A Human Skin Temperature Prediction Method and System Integrating Informer and STGCN

Abstract

The present application relates to the technical field of human body temperature prediction, and provides a human skin temperature prediction method and system that integrates Informer and STGCN. According to the topological structure among multiple measurement points of the skin temperature data of the target object obtained, based on the spatio-temporal graph convolutional network, the spatio-temporal features of the measurement points are extracted, and based on the first Informer module with self-attention mechanism, according to the data matrix of the skin temperature data and the data matrix of the obtained environmental parameter data respectively, the temperature time series features and environmental time series features of the measurement points are extracted respectively; then, the spatio-temporal features, temperature time series features, and environmental time series features of the measurement points are fused in the second Informer module with self-attention mechanism according to the time series, and the obtained feature matrix is input into the fully connected neural network to predict the skin temperature of the target measurement point.

Description

Technical Field

[0001] This application relates to the technical field of human body temperature prediction, and particularly to a human skin temperature prediction method and system that integrates Informer and STGCN. Background Art

[0002] When the ambient temperature is too low, it is easy to cause damage to the human skin (such as frostbite). In a low-temperature environment, the blood vessels in the human skin are prone to spasms, causing tissue damage and thus frostbite; at the same time, the peripheral blood vessels of the human body will constrict, causing an increase in blood resistance and blood pressure, leading to an increase in the heart load, and the probabilities of cerebral hemorrhage and myocardial infarction will also increase significantly. Therefore, in an extremely low-temperature environment, there is a situation of increasing the frostbite risk for outdoor workers, posing a major threat to the life safety and health of the human body.

[0003] Therefore, there is an urgent need to provide a technical solution to address the deficiencies of the above-mentioned existing technologies. Summary of the Invention

[0004] The purpose of this application is to provide a human skin temperature prediction method and system that integrates Informer and STGCN to solve or alleviate the problems existing in the above-mentioned existing technologies.

[0005] To achieve the above purpose, this application provides the following technical solutions:

[0006] This application provides a human skin temperature prediction method that integrates Informer and STGCN, including: Step S101, based on the topological structure between multiple measurement points corresponding to the skin temperature data of the target object, and based on the spatio-temporal graph convolutional network, extract the spatio-temporal features of the measurement points;

[0007] And, based on the first Informer module with self-attention mechanism, respectively according to the data matrix of the skin temperature data and the data matrix of the environmental parameter data corresponding to the skin temperature data obtained , to respectively extract the temperature time-series features , environmental time-series features of the measurement points;

[0008] Step S102, fuse the spatio-temporal features , the temperature time-series features , and the environmental time-series features in sequence in the second Informer module with self-attention mechanism, and input the obtained feature matrix into a fully connected neural network to predict the skin temperature of the measurement points of the target object.

[0009] Preferably, in step S101, for the adjacency matrix constructed based on the topological structure between multiple measurement points corresponding to the skin temperature data of the target object perform normalization processing to obtain a normalized adjacency matrix , and based on graph convolution operation, according to the formula:

[0010]

[0011] determine the skin temperature feature matrix of the measurement point at time to construct the spatio-temporal feature of the measurement point ; ;

[0012] In the formula, is the ReLU activation function; is the skin temperature feature matrix of the target object at time, is the weight matrix corresponding to the skin temperature feature matrix ;

[0013] Preferably, in step S101, according to the formula:

[0014]

[0015] extract the temperature temporal feature and environmental temporal feature of the measurement point;

[0016] In the formula, is the data matrix of the skin temperature data, is the weight matrix of the query matrix corresponding to the data matrix of the skin temperature data, is the weight matrix of the key matrix corresponding to the data matrix of the skin temperature data, is the weight matrix of the value matrix corresponding to the data matrix of the skin temperature data; is the vector dimension in the key matrix corresponding to the data matrix of the skin temperature data;

