Method and system for health management of flight personnel with abnormal vestibular function examination

Abstract

The application discloses a health management method and system for flight personnel with abnormal vestibular function, and relates to the technical field of flight personnel management. The management method comprises the following steps: obtaining a multi-source health data set of a target flight personnel, constructing a load characteristic variable, and synchronously storing; inputting the load characteristic variable into a pre-constructed linear evaluation model after performing a screening normalization processing action; the technical key points are that a linear weighting model is adopted in the running management method, a preliminary flight adaptability is output, and a flight threshold is set, on the one hand, the rapid grading discrimination of the health adaptability of the flight personnel is realized, the continuous operation efficiency of the high-confidence flight personnel is ensured, on the other hand, the preliminary evaluation mechanism is used as a condition triggering tool, the related process of VEMP supplementary examination is activated under the corresponding condition, the problems of low efficiency and heavy medical burden caused by the one-size-fits-all in the traditional medical screening are solved, and the effect of on-demand detection and precise starting resource optimization configuration is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of flight personnel management, in particular to a method and system for health management of flight personnel with abnormal vestibular function. BACKGROUND

[0002] Flight personnel management is a comprehensive management system for airlines or aviation agencies to ensure flight safety and operational efficiency, including the qualification, training, scheduling, health and career development of pilots. In the prior art, the license status, fatigue level and technical ability of flight personnel are monitored by relying on the flight operation management system (OMS) and the aviation personnel health record system, combined with periodic medical examination, flight hour statistics and simulator assessment. Flight tasks are arranged and training cycles are tracked through manual review combined with the basic information system.

[0003] In the prior art, the function of the semicircular canal is often checked in the selection and medical identification of flight personnel and astronauts, but the semicircular canal only senses angular acceleration, lacks the evaluation of the utricle and saccule functions for sensing linear acceleration and gravity, and the existing flight personnel health management relies on periodic medical examination and flight hour statistics, the evaluation methods are scattered and lack of system integration. For example, the vestibular function test is usually conducted independently, which is disconnected with flight load and physiological indicators, resulting in one-sided evaluation results. The aviation doctor mainly determines whether to stop flying according to single abnormal value, such as limiting flight when the DAR value exceeds the standard, ignoring the consideration of overall health data and overall state corresponding to different targets, which is easy to cause excessive intervention or missed judgment. In addition, all pilots need to receive a full set of tests regularly, regardless of the actual risk level, which not only increases the medical cost, but also increases the burden of personnel. In addition, the traditional method uses linear threshold or experience judgment, which is difficult to capture the complex interaction between the otolith function and autonomic nervous regulation, lacks dynamic updating and evidence fusion mechanisms, and the decision-making basis is insufficient, the overall risk prediction ability is weak, and it is difficult to maintain the balance of precise and personalized management of health data. SUMMARY

[0004] To achieve the above purpose, the present application realizes the following technical solutions:

[0005] The method for health management of flight personnel with abnormal vestibular function comprises:

[0006] Step 1: Obtain the multi-source health data set of the target flight personnel, construct the load characteristic variable, and store it synchronously.

[0007] Step two: according to the load characteristic variable, after performing the screening normalization processing action, input into the pre-constructed linear evaluation model, output the preliminary flight fitness, and compare it with the set flight fitness threshold, when the preliminary flight fitness exceeds the flight fitness threshold, it is determined that the preliminary flight fitness is entered into step three; otherwise, it enters step four;

[0008] Step three: for the target flight personnel who are determined to be preliminarily fit to fly, start the periodic data extraction task, detect the trend change rate of the calibration data, and when the trend change rate of Q consecutive times exceeds the warning amplitude, generate a health risk warning prompt signal, and perform the inspection and diagnosis action;

[0009] Step four: obtain supplementary data and perform standardization coding to form a structured data set, input the structured data set and the preliminary flight fitness into the pre-constructed nonlinear weighted fusion model, output the corrected flight fitness, compare it with the set correction threshold, and generate the corresponding decision suggestion according to the comparison result.

[0010] Further, the multi-source health data set at least includes: historical flight data, physical examination data and basic physical sign data; wherein the historical flight data at least includes: cumulative flight time, recent flight frequency and historical average weekly flight frequency; the physical examination data at least includes: peak nystagmus speed of cold and hot test and bilateral asymmetry ratio; the basic physical sign data at least includes: resting heart rate, low frequency to high frequency ratio and blood oxygen saturation.

[0011] Further, before constructing the load characteristic variable, it also includes a data cleaning action, and the data cleaning action at least includes: missing value filling and abnormal value identification and elimination; the process of constructing the load characteristic variable is: constructed according to the multi-source health data set, including: flight load index FLI, inspection stability coefficient VSC and physiological adjustment coefficient PRI.

[0012] Further, the screening normalization processing action performed is: if any type of value in the flight load index FLI, the inspection stability coefficient VSC and the physiological adjustment coefficient PRI does not belong to [0, 1], the normalization function processing is performed to control the value in the range of [0, 1]; the linear evaluation model adopts the weighted summation method to obtain the preliminary flight fitness F1.

