CN116701977A - A Human Body Temperature Data Fitting Method Based on Clustering Algorithm and Neural Network - Google Patents

Abstract

The present invention provides a human body temperature data fitting method based on a clustering algorithm and a neural network, comprising: step S1, within a preset period of time, continuously collecting real-time temperature data of a plurality of experimenters when running on a treadmill; In step S2, the preset time period is divided into multiple sub-time periods, and for each sub-time period, each real-time temperature data in the sub-time period is statistically obtained as the cluster data set corresponding to the sub-time period; in step S3, for each cluster class data set, clustering each real-time temperature data in the cluster class data set to obtain the corresponding cluster center point; step S4, for each experimenter, the cluster center point corresponding to the experimenter and the corresponding cluster center point The sub-time period is input into the pre-built neural network model, and the corresponding human body temperature data fitting curve is obtained. The beneficial effect is that the present invention can effectively reduce errors by extracting cluster central points through a clustering algorithm and a neural network and performing curve fitting.

Description

A Human Body Temperature Data Fitting Method Based on Clustering Algorithm and Neural Network

technical field

The present invention relates to the technical field of human body temperature analysis, in particular to a human body temperature data fitting method based on a clustering algorithm and a neural network.

Background technique

The surface temperature mechanism of the human chest skin is very complicated, mainly affected by the environment and skin heat dissipation. Under the action of thermoregulation, the body temperature of the human body remains constant. The general trend of chest surface temperature will increase with the increase of exercise time, and it will stabilize within a certain range due to the heat production and heat dissipation mechanism at a certain time.

At present, when analyzing and fitting human body temperature data, it is usually by directly taking the temperature value at each time point, but there are many temperature values at each time point, and it is difficult to obtain representative points, resulting in the human body obtained after fitting. The temperature data fitting curve has a large error.

In Chinese Patent Publication No. CN102736911A, a method of data fitting is disclosed, including: obtaining the coordinate values of each sample point to be fitted; performing a differential operation on the coordinate values of each point; generating a fitting Function; obtain the special point that the target curve passes through, integrate the fitting function, and obtain the target curve according to the integral formula and the coordinate value of the special point.

However, if there is a large deviation between a certain coordinate point and other coordinate values in the above fitting method, a large anomaly will occur, and the fitting curve will be integrated again later, and the coordinate point with a large deviation will also affect the integration. , which seriously affects the fitting effect, that is to say, the existing human body temperature data fitting methods all have the problems of poor fitting effect and large error.

Contents of the invention

The problem to be solved by the present invention is to provide a human body temperature data fitting method based on a clustering algorithm and a neural network, which can improve the fitting effect of the human body temperature data fitting curve and reduce errors.

In order to solve the above problems, the present invention provides a method for fitting human body temperature data based on clustering algorithm and neural network, comprising the following steps:

Step S1, within a preset period of time, continuously collect real-time temperature data of multiple experimenters when they continue to run on the treadmill;

Step S2, dividing the preset time period into a plurality of sub-time periods, and for each of the sub-time periods, statistically obtain the real-time temperature data in the sub-time periods as clusters corresponding to the sub-time periods class dataset;

Step S3, for each cluster data set, use a clustering algorithm to perform clustering processing on each of the real-time temperature data in the cluster data set to obtain a corresponding cluster center point;

Step S4, for each of the experimenters, input each of the cluster center points corresponding to the experimenter and the sub-time periods corresponding to each of the cluster center points into the pre-built neural network model to obtain the corresponding The human body temperature data fitting curve.

Preferably, in the step S1, a plurality of healthy males aged between 23-25 years old, with a height between 174-176 cm and a weight between 65-75 kg are selected as the experimenters.

Preferably, in the step S1, the real-time temperature data of each experimenter when they continue to run at 6km/h, 7.5km/h and 9km/h are respectively collected.

Preferably, in the step S2, the preset time period is set to 10 minutes, and the preset time period is evenly divided into a plurality of sub-time periods of 1 minute, and each sub-time period is measured every 1 minute. The temperature at the wrist of the experimenter is used as the real-time temperature data.

