A non-invasive human core temperature prediction method and system based on deep learning
Through deep learning technology, different conditions are simulated in the experimental chamber of artificial environments, regression models are constructed and Kalman filters and long-term prediction networks are used to achieve non-invasive measurement of human core temperature, solving the high invasiveness and inaccurate measurement in the existing technology. Problem, improving the accuracy and applicability of measurements.
Patent Information
- Application Number
- CN202411278717.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-12-27
- Estimated Expiration
- 2044-09-12
AI Technical Summary
The prior art has problems such as high invasiveness, high inapplicability and inaccurate measurement results when measuring the temperature of the human body core, especially in dynamic monitoring environments.
Using a non-invasive human core temperature prediction method based on deep learning, a regression model between skin temperature, heart rate and core temperature is constructed through artificial environment experimental chambers, and data is performed using Kalman filters and long-term prediction networks. Correct and predict to achieve core temperature measurements without direct invasion of the human body.
It improves the safety and comfort of core temperature measurement, reduces the difficulty of measurement, enhances the accuracy and applicability of measurement results, and supports human heat stress warning and auxiliary decision-making.
Smart Images
Figure CN118797574B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of intelligent protection technology, and in particular to a non-invasive human core temperature prediction method and system based on deep learning. Background Art
[0002] Continuous monitoring of human core temperature is essential for assessing the body's heat stress response during strenuous work. Although personal physiological monitoring technology has advanced to the point where multi-parameter sensor systems can be used to collect data over a long period of time in a variety of environments, accurately measuring core body temperature remains a challenge.
[0003] Currently, medical-grade core temperature measurement using pulmonary artery temperature is limited to clinical use. The rectal and esophageal probe methods traditionally used in laboratories are not sufficient for environments that require dynamic monitoring. In contrast, the use of core temperature capsules has higher effectiveness and practicality in actual working environments, but this method still has certain limitations: first, due to medical contraindications, core temperature capsules are not suitable for everyone; second, the measurement results may be inaccurate after drinking too cold or too hot liquids.
[0004] Therefore, there is an urgent need to provide a technical solution to the above-mentioned deficiencies in the prior art. Summary of the invention
[0005] The purpose of this application is to provide a non-invasive human core temperature prediction method and system based on deep learning to solve or alleviate the problems existing in the above-mentioned prior art.
[0006] In order to achieve the above objectives, this application provides the following technical solutions:
[0007] The present application provides a non-invasive human core temperature prediction method based on deep learning, including: step S101, based on an artificial environment test chamber, conducting a warm manikin experiment in a simulated environment, constructing a human thermal response data set to invert the regression model between the skin simulated temperature, simulated heart rate and core simulated temperature in the human thermal response data set; step S102, based on the regression model, automatically adjusting the model parameters in the Kalman filter through a deep learning algorithm; step S103, based on the Kalman filter after the model parameters are adjusted, optimizing and correcting the measured data of the target individual collected in real time to obtain the measured corrected data of the target individual; wherein the measured data includes skin temperature and heart rate data, and the measured corrected data includes skin temperature correction data and heart rate correction data; step S104, according to the heart rate correction data and skin temperature correction data of the target individual, based on the core temperature prediction model of the long time series prediction network, predicting the core temperature of the target individual.
[0008] Preferably, in step S101, based on the human body thermal response data set, a regression model between the core simulated temperature and the simulated heart rate, the skin simulated temperature, and the environmental simulation parameters is inverted:
[0009]
[0010] In the formula, is the simulated temperature of the core, A simulated temperature for the skin, is the simulated heart rate, is the simulated metabolic rate of the human body; Simulating parameters for the environment; are inversion coefficients.
[0011] Preferably, in step S102, the parameters in the Kalman filter are automatically adjusted based on the regression model through a deep learning algorithm, including: constructing a sample data set of the Kalman filter based on the regression model; wherein the sample data set includes: skin sample temperature, sample heart rate and core sample temperature; according to the sample data set, based on a deep learning algorithm, the model parameters in the Kalman filter are automatically adjusted by minimizing the prediction error of the Kalman filter function.
[0012] Preferably, in step S103, according to the formula:
[0013]
[0014] Determine the measured correction data of the target individual ;
[0015] In the formula, The target individual at the current moment The prior state estimate of is the state transfer matrix, For the target individual at time The posterior state estimate of is the control matrix, is the control input; The target individual at the current moment The posterior state estimate of is the Kalman gain, For the current moment The measured data, is the observation model matrix; is the posterior estimated covariance matrix, is the a priori estimated covariance matrix; is the covariance matrix of the observation noise.