[0017] is the data matrix of the environmental parameter data corresponding to the skin temperature data; is the weight matrix of the query matrix corresponding to the data matrix of the environmental parameter data corresponding to the skin temperature data, is the data matrix of the environmental parameter data corresponding to the skin temperature data The weight matrix of the corresponding key matrix is the data matrix of the environmental parameter data corresponding to the skin temperature data The weight matrix of the corresponding value matrix is the data matrix of the environmental parameter data corresponding to the skin temperature data The vector dimension in the corresponding key matrix

[0018] Preferably, in step S102, the spatial-temporal features and temperature-temporal features and environmental-temporal features of the measurement points are respectively calculated with the Pearson correlation coefficient with the skin temperature data, and according to the Pearson correlation coefficient the importance ratios of the spatial-temporal features and temperature-temporal features and environmental-temporal features of the measurement points are calculated

[0019] According to the importance ratios of the spatial-temporal features and temperature-temporal features and environmental-temporal features of the measurement points, the spatial-temporal features and temperature-temporal features and environmental-temporal features of the measurement points are sorted by importance to construct a judgment matrix ;

[0020] According to the judgment matrix the weight vectors of the spatial-temporal features and temperature-temporal features and environmental-temporal features of the measurement points are respectively calculated to fuse the features and temperature-temporal features and environmental-temporal features of the measurement points to obtain a feature matrix .

[0021] Preferably, according to the formula:

[0022]

[0023] the importance ratios of the spatial-temporal features and temperature-temporal features and environmental-temporal features of the measurement points are calculated and the judgment matrix is constructed; where , are all positive integers, is the diagonal element of the judgment matrix representing the importance ratio of the -th time series feature to the -th time series feature of the measurement point; representing the importance ratio of the -th time series feature to the -th time series feature of the measurement point.

[0024] Preferably, according to the spatial and temporal characteristics , temperature time series characteristics , and environmental time series characteristics of the measurement point, perform importance ranking, and make the spatial and temporal characteristics , temperature time series characteristics , and environmental time series characteristics of the measurement point correspond one by one to the elements of each row of the judgment matrix , temperature time series characteristics , and environmental time series characteristics of the measurement point.

[0025] Preferably, according to the formula:

[0026]

[0027] calculate the weight vectors of the spatial and temporal characteristics , temperature time series characteristics , and environmental time series characteristics of the measurement point respectively; in the formula, is the weight vector of the time series characteristics corresponding to the -th row and -th column of the judgment matrix , and is the -th row and -th column element of the judgment matrix are all positive integers.

[0028] Preferably, according to the weight vectors of the spatial and temporal characteristics , temperature time series characteristics , and environmental time series characteristics of the measurement point, for the spatial and temporal characteristics , temperature time series characteristics , environmental temporal characteristics Perform weighted fusion to obtain the feature matrix of the measurement point .

[0029] The embodiment of the present application also provides a human skin temperature prediction system that integrates Informer and STGCN, including:

[0030] A feature extraction unit configured to, based on the topological structure between multiple measurement points corresponding to the skin temperature data of the target object obtained, extract the spatio-temporal features of the measurement points based on the spatio-temporal graph convolutional network ; and a first Informer module based on the self-attention mechanism, respectively according to the data matrix of the skin temperature data and the data matrix of the environmental parameter data obtained corresponding to the skin temperature data , to respectively extract the temperature temporal features of the measurement points , environmental temporal characteristics ;

[0031] A temperature prediction unit configured to fuse the spatio-temporal features , the temperature temporal features , the environmental temporal features in a second Informer module based on the self-attention mechanism according to the time sequence, and input the obtained feature matrix into a fully connected neural network to predict the skin temperature of the measurement points of the target object.

[0032] Beneficial effects:

[0033] The human skin temperature prediction method that integrates Informer and STGCN provided by the embodiment of the present application, according to the topological structure between multiple measurement points corresponding to the skin temperature data of the target object obtained, extracts the spatio-temporal features of the measurement points based on the spatio-temporal graph convolutional network , and based on the first Informer module of the self-attention mechanism, respectively according to the data matrix of the skin temperature data and the data matrix of the environmental parameter data obtained corresponding to the skin temperature data , respectively extract the temperature temporal features of the measurement points , environmental temporal characteristics ; then, the spatio-temporal features of the measurement points , temperature temporal features , environmental temporal characteristics Fuse in the second Informer module based on the self-attention mechanism according to the time sequence, and input the obtained feature matrix into the fully connected neural network to predict the skin temperature of the target measurement point.