[0013] Further, when the trend change rate of Q consecutive times exceeds the warning amplitude, the inspection and diagnosis action is performed; wherein the value of Q is a positive integer greater than 0, and the content of the inspection and diagnosis action performed is: notifying the workstation and checking the target flight personnel.

[0014] Further, the supplementary data is obtained by pre-vestibular myogenic evoked potential VEMP examination, and the supplementary data at least includes: cVEMP latency, cVEMP amplitude, oVEMP latency, oVEMP amplitude, left and right amplitude difference ratio, and waveform distinguishability score; wherein, left and right amplitude difference ratio = | left amplitude - right amplitude | / average amplitude; the waveform distinguishability score corresponds to the scoring standard: 0 = no waveform, 1 = fuzzy, 2 = clear; the structured data set formed is X_vemp ∈ R d ; wherein, d is the feature dimension, corresponding to each supplementary data.

[0015] Further, the operation basis of the pre-constructed nonlinear weighted fusion model is: F2 = α × F1 + (1 - α) × M(X_vemp); wherein, α is the fusion weight coefficient, the value range is (0, 1); M(X_vemp) is a nonlinear mapping function, realized by a multilayer perception machine MLP, used to extract high-order features from X_vemp, and output a health adaptation score in the interval [0, 1], that is, a modified flight fitness F2.

[0016] Further, the architecture of the multilayer perception machine MLP includes: an input layer: 6 neurons, corresponding to 6-dimensional vemp features, that is, supplementary data;

[0017] a hidden layer: 10 neurons, using ReLU activation function, for enhancing nonlinear fitting ability;

[0018] an output layer: 1 neuron, without activation function, connected to a Sigmoid function, for controlling the output in the range [0, 1].

[0019] Further, the comparison process of the modified flight fitness F2 and the set modified threshold T2 is: when the modified flight fitness F2 exceeds the modified threshold T2, it is determined that: the flight is recoverable, and a first decision suggestion is generated; otherwise, it is determined that: intervention is needed, and a second decision suggestion is generated;

[0020] Wherein, the content of the first decision suggestion is: to suggest tracking the recovery flight task, and to review once according to the demand, and the review direction is selected as: VEMP and basic sign data; the content of the second decision suggestion is: to generate a medical intervention suggestion, including but not limited to: vestibular rehabilitation training, drug treatment and short-term flight suspension, and to complete the push.

[0021] The vestibular function examination abnormal flight personnel health management system includes: a health acquisition and processing module: obtaining a multi-source health data set of the target flight personnel, constructing a load characteristic variable, and synchronously storing;

[0022] Preliminary flight fitness evaluation module: after the screening and normalization processing action is performed according to the load characteristic variable, the preliminary flight fitness is output by inputting into the pre-constructed linear evaluation model, and the preliminary flight fitness is compared with the set flight threshold, when the preliminary flight fitness exceeds the flight threshold, it is determined that the preliminary flight is qualified and step three is entered, otherwise, step four is entered;

[0023] Health management monitoring module: for the target flight personnel whose determination result is preliminary flight fitness, a periodic data extraction task is started, the trend change rate of the calibration data is detected, when the trend change rate of consecutive Q times exceeds the early warning amplitude, a health risk early warning prompt signal is generated, and an inspection and diagnosis action is performed;

[0024] Modified flight fitness evaluation module: supplementary data is obtained and standardized coding is performed to form a structured data set, the structured data set and the preliminary flight fitness are input into the pre-constructed nonlinear weighted fusion model, the modified flight fitness is output, and the modified flight fitness is compared with the set modified threshold, and the corresponding decision suggestion is generated according to the comparison result.

[0025] The application provides a health management method and system for flight personnel with abnormal vestibular function, which has the following beneficial effects:

[0026] (1) The scheme realizes high-dimensional fusion and quantitative representation of multi-source health data of flight personnel by constructing composite characteristic variables, improves the information density and calculability of health state evaluation, and at the same time, the feature engineering method systematically integrates flight behavior, vestibular function and autonomic nervous regulation ability, solves the technical problems of fragmentation and strong subjectivity of traditional evaluation methods, and realizes individualized and dynamic health risk screening according to data;

[0027] (2) The scheme outputs the preliminary flight fitness based on the linear weighted model and sets the flight threshold, on the one hand, realizes the rapid classification and discrimination of the health adaptability of flight personnel, guarantees the continuous operation efficiency of high-confidence flight personnel, and on the other hand, uses the preliminary evaluation mechanism as a condition triggered tool to activate the related process of VEMP supplementary examination under the corresponding condition, solves the problems of low efficiency and heavy medical burden caused by one-size-fits-all in traditional medical screening, and achieves the effect of on-demand detection and precise starting of resource optimization configuration;

[0028] (3) The scheme adopts a nonlinear weighted fusion model, which not only maintains the continuity of historical health data, but also enhances the response ability to new functional abnormalities, generates decision suggestions after setting the modified threshold, realizes the closed-loop management from screening to intervention, solves the problems of high misjudgment rate and insufficient decision basis in traditional health evaluation, which leads to excessive flight suspension or risky release, and embodies the dynamic balance mechanism of the scheme between risk control and operation guarantee;

