JP2021157385A - Device for estimating body condition using training property stress information, program and method - Google Patents

Abstract

To provide a body condition estimation device for estimating a body condition of an object in a time point or a period in the future from a time point or a period related to training.SOLUTION: A body condition estimation device is a device for estimating a body condition determined using a measured value related to a body, and on the basis of body condition information of an object determined for one time point or period, and training property stress information being stress information related to training of a body executed by the object in the one time point or period, uses a body condition estimation model related to learned training, for determining body condition information of the object in a time point or a period in the future from the one time point or period. When it is determined that, there is also required, non training property stress information, using a body condition estimation model related to learned training and non-training, body condition information of the object in the time point or period in the future from the one point or period, is determined, preferably.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for estimating the physical condition of a subject from the information relating to the subject.

In recent years, sports have attracted a great deal of attention as they bring various benefits through participation in competitions, training for them, and human and cultural exchanges derived from them, and efforts to do so are also active. It has become.

Under such circumstances, various methods have been proposed and implemented in order to improve the competitiveness (performance) of athletes. For example, Non-Patent Document 1 including the inventor of the present application as an author adjusts the training content of the day based on the measurement result of the heart rate variability at the time of waking up for an elite runner, and improves the performance of the elite runner. The results are disclosed.

In addition, Non-Patent Document 2 visualizes the physical condition of the day for athletes based on the heart rate variability at the time of waking up, and determines the training content of the day based on the visualized physical condition, and as a result, the performance is improved. It is disclosed. In other words, it shows that it is important to manage the physical condition of athletes in order to improve their performance.

Mitsuka T., Takayama F. and Nabekura Y., “Fine-tuning training strategy based on heart rate variability: A case study of a middle-distance runner”, Sport Performance & Science Reports, 62, v1, pp. 1-3 , 2019 Vesterinen V, Nummela A, Heikura I, Laine T, Hynynen E, Botella J and Hakkinen K. “Individual Endurance Training Prescription with Heart Rate Variability”, Medicine and Science in Sports and Exercise, 48 (7), pp. 1347-1354 , 2016

As described above, as can be understood from the contents disclosed in Non-Patent Documents 1 and 2, adjustment of training contents is effective for improving the performance of athletes, and for that purpose, physical condition management of athletes is important. Has been demonstrated. However, in the past, the physical condition of interest was only at the time of training, and the training content was determined according to the physical condition of the day.

However, since the inventor of the present application is the key to improving the performance, if the future physical condition of the athlete can be predicted, the inventor of the present application may use the prediction result to, for example, adjust the training content more appropriately. I thought that it would be possible, and as a result, it would be possible to further improve performance.

Therefore, an object of the present invention is to provide a physical condition estimation device, a program, and a method capable of estimating the physical condition of a target at a time or period in the future in view of the time or period related to training.

According to the present invention, it is a physical condition estimation device that estimates a physical condition determined by using a measured value related to the body.
Learned based on the physical condition information of the subject determined for one time point or period and the training stress information which is the stress information related to the physical training performed by the subject at the one time point or period. A physical condition estimation device having a physical condition information determining means for determining the physical condition information of the target at a time point or period that is in the future with respect to the one time point or period is provided by using the "physical condition estimation model for training".

Here, it is also preferable that the physical condition information includes an amount determined by using the measured value related to the measurement of the heartbeat.

Further, as an embodiment of the physical condition estimation device according to the present invention, the physical condition information determining means is used.
When it is determined that non-training stress information, which is information on stress other than the stress related to the physical training, is also necessary based on the information on the likelihood or error obtained for the determined physical condition information, a certain time point or period Based on the physical condition information of the estimation target determined for, the training stress information related to the time point or period, and the non-training stress information related to the time point or period, the learned "training and" It is also preferable to use the "non-training physical condition estimation model" to determine the physical condition information of the estimation target at a time point or period that is in the future from the certain time point or period.

Further, as another embodiment of the physical condition estimation device according to the present invention, the physical condition information determining means is used.
Non-information on physical condition related to one time point or period, relevant training stress information related to the one time point or period, and information on stress other than stress related to the physical training at the one time point or period. Based on the trainable stress information, the learned "physical condition estimation model for training and non-training" is used to determine the physical condition information of the target at the time or period that is in the future from the one time or period.
Obtained information on the likelihood or error obtained from the physical condition information determined using the "training-related physical condition estimation model" and the physical condition information determined using the "training and non-training physical condition estimation model". Based on the comparison with the information on the likelihood or error obtained, it was decided to adopt one of these physical condition estimation models.
When adopting the physical condition information of the estimation target determined for a certain time point or period, the training stress information related to the certain time point or period, and the "physical condition estimation model related to training and non-training", it is applicable. Based on the non-training stress information related to the time point or period, the physical condition information of the estimation target at the time point or period that is in the future from the time point or period is determined by using the physical condition estimation model decided to be adopted. It is also preferable to do so.