[0034] Therefore, through the topological structure of multiple measurement points with physiological connection relationships, based on the spatio-temporal graph convolutional network and the Informer module based on the self-attention mechanism, comprehensively extract the spatial-temporal features, temperature temporal features, and environmental temporal features, and fuse them in the Informer module based on the self-attention mechanism according to the time sequence, effectively establishing the spatial dependence between different measurement points, overcoming the constraints of the existing technology data quality on the model performance and the insufficient applicability of the traditional human heat balance equation method in dynamic scenarios, enhancing the generalization ability of skin temperature prediction, using the obtained multi-dimensional features as the input of the fully connected neural network, providing sufficient information for subsequent skin temperature prediction, and effectively improving the accuracy and robustness of temperature prediction. Brief Description of the Drawings

[0035] The specification drawings forming a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. Among them:

[0036] Figure 1 It is a schematic flowchart of a method for predicting human skin temperature by fusing Informer and STGCN according to some embodiments of this application;

[0037] Figure 2 It is a schematic diagram of multiple measurement points of a target object according to some embodiments of this application;

[0038] Figure 3 For Figure 2 The corresponding topological structure schematic diagram of the multiple measurement points shown;

[0039] Figure 4 It is a schematic diagram of the structure of the Informer module based on the self-attention mechanism according to some embodiments of this application;

[0040] Figure 5 It is a logic diagram of a method for predicting human skin temperature by fusing Informer and STGCN according to some embodiments of this application;

[0041] Figure 6 It is a schematic diagram of the structure of a system for predicting human skin temperature by fusing Informer and STGCN according to some embodiments of this application. Detailed Embodiments

[0042] 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 rather than a limitation thereof. In fact, those skilled in the art will appreciate that modifications and variations can be made to the present application without departing from the scope or spirit thereof. For example, features shown or described as part of one embodiment can be used in another embodiment to yield 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 shall fall within the scope of protection of the embodiments of the present invention.

[0043] In an existing technology for predicting human skin temperature, by collecting videos of the subject's skin changing over time and the true skin temperature measured by a temperature sensor, a feature matrix is generated using video sampling and saturation extraction. At the same time, interpolation processing is performed on the true skin temperature to generate labels, thereby constructing a training set. Subsequently, the constructed training set is input into an improved ResNet50V2 network for training to obtain an optimized model. In the improved ResNet50V2 network, the feature outputs of the convolutional layer, residual block one, residual block two, and residual block three are extracted and globally pooled respectively to form feature matrices of four feature extraction paths. At the same time, these features are concatenated with the feature matrix extracted from the high-level of the original network.

[0044] In another existing technology for predicting human skin temperature, through simulation experiments in an artificial environmental test chamber, a human thermal response dataset is constructed, which includes parameters such as skin temperature, heart rate, and core temperature. Then, based on this dataset, the relationship between the simulated skin temperature, simulated heart rate, and simulated core temperature is inversely derived using a regression model. At the same time, through a deep learning algorithm, the model parameters in the Kalman filter are automatically adjusted to optimize and correct the measured data of the target individual. A long-time series prediction network model is used to accurately predict the core temperature of the target individual.

[0045] The above prediction technologies for human skin temperature, on the one hand, mainly rely on physiological data such as skin temperature and heart rate. However, in a complex low-temperature environment, the influence of environmental factors on skin temperature cannot be ignored. However, in the existing publicly available technologies, the influence of environmental characteristics (such as temperature, humidity, wind speed, etc.) on skin temperature prediction by fusing environmental factors is not clearly defined. On the other hand, when predicting human skin temperature, the measurement points are treated as independent individuals, lacking consideration of the spatial relationship between each measurement point. However, the temperature changes between different parts of the human body are interrelated. The lack of spatial dependence in the prediction process makes the prediction results of the model inaccurate.