[0029] (4) The scheme cooperates with the multi-layer perception structure, realizes the scale unification and comparability with the preliminary flight adaptability, on the one hand ensures the numerical stability and semantic consistency of the model output, on the other hand enhances the sensitivity to the abnormal function of the otolith organ through the nonlinear mapping, realizes the related technical effects of high precision, interpretability and low misjudgment, solves the technical problems that the traditional linear model is difficult to capture the complex physiological interaction effect. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 It is a whole flowchart of the flight personnel health management method for the vestibular function examination abnormality in the application;

[0031] Figure 2 It is a core flowchart of the flight personnel health management method for the vestibular function examination abnormality in the application. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0033] Embodiment 1:

[0034] Please refer to Figure 1 and Figure 2 The flight personnel health management method for the vestibular function examination abnormality is provided in the embodiment, the selection and health related information of the flight personnel or flight student are managed, the scheme recorded in the method relates to the target task and research content, and the summary is as follows:

[0035] In medical selection, flight students with good vestibular function are the prerequisite for flight; in medical identification, vestibular disease is one of the main reasons for flight stop, therefore, aerospace medical workers pay high attention to the evaluation of flight students' vestibular function; for example, flight students play a very important role in the process of flight, especially in the process of take-off, encountering bumps and landing, which depends on the otolith organ to judge the direction of up and down and acceleration; but the function of semicircular canal is used in the professional selection and medical identification of flight personnel and astronauts (i.e. corresponding flight personnel), and the semicircular canal only senses angular acceleration, lacking the function of utricle and saccule to sense linear acceleration and gravity; in order to comprehensively and systematically evaluate the vestibular receptor function of flight students, it is necessary to use the otolith organ examination method commonly used in clinical practice, i.e. vestibular myogenic evoked potential VEMP, in aviation medicine; the scheme recorded in the embodiment mainly investigates the comprehensive situation of flight students under different conditions, and provides a more authoritative objective detection means for the selection and training of flight personnel in the flight industry; the specific scheme is as follows:

[0036] S1, flight personnel health data acquisition and preprocessing:

[0037] S1.1, multi-source health data acquisition:

[0038] The multi-source health data set corresponding to the target flight personnel is acquired, and synchronous storage is completed; wherein the multi-source health data set includes: historical flight data, physiological examination data and basic physical sign data;

[0039] Specifically, the target flight personnel can be the flight student, astronaut or other type of flight personnel mentioned above, in the embodiment, the flight personnel generally refers to: flight student;

[0040] The historical flight data in the multi-source health data set includes: cumulative flight time, recent flight frequency, single longest flight time, and historical average weekly flight frequency, etc., which together form a structured flight behavior data table. The historical flight data is usually obtained from the existing onboard operation management system (OMS) of the corresponding airline, and is synchronized regularly through an API interface to complete the storage action. It should be noted that the recent flight frequency in this embodiment refers to the flight frequency in the past 7 days. In this embodiment, the physiological examination data refers to the results of past and recent semicircular canal function tests, which are obtained. Therefore, the physiological examination data includes the peak nystagmus velocity and the bilateral asymmetry ratio of the cold and hot test, which are used as quantitative indicators. The data format is standardized to a unified unit and timestamp, and can also be stored through data encryption transmission. The basic vital sign data includes resting heart rate, low-to-high frequency ratio, and blood oxygen saturation, which represent some routine basic parameters monitored daily, and are therefore basic vital sign data. In summary, all data mentioned in the multi-source health data set can be encrypted using AES-256 during transmission and stored in a private database that meets the standard to avoid data leakage.

[0041] S1.2, data cleaning and feature engineering processing:

[0042] This step is also a routine step in the data collection and processing process. After the multi-source health data set is cleaned, load feature variables are constructed for use as input sources for subsequent models.

[0043] The data cleaning action includes filling in missing values and identifying and removing outliers in the multi-source health data set. Specifically, multiple imputation methods are used to fill in missing values, and the conditional distribution of missing items can also be estimated based on the Bayesian framework to ensure the effectiveness of statistical inference. The isolation forest algorithm is used to detect outliers, with a contamination factor of 0.1, to identify and label data points that deviate from the normal mode and remove them. In actual use, after identifying and labeling data points that deviate from the normal mode, expert review can be added to determine whether to remove them, but in this embodiment, direct removal is chosen to improve work efficiency.