Further, as a further embodiment of the physical condition estimation device according to the present invention, the physical condition information determining means is used.
Non-information on physical condition related to one time point or period, relevant training stress information related to the one time point or period, and information on stress other than stress related to the physical training at the one time point or period. Based on the trainable stress information, the learned "physical condition estimation model for training and non-training" is used to determine the physical condition information of the target at the time or period that is in the future from the one time or period.
Based on the physical condition information related to one time point or period and the non-training stress information related to the one time point or period, the one time point using the learned "physical condition estimation model related to non-training". Or, determine the physical condition information of the target at a time or period that will be in the future in terms of the period,
Obtained information on the likelihood or error obtained from the physical condition information determined using the "training-related physical condition estimation model" and the physical condition information determined using the "training and non-training physical condition estimation model". Based on the comparison between the information on the likelihood or error obtained and the information on the likelihood or error obtained for the physical condition information determined using the "non-training physical condition estimation model", these physical condition estimation models Decided to adopt one of them,
When adopting the physical condition information of the estimation target determined for a certain time or period and at least the "physical condition estimation model for training" or the "physical condition estimation model for training and non-training", at that certain time or period The training stress information and at least the non-training stress information related to a certain time point or period when the "physical condition estimation model related to training and non-training" or the "physical condition estimation model related to non-training" is adopted. It is also preferable to determine the physical condition information of the estimation target at a time point or period that is in the future from the certain time point or period by using the physical condition estimation model decided to be adopted based on the above.

Further, in the physical condition estimation device according to the present invention, the non-training stress information has at least a predetermined correlation with the measured value related to the body, which is the objective variable, from among the plurality of non-training stress information candidates. It is also preferable that the candidate corresponding to the explanatory variable is selected and determined.

In addition, the physical condition estimation model used or adopted in the physical condition estimation device according to the present invention is one that has been learned from learning data related to a plurality of objects that are similar to each other in the measured values related to the body, or the plurality of objects. It is also preferable that the data is adjusted by the learning data of the estimation target after being trained by the learning data of the above.

Further, it is also preferable that the physical condition estimation model used or adopted in the physical condition estimation device according to the present invention is an estimation model based on a mixed normal distribution, a Bayesian model, or a neural network estimation model.

Furthermore, as yet another embodiment of the physical condition estimation device according to the present invention, the main body condition estimation device is
When the determined physical condition information and the training stress information are used, the training stress information at a predetermined time or period which is the past from the time or period related to the physical condition information and the non-training. When sexual stress information is used, the training stress and / or the non-training stress is based on the non-training stress information at the predetermined time or period which is the past from the predetermined time or period. It is also preferable to further have an advice information generating means for outputting the advice information relating to the decrease, maintenance or increase.

According to the present invention, it is also a physical condition estimation program that functions a computer that estimates a physical condition determined by using measured values related to the body.
Learned based on the physical condition information of the subject determined for one time point or period and the training stress information which is the stress information related to the physical training performed by the subject at the one time point or period. Provided by a physical condition estimation program that makes a computer function as a physical condition information determination means for determining the physical condition information of the target at a time point or period that is in the future from the one time point or period by using the "physical condition estimation model related to training". Will be done.

According to the present invention, further, it is a physical condition estimation method in a computer that estimates a physical condition determined by using measured values related to the body.
Physical condition information of the subject determined for one time point or period, training stress information which is information on stress related to physical training performed by the subject at the one time point or period, and corresponding correct answer data. With the step of generating a "physical condition estimation model related to training" by performing learning processing using the physical condition information of the target at a time point or period that is in the future from the one time point or period. ,
Using the "training-related physical condition estimation model" generated based on the physical condition information of the estimation target determined for a certain time point or period and the training stress information of the estimation target related to the certain time point or period. Therefore, a physical condition estimation method including a step of determining the physical condition information of the estimation target at a time point or period that is in the future with respect to the certain time point or period is provided.

According to the physical condition estimation device, the program and the method of the present invention, the physical condition of the target at the time or period in the future can be estimated from the time or period related to the training (training).

It is a functional block diagram which shows the functional structure in one Embodiment of the physical condition estimation apparatus by this invention. It is a schematic diagram for demonstrating the embodiment of the model selection process in the model selection part which concerns on this invention. It is a flowchart which shows the outline of one Embodiment in the physical condition estimation method by this invention.

