JP6849106B2 - Stress estimation device and stress estimation method using biological signals - Google Patents

Description

The present invention relates to a stress estimation device and a stress estimation method using biological signals.

Chronic stress is the stress that accumulates when a person is exposed to various stressors over a long period of time. Long-term stress can lead to mental illness such as depression. Occupational stress is an example of long-term stress. Occupational stress is stress that accumulates, especially when workers are exposed to various stressors during work. Depression caused by occupational stress reduces the productivity of workers. Therefore, early detection and prevention of depression and the like are important. Various long-term stress estimation techniques have been proposed for early detection and prevention of depression and the like.

An example of a long-term stress estimation system is described in Non-Patent Documents 1, 2, 3 and 4. For example, Non-Patent Document 1 and Non-Patent Document 2 disclose a technique for identifying the degree of stress by using a feature amount obtained by statistically processing a biological signal of a subject over a long period of one month. The feature amount is obtained by statistically processing the biological signal according to the action time such as sitting, walking, running, and sleeping. Then, a person with a high score and a person with a low score in the PSS (Perceived Stress Scale) questionnaire, which is often used as an index of long-term stress, are identified.

Non-Patent Document 3 discloses a technique for estimating the happiness and productivity of a user based on the frequency distribution of the duration of physical activity of a person. Non-Patent Document 4 contains statistics (mean, standard deviation,) of actions such as walking, conversation, and desk work, and their action time (action time itself, the ratio of the action time in one day, and the number of actions). And median) and correlation analysis between personality and depression indicators are shown.

A. Sano, "Measuring College Students' Sleep, Stress, Mental Health and Wellbeing with Wearable Sensors and Mobile Phones", Massachusetts Institute of Technology, 2015 A. Sano et al., "Recognizing academic performance, sleep quality, stress level, and mental health using personality traits, wearable sensors and mobile phones", in Wearable and Implantable Body Sensor Networks (BSN), 2015 IEEE 12th International Conference on, 2015, pp. 1-6 K. Yano et al., "Measuring Happiness Using Wearable Technology", Hitachi Review, vol. 64, no. 8, pp. 97-104, 2015 J. E. Guevara, R. Onishi, H. Umemuro, K. Yano, and K. Ara, "Personality and mental health assessment: A sensor-based behavior analysis", in ACHI 2011, 2011

The techniques described in each non-patent document perform statistical processing over the entire period (for example, one month) in which stress should be estimated when calculating features for stress estimation from biological signal data. Therefore, a sufficiently high estimation accuracy cannot be obtained.

The reason for this is the overall stress during the period (1 month) in which stress should be estimated in the stress response items of the PSS questionnaire and the simple occupational stress questionnaire, which are psychologically recognized to reflect long-term stress. This is because the experience is not asked, and the existence and frequency of short-term stress experiences during that period are asked. As described above, in the process of estimating the long-term stress, the biological signal information in the time unit having a finer particle size than the period in which the stress should be estimated is not fully utilized. As a result, it is difficult to accurately estimate scores in long-term stress indicators such as PSS questionnaires and simple occupational stress questionnaires.

An object of the present invention is to provide a stress estimation device and a stress estimation method capable of estimating long-term stress with high accuracy.

The stress estimation device according to the present invention connects the biological signal data collected from the subject of stress estimation for the entire period in which the stress should be estimated to generate the biological signal for the entire period, and generates the biological signal data for the entire period. A biological signal component means for generating a plurality of short-term biological signals by connecting over a plurality of short-terms shorter than a period, and a stress feature amount calculated from the whole-period biological signal to obtain a full-period feature amount, which is short-term. It is provided with a stress feature amount calculating means for calculating a stress feature amount from an inter-biological signal and using it as a short-term feature amount, and a stress score estimating means for estimating a stress score from a full-period feature amount and a short-term feature amount.

In the stress estimation method according to the present invention, the biological signal data collected from the subject of the stress estimation is concatenated over the entire period in which the stress should be estimated to generate the biological signal for the entire period, and the biological signal data is collected. Multiple short-term biological signals are generated by connecting them over a plurality of short-terms shorter than the period, and the stress feature amount is calculated from the whole-period biological signal to obtain the full-period feature amount, and the stress is obtained from the short-term biological signal. The feature amount is calculated and used as a short-term feature amount, and the stress score is estimated from the full-period feature amount and the short-term feature amount.

The stress estimation program according to the present invention generates a biological signal by connecting the biological signal data collected from the subject of the stress estimation to a computer over the entire period in which the stress should be estimated, and generates the biological signal data for the entire period. Is processed to generate a plurality of short-term biological signals by concatenating each of a plurality of short-terms shorter than the entire period, and a stress feature amount is calculated from the whole-period biological signal to obtain a full-period feature amount. A process of calculating a stress feature amount from an inter-biological signal to obtain a short-term feature amount and a process of estimating a stress score from a full-period feature amount and a short-term feature amount are executed.

According to the present invention, long-term stress can be estimated with high accuracy.