[0044] Build load characteristic variable: according to multi-source health data set construction, including: flight load index FLI, check stability coefficient VSC and physiological adjustment coefficient PRI; wherein, FLI=af×(1+fp / ha); af represents cumulative flight time, unit: hour, fp represents flight frequency in recent 7 days, ha represents historical average weekly flight frequency; the design logic of this variable is: to comprehensively reflect the superposition effect of long-term cumulative load and short-term work intensity of flight personnel; cumulative flight time reflects the total amount of occupational exposure, which is the basic index for evaluating physiological aging and chronic fatigue; the ratio of flight frequency in recent 7 days to historical average weekly flight frequency is used to quantify whether the recent work rhythm deviates from the individual norm; if the ratio is greater than 1, it means that the recent flight density is higher than the average level, which may cause acute fatigue accumulation; through multiplication coupling, the nonlinear amplification of long-term load and short-term overload double pressure is realized, which more truly simulates the actual physiological stress state; for example, although a pilot has a long total flight time, but the recent flight is sparse, the flight load index FLI will not increase significantly, avoiding misjudgment as a high load state; wherein, VSC=pn / da; in the formula, pn represents peak nystagmus velocity, da represents bilateral asymmetry ratio; the logic is as follows: based on the results of the function test of the semicircular canal, the peak nystagmus velocity pn reflects the overall strength of the vestibular response, the higher the peak nystagmus velocity pn, the better the vestibular sensitivity, da is used to evaluate the symmetry of the function of the left and right semicircular canals, the higher the value, the greater or more serious the probability of unilateral functional damage, usually da is in the range of <25%, divide pn by da to construct a ratio type index of efficiency / imbalance, VSC is significantly improved when pn is high and da is low, indicating that the vestibular system responds strongly and symmetrically, and the stability is high; on the contrary, if da increases, even if pn is normal, VSC will be inhibited, which reflects the intrinsic instability of the system, effectively combining the dual dimensions of function strength and structural symmetry, which is better than the traditional single index independent judgment; wherein, PRI=a1×hrv_z+a2×(1-|hx_z|)+a3×so_g; in the formula, the value range of a1~a3 is [0, 1], hrv_z represents the value of LF / HF ratio Z-score standardization value after taking the inverse, because high LF / HF indicates sympathetic hyperactivity, so it needs to be taken inversely; wherein, LF / HF represents low frequency high frequency ratio, hx_z represents Z-score standardized value of resting heart rate, so_g represents normalized value of blood oxygen saturation, for example: if 95% is normalized to get 0.95; the overall design logic is weighted summation, which integrates basic physical data to reflect the autonomic nervous regulation ability.

[0045] By constructing flight load index, checking stability coefficient, and physiological adjustment coefficient, and other related composite characteristic variables, high-dimensional fusion and quantitative representation of multi-source health data of flight personnel are realized, and the information density and calculability of health status evaluation are improved. At the same time, the feature engineering method breaks through the limitation of relying on only a single index or simple threshold judgment in the traditional health management system, systematically integrates flight behavior, vestibular function and autonomic nervous regulation ability, solves the technical problems of fragmentation and strong subjectivity of traditional evaluation methods, realizes individualized and dynamic health risk screening based on data, and reflects the advantages of structured modeling in data-driven health management.

[0046] S2, preliminary flight fitness evaluation processing:

[0047] S2.1, define a preliminary flight fitness evaluation function:

[0048] According to the load characteristic variable, after performing the screening normalization processing action, input into the pre-constructed linear evaluation model, output the preliminary flight fitness, which is used to preliminarily evaluate whether the target flight personnel is fit to fly;

[0049] Among them, the screening normalization processing action is:

[0050] If any type of value in the flight load index FLI, the checking stability coefficient VSC and the physiological adjustment coefficient PRI does not belong to [0, 1], then the normalization function is processed, so that its value range is controlled in [0, 1], and the specific way adopted in this embodiment is: Min-Max Scaling to [0, 1] interval;

[0051] Input: FLI, VSC and PRI after performing the screening normalization processing action;

[0052] Process: F1=w1xFLI+w2xVSC+w3xPRI;

[0053] Output: preliminary flight fitness F1;

[0054] It should be noted that in the above formula, the linear weighted sum method is adopted, and the value range of w1-w3 is [0, 1], and the FLI, VSC and PRI mentioned in the formula are also after performing the screening normalization processing action, but in this embodiment, since the original values of FLI, VSC and PRI are all [0, 1], the FLI, VSC and PRI after performing the screening normalization processing action are the same as the original values, so the abbreviation formula representing the flight load index, the checking stability coefficient and the physiological adjustment coefficient does not change;

[0055] S2.2, set the preliminary flight fitness threshold constraint condition:

[0056] A flight suitability threshold T1 is set. When the preliminary flight suitability F1 exceeds the flight suitability threshold T1, it is determined that the preliminary flight suitability is met, and the next step S3 is performed. Otherwise, the modified flight suitability evaluation mechanism in S4 is triggered. For example, if a target flight personnel A has FLI, VSC and PRI of 0.82, 0.68 and 0.75 respectively, and w1-w3 take values of 0.4, 0.5 and 0.1 respectively, then F1=0.4×0.82+0.5×0.68+0.1×0.75=0.743. At this time, F1 is compared with T1 which takes a value of 0.7. Since 0.743 exceeds 0.7, it is determined that the preliminary flight suitability is met.

[0057] It should be noted that the setting of the flight suitability threshold T1=0.7 is based on statistical analysis and clinical verification of historical flight personnel health data. A specific operation example can be: through the retrospective modeling of personnel who have experienced vestibular-related flight incidents such as spatial disorientation or motion sickness-induced operational abnormalities in the past three years, the F1 distribution of their health assessment before the incident is calculated. It is found that individuals with F1<0.7 account for 89%. At the same time, in the normal flight group, the coverage rate of F1≥0.7 is more than 92%. Combined with ROC curve analysis, when T1=0.7, the model has a recognition sensitivity of 0.86 for unsuitable flight status, a specificity of 0.83, the largest Youden index, and the best discrimination performance. Therefore, this flight suitability threshold achieves a reasonable balance between ensuring flight safety and avoiding excessive restrictions, meeting the reasonable requirements of interpretability and practicality.