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

[Physical condition estimation device]
FIG. 1 is a functional block diagram showing a functional configuration according to an embodiment of the physical condition estimation device according to the present invention.

The physical condition estimation device 1 of the present embodiment shown in FIG. 1 is
(A) From the training stress information database (DB) 2 which is installed outside and stores "training stress information (training stress information)" generated from data collected by a fitness monitor or the like. For example, by communicating, "training stress information" related to an estimation target (for example, a certain player) is acquired, and
(B) From the non-training stress information DB3, which is also installed externally and stores "non-training stress information (non-training stress information)" generated from the data collected by the activity tracker or the like. For example, by communicating, "non-training stress information" related to the estimation target (the athlete concerned) is acquired, and
(C) From the heart rate measurement information DB4, which is also installed externally and stores "heart rate measurement information" generated from the heart rate measurement value data at the time of waking up collected by a heart rate measurement device or the like, for example, by communication. Acquire "heart rate measurement information" related to the estimation target (the athlete),
Using these acquired information, it is possible to estimate the "physical condition" of the estimation target (the athlete) at a future time or period (for example, when waking up the next day) in view of the time or period related to this information. It is a device.

Here, "trainable stress information" is information on stress related to physical training (training), for example.
(A) Time of training, (b) Intensity of training (for example, heart rate during training or its increase), (c) Time of training, and (d) Style of training (d) For example, running, biking, boating, or strength training)
The information related to is applicable to it.

Further, "non-training stress information" is stress information other than stress related to physical training (training), and is, for example, lifestyle-related information that can be calculated by an activity tracker or the like.
Information on (a) activity, (b) steps, (c) energy intake, (d) energy ratio, (e) sleep time, (f) wake-up time, and (g) sleep efficiency is "non-training". It can be listed as "stress information", and as information that can be collected from questionnaire apps, diary apps, etc.
(H) Subjective stress level, (i) Working hours, (j) Work relationships, (k) Relationships with family members or cohabitants, and (l) Implementation status of stress reduction activities (eg yoga, family) Chat time with friends, etc., whether or not to eat out, and the content and degree of leisure activities)
The information related to the above may be referred to as "non-training stress information".

Further, the "physical condition" is an amount determined by using the measured value related to the body, and in the present embodiment, the "heart rate measurement information" of the above (c), more specifically, the heart rate variability at the time of waking up (HRV, Heart Rate Variability) The amount is determined from the information. For example, the HRV value (for example, the Ln_RMSSD value described later) itself can be used as the physical condition value, or the range of the HRV value is divided into a plurality of sections and the physical condition value (physical condition information) is associated with each section. The physical condition value associated with the section to which the target HRV value belongs may be used as a value representing the “physical condition” of the target.

Similarly, in FIG. 1, the physical condition estimation device 1 specifically
(A) "Physical condition information" of the estimation target determined for one time point or period, and "trainability" related to the training (training) carried out by the estimation target (for example, a certain athlete) at the one time point or period. Based on the "stress information", using the learned "physical condition estimation model for training", "determine the physical condition information" of the estimation target (the athlete) at the time or period that will be in the future from the one time or period. Physical condition information determination unit 115
It is characterized by having.

Further, as a preferred embodiment of the physical condition estimation device 1, the physical condition information determination unit 115 may be used.
(B1) When it is determined that the "non-training stress information" of the estimation target (the athlete) is also necessary for the estimation process based on the information related to the likelihood or error obtained for the determined "physical condition information".
(B2) "Physical condition information" of the estimation target determined for a certain time or period, "training stress information" of the estimation target (the athlete) for the certain time or period, and the certain time or period Based on the "non-training stress information" of the estimation target (the athlete), using the learned "physical condition estimation model for training and non-training", the time or period that will be in the future from the certain time or period. It is also preferable to determine the "physical condition information" of the estimation target (the athlete) in.

In this way, according to the physical condition estimation device 1, for example, the "training stress information" of the estimation target (for example, a certain athlete) in a certain day is acquired, and the "training stress information" is used as the "physical condition" of the estimation target in the same day. By inputting to the learned "physical condition estimation model for training" together with ", for example, it is possible to estimate the" physical condition "of the estimation target (the athlete) on the next day.

In this respect, it has been difficult to predict the future "physical condition" in the past, but the inventor of the present application, especially for athletes, has changed the "physical condition" from the past "physical condition" at a certain point in time or period. Obtained the knowledge that it is greatly affected by training stress in many cases, and based on this knowledge, invented the estimation process by the above-mentioned physical condition estimation device 1, thereby relating to training (training). It has made it possible to estimate the "physical condition" of the estimation target at a time point or period that is in the future in view of the time point or period.