[0046] Based on this, the embodiments of the present application propose a human skin temperature prediction method that combines Informer and STGCN to overcome the problems of the large restriction of the data quality on the model performance in the prior art and the insufficient applicability of the traditional human heat balance equation method in dynamic scenarios. Through the powerful non-linear modeling ability of deep learning algorithms, it adapts to diverse input data (including data of subjects with different heights, weights, ages, genders and different environmental data), mines complex patterns and feature associations in the data, enhances the generalization ability of the model, and improves the accuracy of temperature prediction and the robustness of the model. At the same time, through multi-task learning and data augmentation techniques, the prediction errors caused by noise and inconsistency are reduced, and diverse dynamic working scenarios are adapted to overcome the limitations of traditional methods and achieve more efficient and accurate temperature prediction.

[0047] As Figures 1 to 5 shown, the human skin temperature prediction method that combines Informer and STGCN includes:

[0048] Step S101, according to the topological structure between multiple measurement points corresponding to the skin temperature data of the target object, based on the spatio-temporal graph convolutional network, extract the spatio-temporal features of the measurement points; and, based on the first Informer module of the self-attention mechanism, respectively according to the data matrix of the skin temperature data and the data matrix of the environmental parameter data obtained corresponding to the skin temperature data, extract the temperature temporal features of the measurement points and the environmental temporal features respectively.

[0049] By collecting the skin temperature data of multiple measurement points with physiological connection relationships of the target object (such as the head, hands, arms, chest, thighs, calves, feet, etc.), combined with external conditions such as environmental temperature, humidity, wind speed (environmental parameter data), extract spatio-temporal features such as space, environment, temperature related to the environment where the target object is located, and provide multi-dimensional data input for the skin temperature prediction of the target object at future moments.

[0050] According to the adjacency relationship between multiple measurement points with physiological connection relationships, construct the topological structure between multiple measurement points of the target object to effectively capture the spatial dependence between different measurement points and provide spatial dependence information for the skin temperature prediction of the target object at future moments. Here, according to the relative positions between the measurement points, establish the spatial dependence between the skin temperature measurement points. For example, there is a strong temperature transfer relationship between adjacent parts such as the head and arms, hands and arms, feet and calves, thighs, etc. Through the relative positions and connectivity between the measurement points, establish the spatial relationship between the nodes of the topological structure.

[0051] In this application, the spatial adjacency relationship between the skin temperature measurement points of the target object is described by an adjacency matrix. Specifically, the adjacency matrix is as follows:

[0052]

[0053] Among them, for the measurement points that do not have a direct adjacency relationship, the corresponding elements in the adjacency matrix are 0, and for the measurement points that have a direct adjacency relationship, the corresponding elements in the adjacency matrix are 1. Then, the adjacency matrix is normalized to effectively ensure the balance of the influence of the degree of each node (measurement point) in the topological structure on the spatio-temporal graph convolutional network for feature extraction. Specifically, according to the formula:

[0054]

[0055] the adjacency matrix is normalized to obtain the normalized adjacency matrix . In the formula, the degree matrix is a diagonal matrix, and each diagonal element represents the degree of the corresponding measurement point (the number of direct connections of a measurement point to other measurement points). According to the adjacency matrix , the degree of each node (measurement point) can be calculated (for example, if the head and chest are connected, the degree of the head is 1; if the chest is connected to the head and the forearm and thigh, the degree of the chest is 3), and the corresponding degree matrix is constructed. Specifically,

[0056]

[0057] After obtaining the normalized adjacency matrix , based on the graph convolution operation, according to the formula:

[0058]

[0059] the skin temperature feature matrix of the measurement point at time is determined ; in the formula, is the ReLU activation function; is the skin temperature feature matrix of the target object at time, and its elements are the skin temperature data of each measurement point at time, is the weight matrix corresponding to the skin temperature feature matrix . Among them, in the graph convolutional neural network, the features of each measurement point are combined with the features of neighbor nodes to update the temperature features of each measurement point, and the weight matrix is automatically learned by combining neighbor node information.Adjust the features, and finally obtain the updated features of each measurement point Furthermore, integrate the skin temperature feature matrices at all historical moments of the measurement points to obtain the spatio-temporal features of the measurement points .