[0016] Preferably, in step S104, the measured corrected data is converted into an embedding vector through an embedding layer as an input feature sequence for the core temperature prediction model, and based on the ProbSparse self-attention mechanism, the attention weight of the input feature sequence is calculated to predict the core temperature of the target individual.
[0017] Preferably, the method of converting the measured corrected data into an embedded vector through an embedding layer includes: performing data cleaning on the measured corrected data; performing feature selection on the cleaned data, and performing encoding conversion on the feature selected data to obtain numerical encoded data; initializing an embedding matrix based on the embedding layer of the long time series prediction network, and converting the encoded data into the embedded vector.
[0018] Preferably, the encoding conversion of the feature selected data includes: encoding conversion of the feature selected data based on one-hot encoding or label encoding.
[0019] Preferably, in step S104, according to the formula:
[0020]
[0021] Calculate the attention weight of the input feature sequence ; In the formula, is the query matrix of the input feature sequence, is the key matrix that matches the query matrix, is the value matrix of the input feature sequence, is the dimension of the input feature sequence; wherein, according to the input feature sequence, the query matrix is constructed based on the sparsity measurement .
[0022] Preferably, constructing the query matrix based on the sparsity measurement according to the input feature sequence includes, according to the formula:
[0023]
[0024] Determine the first query vector Sparsity measurement of ;
[0025] Wherein, the query vector Through the The input feature sequence is linearly transformed to obtain; For the bond matrix The number of vectors in , , are all positive integers; For the bond matrix Middle vectors.
[0026] The embodiment of the present application also provides a non-invasive human core temperature prediction system based on deep learning, including: a simulation regression unit, configured to conduct a warm body family experiment in a simulated environment based on an artificial environment experimental cabin, to construct a human thermal response data set, to invert a regression model between the skin simulation temperature, the simulated heart rate and the core simulation temperature in the human thermal response data set;
[0027] A filter parameter adjustment unit, configured to automatically adjust model parameters in the Kalman filter through a deep learning algorithm based on the regression model;
[0028] The target data correction unit is configured to optimize and correct the measured data of the target individual collected in real time based on the Kalman filter after the model parameters are adjusted, so as to obtain the measured corrected data of the target individual; wherein the measured data includes skin temperature and heart rate data, and the measured corrected data includes skin temperature corrected data and heart rate corrected data;
[0029] The core temperature prediction unit is configured to predict the core temperature of the target individual based on the heart rate correction data and skin temperature correction data of the target individual and the core temperature prediction model of the long time series prediction network. Beneficial Effects
[0030] In the non-invasive human core temperature prediction method based on deep learning provided in the embodiment of the present application, first, a warm manikin experiment is conducted in a simulated environment based on an artificial environment test chamber to construct a human thermal response data set to invert the regression model between the skin simulated temperature, simulated heart rate and core simulated temperature in the human thermal response data set; and based on the regression model, the model parameters in the Kalman filter are automatically adjusted through a deep learning algorithm; then, based on the Kalman filter after the model parameters are adjusted, the measured data of the target individual collected in real time are optimized and corrected, and based on the measured corrected data of the target individual obtained by the optimized correction, the core temperature prediction model of the long time series prediction network is used to predict the core temperature of the target individual.