In addition, the inventor of the present application states that HRV, especially in athletes, is not limited to daily training, but also various stresses in daily life, such as attending important meetings and parties, and long-term movements for battles. It was found that it can be influenced not a little by such factors.

Therefore, although the physical condition estimation device 1 is one of the embodiments shown in the above (B1) and (B2), the physical condition is estimated by performing the physical condition estimation process using "non-training stress information" if necessary. It is also possible to improve the estimation accuracy of. Here, there are other modes of use of the "non-training stress information" other than the above, and details thereof will be described later as other embodiments.

Furthermore, the inventor of the present application has found that the factors that can affect HRV differ considerably among individuals, especially among individual athletes. For example, it is known that a factor that lowers the HRV value for one player may be a factor that raises the HRV value for another player.

Therefore, it is also preferable that the above-mentioned "physical condition estimation model for training" and the above-mentioned (B2) "physical condition estimation model for training and non-training" are constructed for each individual athlete, for example. Of course, depending on the degree of common factors that affect HRV, it is possible to construct a physical condition estimation model for a certain sports team, and further, a physical condition estimation model for general athletes who perform one sport.

Of course, the estimation target of "physical condition" is not limited to athletes. If the target is to be trained for some purpose and the "physical condition" can be obtained from the measured values related to the body, the physical condition estimation device 1 is used to estimate the future "physical condition" of the target. Is possible.

[Device function configuration, physical condition estimation program]
Similarly, according to the functional block diagram of FIG. 1, the physical condition estimation device 1 includes a communication interface unit 101, a trainable stress information storage unit 102, a non-trainable stress information storage unit 103, a heart rate measurement information storage unit 104, and the like. It has a keyboard (KB) 105, a display (DP) 106, and a processor memory.

Here, this processor memory stores one embodiment of the physical condition estimation program according to the present invention, has a computer function, and executes the physical condition estimation process by executing the physical condition estimation program. do. From this, the physical condition estimation device 1 may be a server, a cloud server, or a dedicated device for physical condition estimation, but is equipped with a physical condition estimation program according to the present invention, for example, a personal computer (PC), a notebook type, or a notebook type. It can also be a tablet computer, a smartphone, or the like.

Further, the processor memory includes a trainable stress feature amount generation unit 111, a non-trainable stress feature amount generation unit 112, a physical condition information generation unit 113, a physical condition estimation model construction unit 114, and a physical condition including a model selection unit 115a. It has an information determination unit 115, an advice information generation unit 116, a communication control unit 121, and an input / output control unit 122.

It should be noted that these functional components can be regarded as the functions of the physical condition estimation program stored in the processor memory. Further, the processing flow shown by connecting the functional components of the physical condition estimation device 1 in FIG. 1 with arrows is also understood as an embodiment of the physical condition estimation method according to the present invention.

By the way, among the above-mentioned functional components, the trainable stress feature amount generation unit 111, the non-trainable stress feature amount generation unit 112, the physical condition information generation unit 113, the physical condition estimation model construction unit 114, and the advice information generation unit At least one or all of 116 may be provided outside the physical condition estimation device 1. That is, if the physical condition estimation device 1 includes the physical condition information determination unit 115, it is possible to carry out the physical condition estimation process by taking in a part or all of the information necessary for the process from the outside.

Similarly, in the functional block diagram of FIG. 1, the trainable stress information storage unit 102 is trainable (training) for each athlete to be estimated, for example, from the trainable stress information DB2 via the communication interface unit 101 and the communication control unit 121. Gender) Acquires stress information, stores and manages it.

Here, as described above, the training stress information includes, for example, (a) the time during which the training (training) was performed, (b) the intensity of the training performed (for example, the heart rate during training or its increase), and (c). At least one, preferably all, of the information relating to the time of day the training was performed and (d) the mode of training performed (eg, running, biking, boating, strength training, etc.).

The training stress feature amount generation unit 111 acquires the training stress information for each athlete to be estimated, for example, from the training stress information storage unit 102, and generates the training stress feature quantity of the athlete. For example, in each of the above (a) to (d), the range of numerical values expressing the degree and content is divided into a plurality of sections, points are linked to each section, and the numerical value of the player to be estimated is concerned. The score associated with the section to which the athlete belongs can be used as the training stress feature amount (component) of the athlete.