[0060] The skin temperature data of the target object is obtained by measuring multiple measurement points at multiple moments. Each measurement point has a measured temperature value at each moment. For example, by measuring the temperature of the head, chest, arms, hands, thighs, calves, and feet of the target object at multiple moments, the skin temperature data of the target object is obtained, and the data matrix of the skin temperature data is correspondingly constructed , as follows:

[0061]

[0062] In the formula, represents the skin temperature measurement value of the th measurement point at the moment

[0063] Then, the first informer module (informer-1) based on the self-attention mechanism learns the dependence relationship between the data (temperature measurement values) at each moment and the data (temperature measurement values) at other moments. Through multiple self-attention mechanisms stacked together, deeply mine the complex and long-term dependence relationship of skin temperature changing over time. Input the data matrix of skin temperature data into informer-1, and through the self-attention learning of each layer, judge the correlation between the data (temperature measurement values) at any moment and the data (temperature measurement values) at other moments, and generate high-level temporal features. Specifically, according to the formula:

[0064]

[0065] Extract the temperature temporal features of the measurement points to obtain the variation law and long-term dependence of skin temperature data over time. In the formula, is the data matrix of skin temperature data, is the weight matrix of the query matrix corresponding to the data matrix of skin temperature data , is the weight matrix of the key matrix corresponding to the data matrix of skin temperature data , is the weight matrix of the value matrix corresponding to the data matrix of skin temperature data ; is the vector dimension in the key matrix corresponding to the data matrix of skin temperature data .

[0066] While measuring the skin temperature data of the target object, record the environmental variables (such as temperature, humidity, wind speed, etc.) at each moment respectively, and construct a data matrix of environmental parameter data , as follows:

[0067]

[0068] In the formula, represents the environmental measurement value of the th environmental variable at the moment.

[0069] Then, based on the first informer module (informer-1) of the self-attention mechanism, learn the dependence relationship between the data (environmental measurement values) at each moment and the data (environmental measurement values) at other moments. By stacking multiple self-attention mechanisms together, deeply mine the complex and long-term dependence relationship of environmental variables changing over time. Input the data matrix of environmental parameter data into informer-1. After self-attention learning of each layer, judge the correlation between the data (environmental measurement values) at any moment and the data (environmental measurement values) at other moments, and generate high-level temporal features. Specifically, according to the formula:

[0070]

[0071] Extract the environmental temporal features of the measurement point to obtain the variation law and long-term dependence of environmental parameter data over time. In the formula, is the data matrix of environmental parameter data corresponding to the skin temperature data; is the weight matrix of the query matrix corresponding to the data matrix of environmental parameter data corresponding to the skin temperature data, is the weight matrix of the key matrix corresponding to the data matrix of environmental parameter data corresponding to the skin temperature data, is the weight matrix of the value matrix corresponding to the data matrix of environmental parameter data corresponding to the skin temperature data; is the vector dimension in the key matrix corresponding to the data matrix of environmental parameter data corresponding to the skin temperature data.

[0072] Step S102, the spatio-temporal features , the temperature temporal features , the environmental temporal features Fuse in the second Informer module based on the self-attention mechanism according to the time sequence, and input the obtained feature matrix into a fully connected neural network to predict the skin temperature of the measurement points of the target object.

[0073] In this application, through the spatio-temporal features reflect the spatial dependence relationship between the skin temperature measurement points of the target object, and through the temperature time series features reflect the law of the skin temperature of the target object changing with time, and through the environmental time series features reflect the influence of changes in environmental factors on the change of the skin temperature of the target object over time, and characterize the change law of the environmental variables in the area where the target object takes temperature measurements. Then, the extracted spatio-temporal features 、temperature time series features 、environmental time series features are fused in the second Informer module (informer-2) based on the self-attention mechanism according to the time sequence.

