阅读说明：本技术 一种基于多源数据信息的疲劳状态因果网络方法 (fatigue state causal network method based on multi-source data information ) 是由 任长娥 袁超 杜涛 王岩 于 2019-09-26 设计创作，主要内容包括：本发明公开一种基于多源数据信息的疲劳状态因果网络方法。步骤如下：一、采集与人体疲劳状态相关的呼吸、心率数据；二、针对每一项生理数据,使用外推拟合方法去除野值；三、任选两项生理数据,检验该两项数据之间是否具有相关关系；四、对于具有相关关系的两项生理数据,采用格兰杰因果分析方法检验二者之间是否具有因果关系；五、循环选择两组生理数据,运用第四步中的格兰杰因果分析方法；六、遍历所有生理数据后,建立变量间的因果关系网络。本发明的方法具有精度高、计算量小、兼容性强等优点。可用于汽车、轮船、飞机等载体工作人员的疲劳状态因果分析与检测,提高基于多源数据信息融合的疲劳状态因果分析准确率,具有较高应用价值。(The invention discloses fatigue state causal network methods based on multi-source data information, which comprises the following steps of collecting respiration and heart rate data related to human fatigue states, secondly removing outliers by using an extrapolation fitting method aiming at each physiological data, thirdly selecting two optional physiological data to check whether the two data have a correlation relationship, fourthly checking whether the two physiological data have the causal relationship by using a Glange causal analysis method for the two physiological data with the correlation relationship, fifthly circularly selecting two groups of physiological data, applying the Gegere causal analysis method in the fourth step, and sixthly establishing a causal relationship network among variables after traversing all the physiological data.)

1, A method for causal network of fatigue state based on multi-source data information, which is characterized by comprising the following steps:

, collecting the respiration and heart rate data relative to the fatigue state of human body;

secondly, removing outliers by using an extrapolation fitting method aiming at each physiological data;

selecting two items of physiological data optionally, and checking whether the two items of data have a correlation;

fourthly, for two items of physiological data with correlation, a Glange causal analysis method is adopted to test whether the two sets of data have causal relationship;

step five, circularly selecting two groups of physiological data, and applying the Glangel causal analysis method in the step four;

and sixthly, establishing a causal relationship network among the variables after traversing all the physiological data.

2. The multi-source data information-based causal network method of fatigue states of claim 1, wherein: the respiratory data comprises: respiratory quotient, respiratory rate, tidal flow volume, ventilation, oxygen consumption per minute, carbon dioxide consumption per minute, end-tidal oxygen partial pressure, end-tidal carbon dioxide partial pressure, energy consumption per minute; the heart rate data comprises: heart rate.

3. The multi-source data information-based causal network method of fatigue states of claim 1, wherein: the fourth step is specifically realized as follows: after the third step of test, if the correlation coefficient between the two physiological indexes is rho_XYAnd if the correlation exists between the two variables, the causal relationship between the two variables is further tested .

4. The multi-source data information-based causal network method of fatigue status of claim 3, wherein: the causal relationship is tested as follows:

assuming that X (t) and Y (t) are two random stationary time series with correlation, the variable Y (t) can be expressed by its autoregressive equation and its interaction with X (t), as:

wherein p and q are respectively hysteresis orders in the regression equations X (t) and Y (t), i and j are hysteresis terms, i is an integer between 1 and p, j is an integer between 1 and q, X (t-i) is the ith hysteresis term of X (t), Y (t-j) is the jth hysteresis term of Y (t), α_YX(i) And α_YY(i) Is the coefficient estimation value of Y (t), e (t) is white noise of Y (t), if α_YX(i) (i-1, 2.. p) statistically significantly differs from zero overall, indicating that x (t) is the cause of glandor of y (t), if α_YX(i) P) statistically significantly equals zero overall, indicating that x (t) and y (t) are not causal.

5. The multi-source data information-based causal network method of fatigue states of claim 1, wherein: according to the method, a causal network diagram among various physiological indexes is obtained according to the tested Glanberg causal relationship.

Technical Field

The invention relates to fatigue state causal network methods based on multi-source data information, which can be used for fatigue state causal analysis and detection of carrier workers such as automobiles, ships, airplanes and the like, and improve fatigue state causal analysis accuracy based on multi-source data information fusion.

Background

When working under special circumstances, some physiological index values of people can change, and the change of the physiological data can cause some physiological diseases. For example, working in a closed environment for a long time may be more prone to fatigue, depression, and the like. The fatigue state is an early warning for certain physiological sub-health states, so that the research on the fatigue state of a person in a special working environment is helpful for predicting the health state of the person so as to reasonably improve the working conditions of the person, and the method has important significance.

The machine learning method proposed by the thesis of driver fatigue detection method research based on deep learning and facial multi-source dynamic behavior fusion can judge whether a human body is in a fatigue state or not by learning a data set of certain physiological indexes. But the operation process is complicated and the causal relationship between the physiological indexes cannot be given.