It is a block diagram which shows the 1st Embodiment of a stress estimation apparatus. It is a flowchart which shows the operation of the stress estimation apparatus of 1st Embodiment. It is a block diagram which shows the 2nd Embodiment of a stress estimation apparatus. It is a flowchart which shows the operation of the stress estimation apparatus of 2nd Embodiment. It is a block diagram which shows the 3rd Embodiment of a stress estimation apparatus. It is a flowchart which shows the operation of the stress estimation apparatus of 3rd Embodiment. It is explanatory drawing which shows the overall configuration example of the stress estimation apparatus of 1st Example. It is a block diagram which shows the configuration example of an analysis server. It is explanatory drawing which shows the performance index of the 1st Example model and the reference model. It is a block diagram which shows the overall structure example of the stress estimation apparatus of 2nd Example. It is a block diagram which shows the main part of a stress estimation apparatus. It is a block diagram which shows the main part of another aspect of a stress estimator. It is a block diagram which shows the main part of still another aspect of a stress estimator. It is a block diagram which shows the main part of another aspect of a stress estimator.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

Embodiment 1.
FIG. 1 is a block diagram showing a first embodiment of the stress estimation device. In the first embodiment, the stress estimation device is realized by the information processing server 100. The information processing server 100 includes a biological signal storage unit 101, a biological signal component unit 102, a full-period biological signal storage unit 103, a short-term biological signal storage unit 104, a stress feature amount calculation unit 105, and a stress score estimation unit. 106 and a stress score output unit 107 are included.

The biological signal configuration unit 102, the stress feature amount calculation unit 105, and the stress score estimation unit 106 are set to 1 in the information processing server 100 according to, for example, a stress estimation program stored in a storage unit (not shown) in the information processing server 100. This can be achieved by having one or more processors (eg, CPU: Central Processing Unit) perform the processing. The biological signal storage unit 101, the full-period biological signal storage unit 103, and the short-term biological signal storage unit 104 can be realized by a storage device (not shown) in the information processing server 100.

The storage unit in which the stress estimation program is stored is, for example, a non-temporary non-volatile memory such as a ROM (Read Only Memory), a flash memory, or a hard disk. Further, the storage device that realizes the biological signal storage unit 101, the biological signal storage unit 103 for the entire period, and the biological signal storage unit 104 for a short period of time is, for example, a hard disk, a flash memory, or an SSD (Solid State Drive).

Further, in the present embodiment, the stress estimation device is realized by the information processing server 100, but the components of the stress estimation device can also be realized by a hardware circuit.

The biological signal storage unit 101 stores the biological signal data collected from the subject of stress estimation. The biological signal component 102 uses the biological signal data stored in the biological signal storage unit 101. The biological signal component 102 connects the biological signal data over a predetermined period. In the present embodiment, the predetermined period includes a total period for which the stress score should be calculated and a period shorter than the total period (short period: for example, one day). For example, when the entire period is one month, the biological signal component 102 concatenates the biological signal data for one month to obtain a biological signal for the entire period. When the short period is one day, the biological signal data of each day from the first day to the last day of January are concatenated to obtain a short-term biological signal for each day.

Hereinafter, as an example, the entire period will be one month and the short period will be one day. _{Let the} sequence P N represented by the equation (1) be a certain biological signal for one day on the Nth day. The "certain biological signal" may be any one-dimensional signal such as a heartbeat, a pulse wave, a numerical value of skin conductivity, and an acceleration in the X-axis direction. In the sequence, p _n represents the nth biological signal data itself. D is the number of biological signals for one day. For example, if the sampling rate is SR [Hz], then D = 24 * 3600 * SR, except in exceptional cases where leap seconds are inserted.

The biological signal (biological signal for one month) Q for the entire period is expressed by Eq. (2).

In equation (2), M indicates the number of days in one month. M can be any number of 28, 29, 30, and 31 depending on the situation. The biological signal data stored in the biological signal storage unit 101 _{is R L} , which is a set of _{fragmented p n as shown below.} Where E _L > S _L.

For example, suppose that the biological signal data on the specific N'day is R _L' , R _L'+1 and R _L'+2 . At this time, by connecting RL', RL'+1 and RL'+2 in order, the biological signal P _{N'of the N'day} is generated.

There may be an empty period (the period during which the collection of _{R L} is stopped) between the time when each R _{L is collected.} In that case, information indicating the existence of an empty period is also recorded in the biological signal storage unit 101. In this case, P _N'has fewer components than D.

Also, when there is an empty period between the first data p _{(N'-1) D} of the N'day and the first data p _{SL'of R L'} , the biological signal storage unit 101 can be used. Information indicating the existence of such an empty period is recorded. Even if there is a blank period between the last data on the N'day and the last data on R _L'+2 , the biological signal storage unit 101 records information indicating the existence of such a blank period. Will be done.

In this way, _{along with the data P N'for} one day of the N'day, the information of the part where the data is missing is also recorded. Let the record be _{N P N'} . If the daily break position _{exists in the middle of R L} , the biological signal component 102 performs a process of cutting _{R L.}

As described above, the biological signal component 102 can configure P N'and _{N P N'from an appropriate combination of R L.} Similarly, the biological signal configuration unit 102, it is also possible to configure the Q from the combination of suitable R _L. As for Q, the information NQ of the part where the data is missing is recorded in the biological signal storage unit 101. In this case, Q has fewer components than MD.

The process described above is referred to herein as "consolidation". In the present embodiment, the total period is one month and the short period is one day, but these two periods may be of any length as long as the short period is twice as long as the total period or less.

The whole-period biological signal storage unit 103 stores the biological signal for the entire period (biological signal for the entire period) output by the biological signal component unit 102. The short-term biological signal constituent unit 104 stores a short-term biological signal (short-term biological signal) output by the biological signal constituent unit 102.

The stress feature amount calculation unit 105 calculates the stress feature amount for each of the full-period biological signal and the short-term biological signal, and outputs the stress feature amount as the full-period feature amount and the short-term feature amount. As the stress features, as listed in Non-Patent Documents 1, 2, 3 and 4, the features statistically processed according to the action time such as sitting, walking, running, desk work, dialogue and sleep (for example, average). , Variance, median, power spectral density, and components of the histogram of the number of peaks over a period of time, such as 30 seconds) are preferably used. In addition, a frequency distribution of the duration of activities above a predetermined threshold is also preferably used.