[0031] Therefore, through simulation experiments in artificial environment test cabins, physiological simulation data of the human body under different physiological states and environmental conditions are collected, and a regression model between skin temperature, heart rate and core temperature is established based on this. The regression model is used to calculate a sample data set that can comprehensively cover different ages, genders, environments and other conditions, so that the model parameters of the Kalman filter are automatically adjusted through a deep learning algorithm according to the sample data set, so that the Kalman filter after the model parameters are adjusted has a wider applicability and higher accuracy; the measured skin temperature and measured heart rate data of the target individual are corrected with the help of a Kalman filter with wide applicability and higher accuracy, and the core temperature of the target individual is predicted based on the long time series prediction grid through the corrected skin temperature and heart rate data, effectively avoiding invasive measurement of the human body when obtaining the human core temperature, and realizing accurate prediction of the human core temperature through the human skin temperature and heart rate without directly invading the human body, thereby reducing the difficulty of measuring the human core temperature, so as to support human heat stress early warning response and auxiliary decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] The drawings constituting part of the present application are used to provide a further understanding of the present application. The exemplary embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. Among them:
[0033] Figure 1 A schematic flow chart of a non-invasive human core temperature prediction method based on deep learning according to some embodiments of the present application;
[0034] Figure 2 A technical roadmap for a non-invasive human core temperature prediction method based on deep learning according to some embodiments of the present application;
[0035] Figure 3 A schematic diagram of the model structure of the long time series prediction network Informer provided according to some embodiments of the present application;
[0036] Figure 4 A comparative schematic diagram of core temperature prediction provided according to some embodiments of the present application;
[0037] Figure 5 This is a schematic diagram of the structure of a non-invasive human core temperature prediction system based on deep learning according to some embodiments of the present application. DETAILED DESCRIPTION
[0038] The present application will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments. Each example is provided by way of explanation of the present application and does not limit the present application. In fact, it will be clear to those skilled in the art that modifications and variations may be made in the present application without departing from the scope or spirit of the present application. For example, a feature shown or described as a part of an embodiment may be used in another embodiment to produce yet another embodiment. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the embodiments of the present invention should belong to the scope of protection of the embodiments of the present invention.
[0039] At present, the measurement of human core temperature (such as pulmonary artery temperature measurement, rectal and esophageal probe method) cannot avoid intrusion into the human body, and is affected by medical contraindications or subjective and objective factors of the target individual, and the validity of the measurement results cannot be guaranteed. Based on this, this application proposes a non-invasive human core temperature prediction method based on deep learning, which covers the data variability of various physiological and environmental conditions, and optimizes and adjusts the Kalman filter through deep learning. Then, the Kalman filter is used to update and correct the skin temperature and heart rate of the target individual in real time, and the core temperature of the target individual is predicted using a long time series prediction network. Compared with traditional core temperature measurement, no invasive operation is required, which improves the safety and comfort of human core temperature measurement.
[0040] like Figures 1 to 4 As shown, the non-invasive human core temperature prediction method based on deep learning includes:
[0041] Step S101: Conduct a warm manikin experiment in a simulated environment based on an artificial environment test chamber to construct a human thermal response data set to invert a regression model between skin simulated temperature, simulated heart rate and core simulated temperature in the human thermal response data set.
[0042] In this application, the artificial environment test chamber can simulate and adjust the environmental conditions such as temperature, humidity, wind speed, gas pressure, etc. by artificial environment test chamber, and simulate various environmental climate conditions to provide a comparative test environment for the thermal manikin experiment. The thermal manikin is a simulation body that simulates the energy exchange between the human body and the environment. By setting different parameters of the thermal manikin, different genders, ages, physiological states, etc. of the human body can be simulated.
[0043] With the help of artificial environment experiment chambers and warm-body manikin experiments, different physiological and environmental conditions are simulated, and physiological simulation data covering different environmental conditions, different ages, different genders, and different physiological states are collected. Through the diverse distribution of demographic characteristics, the diversity and representativeness of the simulation data are ensured, effectively enhancing the wide applicability and high accuracy of the model.
[0044] Core temperature is usually the temperature inside the human body, such as the temperature of the chest, abdomen, and central nervous system. Maintaining core temperature is a necessary condition to ensure the normal metabolism and life activities. Skin temperature is the temperature of the body surface. In daily life, the body surface temperature is usually measured, such as the temperature of the armpit, mouth, forehead, etc. These temperatures are affected by the surrounding environment and personal factors and cannot always accurately represent the core temperature of the human body. Changes in heart rate are an indicator of changes in body temperature. For example, when exercising or the outside temperature rises, the body's metabolic activity increases, heat production increases, and the heart rate may increase; conversely, in a cold environment, in order to reduce heat loss, the heart rate may decrease.
[0045] Based on this, in this application, after obtaining physiological simulation data of the human body under different physiological states and environmental conditions through artificial environment test chambers and warm manikin experiments, a regression model between skin temperature, heart rate, and core temperature is established based on the inversion calculation method. Specifically, based on the human body thermal response data set, a regression model between the core simulation temperature and the simulated heart rate, skin simulation temperature, and environmental simulation parameters is inverted.