[0074] First, calculate the Pearson correlation coefficients 、temperature time series features 、environmental time series features of the measurement points and the skin temperature data ; among them, 、 、 , characterizes the correlation between the spatio-temporal features of the measurement points and the skin temperature data, characterizes the correlation between the temperature time series features of the measurement points and the skin temperature data, characterizes the correlation between the environmental time series features of the measurement points and the skin temperature data.

[0075] Then, according to the Pearson correlation coefficients 、temperature time series features 、environmental time series features of the measurement points and the skin temperature data , according to the formula:

[0076]

[0077] Calculate the importance ratios of the spatio-temporal features 、temperature time series features 、environmental time series features of the measurement points. In the formula, , are all positive integers; The importance ratio of the th temporal feature representing the measurement point to the th temporal feature.

[0078] Next, according to the importance ratios of the spatial-temporal features , temperature temporal features , and environmental temporal features of the measurement point, sort the spatial-temporal features , temperature temporal features , and environmental temporal features of the measurement point in order of importance, and construct a judgment matrix . Among them, take the importance ratio as the element in the judgment matrix , and the diagonal elements in the judgment matrix are reciprocals of each other, that is:

[0079]

[0080] The importance ratio of the th temporal feature representing the measurement point to the th temporal feature.

[0081] In the judgment matrix , the elements in each row are sorted according to the importance ratios of the spatial-temporal features , temperature temporal features , and environmental temporal features , and are respectively corresponding to the spatial-temporal features , temperature temporal features , and environmental temporal features . That is to say, according to the spatial-temporal features , temperature temporal features , and environmental temporal features of the measurement point, sort them in order of importance, and make the spatial-temporal features , temperature temporal features , and environmental temporal features of the measurement point correspond one by one to the elements in each row of the judgment matrix .

[0082] For example, when sorting the spatial-temporal features , temperature temporal features , and environmental temporal features in descending order of importance ratio, the judgment matrix The first row element in it is the data corresponding to the time series feature with the largest importance ratio, and the last row element is the data corresponding to the time series feature with the smallest importance ratio;

[0083] Furthermore, according to the judgment matrix in it, for the time series features corresponding to each row element, calculate the weight vectors of the corresponding time series features respectively. That is, according to the judgment matrix , calculate the weight vectors of the spatial time series feature , the temperature time series feature , and the environmental time series feature of the measurement point respectively. Specifically, according to the formula:

[0084]

[0085] calculate the weight vectors of the spatial time series feature , the temperature time series feature , and the environmental time series feature of the measurement point respectively; in the formula, is the weight vector of the time series feature corresponding to the th row and th column of the judgment matrix , is the th row and th column element in the judgment matrix , ; are all positive integers.

[0086] Finally, according to the weight vectors of the spatial time series feature , the temperature time series feature , and the environmental time series feature of the measurement point, perform weighted fusion on the spatial time series feature , the temperature time series feature , and the environmental time series feature of the measurement point to obtain the fusion feature of the measurement point, that is, the feature matrix ; and input the obtained feature matrix into the fully connected neural network to predict the skin temperature of each measurement point of the target object at any future moment.

[0087] Therefore, through the topological structure of multiple measurement points with physiological connection relationships, based on the spatio-temporal graph convolutional network and the informer-1 module based on the self-attention mechanism, the comprehensive extraction of spatial-temporal features, temperature temporal features, and environmental temporal features is carried out, and the fusion is performed in the informer-2 module based on the self-attention mechanism according to the time sequence, effectively establishing the spatial dependence between different measurement points, overcoming the problems such as the restriction of the existing technology data quality on the model performance and the insufficient applicability of the traditional human body heat balance equation method in dynamic scenarios, enhancing the generalization ability of skin temperature prediction, using the obtained multi-dimensional features as the input of the fully connected neural network, providing sufficient information for subsequent skin temperature prediction, and effectively improving the accuracy and robustness of temperature prediction.