In the stress score estimation unit 106, the full-term feature amount and the short-term feature amount output from the stress feature amount calculation unit 105 are input. The stress score estimation unit 106 estimates the stress score from the full-term feature amount and the short-term feature amount. The stress score is a score of stress accumulated over a long period of time. For example, it is a score that is recognized to reflect psychological stress, such as the score of a PSS questionnaire. In order to accurately estimate the stress score, the stress score estimation unit 106 sufficiently uses a database in which biological signal data acquired from many subjects in advance and stress scores, which are teacher data, are stored as a set. Use the machine learning model etc. learned in.

The stress score estimation unit 106 may estimate the PSS score itself as the stress score, or may use the stress score or the like to perform classification. For example, the stress score estimation unit 106 sets two or more classes by setting thresholds for one or more stress scores. Then, the stress score estimation unit 106 classifies the stress score using the threshold value. In this case, the calculated stress score is a number or the like that specifies the classification class.

The stress score estimation unit 106 outputs the estimated value of the stress score to the stress score output unit 107.

In the present embodiment, by realizing the above-mentioned functions, a stress score that simultaneously reflects the overall tendency of stress and the tendency of finer time change of stress is estimated. In other words, not only the stress experience for the entire period for which stress should be estimated, but also information on the frequency of stress experiences for each short period, which is psychologically important, can be obtained, so that a more accurate stress score can be estimated. is there.

Next, the operation of the present embodiment will be described with reference to the flowchart of FIG.

First, the biological signal constituent unit 102 acquires time-series biological signal data from the biological signal storage unit 101 (step A1). The biological signal component 102 connects the biological signal data based on the time series information included in the biological signal data (step A2).

Further, the biological signal component 102 sets the short-term biological signal delimiter position from the time-series information of the biological signal data (for example, when the short period is one day, the end of the day is 24:00, and the short period is one week. At one point, when reading (24:00 on Sunday, the end of the week) (step A3), data concatenation is stopped (step A6). Then, the biological signal constituent unit 102 stores the connected biological signal data as a short-term biological signal in the short-term biological signal storage unit 104 (step A8).

Further, when the biological signal component 102 reads the final position of the entire period of the stress estimation period (for example, 24:00 on the 30th day or the 31st day when the total period is one month), the data (step A4). The connection is stopped (step A7). Then, the biological signal constituent unit 102 stores the connected biological signal data as a biological signal for the entire period in the biological signal storage unit 103 for the entire period (step A9).

The full-period biological signal and the short-term biological signal stored by the full-period biological signal storage unit 103 and the short-term biological signal storage unit 104 are input to the stress feature amount calculation unit 105. The stress feature amount calculation unit 105 calculates the stress feature amount for each of the whole-period biological signal and the short-term biological signal. As listed in Non-Patent Documents 1, 2, 3 and 4, for example, the stress feature calculation unit 105 statistically processes stress according to action time such as sitting, walking, running, desk work, dialogue and sleep. Features (eg, mean, variance, median, power spectral density, and components of the histogram of the number of peaks over a period of time, such as 30 seconds, etc.) are calculated (steps A10, A11). The stress feature amount calculation unit 105 may calculate the frequency distribution of the duration of the activity above a certain threshold value as the stress feature amount.

Finally, the stress score estimation unit 106 estimates the stress score using the calculated full-term features and short-term features as described above. In order to accurately estimate the stress score, the stress score estimation unit 106 estimates the stress score using the machine learning model or the like as described above (step A12). The stress score output unit 107 outputs the estimated score.

The effect of this embodiment will be described.

In the present embodiment, the stress score is estimated using both the feature amount based on the whole-period biological signal (long-term feature amount) and the feature amount based on the short-term biological signal (short-term feature amount). That is, not only the stress experience for the entire period in which the stress should be estimated, but also the frequency of the stress experience for each short period of psychological importance can be obtained. Therefore, long-term stress is estimated with higher accuracy.

Embodiment 2.
Next, a second embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 3 is a block diagram showing a second embodiment of the stress estimation device. In the stress estimation device of the second embodiment, a short-term feature amount difference calculation unit 108 is added to the stress estimation device of the first embodiment shown in FIG. In the second embodiment as well, the case where the stress estimation device is realized by the information processing server 100 will be taken as an example.

Components other than the short-term feature amount difference calculation unit 108 in FIG. 2, that is, the biological signal storage unit 101, the biological signal component 102, the full-period biological signal storage unit 103, the short-term biological signal storage unit 104, and the stress feature amount calculation. The configuration and function of the unit 105, the stress score estimation unit 106, and the stress score output unit 107 are the same as the configuration and function of the corresponding elements shown in FIG.

The short-term feature amount difference calculation unit 108 executes the difference calculation for a part of the feature amount based on the short-term feature amount calculated by the stress feature amount calculation unit 105. Difference calculation is effective in the following cases.

For example, when the user is a worker and long-term stress estimation is carried out as a system of the company where he / she works, the biological signal may be acquired only on the working day. In such a case, the state of stress accumulation on the weekend, which is presumed to be important as a period for alleviating stress accumulation, cannot be obtained from the biological signal. Therefore, the short-term feature difference calculation unit 108 calculates the difference between Friday on the weekend and Monday after the beginning of the week. The short-term feature amount difference calculation unit 108 also outputs the difference as a feature amount to the stress score estimation unit 106.

That is, the short-term feature difference calculation unit 108 calculates, for example, the difference between the short-term features before and after the biological signal data non-acquisition period, and outputs the calculation result to the stress score estimation unit 106 as an additional short-term feature. Output.

The short-term feature difference calculation unit 108 also outputs the short-term feature input from the stress feature calculation unit 105 to the stress score estimation unit 106. However, the stress feature amount calculation unit 105 may be configured to directly output the short-term feature amount to the stress score estimation unit 106.