[0046]
[0047] In the formula, is the core simulated temperature, Simulate temperature for the skin, To simulate heart rate, To simulate the metabolic rate, ; is the environmental simulation parameter; are inversion coefficients.
[0048] in, is the simulated metabolic rate for men, is the simulated metabolic rate of women. Here, according to the formula:
[0049]
[0050]
[0051] Determine the simulated metabolic rate for men separately and simulated metabolic rates in women In this way, the constructed regression model can cover people of different ages, genders, physiological states, and environmental conditions, and has a wider applicability.
[0052] Step S102: Based on the regression model, the model parameters in the Kalman filter are automatically adjusted through a deep learning algorithm.
[0053] In this application, a sample data set of the Kalman filter is constructed based on the regression model. Among them, the sample data set includes: skin sample temperature, sample heart rate and core sample temperature. Therefore, through simulation experiments in artificial environment test chambers, physiological simulation data covering human bodies under different physiological states and environmental conditions are collected, and a regression model between skin temperature, heart rate and core temperature is established accordingly. The regression model is used to calculate sample data sets that can comprehensively cover conditions of different ages, genders, environments, etc., so as to automatically adjust the model parameters of the Kalman filter through a deep learning algorithm according to the sample data set.
[0054] Specifically, according to the sample data set, based on the deep learning algorithm, the Kalman filter function is minimized by the prediction error method to automatically adjust the model parameters in the Kalman filter. Here, the model parameters of the Kalman filter can be optimized by maximum likelihood estimation, adaptive filtering, Bayesian optimization, genetic algorithm, gradient descent method, etc.
[0055] In this application, the system state of the Kalman filter is recursively estimated by using a machine learning method and the powerful fitting ability of a neural network. The change of the system state is described by a linear equation, and the noise in the observed data is quantified by a covariance matrix. By continuously updating the process noise covariance matrix Q and the observation noise covariance matrix R, the model parameters of the Kalman filter are optimized and adjusted to adapt to changing environmental conditions and different individual differences, so that the Kalman filter after the model parameters are adjusted has a wider applicability and higher accuracy.
[0056] In this application, after creating the Kalman filter, a loss function is defined to measure the prediction accuracy of the Kalman filter. The loss function uses the average value of the sum of squares of the differences between the predicted value and the true value (i.e., the mean square error, ) to measure. Among them,
[0057]
[0058] In the formula, is the number of samples in the sample data set, , are all positive integers; The first The core sample temperature, is the value predicted by the Kalman filter and The corresponding core temperature prediction value.
[0059] In the process of optimizing the model parameters of the Kalman filter, the minimize function from the scipy.optimize library is used to perform the parameter optimization task. Specifically, the minimize function takes the loss function as the optimization target and iteratively adjusts the model parameters of the Kalman filter to find the parameter set that minimizes the objective function value. Here, an automatic differentiation library (such as NumPy or TensorFlow in Python) is used to calculate the function gradient and the Hirsch matrix to accurately estimate the change in the gradient or function value.
[0060] Step S103: Based on the Kalman filter after the model parameters are adjusted, the measured data of the target individual collected in real time is optimized and corrected to obtain the measured corrected data of the target individual.
[0061] The model parameters of the Kalman filter are optimized and adjusted through a deep learning algorithm, making the adjusted Kalman filter more widely applicable and more accurate. It can make targeted corrections to the measured data of target individuals under the same age, gender, environment and other conditions, so as to avoid the influence of subjective and objective factors (such as drinking too cold or too hot liquids) when predicting the core temperature of the target individual, thereby improving the accuracy of the core temperature prediction of the target individual.
[0062] The skin temperature of the target individual is measured by contact or non-contact technology, and the heart rate data of the target individual is measured by an optical sensor or a heart rate monitor; then, the measured skin temperature and heart rate data of the target individual are input into a Kalman filter after the model parameters are adjusted, and the Kalman filter after the model parameters are adjusted performs optimized correction on the input skin temperature and heart rate data of the target individual to obtain skin temperature corrected data and heart rate corrected data of the target individual.
[0063] Specifically, according to the formula:
[0064]
[0065] Determine the measured correction data of the target individual ;
[0066] In the formula, For the target individual at the current moment The prior state estimate of is the state transfer matrix, For the target individual at time The posterior state estimate of is the control matrix, is the control input; For the target individual at the current moment The posterior state estimate of is the Kalman gain, For the current moment The measured data, is the observation model matrix; is the posterior estimated covariance matrix, is the a priori estimated covariance matrix; is the covariance matrix of the observation noise.