Preferably, said step S3 includes:

Step S31, judging whether an algorithm instruction input from outside is received:

If so, turn to step S32;

If not, then turn to step S33;

Step S32, randomly obtain a plurality of the real-time temperature data in each of the cluster data sets as the initial cluster center points, and use the remaining real-time temperature data as data points, according to each of the initial cluster center points and each Euclidean distance between the data points iteratively calculates the corresponding center point of the cluster, and then turns to step S4;

Step S33, using each of the cluster data sets as the initial cluster number, randomly obtaining a plurality of the real-time temperature data from each of the initial cluster numbers as the initial cluster center point, and using the remaining real-time temperature data As data points, the corresponding cluster center points are obtained according to the Euclidean distance and similarity between each of the initial cluster center points and each of the data points.

Preferably, said step S32 includes:

Step S321, randomly acquiring a plurality of the real-time temperature data in each of the cluster data sets as the initial cluster center point, and using the remaining real-time temperature data as the data points;

Step S322, for each of the data points, based on the Euclidean distance between the data point and each of the initial cluster center points, assign the data point to the initial cluster center with the closest Euclidean distance The cluster data set where the point is located;

Step S323, converting each of the initial cluster center points into the corresponding data points, and calculating the data points and the remaining data points for each of the data points in each of the cluster data sets The sum of the Euclidean distances between the points, the data point with the smallest sum of Euclidean distances is used as the center point of the second-generation clustering;

Step S324, obtain the corresponding sum of squares of the first clustering error according to each of the second-generation clustering center points and each of the data points, and return each of the second-generation clustering center points as the initial clustering center point to the Step S322 is described, and the data point with the smallest sum of Euclidean distances in the step S323 is used as the three-generation clustering center point, and the corresponding second clustering error is obtained according to each of the three-generation clustering center points and each of the data points sum of square;

Step S325, judging whether the second clustering error sum of squares is equal to the first clustering error sum of squares:

If not, each of the three-generation cluster center points is used as the initial cluster center point and returns to step S322;

If yes, each of the three-generation cluster center points is used as the cluster center point.

Preferably, said step S33 includes:

Step S331, using each of the cluster data sets as the initial cluster number, randomly obtaining a plurality of the real-time temperature data from each of the initial cluster numbers as the initial cluster center point, and using the remaining The real-time temperature data is used as the data point;

Step S332, for each of the data points, based on the Euclidean distance between the data point and each of the initial cluster center points and the principle of negative correlation between the Euclidean distance and the similarity, the data points are divided into Assigning to the cluster data set where the initial cluster center point with the closest Euclidean distance is located;

Step S333, converting each of the initial cluster center points into the corresponding data points, and for each of the cluster data sets, using the average value of each of the data points in the cluster data sets as the cluster center point.

Preferably, in the step S4, the neural network model uses the time attribute corresponding to the sub-time period as an input layer input, and uses the temperature attribute corresponding to each of the cluster center points as an output layer output to obtain the Human body temperature data fitting curve.

The present invention has the following beneficial effects: Compared with the differential integral data fitting method, the present invention uses a clustering algorithm to cluster the real-time temperature data in the cluster data set to take out the cluster center point to ensure the representativeness of the cluster center point. Compared with the traditional method, the error of directly replacing the representative point is smaller, and then the neural network model is used to fit the center point of the cluster to obtain the fitting curve of the human body temperature data, which makes it more in line with the fitting strategy of the nonlinear function to improve the fitting effect.

Description of drawings

Fig. 1 is a flow chart of steps of the present invention;

Fig. 2 is the specific flowchart of step S3 of the present invention;

Fig. 3 is the specific flowchart of step S32 of the present invention;

Fig. 4 is the specific flowchart of step S33 of the present invention;

Fig. 5 is the processing result schematic diagram that the real-time temperature data under 6km/h of the present invention obtains through K-medoids algorithm processing;

Fig. 6 is the processing result schematic diagram that the real-time temperature data under 7.5km/h of the present invention obtains through K-means algorithm processing;

Fig. 7 is the processing result schematic diagram that the real-time temperature data under 9km/h of the present invention obtains through K-medoids algorithm processing;

Fig. 8 is a schematic diagram of the neural network structure of the present invention;

Fig. 9 is a schematic diagram of the fitting results of the present invention for fitting the center point of the cluster under 6km/h;

Fig. 10 is the fitting result schematic diagram of the present invention to 7.5km/h lower cluster central point is fitted;

Fig. 11 is a schematic diagram of the fitting results of the present invention for fitting the center points of the clusters at 9 km/h.