Next, the operation of the present embodiment will be described with reference to the flowchart of FIG.

The processes of steps A1 to A12 shown in FIG. 4 are the same as the processes shown in FIG.

In step B1, the short-term feature amount difference calculation unit 108 executes the difference calculation as described above for the feature amount based on the short-term biological signal calculated by the stress feature amount calculation unit 105.

In the present embodiment, the stress score estimation unit 106 also considers the difference input from the short-term feature amount difference calculation unit 108 when estimating the stress score. It is considered that the difference reflects the degree of reduction or increase of stress during the period when the biological signal data is not acquired. The stress score estimation unit 106 can estimate, for example, the frequency of recovery from a stress experience based on the difference. The stress score estimation unit 106 can perform more accurate stress estimation by adding the difference to the stress score.

In the present embodiment, in addition to the feature amount based on the biological signal for the entire period and the feature amount based on the biological signal for a short period of time, information on the difference regarding the feature amount based on the biological signal for a short period of time can be obtained. Information on the frequency of recovery from stress experiences is available. That information is also used to achieve higher estimation accuracy.

Embodiment 3.
Next, a third embodiment of the present invention will be described with reference to the drawings.

FIG. 5 is a block diagram showing a third embodiment of the stress estimation device. In the stress estimation device of the third embodiment, a feature amount calculation unit 109 is added to the stress estimation device of the second embodiment shown in FIG. Also in the third embodiment, the case where the stress estimation device is realized by the information processing server 100 will be taken as an example.

Components other than the feature amount calculation unit 109 in FIG. 5, that is, the biological signal storage unit 101, the biological signal component unit 102, the whole period biological signal storage unit 103, the short-term biological signal storage unit 104, the stress feature amount calculation unit 105, The stress score estimation unit 106, the stress score output unit 107, and the short-term feature amount difference calculation unit 108 are configured in the same manner as the elements shown in FIG.

The feature amount calculation unit 109 calculates all or part of the short-term feature amount calculated by the stress feature amount calculation unit 105, and all or part of the short-term feature amount difference calculated by the short-term feature amount difference calculation unit 108. As a target, the calculation of the feature amount (feature value) or the detection of the feature value is executed. Examples of the feature value include a maximum value, a minimum value, a second maximum value, a second minimum value, a first quartile, a third quartile, and a standard deviation.

The stress score estimation unit 106 includes not only the difference between the short-term feature amount and the short-term feature amount, but also all or part of the short-term feature amount, or all or part of the short-term feature amount difference (maximum). By considering the values, minimums, second maximums, second minimums, first quartiles, third quartiles, and standard deviations), more clearly psychologically important stresses. You can get information about the experience and its frequency. The stress score estimation unit 106 may consider both the features of all or part of the short-term features and all or part of the difference between the short-term features.

Next, the operation of the present embodiment will be described with reference to the flowchart of FIG.

The processing of steps A1 to A12 and B1 in FIG. 6 is the same as the processing shown in FIG.

In steps C1 and C2, the feature amount calculation unit 109 targets the difference between the short-term feature amount calculated by the stress feature amount calculation unit 105 and the short-term feature amount calculated by the short-term feature amount difference calculation unit 108. The feature values are calculated for all or part of each.

The significance of calculating the feature value will be explained. Take as an example the case where the feature amount (feature value) is proportional to the strength of a certain stress experience. It is assumed that the feature amount is the same for user A and user B during the entire period in which long-term stress should be estimated. Further, the case where each short period is (1/8) of the total period for which long-term stress should be estimated is taken as an example.

As an example, the short-term feature amount of user A is "20, 1, 19, 2, 18, 3, 17, 4", and the short-term feature amount of user B is "11, 10, 11, 10, 11, It is assumed that it was 10, 11, 10 ". From these figures, it can be said that it is User A who frequently experienced strong stress for a short period of time. Further, it can be said that the user B does not have as much stress experience as the user A experienced. Therefore, in this example, the PSS questionnaire score is expected to be higher for User A. Here, as an example, the entire period is divided into eight to make it a short period, but conditions such as how many short periods the entire period is divided into can be arbitrarily set according to the situation.

The maximum, minimum, second maximum, and second minimum of short-term features are to estimate how strong the stress experience was during the entire period during which long-term stress should be estimated. Useful for. In the above example of user A and user B, the maximum value of user A is "20" and the second maximum value is "19". The maximum value of user B is "11", and the second maximum value is also "11". The reason why not only the maximum value but also the minimum value is used is that the individual features include not only those proportional to the stress experience but also those that are inversely proportional or inversely proportional. ..

The first and third quartiles help estimate the frequency of intense stress experiences during the entire period during which long-term stress should be estimated. In the above example of user A and user B, the first quartile of user A is "19" and the first quartile of user B is "11". The reason why not only the first quartile but also the third quartile is used as a feature amount is that, as in the case of the maximum value and the minimum value, some of the individual feature amounts are proportional to the stress experience. This is because it includes not only those that do, but also those that are inversely proportional or inversely proportional.

The standard deviation (variation in stress experience) helps to estimate the difference between a weak stress experience and a strong stress experience during the entire period during which long-term stress should be estimated. Even if the features are the same for the entire period, the large difference between the weak stress experience and the strong stress experience means that there were frequent and relatively strong stress experiences. For example, in the example of user A and user B, the standard deviation of user A is 8.64 and the standard deviation of user B is 0.53.

Further, when the difference calculation is executed for four pairs of two adjacent short-term features for the short-term features of user A and user B, "19, 17, 15, 13" for user A are executed. , "1, 1, 1, 1" for user B. When the maximum value, the second maximum value, the first quartile, the standard deviation, and the like are calculated for these, the user A is higher.