[0067] Here, the Kalman filter after adjusting the model parameters predicts the current state based on the state estimate of the previous moment, and uses the current observation data, combined with the prior state estimate, to adjust the predicted value through the Kalman gain, thereby obtaining the posterior state estimate, that is, the measured corrected data. In this process, each time step is repeated to achieve recursive estimation and update of the system state.
[0068] Step S104: predicting the core temperature of the target individual based on the core temperature prediction model of the long time series prediction network according to the heart rate correction data and skin temperature correction data of the target individual.
[0069] In this application, a core temperature prediction model based on the long time series prediction model - Informer is constructed. The prediction time step of the core temperature prediction model is set to 0, that is, no advance prediction of the core temperature is performed, and only the real-time core temperature is predicted. In the core temperature prediction model training phase, the data in the sample data set is used for training, and the prediction performance of the training model is tested, and the parameters of the core temperature prediction model are adjusted. Finally, the parameters of the core temperature prediction model are set as follows: the initial number of iterations (epoch) is set to 20, the patience is set to 3, the activation function is GELU, the initial learning rate is 0.00001, the loss function is the mean square error (MSE), the token length is 0, the input sequence length is 20, the prediction sequence length is 0, the number of attention samples is 5, the model dimension is 512, the number of multi-head attention heads is 8, the number of encoder layers is 2, and the number of decoder layers is 1.
[0070] The heart rate correction data and skin temperature correction data of the target individual are obtained as the input of the core temperature prediction model based on the long time series prediction network. Here, the input measured correction data is converted into an embedding vector through the embedding layer as the input feature sequence of the core temperature prediction model. Specifically, first, the input measured correction data is cleaned, such as missing values and outliers, to ensure the data quality of the input data; then, feature selection is performed on the cleaned data to select a feature subset from the input data that is useful for predicting the core temperature, thereby improving the performance of the core temperature prediction model and reducing the computational complexity. For example, feature selection is performed on the data through the correlation coefficient method, the chi-square test method, the principal component analysis method, the recursive feature elimination method, etc.
[0071] Next, the feature selected data is converted to convert the input discrete data into numerical data. Specifically, the encoding conversion of the feature selected data is implemented based on one-hot encoding or label encoding. Here, the categorical variables in the feature selected data are converted, and each category in the categorical variable is created as a one-hot variable, and the categorical variables are converted using the one-hot encoding method, such as using the get_dummies function of the pandas library, or the OneHotEncoder class of the scikit-learn library. By repeating the above process for each categorical variable in the data set, a set of one-hot encoding columns is generated for each variable, and the converted one-hot encoding columns are merged with other numerical features in the original data set (feature selected data), and after merging the data, the original categorical variable columns are deleted from the data set to avoid multicollinearity problems in the data. Thus, through encoding conversion, the data format is effectively ensured to be suitable for data analysis of long time series prediction networks.
[0072] Finally, based on the embedding layer of the long time series prediction network, the embedding matrix is initialized to convert the encoded data into an embedding vector. The rows of the embedding matrix correspond to the number of different features in the encoded data, and the columns of the embedding matrix correspond to the embedding dimension of each feature in the encoded data.
[0073] After converting the measured and corrected data into an embedding vector, it is used as the input feature sequence of the core temperature prediction model, and based on the ProbSparse self-attention mechanism, the attention weight of the input feature sequence is calculated to predict the core temperature of the target individual.
[0074] The attention weight of the input feature sequence determines the contribution of each input sequence element (skin temperature corrected data and heart rate corrected data) to the output element (core temperature). Here, the query matrix (sparse matrix) constructed based on the sparsity measurement evaluates the distribution and proportion of non-zero elements (or large weights) in the assistant doctor weight matrix. Furthermore, through the sparsity measurement, the important parts of the attention weight are identified, thereby improving the computational efficiency. Specifically, according to the formula:
[0075]
[0076] Determine the query matrix The query vector Sparsity measurement of . Among them, the query vector Through the The input feature sequence is linearly transformed; is the key matrix The number of vectors in , , are all positive integers; is the key matrix Middle vectors.