[0088] The embodiment of the present application also provides a human skin temperature prediction system integrating informer and STGCN, as Figure 6 shown. The prediction system includes:

[0089] A feature extraction unit 601, configured to extract the spatial-temporal features of the measurement points based on the spatio-temporal graph convolutional network according to the topological structure between multiple measurement points corresponding to the skin temperature data of the target object obtained; and, a first informer module based on the self-attention mechanism, respectively according to the data matrix of the skin temperature data and the data matrix of the environmental parameter data obtained corresponding to the skin temperature data, to respectively extract the temperature temporal features of the measurement points and the environmental temporal features ;

[0090] A temperature prediction unit 602, configured to fuse the spatial-temporal features , temperature temporal features , and environmental temporal features in the second informer module based on the self-attention mechanism according to the time sequence, and input the obtained feature matrix into the fully connected neural network to predict the skin temperature of the measurement points of the target object.

[0091] The human skin temperature prediction system integrating informer and STGCN provided by the embodiment of the present application can implement the steps and processes of the human skin temperature prediction method integrating informer and STGCN in any of the above embodiments, and achieve the same technical effects, which will not be elaborated here one by one.

[0092] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation on the present invention.

[0093] In addition, the terms "first" and "second" are only used for descriptive purposes and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of such features. In the description of the present invention, the meaning of "a plurality" is at least two, such as two, three, etc., unless otherwise specifically and clearly defined.

[0094] In the present invention, unless otherwise clearly specified and defined, the terms "mounted", "connected", "coupled", "fixed", etc. shall be construed in a broad sense. For example, it may be a fixed connection, a detachable connection, or integrated; it may be a mechanical connection, an electrical connection, or communicable with each other; it may be directly connected, or indirectly connected through an intermediate medium, and it may be the internal communication of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0095] In the present invention, unless otherwise clearly specified and defined, the first feature being "on" or "under" the second feature may be that the first and second features are in direct contact, or the first and second features are indirectly in contact through an intermediate medium. Moreover, the first feature being "above", "over" and "on top of" the second feature may be that the first feature is directly above or obliquely above the second feature, or merely indicates that the first feature has a higher horizontal height than the second feature. The first feature being "under", "beneath" and "underneath" the second feature may be that the first feature is directly below or obliquely below the second feature, or merely indicates that the first feature has a lower horizontal height than the second feature.

[0096] In the present invention, terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in connection 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 can be combined in any one or more embodiments or examples in a suitable manner.

[0097] The foregoing are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A human skin temperature prediction method integrating informer and STGCN, characterized in that: include: Step S101: Based on the topological structure between multiple measurement points corresponding to the skin temperature data of the target object, the spatial temporal characteristics of the measurement points are analyzed based on the spatiotemporal graph convolution network. Perform extraction; And, the first informer module based on the self-attention mechanism, according to the data matrix of the skin temperature data and a data matrix of environmental parameter data corresponding to the skin temperature data obtained , to respectively analyze the temperature timing characteristics of the measurement points , Environmental time series characteristics Perform extraction; The adjacency matrix constructed based on the topological structure between multiple measurement points corresponding to the skin temperature data of the target object is Perform normalization to obtain the normalized adjacency matrix , and based on the graph convolution operation, according to the formula: Determine the measuring point at Skin temperature feature matrix at the moment , to construct the spatial temporal characteristics of the measurement points ; In the formula, is the ReLU activation function; For the target object The skin temperature feature matrix at the moment, is the skin temperature feature matrix The corresponding weight matrix; According to the formula: The temperature time series characteristics of the measurement point , Environmental time series characteristics Extraction; where, is the data matrix of the skin temperature data, The data matrix for the skin temperature data The weight matrix of the corresponding query matrix, The data matrix for the skin temperature data The weight matrix corresponding to the key matrix, The data matrix for the skin temperature data The weight matrix corresponding to the value matrix; The data matrix for the skin temperature data The corresponding vector dimensions in the key matrix; A data matrix of environmental parameter data corresponding to the skin temperature data; is a data matrix of environmental parameter data corresponding to the skin temperature data The corresponding weight matrix of the query matrix, is a data matrix of environmental parameter data corresponding to the skin temperature data The weight matrix corresponding to the bond matrix, is a data matrix of environmental parameter data corresponding to the skin temperature data The weight matrix corresponding to the value matrix; is a data matrix of environmental parameter data corresponding to the skin temperature data The corresponding vector dimensions in the key matrix; Step S102: The spatial temporal features , the temperature time series characteristics , the environmental time series characteristics The features are fused in a second informer module based on a self-attention mechanism in time sequence, and the obtained feature matrix is ​​input into a fully connected neural network to predict the skin temperature of the measuring point of the target object.