Detailed ways

In order to make the above objects, features and advantages of the present invention more comprehensible, specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, a human body temperature data fitting method based on a clustering algorithm and a neural network is now provided, as shown in Figure 1, comprising the following steps:

Step S1, within a preset period of time, continuously collect real-time temperature data of multiple experimenters when they continue to run on the treadmill;

In step S2, the preset time period is divided into multiple sub-time periods, and for each sub-time period, each real-time temperature data in the sub-time period is statistically obtained as a cluster data set corresponding to the sub-time period;

Step S3, for each cluster data set, use a clustering algorithm to cluster each real-time temperature data in the cluster data set to obtain a corresponding cluster center point;

Step S4, for each experimenter, input each cluster center point corresponding to the experimenter and the sub-time period corresponding to each cluster center point into the pre-built neural network model to obtain the corresponding human body temperature data fitting curve.

Specifically, in this embodiment, compared with differential integral data fitting, this embodiment performs clustering algorithm processing on the cluster data set, and takes out the cluster center point (the cluster center point is the most representative point in the cluster data ), and then fit it to the neural network data, it is more representative to extract the cluster center point through the clustering algorithm, and the data fitting effect is improved through the neural network fitting, which is more suitable for the fitting of nonlinear functions.

In a preferred embodiment of the present invention, in step S1, multiple healthy males aged between 23-25 years old, 174-176 cm in height, and 65-75 kg in weight are selected as experimenters.

Specifically, in this example, 20 normal healthy male experimenters were selected, aged 24±1 years old, 175±1cm in height, and 70±5kg in weight, and all experimenters met three conditions: (1) no cardiopulmonary Diseases and movement disorders; (2) No colds and other current conditions during the test period; (3) Volunteer to complete the entire experiment.

Preferably, all test processes are carried out in a constant temperature and humidity laboratory. In order to avoid the impact of human body biorhythm changes on the test, the experiment is arranged in the afternoon. The experimental environment used in this embodiment is at room temperature of 20 ± 1 degrees Celsius, and the humidity is It is carried out under the condition of 40±5%, the indoor wind speed is negligible, the body temperature is measured by a temperature tester, and the sweat volume is collected by the local collection method.

In a preferred embodiment of the present invention, in step S1, the real-time temperature data of each experimenter when running continuously at 6km/h, 7.5km/h and 9km/h are respectively collected.

In a preferred embodiment of the present invention, in step S2, the preset time period is set to 10 minutes, and the preset time period is evenly divided into multiple sub-time periods of 1 minute length, and each experimenter is measured every 1 minute The temperature at the wrist serves as real-time temperature data.

Specifically, in this embodiment, the treadmill is set to 6km/h, 7.5km/h and 9km/h speeds for testing, and the running test lasts 10 minutes, and the temperature at the wrist of the experimenter is tested every 1 minute, and the real-time temperature data is recorded .

In a preferred embodiment of the present invention, as shown in Figure 2, step S3 includes:

Step S31, judging whether an algorithm instruction input from outside is received:

If so, turn to step S32;

If not, then turn to step S33;

Step S32, randomly obtain a plurality of real-time temperature data in each cluster data set as the initial cluster center point, and use the remaining real-time temperature data as data points, according to the Euclidean distance between each initial cluster center point and each data point Iteratively calculate the corresponding cluster center point, then turn to step S4;

In step S33, each cluster data set is used as the initial cluster number, a plurality of real-time temperature data are randomly obtained from each initial cluster number as the initial cluster center point, and the rest of the real-time temperature data are used as data points, according to each initial cluster number The Euclidean distance and similarity between the class center point and each data point get the corresponding cluster center point.

Specifically, in this embodiment, what step S32 adopts is the K-medoids clustering algorithm, and evaluating the clustering quality of a cluster is to use a cost function and the algorithm to iterate repeatedly to find the most suitable cluster division and cluster center point , step S33 uses the K-means algorithm, which selects the cluster center point by calculating the average value of the cluster center point, which is very sensitive to the outliers, so that the selected cluster center point may not exist, and the K-medoids algorithm is In the iterative process, when selecting the cluster center point, the sample point near the cluster center point is used as the selection object, thereby eliminating the sensitivity to the outlier point. Both have their own advantages and disadvantages, and they can be oriented for processing separately.