Further, for example, when the "intensity of the training performed" in (b) above is featured, the training stress feature amount TSF related to the training intensity is calculated by the following equation (1) TSF = training time (minutes) x "intensity".
It may be calculated by. Here, the value of "intensity" is, for example, 1 when the heart rate measurement value is 50 to 59% of the maximum heart rate, 2 when the heart rate measurement value is 60 to 69% of the maximum heart rate, and the heart rate. 3 if the number measurement is 70-79% of the maximum heart rate, 4 if the heart rate measurement is 80-89% of the maximum heart rate, 4 if the heart rate measurement is 90-100% of the maximum heart rate Can be 5.

Regarding the calculation of such feature TSF, for example, Non-Patent Documents: Sanders et al., “Methods of Monitoring Training Load and Their Relationships to Changes in Fitness and Performance in Competitive Road Cyclists”, International journal of sports physiology and It is described in performance 12 (5), pp.668-675, 2017.

As a modification, the trainable stress feature TSF may include the time interval from the training end time to the next physical condition determination time (for example, the HRV measurement time when waking up the next day) as a coefficient in the above equation (1). preferable. This takes into account the fact that, for example, training stress increases when training is performed immediately before going to bed, in order to calculate the feature amount.

Further, as a further modification mode, the "intensity" in the above equation (1) is not a heart rate measurement value, but is, for example, a subjective exercise intensity value obtained from a questionnaire (a) or a subjective exercise intensity value.
(B) It is also possible to calculate based on the speed value related to training, which is calculated from the displacement information of the target obtained by the GPS (Global Positioning System) built in the fitness monitor.

Similarly, in the functional block diagram of FIG. 1, the non-training stress information storage unit 103 receives, for example, non-training for each athlete to be estimated from the non-training stress information DB 3 via the communication interface unit 101 and the communication control unit 121. Acquires, stores and manages sexual (non-training) stress information.

Here, as the non-training stress information, as described above,
(A) Lifestyle-related information, such as (a) activity, (b) steps, (c) energy intake, (d) energy ratio, (e) sleep time, (f) wake-up time, and ( g) Information on sleep efficiency, as well as
(B) Information that can be collected from questionnaire apps, diary apps, etc., such as (h) subjective stress level, (i) working hours, (j) work relationships, (k) with family members or cohabitants. At least one of the information related to human relations and (l) implementation status of stress reduction activities (for example, yoga, chat time with family and friends, presence / absence of eating out, content / degree of leisure activities), preferably. All may be adopted.

By the way, the lifestyle-related information that can be collected by, for example, the active tracker in (a) above is regarded as an objective index of non-training stress, while the information that can be collected from the questionnaire application, diary application, etc. in (a) above. Can be regarded as a subjective indicator of untrained stress.

The non-training stress feature generation unit 112 acquires non-training stress information for each athlete to be estimated, for example, from the non-training stress information storage unit 103, and generates the non-training stress feature of the athlete. do. For example, in each of the above (a) to (l), the range of numerical values expressing the contents is divided into a plurality of sections, points are linked to each section, and the numerical value of the player to be estimated belongs to. The score associated with the section can be used as the non-training stress feature amount (component) of the athlete.

Further, for example, in the case of quantifying the "subjective stress degree" of the above (h), for example, when the athlete to be estimated becomes aware of some kind of stress, the questionnaire application installed in the terminal carried by the player himself / herself is asked. You may enter the specific type of the stress (for example, "presentation for a large number of people", "preparation of materials", etc.) and the degree (for example, a score of 0 to 100 points). The non-training stress feature amount generation unit 112 can generate a non-training stress feature amount (component) by using the information (type of stress and its score) input to the questionnaire application.

As described above, the non-training stress information (or feature amount) has a wide range of defense, and depending on the estimation target, measurement items related to the body (for calculating the physical condition value), HRV in the present embodiment, There are not a few things that do not affect the stress, and there are often significant individual differences (for example, individual differences of athletes) in the degree of influence. Therefore, multiple regression analysis or the like is performed on the estimation target in advance, and from among a plurality of non-training stress information (or feature quantity) candidates, at least the correlation coefficient with the HRV value, which is the objective variable, or the feature quantity. It is also preferable to select a candidate corresponding to the explanatory variable whose explanatory power for the objective variable is equal to or greater than a predetermined value and determine the non-training stress information (or feature amount).

Here, the other training stress information (or feature amount) can also be selected based on the correlation with the HRV value. However, since training stress information is often a factor that affects physical condition, especially in athletes, such selection processing may be omitted.

Similarly, in the functional block diagram of FIG. 1, the heart rate measurement information storage unit 104 acquires heart rate measurement information for each player to be estimated, for example, from the heart rate measurement information DB4 via the communication interface unit 101 and the communication control unit 121. , Save and manage.