In the present embodiment, the feature amount is calculated for the difference between the short-term feature amount and the short-term feature amount. It is more clearly psychologically important by using the feature values related to the difference between the short-term feature and the short-term feature as well as the information on the difference between the full-term feature, the short-term feature, and the short-term feature. Index can be obtained.

Specifically, the stress score estimation unit 106 also considers the result value (feature value) calculated by the feature amount calculation unit 109 when estimating the stress score. That is, the feature values (maximum value, minimum value, second maximum value, second minimum value, 14th) relating to all or part of the short-term feature amount or all or part of the difference of the short-term feature amount. Quantiles, third quartiles, and standard deviations, etc.) are also considered. The stress score estimation unit 106 can perform more accurate stress estimation by considering them as well.

Example 1.
Next, the first embodiment will be described with reference to FIGS. 7 and 8. The first embodiment is an embodiment corresponding to the first embodiment. In the following description, specific conditions in an experiment actually performed to confirm the effect of the first embodiment are also referred to.

FIG. 7 is an explanatory diagram showing an overall configuration example of the stress estimation device of the first embodiment. In the example shown in FIG. 7, the analysis server 400 as the information processing server is configured to be able to communicate with the biological signal sensors 420A and 420B and the information display devices 430A and 430B via the communication means 410a, 410b, 410c and 410d. Has been done.

The biological signal sensors 420A and 420B acquire the biological signals of the user who is a worker during working hours. As the user's biological signals, skin surface electrical activity that reflects the user's sweating, triaxial acceleration that reflects the user's body movement, and user's body temperature, as listed in Non-Patent Documents 1, 2, 3, and 4. Skin surface temperature) Pulse wave, heartbeat, voice and the like.

As an example, the biological signal sensors 420A and 420B are wristband type sensors as described in Non-Patent Documents 1 and 2. However, sensors such as badge type or employee ID type as described in Non-Patent Documents 3 and 4 can also be suitably used as biological signal sensors 420A and 420B.

In the actual experiment, Empatica's E4 sensor was used. The E4 sensor acquires skin conductivity, triaxial acceleration, pulse wave, and skin surface temperature data at sampling rates of 4 Hz, 32 Hz, 64 Hz, and 1 Hz, respectively. The acquired data is stored in the built-in memory of the biological signal sensors 420A and 420B.

The communication means 410a, 410b, 410c transmits the biological signal data acquired by the biological signal sensors 420A, 420B to the analysis server 400.

In the actual experiment, the E4 sensor is connected to the personal computer by the attached USB (Universal Serial Bus) cable as the communication means 410a and 410b. The personal computer uploads the biometric signal data to Empatica Cloud via the wireless LAN (Local Area Network) as the communication means 410c and the Internet as the communication means 410d by the installed dedicated software. The analysis server 400 downloads the biological signal data from Empatica Cloud.

FIG. 8 is a block diagram showing a configuration example of the analysis server. The analysis server 400 includes a communication interface 111, a biological signal storage unit 101, a biological signal component unit 102, a full-period biological signal storage unit 103, a short-term biological signal storage unit 104, a stress feature amount calculation unit 105, and a stress score estimation unit 106. , The stress score output unit 107, and the stress score storage unit 110.

The biological signal configuration unit 102, the stress feature amount calculation unit 105, and the stress score estimation unit 106 follow a processor (for example, a CPU) in the analysis server 400 according to a program stored in a storage unit (not shown) in the analysis server 400, for example. ) Can be realized by executing the process. The biological signal storage unit 101, the full-period biological signal storage unit 103, the short-term biological signal storage unit 104, and the stress score storage unit 110 can be realized by a storage device (not shown) in the analysis server 400.

The biological signal storage unit 101 stores the biological signal data received by the communication interface 111. The biological signal component 102 generates a biological signal for the entire period using the biological signal data stored in the biological signal storage unit 101. The biological signal component 102 also generates a biological signal for a short period of time. The full-term biological signal and the short-term biological signal are stored in the full-period biological signal storage unit 103 and the short-term biological signal storage unit 104.

In the actual experiment, the total period was one month and the short period was one week.

The stress feature amount calculation unit 105 calculates the stress feature amount from the biological signal for the entire period of one month. In addition, the stress feature amount calculation unit 105 calculates the stress feature amount from a short-term biological signal every week. As the stress features, as described above, the features statistically processed according to the action time such as sitting, walking, running, desk work, dialogue and sleep (for example, mean, variance, median, power spectral density, and 30). A component of a histogram of the number of peaks in a certain period such as seconds) or a frequency distribution of the duration of activity above a certain threshold value is preferably used.

In an actual experiment, as stress features, the average, dispersion, median value, power spectral density, and a certain period such as 30 seconds, which are disclosed in Non-Patent Document 1, such as sweating, body movement, and skin surface temperature, are used. All of the components of the peak number histogram of were calculated. In addition, for them, calculations were performed for the three active states of sitting, walking and running, and the total of all of them. In Non-Patent Document 1, it is said that the feature amount is calculated for the whole sleep and the first, second, third, and fourth quarters during sleep. However, in this embodiment, since the target is a worker and only the data during working hours is acquired, the data during sleep is not used. Moreover, in this example, the heartbeat that can be acquired by Empatica E4 was not used. Further, the power spectral density, the histogram element of the number of peaks in a certain period, and the like depend on the magnitude of the data acquisition time. Therefore, they are normalized by dividing them by the actual data acquisition time (pure data acquisition time excluding all the time when data could not be acquired properly due to a mounting error or the like and the time when the data was not mounted).

The stress score estimation unit 106 estimates long-term stress using the input long-term features for one month and short-term features for each week.