In a preferred embodiment of the present invention, as shown in Figure 3, step S32 includes:

Step S321, randomly obtain a plurality of real-time temperature data in each cluster data set as the initial cluster center point, and use the remaining real-time temperature data as data points;

Step S322, for each data point, based on the Euclidean distance between the data point and each initial cluster center point, assign the data point to the cluster data set where the initial cluster center point with the closest Euclidean distance is located;

Step S323, convert each initial cluster center point into a corresponding data point, and calculate the sum of the Euclidean distances between the data point and the rest of the data points for each data point in each cluster data set, and calculate the sum of the Euclidean distances The smallest data point is used as the center point of the second-generation clustering;

Step S324, obtain the corresponding sum of squares of the first clustering error according to each second-generation cluster center point and each data point, return to step S322 with each second-generation cluster center point as the initial cluster center point, and set the Euclidean sum of squares in step S323 The data point with the smallest sum of distances is used as the center point of the three-generation clustering, and the corresponding sum of squares of the second clustering error is obtained according to each three-generation clustering center point and each data point;

Step S325, judging whether the second clustering error sum of squares is equal to the first clustering error sum of squares:

If not, then use each three-generation clustering central point as the initial clustering central point and return to step S322;

If so, each three-generation cluster center point is used as the cluster center point.

Specifically, in this embodiment, the evaluation of the quality of the cluster center points uses the error square of the Euclidean distance, which is defined as follows:

in,

x represents the real-time temperature data in each cluster data set C _j ;

O _j represents the cluster center point.

In a preferred embodiment of the present invention, as shown in Figure 4, step S33 includes:

Step S331, using each cluster data set as the initial cluster number, randomly obtaining a plurality of real-time temperature data from each initial cluster number as the initial cluster center point, and using the remaining real-time temperature data as data points;

Step S332, for each data point, based on the Euclidean distance between the data point and each initial clustering center point and the principle of negative correlation between the Euclidean distance and the similarity, assign the data point to the initial cluster with the closest Euclidean distance The cluster data set where the cluster center point is located;

In step S333, each initial cluster center point is converted into a corresponding data point, and for each cluster data set, the average value of each data point in the cluster data set is used as the cluster center point.

Specifically, in this embodiment, the K-means algorithm measures the similarity of different data objects by selecting an appropriate distance formula. The distance between data is inversely proportional to the similarity. It can be considered that the smaller the similarity, the larger the distance , the distance here is Euclidean distance, and the formula of Euclidean distance is as follows:

in,

x represents a data point;

C _i represents the i-th cluster center point;

m represents the dimension of the data point;

x _j , C _ij represent the attribute value of the jth dimension of the data point x and the cluster center point C _i .

Preferably, the similarity SSE formula is as follows:

Preferably, the real-time temperature data collected under 6km/h is processed by K-medoids algorithm, and the processing result is as shown in Figure 5, and the real-time temperature data collected under 7.5km/h is processed by K-means algorithm, and the processing result As shown in Figure 6, the K-medoids algorithm is used to process the real-time temperature data collected at 9km/h, and the processing results are shown in Figure 7.

In a preferred embodiment of the present invention, in step S4, the neural network model uses the time attribute corresponding to the sub-time period as the input layer input, and uses the temperature attribute corresponding to each cluster center point as the output layer output to obtain the human body temperature data by fitting Curve fitting.

Specifically, in this embodiment, the neural network structure shown in Figure 8 is used to fit the cluster center point of 6km/h, and the fitting result is shown in Figure 9, and the cluster center point of 7.5km/h is fitted Fitting, the fitting result is shown in Figure 10, and the 9km/h cluster center point is fitted, and the fitting result is shown in Figure 11.

Although the present disclosure is disclosed as above, the protection scope of the present disclosure is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present disclosure, and these changes and modifications will all fall within the protection scope of the present invention.