Here, in the present embodiment, as this heart rate measurement information, an electrocardiogram collected by an electrocardiograph (heart rate measuring device) or an HRV generated based on an RR interval (RRI) measured by a heart rate monitor (heart rate measuring device) is generated. (Heart rate variability) information is adopted. As a modification, HRV information as pulse (heartbeat) measurement information may be generated and adopted based on the PP interval (PPI) measured by the photoelectric volumetric pulse wave recording method.

Further, the generation of HRV information as described above is usually carried out by using any one of time domain analysis, frequency analysis, and non-linear analysis. Of these, especially in the case where an athlete is the measurement target, it is common to generate HRV information by "time domain analysis" using the standard deviation of RRI or the like. In this case, specifically, Ln_RMSSD, which is the logarithm of the root mean square of the difference between two RRIs adjacent to each other, is often used as the HRV information.

Of course, as HRV, various known indexes other than the above-mentioned Ln_RMSSD can be adopted, but Ln_RMSSD takes a value that strongly reflects the functional status of the autonomic nerve, particularly the parasympathetic nerve, and is used in short-time measurement. Since it has excellent reproducibility, it is a preferable index for defining physical condition.

When the autonomic nerves are functioning normally, the RR interval has a fluctuation component synchronized with respiration and blood pressure, and therefore a considerable amount of HRV is observed, but on the other hand, due to excessive fatigue or drug collection, etc. If the autonomic nervous function is abnormal, HRV may disappear.

In the present embodiment, the physical condition information generation unit 113 acquires HRV information for each athlete to be estimated, for example, acquired from the heart rate measurement information storage unit 104, and generates physical condition information (physical condition value) of the athlete. In particular,
(A) Ln_RMSSD value at the time of waking up on a certain day,
(B) The Ln_RMSSD value at the time of waking up on the certain day, the average value of the Ln_RMSSD value at the time of waking up in the most recent past several days from the certain day, and (c) the Ln_RMSSD value at the time of waking up on the relevant day. , And the reciprocal of the coefficient of variation in the Ln_RMSSD value at the time of waking up in the last few days from that one day (= (mean value) / (standard deviation))
At least one of them can be a physical condition value (physical condition information). By the way, in this case, it is judged that the smaller the physical condition value is, the worse the physical condition is.

Alternatively, the range of numerical values (sets) related to the above (a), (b) and (c) is divided into a plurality of sections, and the physical condition value (physical condition information) is associated with each section, and the value of the athlete (physical condition information) ( The physical condition value (physical condition information) associated with the section to which the group) belongs may indicate the physical condition of the athlete.

Similarly, in the functional block diagram of FIG. 1, the physical condition estimation model construction unit 114
(A) A set of a trainable stress feature amount generated for one time point or period and a physical condition value (physical condition information) as correct answer data generated for the one time point or period is used as teacher data for "training". Build a "physical condition estimation model"
(B) As the trainable stress feature generated for one time point or period, the non-trainable stress feature generated for that one time point or period, and the correct answer data generated for that one time point or period. A "physical condition estimation model for training and non-training" was constructed using the pair with the physical condition value (physical condition information) of
(C) The set of the non-training stress feature amount generated for one time point or period and the physical condition value (physical condition information) as the correct answer data generated for the one time point or period is used as the teacher data, and is "non-training". Build a "physical condition estimation model for training".

Here, the physical condition estimation model construction unit 114 will be described in detail later, but depending on the content of the physical condition information determination process performed by the physical condition information determination unit 115, the above-mentioned “physical condition estimation model related to training” (a) , And when the above-mentioned (b) "Physical condition estimation model for training and non-training" is also used, "Physical condition estimation model for training" and "Physical condition for training and non-training" When the "estimation model" is constructed and the "non-training physical condition estimation model" is also used, the "training-related physical condition estimation model", "training and non-training physical condition estimation model", And all of the "non-training physical condition estimation model" may be constructed.

Further, any of the physical condition estimation models (a) to (c) above can be an estimation model based on a mixed normal distribution, another Bayesian model, or a neural network estimation model, or various other machine learning algorithms. It is also possible to use an estimation model based on.

Further, all of the physical condition estimation models (a) to (c) may be constructed for a predetermined group (for example, a specific sports team) as described above, and the constituent units of the group (for example, a specific sports team) may be constructed. It may be constructed for one athlete), and may be constructed for general application fields (for example, general athletes who play one competition).

Further, all of the physical condition estimation models (a) to (c) above can be trained by learning data relating to, for example, a plurality of objects (for example, such a plurality of athletes) similar to each other in HRV. .. Specifically, for example, a plurality of objects having similar HRV baselines (reference values) are extracted by, for example, grastering and classifying HRV time series data (multiple objects belonging to the same cluster are extracted). However, the physical condition estimation model may be constructed using the extracted training data related to the plurality of objects. Incidentally, in this case, the constructed physical condition estimation model can be a model common to the objects belonging to the cluster.