[0058] Based on the linear weighted model, the preliminary flight suitability is output and the flight suitability threshold is set. On the one hand, it realizes the rapid classification and discrimination of the health adaptability of flight personnel, ensuring the continuous operation efficiency of high-confidence flight personnel. On the other hand, the preliminary evaluation mechanism is used as a conditional trigger to activate the relevant procedures of VEMP supplementary examination under corresponding conditions, avoiding the waste of resources caused by high-cost otolith function screening for all personnel, solving the problem of low efficiency and heavy medical burden caused by the one-size-fits-all approach in traditional medical screening, and realizing the role of on-demand detection and precise activation of resource optimization, so that the overall health management scheme achieves effective hierarchical progressive logic.

[0059] S3, health management monitoring:

[0060] S3.1, dynamic monitoring strategy generation:

[0061] For the target flight personnel with the preliminary flight eligibility determination result, a periodic data extraction task is started, and a sliding window algorithm is used to detect the trend change rate of the calibration data; wherein the data extracted in the periodic data extraction task includes: physiological examination data and basic physical sign data, i.e. calibration data; the periodicity is defined as: data extraction is triggered once every T days; wherein the value of T is a positive integer greater than 0, and in this embodiment, T = 14; therefore, when detecting the trend change rate of the calibration data subsequently, it means that the trend change rate of each type of index contained in the physiological examination data and the basic physical sign data;

[0062] S3.2, early warning prompt measures:

[0063] When the trend change rate of any index contained in the physiological examination data and the basic physical sign data exceeds the target early warning amplitude for Q consecutive times, a health risk early warning prompt signal is generated, and a check diagnosis action is performed; wherein the value of Q is a positive integer greater than 0, and in this embodiment, Q is 2; the target early warning amplitude is usually 5%; for example: the trend change rates of the blood oxygen saturation of a certain target flight personnel B for 2 consecutive times are 6% and 8% respectively, which indicates a deteriorating trend, so a physical examination or diagnosis is needed, at this time a health risk early warning prompt signal is generated, and the check diagnosis action is performed as a normal selection means; specifically, the content of the check diagnosis action performed is: informing the medical personnel to perform a physical examination on the target flight personnel in the workstation.

[0064] S4, correction of flight eligibility evaluation mechanism:

[0065] S4.1, trigger supplementary data instruction:

[0066] When the preliminary flight eligibility F1 does not exceed the flight threshold T1, it is determined that the preliminary health adaptability screening is failed, in order to avoid misjudgment due to the limitation of a single evaluation model, step S4 is started, which corresponds to the content recorded in the scheme recorded in the foregoing summary of the target task and research content recorded in the method, i.e. adding an otolith organ examination method; obtaining supplementary data and standardizing coding to form a structured data set;

[0067] Specifically, the supplementary data is obtained by pre-vestibular myogenic evoked potential VEMP examination, and the supplementary data includes: the latency and amplitude of the p13 / n23 wave of cVEMP, i.e. cVEMP latency and cVEMP amplitude; the latency and amplitude of the n10 / p15 wave of oVEMP, i.e. oVEMP latency and oVEMP amplitude, the symmetry difference of left and right side responses: | left amplitude-right amplitude| / average amplitude), i.e. left and right amplitude difference ratio, waveform recognizability score, the specific scoring standard is: 0 = no waveform, 1 = fuzzy, 2 = clear; wherein the latency unit is: ms, and the amplitude unit is: μV;

[0068] It should be noted that VEMP includes two types of sub-detection, that is, cVEMP: by recording the potential response of the neck muscle group under acoustic stimulation, the function of the saccule-vestibular inferior nerve pathway is evaluated; oVEMP: by recording the potential response of the extraocular muscle, the function of the utricle-vestibular superior nerve pathway is evaluated; these detections are all made in advance, and only the result data obtained by detection is analyzed, and then the analysis result is input into the model constructed subsequently for processing;

[0069] The structured data set formed is X_vemp∈R d ; Wherein d is the feature dimension, in this embodiment: d=6, respectively corresponding to cVEMP latency, cVEMP amplitude, oVEMP latency, oVEMP amplitude, left and right amplitude difference ratio and waveform distinguishability score; The structured data set of this part is also used as the input of the subsequent model;

[0070] S4.2, correction of flightworthiness model reconstruction:

[0071] According to the structured data set and the preliminary flightworthiness F1, input into the pre-constructed nonlinear weighted fusion model, output the corrected flightworthiness F2, used for the final decision whether the target flight personnel is recoverable or needs intervention;

[0072] Wherein, input: X_vemp and F1;

[0073] Process: F2=α×F1+(1-α)×M(X_vemp);

[0074] Output: corrected flightworthiness F2;

[0075] It should be noted that α in the above formula is the fusion weight coefficient, the value range is (0, 1), which represents the degree of confidence in the original evaluation result; When the fusion weight coefficient α is set to 0.6, it means that 60% of the preliminary evaluation confidence is retained, and 40% of the new evidence weight is introduced; M(X_vemp) represents a nonlinear mapping function, which is realized by using a multilayer perception machine MLP, used to extract high-order features from X_vemp and output a health adaptation score in the interval [0, 1];

[0076] The above uses a nonlinear weighted fusion model, retains 60% of the original evaluation confidence while introducing 40% of the new evidence weight, which not only maintains the continuity of historical health data, but also enhances the response ability to new functional abnormalities; At the same time, after setting the correction threshold, two types of decision suggestions are generated, realizing the closed-loop management from screening to intervention, solving the problems of high misjudgment rate and insufficient decision basis in traditional health evaluation, which leads to excessive flight suspension or risky release, improving the scientificity and safety of aviation medical management, and embodying the dynamic balance mechanism between risk control and operation guarantee.