The granger causal relationship analysis can describe the interaction strength of different physiological characteristic data and also can show the causal relationship of the interaction between the different physiological characteristic data, and the granger causal relationship analysis has obvious advantages compared with the traditional analysis methods such as correlation, coherence and mutual information algorithms. Therefore, there is a significant advantage to analyzing the relationship between human physiological characteristic parameters in conjunction with the granger's causal relationship. Because the causal relationship exists directionally and objectively, an expected result can be obtained by manipulating the reason, so that people obtain stronger subjective motility to the natural world, and the research of the causal relationship is of great significance.

Disclosure of Invention

The invention aims to provide fatigue state causal network methods based on multi-source data information, which overcome the defects of the machine learning method and apply the Glandum causal analysis method to various physiological indexes affecting the fatigue state of a human body for analysis.

The technical solution of the invention is as follows:

A method for causal network of fatigue state based on multi-source data information, which is characterized by comprising the following steps:

, collecting physiological data such as respiration and heart rate related to the fatigue state of the human body;

the method specifically comprises the following steps: utilize cardiopulmonary function apparatus to collect respiratory data, utilize the electrocardio monitor to collect heart rate data, above-mentioned respiratory data includes: respiratory quotient (R), respiratory rate (Rf), tidal flow (VT), Ventilation (VE), oxygen consumption per minute (VO)₂) Carbon dioxide consumption per minute (VCO)₂) End-tidal oxygen partial pressure (PetO)₂) End tidal carbon dioxide partial pressure (PetCO)₂) Energy expenditure per minute (EEm); the heart rate data includes: heart Rate (HR).

Secondly, removing outliers aiming at each items of physiological data;

outliers can be removed in particular using an extrapolation fitting method comprising for each individual physiological data of the acquired respiratory and heart rate data, optionally five consecutive measured data from time n to time n +4, denoted as z_n,z_n+1,z_n+2,z_n+3,z_n+4Calculating the estimated value of the n +4 th time according to the data of the first four timesThe following were used:

calculating an estimated value

Measured value z at later and n +4 th time_n+4Making a comparison when the conditions are satisfiedWhen it is, then z is considered_n+4Is a normal value, otherwise z is considered to be_n+4To obtain outliers, the outliers were removed. Where σ is the mean square error and can be obtained by consulting the sensor specification.

Selecting two items of physiological data optionally, and checking whether the two items of data have a correlation;

the method specifically comprises the following steps of selecting two items from the physiological data collected in the step , and checking whether a correlation exists between the two items, wherein the correlation is calculated according to the following formula:

where X and Y represent two sets of variables, respectively, cov (X, Y) represents the covariance of X and Y,

is the standard deviation of the X, and,

is the standard deviation of Y, ρ_XYRepresenting the correlation coefficient of X and Y thresholds e are set when | ρ_XYAnd when the | is more than or equal to e, considering that a correlation exists between the variable X and the variable Y. And (4) performing the correlation calculation on any two physiological data to find out all physiological indexes with correlation.

Fourthly, checking whether the two groups of data which are obtained in the third step and have the correlation have the causal relationship by adopting a Glange causal analysis method;

the method specifically comprises the following steps: if the correlation coefficient between the two physiological indexes is rho_XYWhen there is a correlation between two variables, the causal relationship between them is tested , the causal relationship is tested as follows:

assuming that X (t) and Y (t) are two random stationary time series with correlation, the variable Y (t) can be expressed by its autoregressive equation and its interaction with X (t), as:

wherein p and q are respectively hysteresis orders in the regression equations X (t) and Y (t), i and j are hysteresis terms, i is an integer between 1 and p, j is an integer between 1 and q, X (t-i) is the ith hysteresis term of X (t), Y (t-j) is the jth hysteresis term of Y (t), α_YX(i) And α_YY(i) Is an estimate of the coefficient of Y (t), e (t) is white noise of Y (t) < if α_YX(i) (i-1, 2.. p) statistically significantly differs from zero overall, indicating that x (t) is the glandor cause of y (t): if α_YX(i) P) statistically significantly equals zero overall, indicating that x (t) and y (t) are not causal.

Step five, circularly selecting two groups of physiological data, and applying the Glangel causal analysis method in the step four;

the method specifically comprises the following steps: and (3) for the physiological data with the correlation relationship determined in the third step, testing the Glan causal relationship between any two groups of data by using the Glan causal test method in the fourth step until the causal relationship between all the data is tested.

And sixthly, establishing a causal relationship network among the variables after traversing all the physiological data.

The method specifically comprises the following steps: and drawing a causal relationship network graph among various data for the tested Glan's causal relationship in the fifth step. The causal relationship direction between data is represented by arrows, which point from the "cause" variable to the "effect" variable.

Compared with the prior art, the invention has the advantages that:

(1) the method has the advantages of simple operation, high precision and small calculation amount.