Here, as the baseline of HRV, for example, from "the average value of Ln_RMSSD in the last 2 weeks" to "the average value of Ln_RMSSD in the last 2 weeks" and "the coefficient of variation of Ln_RMSSD in the last 2 weeks" (= (standard deviation) / ( It is possible to set a value obtained by subtracting the multiplication value of "0.5 times the average value))" (for such a setting, for example, non-patent literature: Daniel J. Plews et al., "Heart rate variability in elite triathletes". , is variation in variability the key to effective training? A case comparison ”, European Journal of Applied Physiology, 112, pp.3729-3741, 2012). Since the baseline set in this way generally has individual differences, it is also preferable to calculate for each individual (for example, a player). Furthermore, the physical condition estimation model common within the cluster constructed as described above is adjusted by the learning data related to the estimation target (for example, a specific player belonging to the cluster), and the physical condition model dedicated to the estimation target is adjusted. It is also possible to generate.

Here, specifically, as the above adjustment, the (learned) parameter of the constructed physical condition estimation model is set as the initial value of the model parameter, and the learning data related to the estimation target (for example, a specific player) is used again. Learning (fine tuning) may be performed. In any case, such a model construction method is particularly effective when the accumulated source information for feature quantity generation is small.

Similarly, in the functional block diagram of FIG. 1, the physical condition information determination unit 115 including the model selection unit 115a uses the physical condition estimation model generated as described above to provide future physical condition information of the estimation target (for example, a specific player). Determine the physical condition value (estimate the physical condition). Hereinafter, four forms of determining physical condition information that can be taken at this time will be sequentially described.

<First form>
The physical condition information determination unit 115
(A) The physical condition value of the estimated target (the athlete) determined for a certain time point or period (for example, the estimated day), and
(B) Using the "training-related physical condition estimation model" constructed based on the training stress features generated for the estimation target (the athlete) at the certain time point or period (estimation day), the one concerned Determine the physical condition value of the estimation target (the athlete) at the time or period (for example, when waking up the next day) that is in the future from the time or period (estimated day).

As described above, the inventor of the present application has obtained the finding that the past, for example, the change from the previous day in "physical condition" is greatly affected by the training stress in many cases, and based on this finding. Based on this, he invented this <first form>. This makes it possible to estimate the "physical condition" of the estimation target (the athlete) at the time or period (when waking up the next day) in the future from the time or period (estimated day) related to the training.

<Second form>
The model selection unit 115a of the physical condition information determination unit 115
(A) "Likelihood" obtained for the physical condition values determined in the same manner as in the above <first form> for one time point or period (for example, the estimated day) (which is the output of the "physical condition estimation model for training"). Or, based on the "information related to the error", it is determined whether or not the non-training stress feature amount is also "necessary" for the physical condition estimation.
(B) When it is determined that "it is necessary", the "physical condition estimation model for training and non-training" is determined as the physical condition estimation model to be used. On the other hand, when it is determined that "it is not necessary" (it is not "necessary"), the physical condition information determination unit 115 may determine the physical condition value of the above (a) as the estimated physical condition value. Alternatively, the physical condition value may be determined again at a certain time point or period (estimated day) in the same manner as in the above <first form>.

Here, the "information related to the likelihood or error" in (a) above is the likelihood of the input data related to the estimated distribution when the "physical condition estimation model related to training" is, for example, an estimation model based on a mixed normal distribution or a Bayesian model. The degree can be L1. In this case, the model selection unit 115a expects that the likelihood is improved by considering the information related to the non-training property, for example, when the likelihood L1 is less than the predetermined likelihood threshold L_th, and is "necessary". "Can be determined. Alternatively, it is also possible to use the prediction error obtained for the sampled prediction points as "information on the likelihood or error". In this case, the model selection unit 115a may determine that it is "necessary", for example, when the prediction error exceeds a predetermined prediction error threshold value.

Further, when the "physical condition estimation model related to training" is, for example, a neural network estimation model, the "information related to likelihood or error" in (a) above can be a prediction error E1. In this case, the model selection unit 115a may determine that it is "necessary", for example, when the prediction error E1 exceeds a predetermined prediction error threshold value E_th.