[0077] The specific structure of the multi-layer perception machine (MLP) is as follows:

[0078] Input layer: 6 neurons corresponding to 6-dimensional vemp features, i.e., cVEMP latency, cVEMP amplitude, oVEMP latency, oVEMP amplitude, left-right amplitude difference ratio, and waveform recognizability score;

[0079] Hidden layer: 10 neurons, with ReLU as the activation function, for enhancing the non-linear fitting capability;

[0080] wherein ReLU(x) = max(0, x); in the formula, ReLU(x) represents the output value of any neuron, and x represents the sum of the weighted input and the bias term of a neuron in the hidden layer, i.e., the linear transformation result, and the specific form is as follows:

[0081] wherein d is the feature dimension, u i The initial value of the i-th connection weight from the input layer to the neuron in the hidden layer is randomly initialized, and is optimized in the training process through the back propagation algorithm; x i represents the normalized value of the i-th input feature, i = 1, 2,..., d, and in this embodiment, d = 6, so the maximum value of i is 6; b represents the first bias term of the neuron, which is used to adjust the activation threshold and improve the model fitting capability;

[0082] When the input x > 0, the output is equal to x;

[0083] When the input x ≤ 0, the output is equal to 0;

[0084] The ReLU activation function has the characteristics of non-linearity, sparse activation and high efficiency in calculation, which helps the model to learn complex non-linear relationships and at the same time alleviates the gradient vanishing problem; the principle is as follows: the ReLU function maps the linear combination result to a non-negative value, realizes the gating mechanism, and only when the input is strong enough, the signal is transmitted, and the weak signal is suppressed; this simulates the activation characteristics of biological neurons, and enhances the selective response ability of the model to key features;

[0085] Output layer: 1 neuron without activation function, followed by a Sigmoid function to ensure that the output is in the range of [0, 1];

[0086] wherein the expression of the Sigmoid function is: σ(z) = 1 / (1 + e -z); in the formula, z represents the sum of the weighted input and the bias term of the output layer neuron, that is, the weighted sum of the outputs of all neurons in the previous layer (hidden layer) plus the second bias term, and the form of expression is the same as that of the linear transformation result, and details are not described here; e represents the base number of the natural logarithm, which is the base constant of the exponential function; and σ(z) represents the final output value, that is, M(X_vemp), which ranges from 0 to 1, and is used to quantify the ability level of the otolithic apparatus of the target flight personnel to support the flight task; the action logic is as follows: the Sigmoid function maps any real number input z ∈ (−∞, +∞) to the interval (0, 1), so that it has probability semantics, facilitating weighted fusion with F1; for example, if z is positive and very large, the result of the Sigmoid function tends to 1, indicating good function; if z is negative and very small, the result of the Sigmoid function tends to 0, indicating impaired function; for the running logic of the non-linear weighted fusion model, F1 reflects the macroscopic judgment based on flight load and semicircular canal function, and M(X_vemp) provides microscopic evidence of otolithic apparatus-specific function, and through weighted fusion, the continuity of historical evaluation is retained, and new high-specificity data are introduced, thereby improving the robustness of the final decision.

[0087] By introducing the multi-layer perception to perform non-linear mapping on the supplementary data, high-order features such as latency, amplitude, symmetry and waveform quality are extracted, and the recognition sensitivity to otolithic apparatus dysfunction is significantly enhanced, making up for the technical shortcoming that traditional linear statistical methods are difficult to capture complex physiological interaction effects; the MLP output is normalized to the interval [0, 1] through the Sigmoid function, so that it has the same scale as F1, providing a basis for subsequent weighted fusion, solving the problem of unified quantification of multi-modal health evidence, completing the semantic alignment of microscopic physiological data and macroscopic behavior indicators, and embodying the model interpretability and numerical stability design of the scheme in heterogeneous data fusion.

[0088] In summary, by adopting the above structure design, the input layer receives the standardized structured data set X_vemp; the hidden layer uses ReLU to extract non-linear feature combinations; the output layer ensures M(X_vemp) ∈ (0, 1) through Sigmoid; and finally, a non-linear weighted fusion model is adopted to ensure F2 ∈ (0, 1), realizing the scale unification and comparability with the preliminary flight suitability F1; on the one hand, the numerical stability and semantic consistency of the model output are ensured, and on the other hand, the sensitivity to otolithic apparatus dysfunction is enhanced through non-linear mapping, solving the technical problem that traditional linear models are difficult to capture complex physiological interaction effects, and realizing the related technical effects of high precision, interpretability and low misjudgment.