(2) The method can help people to quickly find the causal relationship network among various physiological data, so that an expected result is obtained by controlling the cause variable. The method can be used for the causal analysis and detection of the fatigue state of carrier workers such as automobiles, ships, airplanes and the like, improves the accuracy of the causal analysis of the fatigue state based on multi-source data information fusion, and has high application value.

Drawings

FIG. 1 is a flow chart of an implementation of the method of the present invention;

FIG. 2 is a causal network diagram of embodiments of the present invention;

Detailed Description

The present invention will now be described more fully hereinafter with reference to the accompanying drawings and examples, in which it is to be understood that the above-described examples are intended to illustrate only some, but not all, of the embodiments of the present invention.

The descriptions related to "", "second", etc. in this invention are for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicit ly indicating the number of technical features indicated.

Referring to fig. 1, fig. 1 is a flow chart of the method of the present invention.

Step S1, using K4b²The cardiopulmonary function tester collects the respiratory data of 9 testees, and the electrocardiograph collects the heart rate data of the testees. The data breath includes: respiratory quotient (R), respiratory rate (Rf), tidal flow (VT), Ventilation (VE), oxygen consumption per minute (VO)₂) Carbon dioxide consumption per minute (VCO)₂) End-tidal oxygen partial pressure (PetO)₂) End tidal carbon dioxide partial pressure (PetCO)₂) The heart rate data comprises Heart Rate (HR) and 10 physiological indexes, 369 data are collected for each physiological index, and the 10 data collected each time are correspondingly stored.

Step S2, for each physiological items in the acquired respiratory and heart rate dataData, optionally five successive measurements from time n to time n +4, denoted as z_n,z_n+1,z_n+2,z_n+3,z_n+4Calculating the estimated value of the n +4 th time according to the data of the first four times

Will estimate the value

With the measured value z at the n +4 th moment_n+4Make a comparison if

Then consider z to be_n+4Is a normal value, otherwise z is considered to be_n+4To obtain outliers, the outliers were removed. Where σ is the mean square error and can be obtained by consulting the sensor specification.

In step S3, it is checked whether there is a correlation between the 10 items of physiological data. The correlation coefficient calculation formula is as follows:

where X and Y represent two sets of variables, respectively, cov (X, Y) represents the covariance of X and Y,

is the standard deviation of the X, and,

is the standard deviation of Y, ρ_XYRepresenting the correlation coefficient of X and Y thresholds e are set, when | r_XYAnd when the | is more than or equal to e, considering that a correlation exists between the variable X and the variable Y. In the embodiment of the present invention, when ρ is set to 0.8, e is set to_XYIn embodiments of the present invention, correlation tests show that the physiological indicators having correlation include VE and VO₂VE and VCO₂,VO₂And VCO₂,EEm and VCO₂EEm and VO₂。

Step S4, adopting a Glange causal analysis method to check whether the two groups of data with the correlation obtained in the third step have causal relationship; the implementation mode is as follows:

the causal relationship is tested only when there is a correlation between the two variables, assuming that X (t) and Y (t) are two correlated random stationary time series obtained in step S3, the variable Y (t) can be expressed by its autoregressive equation and its interaction with X (t), as follows:

wherein p and q are respectively hysteresis orders in the regression equations X (t) and Y (t), i and j are hysteresis terms, i is an integer between 1 and p, j is an integer between 1 and q, X (t-i) is the ith hysteresis term of X (t), Y (t-j) is the jth hysteresis term of Y (t), α_YX(i) And α_YY(i) Is an estimate of the coefficient of Y (t), e (t) is white noise of Y (t) < if α_YX(i) (i-1, 2.. p) statistically significantly differs from zero overall, indicating that x (t) is the glandor cause of y (t): if α_YX(i) In examples of the invention, p and q both take 4 th order hysteresis, and the causal test is performed by Eviews10 software.

In step S5, the data with correlation determined in step S3 is tested for causal relationship between any two groups of data by cyclic application of the granger causal test method in step S4 until all the data are tested for causal relationship.

In step S6, a causal relationship network graph among the physiological indicators having causal relationship, which has been verified in step S5, is drawn, wherein arrows indicate the causal relationship direction among the physiological indicators, and the arrows point to the effect variable from the cause variable, and the causal relationship network graph obtained by embodiments of the present invention is described in FIG. 2.

Referring to fig. 2, in the embodiment of the present invention, the physiological indicators having correlation are as follows through correlation test: VE and VO₂VE and VCO₂,VO₂And VCO₂EEm and VCO₂EEm and VO₂The process proceeds to where the physiological index of 5 pairs is subjected to the Glandum causal relationship test to obtain the results shown in FIG. 2, i.e., VE is VO₂Of the Glan Jie, VCO₂Is the Glandoy cause of VE, VCO₂Is VO₂Of the Glan Jie, VCO₂Is the Glandoy cause of EEm, VO₂Is the Glanbe cause of EEm, which is also VO₂The cause of glandory.

Those skilled in the art will appreciate that the invention may be practiced without these specific details.