In the actual experiment, PSS was estimated by regression analysis as long-term stress. The score calculated from the PSS questionnaire conducted at the end of the experimental period (1 month) is used as teacher data for the user, and the linear regression model is trained using the feature amount calculated by the stress feature amount calculation unit 105. The PSS was estimated using the model. In the estimation, the Leave One-person Out Cross Validation method was used. That is, in order to estimate the PSS score of one user, a trained model was used in which all the other users were used as training data and the PSS scores and features of those users were used.

In addition, the performance of the model was evaluated. At that time, the regularization coefficient for the hyperparameters of the linear regression model was fixed at 0.2. In addition, the training sample was increased by the Over Sampling method for users in the high PSS score area and low sample area users who tend to have a small number of samples.

Specifically, we introduced a process to increase the number of training samples by Over Sampling for the top 20% and bottom 20% of users in the PSS score. Regarding the Over Sampling ratio, 1: 1 (without Over Sampling), 10: 1 (when increasing the training sample of the top 20% and bottom 20% by 10 times), and 1: 0 (top 20% and bottom). The processing of three cases (when not learning the medium 60% sample other than 20%) was executed. Regarding the selection of the feature quantity, the processing of the top 5 ranks, the top 10 ranks, and the top 20 ranks of the correlation coefficient with PSS was executed. For the above 9 cases (3 × 3), the correlation coefficient between the PSS score estimated by linear regression and the actual score was calculated for the three cases of the correlation coefficient with PSS. In addition, the error of the estimated PSS score with respect to the actual score, specifically, RMSE (Rooted Mean Square Error) was calculated as a performance index.

Furthermore, the users with the top 20% of PSS scores were defined as the "high stress group", and the recall curve and the precision rate curve for finding the users in the high stress group were created. Then, the area under the curves (Area Under Curve) was also calculated as a performance index. The model including the above performance index is referred to as the "first embodiment model".

In addition, in order to confirm the effect of the first embodiment, the performance index of a model (referred to as "reference model") using only the full-term features for one month without using the short-term features is also calculated. did. FIG. 9 shows the results when the correlation coefficient is maximum for each of the first embodiment model and the reference model. In the example shown in FIG. 9, the first embodiment model outperforms the reference model for all performance indicators.

The stress score estimated by the stress score estimation unit 106 is recorded in the stress score storage unit 110 via the stress score output unit 107.

The stress score can be supplied to the personal computer via the communication interface 111, the Internet, and the wireless LAN at the request of the user. In the personal computer, the stress score is displayed on the information display devices 430A and 430B.

In this embodiment, the functions of the information display devices 430A and 430B and the functions of the biological signal sensors 420A and 420B are realized by different components, but those functions may be realized by one terminal. ..

Example 2.
Next, a second embodiment will be described with reference to FIG. The second embodiment is also an embodiment corresponding to the first embodiment. FIG. 10 is a block diagram showing an overall configuration example of the stress estimation device of the second embodiment. In the second embodiment, the stress estimation device is constructed in the wearable terminal 500.

The biological signal sensor 420 corresponding to the biological signal sensors 420A and 420B shown in FIG. 7 is built in the wearable terminal 500. Further, the information display device 430 corresponding to the information display devices 430A and 430B shown in FIG. 7 is built in the wearable terminal 500.

The wearable terminal 500 further incorporates an analyzer 510. The analyzer 510 includes a biological signal storage unit 101, a biological signal component 102, a full-period biological signal storage unit 103, a short-term biological signal storage unit 104, a stress feature amount calculation unit 105, a stress score estimation unit 106, and a stress score output unit. Includes 107, and stress score storage 110.

Compared with the first embodiment, in the second embodiment, there is no process of transmitting and receiving signals via the components corresponding to the communication means 410a, 410b, 410c, 410d. Therefore, the function of the stress estimation device is realized even in a situation where communication means cannot be obtained.

The operation of the stress estimation device other than the process of transmitting and receiving signals is the same as the operation of the first embodiment.

Although a wristwatch-type terminal is illustrated as the wearable terminal 500 in FIG. 10, the wearable terminal 500 is not limited to the wristwatch-type terminal, and a spectacle-type terminal or the like can also be preferably used.

As described above, in the above embodiment, long-term stress can be estimated with high accuracy. The reason is that short-term stress, which is psychologically important, is calculated by calculating features for stress estimation not only for the entire period during which long-term stress should be estimated, but also for shorter time units. This is because it is possible to incorporate information on the presence or absence of experience and its frequency into the estimation of long-term stress.

FIG. 11 is a block diagram showing a main part of the stress estimation device. As shown in FIG. 11, the stress estimation device 10 connects the biological signal data collected from the subject of stress estimation over the entire period in which the stress should be estimated to generate a biological signal for the entire period, and the biological signal is generated. It is realized by the biological signal constituent means 12 (in the embodiment, the biological signal constituent unit 102) that connects the signal data over each of a plurality of short periods shorter than the entire period to generate a plurality of short-term biological signals. ) And the stress feature amount calculating means 15 (in the embodiment, the stress feature amount is calculated from the whole period biological signal and used as the whole period feature amount, and the stress feature amount is calculated from the short period biological signal and used as the short period feature amount. It is realized by the stress feature amount calculation unit 105) and the stress score estimation means 16 (in the embodiment, the stress score estimation unit 106) that estimates the stress score from the full-period feature amount and the short-term feature amount. ) And.

FIG. 12 is a block diagram showing a main part of another aspect of the stress estimation device. The stress estimation device 10 shown in FIG. 12 is further generated by a full-period biological signal storage means 13 (in the embodiment, realized by the full-period biological signal storage unit 103) that stores the generated biological signal for the entire period. A short-term biological signal storage means 14 (in the embodiment, realized by the short-term biological signal storage unit 104) for storing the short-term biological signal is provided, and the stress feature amount calculation means 15 provides the biological signal for the entire period. The full-period feature amount is calculated using the full-period biological signal stored in the storage means, and the short-term feature amount is calculated using the short-term biological signal stored in the short-term biological signal storage means.