[0089] S4.3, comprehensive decision generation:

[0090] A correction threshold T2 is set, which is compared with the correction fitness F2. When the correction fitness F2 exceeds the correction threshold T2, it is determined that the flight fitness can be restored, and a first decision suggestion is generated. Otherwise, it is determined that intervention is needed, and a second decision suggestion is generated. It should be noted that the decision and management scheme content at this place are only for data analysis, not directly for people, and the given decision is only a suggestion, which can be selected. Whether to adopt or accept part of the suggestion or not can be determined according to actual needs.

[0091] In the embodiment, the correction threshold T2 can be selected as 0.65. In the historical data backtracking analysis, the correction threshold T2 can make the sensitivity and specificity balance point optimal, that is, the Youden index is maximum, and the risk of missing judgment and misjudgment is considered. Therefore, the correction threshold T2 is not described here. The content of the first decision suggestion can be selected as: it is suggested to track and restore the flight task, and the recheck is performed every 30 days or according to the demand. The recheck direction can be selected as: VEMP and basic sign data, etc. The content of the second decision suggestion can be selected as: a medical intervention suggestion is generated, including but not limited to: vestibular rehabilitation training, drug treatment, short-term flight suspension (short-term is 7 days), etc., and the push is completed. By using the scheme of the above embodiments, the mechanism realizes the accurate re-identification of the personnel who do not meet the preliminary evaluation, reduces unnecessary flight suspension, and significantly reduces the overall screening resource consumption through conditional triggering inspection, realizes the function of generating decision suggestions and accurate management on demand.

[0092] The scheme described in embodiment 1 covers a health management system of four levels of data representation, process control, model learning and decision closed loop in turn, forms a complete technical chain from data processing to health management decision, each link depends on each other and realizes coherent support, and together constitutes an intelligent flight personnel health management system.

[0093] Embodiment 2:

[0094] Based on embodiment 1, the embodiment further provides a flight personnel health management system for vestibular function examination abnormality, which comprises: a health collection and processing module: obtaining a multi-source health data set of a target flight personnel, constructing a load characteristic variable, and synchronously storing;

[0095] A preliminary flight fitness evaluation module: according to the load characteristic variable, after performing the screening and normalization processing action, inputting into the pre-constructed linear evaluation model, outputting the preliminary flight fitness, and comparing the preliminary flight fitness with the set flight threshold. When the preliminary flight fitness exceeds the flight threshold, it is determined that the preliminary flight fitness is met and step three is entered. Otherwise, step four is entered.

[0096] The health management monitoring module: for the target flight personnel with the preliminary flight fitness determination result, a periodic data extraction task is started, the trend change rate of the calibration data is detected, when the trend change rate exceeds the early warning amplitude for Q times in succession, a health risk early warning prompt signal is generated, and a check diagnosis action is performed;

[0097] The modified flight fitness evaluation module: supplementary data is acquired and standardized coding is performed to form a structured data set, the structured data set and the preliminary flight fitness are input into a pre-constructed nonlinear weighted fusion model, and a modified flight fitness is output, which is compared with a set modified threshold value, and a corresponding decision suggestion is generated according to the comparison result.

[0098] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software depends on the specific application and design constraints of the technical solutions.

[0099] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiments according to actual needs.

[0100] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for health management of flight personnel with abnormal vestibular function, characterized by, The method comprises the following steps: Step 1: Obtain a multi-source health data set of the target flight personnel, construct a load characteristic variable, and store synchronously; The multi-source health data set at least comprises historical flight data, physical examination data and basic vital sign data; The historical flight data at least comprises cumulative flight time, recent flight frequency and historical average weekly flight frequency; the physical examination data at least comprises peak nystagmus velocity and bilateral asymmetry ratio of cold and hot test; and the basic vital sign data at least comprises resting heart rate, low-to-high frequency ratio and blood oxygen saturation. Step 2: After performing a screening normalization processing action according to the load characteristic variable, input into a pre-constructed linear evaluation model, output a preliminary flight fitness, and compare it with a set flight fitness threshold, if the preliminary flight fitness exceeds the flight fitness threshold, it is determined that the target flight personnel is preliminarily fit to fly and enters step 3; otherwise, it enters step 4; Step 3: For the target flight personnel determined to be preliminarily fit to fly, start a periodic data extraction task, detect the trend change rate of the calibration data, if the trend change rate of consecutive Q times exceeds a warning amplitude, generate a health risk warning prompt signal, and perform a check and diagnosis action; Step 4: Obtain supplementary data and perform standardization coding to form a structured data set, input the structured data set and the preliminary flight fitness into a pre-constructed nonlinear weighted fusion model, output a corrected flight fitness, compare it with a set correction threshold, and generate a corresponding decision suggestion according to the comparison result; the supplementary data is obtained by pre-vestibular muscle source induced potential VEMP examination, and at least comprises cVEMP latency, cVEMP amplitude, oVEMP latency, oVEMP amplitude, left-right amplitude difference ratio and waveform recognizability score; wherein, the left-right amplitude difference ratio = | left amplitude - right amplitude | / average amplitude; the waveform recognizability score corresponds to a scoring standard of 0 = no waveform, 1 = fuzzy, 2 = clear; the structured data set formed is X_vemp ∈ Rd; wherein, d is the feature dimension, corresponding to each supplementary data.