2. The human skin temperature prediction method integrating informer and STGCN according to claim 1 is characterized in that: In step S102, Calculate the spatial temporal characteristics of the measurement points respectively , Temperature time series characteristics , Environmental time series characteristics Pearson correlation coefficient with the skin temperature data , and according to the Pearson correlation coefficient Calculate the spatial temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics The importance ratio of According to the spatial and temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics The importance ratio of the spatial and temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics Sort by importance to build a judgment matrix ; According to the judgment matrix , respectively calculate the spatial temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics The weight vector is used to calculate the spatial and temporal characteristics of the measurement points. , Temperature time series characteristics , Environmental time series characteristics Perform feature fusion to obtain the feature matrix .

3. The human skin temperature prediction method integrating informer and STGCN according to claim 2 is characterized in that: According to the formula: Calculate the spatial temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics The importance ratio of ; in, , are all positive integers, is the judgment matrix The diagonal elements in Characterizes the measurement point The time series characteristics and The importance ratio of the time series features; Characterizes the measurement point The time series characteristics and The importance ratio of the time series features.

4. The human skin temperature prediction method integrating informer and STGCN according to claim 2 is characterized in that: According to the spatial and temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics Sort the importance of the spatial and temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics With the judgment matrix The row elements of each row correspond one to one to determine the spatial temporal characteristics of the measurement point , Temperature time series characteristics , Environmental time series characteristics The weight vector of .

5. The human skin temperature prediction method integrating informer and STGCN according to claim 4 is characterized in that: According to the formula: Calculate the spatial temporal characteristics of the measurement points respectively , Temperature time series characteristics , Environmental time series characteristics The weight vector of In the formula, for OK The judgment matrix of columns The The weight vector of the time series features corresponding to the row elements, is the judgment matrix The Line The elements of the column, ; All are positive integers.

6. According to claim 1, the method for predicting human skin temperature by integrating informer and STGCN is characterized in that: According to the spatial and temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics The weight vector of the spatial and temporal characteristics of the measurement points , Temperature time series characteristics , Environmental time series characteristics Perform weighted fusion to obtain the feature matrix of the measurement point .

7. A human skin temperature prediction system integrating informer and STGCN, characterized in that: A human skin temperature prediction method integrating informer and STGCN according to any one of claims 1 to 6 is deployed, and the system includes: The feature extraction unit is configured to extract the spatial temporal features of the measurement points based on the topological structure between the multiple measurement points corresponding to the skin temperature data of the target object obtained, based on the spatiotemporal graph convolution network. Extraction; and, based on the first informer module of the self-attention mechanism, respectively according to the data matrix of the skin temperature data and a data matrix of environmental parameter data corresponding to the skin temperature data obtained , to respectively analyze the temperature timing characteristics of the measurement points , Environmental time series characteristics Perform extraction; The temperature prediction unit is configured to convert the spatial temporal characteristics , the temperature time series characteristics , the environmental time series characteristics The features are fused in a second informer module based on a self-attention mechanism in time sequence, and the obtained feature matrix is ​​input into a fully connected neural network to predict the skin temperature of the measuring point of the target object.

Images