FIG. 13 is a block diagram showing a main part of another aspect of the stress estimation device. The stress estimation device 10 shown in FIG. 13 further calculates the difference between the short-term features before and after the biological signal data non-acquisition period, and outputs the calculation result to the stress score estimation means 16 as an additional short-term feature. The short-term feature amount replenishment means 18 (in the embodiment, realized by the short-term feature amount difference calculation unit 108) is provided.

FIG. 14 is a block diagram showing a main part of another aspect of the stress estimator. The stress estimation device 10 shown in FIG. 14 further calculates a feature value from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount output by the short-term feature amount replenishing means 18, and obtains the calculation result. As an additional short-term feature amount, a short-term feature amount calculation means 19 (in the embodiment, realized by the feature amount calculation unit 109) that outputs to the stress score estimation means 16 is provided.

Although some or all of the above embodiments may be described as in the appendix below, the configuration of the present invention is not limited to the following configuration.

(Appendix 1) Biological signal data collected from the subject of stress estimation is concatenated over the entire period in which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from the entire period. A biological signal component means for generating a plurality of short-term biological signals by connecting them over a plurality of short periods, respectively.
A stress feature amount calculating means for calculating a stress feature amount from the whole-period biological signal and using it as a full-time feature amount, and calculating a stress feature amount from the short-term biological signal and using it as a short-term feature amount.
A stress estimation device including a stress score estimating means for estimating a stress score from the whole-period feature amount and the short-term feature amount.

(Appendix 2) An all-period biomedical signal storage means for storing the generated biomedical signal for the entire period,
A short-term biological signal storage means for storing the generated short-term biological signal is provided.
The stress feature amount calculating means calculates the whole period feature amount using the whole period biological signal stored in the whole period biological signal storage means, and stores it in the short period biological signal storage means. The stress estimation device of Appendix 1 that calculates the short-term feature amount using the short-term biological signal.

(Appendix 3) A short-term feature replenishment means that calculates the difference between the short-term features before and after the period during which the biological signal data has not been acquired and outputs the calculation result as an additional short-term feature to the stress score estimation means. Further provided, the stress estimation device of Appendix 1 or Appendix 2.

(Appendix 4) A feature value is calculated from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount output by the short-term feature amount replenishing means, and the calculation result is added to the additional short-term feature. The stress estimation device according to Appendix 3, further comprising a short-term feature amount calculation means that outputs a quantity to the stress score estimation means.

(Appendix 5) The biological signal data is a part or all of the signals of skin electrical conductivity, skin surface temperature, pulse wave, heartbeat, voice, and acceleration. ..

(Appendix 6) Biological signal data collected from the subject of stress estimation is concatenated over the entire period in which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from the entire period. Also short, multiple short-term biosignals are generated by connecting them over each of the short-term, respectively.
The stress feature amount was calculated from the whole period biological signal and used as the whole period feature amount, and the stress feature amount was calculated from the short period biological signal and used as the short period feature amount.
A stress estimation method for estimating a stress score from the full-term features and the short-term features.

(Appendix 7) The generated biomedical signal for the entire period is stored in the biological signal storage means for the entire period.
The generated short-term biological signal is stored in the short-term biological signal storage means, and the generated short-term biological signal is stored.
The whole-period feature amount is calculated using the whole-period biological signal stored in the whole-period biological signal storage means, and the short-term biological signal stored in the short-term biological signal storage means is used. The stress estimation method of Appendix 6 for calculating the short-term feature amount.

(Appendix 8) The difference between the short-term features before and after the period during which the biological signal data has not been acquired is calculated, and the calculation result is used as an additional short-term feature used for estimating the stress score. Appendix 6 or Appendix 7 Stress estimation method.

(Appendix 9) An additional short-term feature value is calculated from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount based on the difference, and the calculation result is used for stress score estimation. The stress estimation method of Appendix 8 as an inter-feature amount.

(Appendix 10) To the computer
Biological signal data collected from the subject of stress estimation is concatenated over the entire period for which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from a plurality of biological signal data shorter than the entire period. The process of generating multiple short-term biological signals by connecting them over a short period of time,
The process of calculating the stress feature amount from the whole period biological signal and using it as the whole period feature amount, and calculating the stress feature amount from the short period biological signal and using it as the short period feature amount.
A stress estimation program for executing a process of estimating a stress score from the full-term feature amount and the short-term feature amount.

(Appendix 11) To the computer
A process of storing the generated biological signal for the entire period in the biological signal storage means for the entire period,
A process of storing the generated short-term biological signal in the short-term biological signal storage means, and
The whole-period feature amount is calculated using the whole-period biological signal stored in the whole-period biological signal storage means, and the short-term biological signal stored in the short-term biological signal storage means is used. The stress estimation program of Appendix 10 for executing the process of calculating the feature amount for a short period of time.

(Appendix 12) To the computer
Addendum 10 or 11 to calculate the difference between the short-term features before and after the period during which the biological signal data has not been acquired, and to use the calculation result as an additional short-term feature used for estimating the stress score. Stress estimation program.

(Appendix 13) To the computer
A feature value is calculated from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount based on the difference, and the calculation result is used as an additional short-term feature amount used for stress score estimation. The stress estimation program of Appendix 12 for executing the processing to be performed.