2. The method for health management of flight personnel with abnormal vestibular function according to claim 1, characterized in that: Before constructing the load characteristic variable, it further comprises a data cleaning action, and the data cleaning action at least comprises missing value filling and abnormal value identification and elimination; the process of constructing the load characteristic variable is: constructed according to the multi-source health data set, including flight load index FLI, examination stability coefficient VSC and physiological adjustment coefficient PRI.

3. The method for health management of flight personnel with abnormal vestibular function according to claim 2, characterized in that: The screening normalization processing action performed is: if any type of value of the flight load index FLI, the examination stability coefficient VSC and the physiological adjustment coefficient PRI does not belong to [0, 1], normalization function processing is performed to control the value in the range of [0, 1]; the linear evaluation model adopts a weighted summation method to obtain the preliminary flight fitness F1.

4. The method for health management of flight personnel with abnormal vestibular function according to claim 1, characterized in that: When the trend change rate of consecutive Q times exceeds the warning amplitude, the check and diagnosis action is performed; wherein, Q is a positive integer greater than 0, and the content of the check and diagnosis action performed is to notify the workstation and check the target flight personnel.

5. The method for health management of flight personnel with abnormal vestibular function according to claim 1, characterized in that: The operation basis of the pre-constructed nonlinear weighted fusion model is: F2= a x F1+ (1-a) x M(X_vemp); In the formula, a is a fusion weight coefficient, the value range is (0, 1); M(X_vemp) is a nonlinear mapping function, which is realized by a multilayer perception MLP, used to extract high-order features from X_vemp, and output a health adaptation score in the interval [0, 1], that is, a modified flightworthiness F2.

6. The method for health management of flight personnel with abnormal vestibular function according to claim 5, characterized in that: The architecture of the multilayer perception MLP includes: an input layer: 6 neurons corresponding to 6-dimensional vemp features, that is, supplementary data; A hidden layer: 10 neurons, using a ReLU activation function, for enhancing nonlinear fitting capability; An output layer: 1 neuron, no activation function, connected to a Sigmoid function, for controlling the output in the range [0, 1].

7. The method for health management of flight personnel with abnormal vestibular function according to claim 1, characterized in that: The comparison process between the modified flightworthiness F2 and the set modified threshold T2 is: when the modified flightworthiness F2 exceeds the modified threshold T2, it is determined that: the flight is recoverable, and a first decision suggestion is generated; otherwise, it is determined that: intervention is needed, and a second decision suggestion is generated; Wherein, the content of the first decision suggestion is: to suggest tracking the recovery flight task, and to review once according to the demand, and the review direction is selected as: VEMP and basic sign data; The content of the second decision suggestion is: to generate a medical intervention suggestion, including but not limited to: vestibular rehabilitation training, drug treatment and short-term flight suspension, and to complete the push.

8. A health management system for flight personnel with abnormal vestibular function examination, characterized in that, It includes: Health acquisition and processing module: acquire a multi-source health data set of the target flight personnel, construct load characteristic variables, and store synchronously; The multi-source health data set at least includes: historical flight data, physiological examination data and basic sign data; Wherein, the historical flight data at least includes: cumulative flight time, recent flight frequency and historical average weekly flight frequency; The physiological examination data at least includes: peak nystagmus velocity of cold and hot test and bilateral asymmetry ratio; The basic sign data at least includes: resting heart rate, low-to-high frequency ratio and blood oxygen saturation; Preliminary flightworthiness evaluation module: after performing screening and normalization processing actions according to the load characteristic variables, input into the pre-constructed linear evaluation model, output the preliminary flightworthiness, and compare it with the set flight threshold, when the preliminary flightworthiness exceeds the flight threshold, it is determined that the preliminary flight is suitable and enters the health management monitoring module; Otherwise, it enters the modified flightworthiness evaluation module; Health management monitoring module: for the target flight personnel whose determination result is preliminary flight, start a periodic data extraction task, detect the trend change rate of the calibration data, when the trend change rate exceeds the warning amplitude for Q consecutive times, generate a health risk warning prompt signal, and perform an inspection and diagnosis action; The modified flightworthiness evaluation module: obtains supplementary data and carries out standardized coding, forms a structured data set, inputs the structured data set and the preliminary flightworthiness into a pre-constructed nonlinear weighted fusion model, outputs the modified flightworthiness, compares the modified flightworthiness with a set of modified threshold values, and generates corresponding decision suggestions according to the comparison results; the supplementary data is obtained by pre-vestibular myogenic evoked potential VEMP examination, and at least includes: cVEMP latency, cVEMP amplitude, oVEMP latency, oVEMP amplitude, left and right amplitude difference ratio and waveform recognizability score; wherein, left and right amplitude difference ratio = | left amplitude - right amplitude | / average amplitude; the waveform recognizability score corresponds to the scoring standard: 0 = no waveform, 1 = fuzzy, 2 = clear; the structured data set formed is X_vemp ∈ Rd; wherein, d is the feature dimension, corresponding to each supplementary data.

Images