, A is the action vector, R is the reward value, and S' is the new state vector at the next moment after executing the action vector. A meta-learning model with a dual-network structure of teacher network and student network is constructed. Based on the meta-training dataset, knowledge from the teacher network is transferred to the student network through knowledge distillation and trained to obtain the meta-trained student network parameters θ. The teacher network is trained using historical data and its loss function includes physical constraint loss. The loss function of the student network includes KL divergence and hard label loss. The student network parameter θ is divided into general layer parameters, task layer parameters, and individual layer parameters by hierarchical parameter decoupling. When a new pilot joins, time-series data of the new pilot for a preset duration is collected as the fine-tuning dataset D_finetune; Load the meta-trained student network parameters θ, and perform finite gradient descent update with the D_finetune as input to obtain personalized student network parameters θ'; The student network updated using parameter θ' yields temperature control instructions and confidence assessment values.

7. The intelligent temperature control system for a flight simulator cockpit according to claim 6, characterized in that, The finite gradient descent update employs dynamic parameter freezing and selective fine-tuning. The dynamic parameter freezing and selective fine-tuning are as follows: the parameter freezing ratio is calculated based on the similarity between the network parameters of new pilots and historical pilots. The parameter freezing ratio is combined with the training time to select and unfreeze parameters of different dimensions in stages, and perform finite gradient descent updates to obtain personalized model parameters.

8. The intelligent temperature control system for a flight simulator cockpit according to claim 6, characterized in that, The student network parameters θ after loading the meta-training are updated by finite gradient descent with D_finetune as input to obtain personalized student network parameters θ'. It also includes feature space alignment and transfer optimization update: minimizing the feature distribution difference between new and old pilots by maximizing the mean difference, and dynamically adjusting the feature weights of the personalized layer with a progressive fusion function, thereby correcting the parameter gradient updated by finite gradient descent.

9. The intelligent temperature control system for a flight simulator cockpit according to claim 7, characterized in that, The formula for calculating the freezing ratio of the parameter is: ； in, The parameter is the freeze ratio. It is a natural exponential function. Here, represents the similarity calculation function, indicating the initial model parameters for the new pilot. Cluster centers of parameters from historical pilot models Similarity measure This is the parameter for freezing adjustment; The parameter freezing ratio is combined with the training duration to select and unfreeze parameters of different dimensions in stages, including: in the initial stage, only the individual layer parameters are unfrozen; in the adaptation stage, the individual layer parameters are unfrozen and some task layer parameters are unfrozen; in the stable stage, all parameters are unfrozen and key parameters are updated using elastic weight regularization.

10. The intelligent temperature control system for a flight simulator cockpit according to claim 1, characterized in that, The IF-THEN rules include: safety criticality rules, energy efficiency priority rules, and steady-state optimization rules.

Images