(Appendix 14) When the stress estimation program is a non-temporary recording medium in which the stress estimation program is stored and the stress estimation program is executed by the processor,
Biological signal data collected from the subject of stress estimation is concatenated over the entire period for which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from a plurality of biological signal data shorter than the entire period. Generate multiple short-term biosignals by linking them over a short period of time.
The stress feature amount was calculated from the whole period biological signal and used as the whole period feature amount, and the stress feature amount was calculated from the short period biological signal and used as the short period feature amount.
The stress score is estimated from the total-period features and the short-term features.

(Appendix 15) When the stress estimation program is executed on the processor,
The generated biological signal for the entire period is stored in the biological signal storage means for the entire period, and the biosignal is stored in the biological signal storage means for the entire period.
The generated short-term biological signal is stored in the short-term biological signal storage means, and the generated short-term biological signal is stored.
The whole-period feature amount is calculated using the whole-period biological signal stored in the whole-period biological signal storage means, and the short-term biological signal stored in the short-term biological signal storage means is used. The short-term feature amount is calculated.
The recording medium of Appendix 14.

(Appendix 16) When the stress estimation program is executed on the processor,
The recording medium of Appendix 14 or Appendix 15, which calculates the difference between the short-term features before and after the period during which the biological signal data has not been acquired, and uses the calculation result as an additional short-term feature used for estimating the stress score.

(Appendix 17) When the stress estimation program is executed on the processor,
A feature value is calculated from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount based on the difference, and the calculation result is used as an additional short-term feature amount used for stress score estimation. The recording medium of Appendix 16 for executing the processing to be performed.

10 Stress estimation device 12 Biological signal constituent means 13 Full-period biosignal storage means 14 Short-term biosignal storage means 15 Stress feature amount calculation means 16 Stress score estimation means 18 Short-term feature amount replenishment means 100 Information processing server 101 Biological signal storage unit 102 Biological signal component unit 103 Full-period biological signal storage unit 104 Short-term biological signal storage unit 105 Stress feature amount calculation unit 106 Stress score estimation unit 107 Stress score output unit 108 Short-term feature amount difference calculation unit 109 Feature amount calculation unit 110 Stress Score storage 111 Communication interface 400 Analysis server 410a, 410b, 410c, 410d Communication means 420, 420A, 420B Biological signal sensor 430, 430A, 430B Information display device 500 Wearable terminal 510 Analysis device

Claims (10)

Biological signal data collected from the subject of stress estimation is concatenated over the entire period for which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from a plurality of biological signal data shorter than the entire period. Biological signal constituent means for generating a plurality of short-term biosignals by connecting them over a short period of time,
A stress feature amount calculating means for calculating a stress feature amount from the whole-period biological signal and using it as a full-time feature amount, and calculating a stress feature amount from the short-term biological signal and using it as a short-term feature amount.
A stress estimation device including a stress score estimating means for estimating a stress score from the whole-period feature amount and the short-term feature amount. An all-period biological signal storage means for storing the generated all-period biological signal,
A short-term biological signal storage means for storing the generated short-term biological signal is provided.
The stress feature amount calculating means calculates the whole period feature amount using the whole period biological signal stored in the whole period biological signal storage means, and stores it in the short period biological signal storage means. The stress estimation device according to claim 1, wherein the short-term feature amount is calculated using the short-term biological signal. A claim further comprising a short-term feature replenishment means for calculating the difference between the short-term features before and after the period during which the biological signal data has not been acquired and outputting the calculation result to the stress score estimation means as an additional short-term feature. 1 or the stress estimation device according to claim 2. A feature value is calculated from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount output by the short-term feature amount replenishing means, and the calculation result is used as an additional short-term feature amount as the stress. The stress estimation device according to claim 3, further comprising a short-term feature amount calculation means for outputting to the score estimation means. The stress according to any one of claims 1 to 4, wherein the biological signal data is a part or all of a signal of skin electrical conductivity, skin surface temperature, pulse wave, heartbeat, voice, and acceleration. Estimator. Biological signal data collected from the subject of stress estimation is concatenated over the entire period for which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from a plurality of biological signal data shorter than the entire period. Generate multiple short-term biosignals by linking them over a short period of time.
The stress feature amount was calculated from the whole period biological signal and used as the whole period feature amount, and the stress feature amount was calculated from the short period biological signal and used as the short period feature amount.
A stress estimation method for estimating a stress score from the full-term features and the short-term features. The generated biological signal for the entire period is stored in the biological signal storage means for the entire period, and the biosignal is stored in the biological signal storage means for the entire period.
The generated short-term biological signal is stored in the short-term biological signal storage means, and the generated short-term biological signal is stored.
The whole-period feature amount is calculated using the whole-period biological signal stored in the whole-period biological signal storage means, and the short-term biological signal stored in the short-term biological signal storage means is used. The stress estimation method according to claim 6, wherein the short-term feature amount is calculated. The sixth or seventh aspect, wherein the difference between the short-term features before and after the period during which the biological signal data has not been acquired is calculated, and the calculation result is used as an additional short-term feature used for estimating the stress score. Stress estimation method. A feature value is calculated from the short-term feature amount calculated from the short-term biological signal or the short-term feature amount based on the difference, and the calculation result is used as an additional short-term feature amount used for stress score estimation. The stress estimation method according to claim 8. On the computer
Biological signal data collected from the subject of stress estimation is concatenated over the entire period for which stress should be estimated to generate a biological signal for the entire period, and the biological signal data is obtained from a plurality of biological signal data shorter than the entire period. The process of generating multiple short-term biological signals by connecting them over a short period of time,
The process of calculating the stress feature amount from the whole period biological signal and using it as the whole period feature amount, and calculating the stress feature amount from the short period biological signal and using it as the short period feature amount.
A stress estimation program for executing a process of estimating a stress score from the full-term feature amount and the short-term feature amount.

Images