[0077] When calculating the attention weight, the value matrix of the input feature sequence is , query matrix of input feature sequence and the key matrix that matches the query matrix Calculate the attention weight of the input feature sequence together. Specifically, by querying the matrix and key matrix The dot product calculation determines the attention score of the input feature sequence , as shown below:
[0078]
[0079] Then, through The attention score of the function on the input feature sequence Normalize and get the attention weight of the input feature sequence. Specifically, according to the formula:
[0080]
[0081] Calculate the attention weight of the input feature sequence In the formula, is the query matrix of the input feature sequence, is the key matrix that matches the query matrix, is the value matrix of the input feature sequence, is the dimension of the input feature sequence.
[0082] Most importantly, the attention weight The corresponding value matrix The weighted sum is performed, and the output value is the core temperature of the target individual. In this way, the core temperature of the target individual is predicted based on the long-term prediction grid through the corrected skin temperature and heart rate data, effectively avoiding invasive measurement of the human body when obtaining the human core temperature, and achieving accurate prediction of the human core temperature through the human skin temperature and heart rate without direct invasion of the human body, reducing the difficulty of measuring the human core temperature, so as to support the human heat stress warning response and auxiliary decision-making.
[0083] In this application, the core temperature of the human body is predicted by a deep learning model, which can comprehensively consider the influence of multiple factors, and adapt to individual differences and environmental changes through continuous learning, efficiently integrate and process multi-source physiological data, ensure the accuracy and reliability of the model, so that the model has the ability to predict the core temperature in a complex and dynamic environment, and predict the core temperature of the human body from multiple physiological parameters such as skin temperature and heart rate measured non-invasively. At the same time, through simulation experiments in artificial environment test chambers, physiological simulation data covering human bodies under different physiological states and environmental conditions are collected, and a regression model between skin temperature, heart rate, and core temperature is established based on this. The regression model is used to calculate sample data sets that can comprehensively cover different ages, genders, environments, and other conditions, so that the model parameters of the Kalman filter can be automatically adjusted through a deep learning algorithm based on the sample data set, so that the Kalman filter after the model parameters are adjusted has a wider applicability and higher accuracy.
[0084] The present application also provides a non-invasive human core temperature prediction system based on deep learning. Figure 5 As shown, including:
[0085] The simulation regression unit 501 is configured to perform a warm body experiment in a simulated environment based on an artificial environment experiment cabin, construct a human thermal response data set, and invert a regression model between the skin simulation temperature, the simulated heart rate, and the core simulation temperature in the human thermal response data set;
[0086] A filter parameter adjustment unit 502 is configured to automatically adjust model parameters in the Kalman filter through a deep learning algorithm based on the regression model;
[0087] The target data correction unit 503 is configured to optimize and correct the measured data of the target individual collected in real time based on the Kalman filter after the model parameters are adjusted, so as to obtain the measured corrected data of the target individual; wherein the measured data includes skin temperature and heart rate data, and the measured corrected data includes skin temperature corrected data and heart rate corrected data;
[0088] The core temperature prediction unit 504 is configured to predict the core temperature of the target individual based on the heart rate correction data and skin temperature correction data of the target individual and the core temperature prediction model of the long time series prediction network.
[0089] The non-invasive human core temperature prediction system based on deep learning provided in the embodiments of the present application can implement the steps and processes of the non-invasive human core temperature prediction method based on deep learning in any of the above embodiments and achieve the same technical effects, which will not be repeated here one by one.
[0090] In the description of the present invention, the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0091] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A non-invasive human core temperature prediction method based on deep learning, characterized in that: include: Step S101, conducting a warm manikin experiment in a simulated environment based on an artificial environment test chamber, constructing a human thermal response data set to invert a regression model between the simulated skin temperature, simulated heart rate and simulated core temperature in the human thermal response data set; Step S102: Based on the regression model, automatically adjust the model parameters in the Kalman filter through a deep learning algorithm; Step S103: Based on the Kalman filter after the model parameters are adjusted, the measured data of the target individual collected in real time is optimized and corrected according to the formula: ； Determine the measured correction data of the target individual ; In the formula, The target individual at the current moment The prior state estimate of is the state transfer matrix, For the target individual at time The posterior state estimate of is the control matrix, is the control input; The target individual at the current moment The posterior state estimate of is the Kalman gain, For the current moment The measured data, is the observation model matrix; is the observation model matrix The transposed matrix of is the posterior estimated covariance matrix, is the a priori estimated covariance matrix; is the covariance matrix of the observation noise; Wherein, the measured data includes skin temperature and heart rate data, and the measured corrected data includes skin temperature corrected data and heart rate corrected data; Step S104: predicting the core temperature of the target individual based on the core temperature prediction model of the long time series prediction network according to the heart rate correction data and skin temperature correction data of the target individual.