Claims (8)

1. The human body temperature data fitting method based on the clustering algorithm and the neural network is characterized by comprising the following steps of:

step S1, continuously collecting real-time temperature data of multiple experimenters when running continuously on a running machine in a preset time period;

step S2, dividing the preset time period into a plurality of sub-time periods, and counting to obtain each real-time temperature data in each sub-time period as a cluster data set corresponding to the sub-time period;

step S3, clustering each real-time temperature data in each cluster data set by adopting a clustering algorithm to obtain a corresponding cluster center point;

and S4, inputting each cluster center point corresponding to each experimenter and the sub-time period corresponding to each cluster center point into a pre-constructed neural network model to obtain a corresponding human body temperature data fitting curve.

2. The method according to claim 1, wherein in the step S1, a plurality of healthy men with ages of 23-25 years, heights of 174-176 cm, and weights of 65-75 kg are selected as the experimenters.

3. The method according to claim 1, wherein in the step S1, the real-time temperature data of each experimenter is collected when running continuously at 6km/h, 7.5km/h and 9 km/h.

4. The human body temperature data fitting method according to claim 1, wherein in the step S2, the preset time period is set to 10 minutes, and the preset time period is divided into a plurality of the sub-time periods of 1 minute duration on average, and the temperature at each of the experimenters' wrists is measured every 1 minute as the real-time temperature data.

5. The method of fitting human body temperature data according to claim 1, wherein the step S3 comprises:

step S31, judging whether an externally input algorithm instruction is received or not:

if yes, go to step S32;

if not, turning to step S33;

step S32, randomly acquiring a plurality of real-time temperature data in each cluster data set to serve as initial clustering center points, taking the rest of the real-time temperature data as data points, iteratively calculating the corresponding cluster center points according to Euclidean distances between the initial clustering center points and the data points, and then turning to step S4;

step S33, taking each cluster data set as an initial cluster number, randomly acquiring a plurality of real-time temperature data in each initial cluster number as initial cluster center points, taking the rest of the real-time temperature data as data points, and obtaining the corresponding cluster center points according to euclidean distance and similarity between each initial cluster center point and each data point.

6. The method of fitting human body temperature data according to claim 5, wherein the step S32 comprises:

step S321, randomly acquiring a plurality of real-time temperature data from each cluster data set to serve as the initial clustering center point, and taking the rest real-time temperature data as the data points;

step S322, for each data point, assigning the data point to the cluster data set where the initial cluster center point with the closest euclidean distance is located, based on the euclidean distance between the data point and each initial cluster center point;

step S323, converting each initial cluster center point into a corresponding data point, and calculating, for each data point in each cluster data set, a sum of euclidean distances between the data point and the rest of the data points, wherein the data point with the smallest sum of euclidean distances is used as a second-generation cluster center point;

step S324, obtaining a corresponding first cluster error square sum according to each second generation cluster center point and each data point, returning each second generation cluster center point to the step S322 as the initial cluster center point, and obtaining a corresponding second cluster error square sum according to each third generation cluster center point and each data point by taking the data point with the smallest euclidean distance sum in the step S323 as a third generation cluster center point;

step S325, determining whether the second cluster error square sum is equal to the first cluster error square sum:

if not, taking each third generation clustering central point as the initial clustering central point and returning to the step S322;

if yes, taking each third generation cluster center point as the cluster center point.

7. The method of fitting human body temperature data according to claim 5, wherein the step S33 comprises:

step S331, taking each cluster data set as the initial cluster number, randomly acquiring a plurality of real-time temperature data from each initial cluster number as the initial cluster center point, and taking the rest of real-time temperature data as the data points;

step S332, for each data point, assigning the data point to the cluster data set where the initial cluster center point with the closest euclidean distance is located, based on the euclidean distance between the data point and each initial cluster center point and the principle of negative correlation between the euclidean distance and the similarity;

step S333, converting each initial cluster center point into a corresponding data point, and regarding each cluster data set, taking an average value of each data point in the cluster data set as the cluster center point.

8. The method according to claim 1, wherein in the step S4, the neural network model takes a time attribute corresponding to the sub-time period as an input layer input, and takes a temperature attribute corresponding to each cluster center point as an output layer output, and the neural network model is fitted to obtain the human body temperature data fitting curve.