2. The non-invasive human core temperature prediction method based on deep learning according to claim 1, characterized in that: In step S101, According to the human body thermal response data set, the regression model between the core simulated temperature and the simulated heart rate, the skin simulated temperature, and the environmental simulation parameters is inverted: ； In the formula, is the simulated temperature of the core, A simulated temperature for the skin, is the simulated heart rate, is the simulated metabolic rate of the human body; Simulating parameters for the environment; are inversion coefficients.
3. The non-invasive human core temperature prediction method based on deep learning according to claim 2, characterized in that: In step S102, automatically adjusting the parameters in the Kalman filter by a deep learning algorithm based on the regression model includes: Based on the regression model, a sample data set of the Kalman filter is constructed; wherein the sample data set includes: skin sample temperature, sample heart rate and core sample temperature; According to the sample data set, based on a deep learning algorithm, the model parameters in the Kalman filter are automatically adjusted by minimizing the prediction error of the Kalman filter function.
4. The non-invasive human core temperature prediction method based on deep learning according to claim 1, characterized in that: In step S104, The measured corrected data is converted into an embedding vector through an embedding layer as an input feature sequence for the core temperature prediction model, and the attention weight of the input feature sequence is calculated based on the ProbSparse self-attention mechanism to predict the core temperature of the target individual.
5. The non-invasive human core temperature prediction method based on deep learning according to claim 4, characterized in that: The step of converting the measured correction data into an embedding vector through an embedding layer includes: Performing data cleaning on the measured correction data; Performing feature selection on the cleaned data, and performing encoding conversion on the feature-selected data to obtain numerically encoded data; Based on the embedding layer of the long time series prediction network, an embedding matrix is initialized, and the encoded data is converted into the embedding vector.
6. The non-invasive human core temperature prediction method based on deep learning according to claim 5, characterized in that: The encoding conversion of the feature selected data includes: encoding conversion of the feature selected data based on one-hot encoding or label encoding.
7. The non-invasive human core temperature prediction method based on deep learning according to claim 4, characterized in that: In step S104, according to the formula: ； Calculate the attention weight of the input feature sequence ; In the formula, is the query matrix of the input feature sequence, is the key matrix that matches the query matrix, is the value matrix of the input feature sequence, is the dimension of the input feature sequence; According to the input feature sequence, the query matrix is constructed based on the sparsity measurement. .
8. The non-invasive human core temperature prediction method based on deep learning according to claim 7, characterized in that: The constructing the query matrix based on the sparsity measurement according to the input feature sequence includes: ； Determine the first query vector Sparsity measurement of ; Wherein, the query vector Through the The input feature sequence is linearly transformed to obtain; For the bond matrix The number of vectors in , , All are positive integers; For the bond matrix Middle vectors.
9. A non-invasive human core temperature prediction system based on deep learning, characterized in that: include: A simulation regression unit is configured to conduct a warm body experiment in a simulated environment based on an artificial environment experiment cabin, construct a human thermal response data set, and invert a regression model between the skin simulation temperature, the simulated heart rate, and the core simulation temperature in the human thermal response data set; A filter parameter adjustment unit, configured to automatically adjust model parameters in the Kalman filter through a deep learning algorithm based on the regression model; The target data correction unit is configured as a Kalman filter based on the adjusted model parameters to optimize and correct the measured data of the target individual collected in real time according to the formula: ； Determine the measured correction data of the target individual ; In the formula, The target individual at the current moment The prior state estimate of is the state transfer matrix, For the target individual at time The posterior state estimate of is the control matrix, is the control input; The target individual at the current moment The posterior state estimate of is the Kalman gain, For the current moment The measured data, is the observation model matrix; is the observation model matrix The transposed matrix of is the posterior estimated covariance matrix, is the a priori estimated covariance matrix; is the covariance matrix of the observation noise; wherein the measured data includes skin temperature and heart rate data, and the measured corrected data includes skin temperature corrected data and heart rate corrected data; The core temperature prediction unit is configured to predict the core temperature of the target individual based on the heart rate correction data and skin temperature correction data of the target individual and the core temperature prediction model of the long time series prediction network.