Next, when it is decided to use the "physical condition estimation model for training and non-training" based on the determination that it is "necessary", the physical condition information determination unit 115 determines.
(C) The physical condition value of the estimated target (the athlete) determined for a certain point in time or period (estimated day), and
(D) The training stress feature amount generated for the estimation target (the athlete) at the certain time point or period (estimation day), and
(E) Using the "training and non-training physical condition estimation model" constructed based on the non-training stress features generated for the estimation target (the athlete) at the certain time point or period (estimation day). , The physical condition value of the estimation target (the athlete) at the time or period (when waking up the next day) in the future from the one time point or period (estimated day) is determined.

As described above, the inventor of the present application states that the HRV to be estimated is not only for daily training but also for various stresses in daily life, such as attendance at important meetings and parties, and even for a war. We obtained the knowledge that it can be affected not a little by moving for a long time, and invented this <second form> based on this knowledge. This makes it possible to more accurately estimate the "physical condition" of the estimation target (the athlete) at a future time point or period (when waking up the next day).

Incidentally, even if the estimated target is, for example, an elite athlete, it is considered that the influence of the non-training stress feature amount on the physical condition is high during the rest period when the stress due to training is low. Therefore, in the <second form>, not only the characteristics / properties of the estimation target itself but also the influence of the environment / time related to the estimation target are taken into consideration to perform more accurate physical condition estimation processing.

Next, the <third form> and the <fourth form> will be described with reference to FIG. 2, which schematically shows an embodiment of the model selection process in the model selection unit 115a.

<Third form>
As shown in FIG. 2, the model selection unit 115a of the physical condition information determination unit 115 is
(A) "Likelihood or error information" (for example) about the physical condition value obtained by using the "training physical condition estimation model" based on the information / features at one time point or period (for example, the estimated day). Likelihood L1 or prediction error E1),
(B) "Likelihood or error information" about the physical condition value obtained by using the "physical condition estimation model for training and non-training" based on the information / feature amount at the one time point or period (estimated day). (For example, likelihood L2 or prediction error E2), the adoption of one of these physical condition estimation models is decided.

Here, when both the "physical condition estimation model related to training" and the "physical condition estimation model related to training and non-training" are, for example, an estimation model based on a mixed normal distribution or a Bayesian model, the likelihood output from these physical condition estimation models You may decide to adopt the physical condition estimation model with the higher likelihood by comparing L1 and the likelihood L2. Alternatively, it is also possible to compare the prediction errors obtained for both sampled prediction points and decide to adopt the physical condition estimation model having the smaller prediction error.

Further, when both the "physical condition estimation model related to training" and the "physical condition estimation model related to training and non-training" are, for example, neural network estimation models, the prediction error E1 and the prediction error E2 output from these physical condition estimation models are calculated. By comparing, it is possible to decide to adopt the physical condition estimation model having the smaller prediction error.

Next, the physical condition information determination unit 115
(C) The physical condition value of the estimated target (the athlete) determined for a certain point in time or period (estimated day), and
(D) The training stress feature amount generated for the estimation target (the athlete) at the certain time point or period (estimation day), and
(E) When the "physical condition estimation model for training and non-training" is adopted, based on the non-training stress feature amount generated for the estimation target (the athlete) at the certain time point or period (estimation day). Using the physical condition estimation model of the person who has been decided to be hired, determine the physical condition value of the estimation target (the athlete) at the time or period (when waking up the next day) that is in the future from the one time point or period (estimated day). It is also preferable to do so.

In this way, the <third form> is not only the "information related to the likelihood or error" of the "physical condition estimation model related to training" as in the <second form>, but also the other "physical condition related to training and non-training". A more appropriate physical condition estimation model can be adopted in consideration of the "information on the likelihood or error" of the "estimation model". As a result, it is possible to more accurately estimate the "physical condition" of the estimation target (the athlete) at a future time point or period (when waking up the next day).

<Fourth form>
As shown in FIG. 2, the model selection unit 115a of the physical condition information determination unit 115 is
(A) "Likelihood or error information" (for example) about the physical condition value obtained by using the "training physical condition estimation model" based on the information / features at one time point or period (for example, the estimated day). Likelihood L1 or prediction error E1),
(B) "Likelihood or error information" about the physical condition value obtained by using the "physical condition estimation model for training and non-training" based on the information / feature amount at the one time point or period (estimated day). (For example, likelihood L2 or prediction error E2),
(C) "Likelihood or error information" about the physical condition value obtained by using the "non-training physical condition estimation model" based on the information / feature amount at the one time point or period (estimated day) (c) For example, the adoption of any one of these physical condition estimation models is decided based on the comparison with the likelihood L3 or the prediction error E3).

Here, when the "physical condition estimation model for training", the "physical condition estimation model for training and non-training", and the "physical condition estimation model for non-training" are, for example, an estimation model based on a mixed normal distribution or a Bayesian model, The likelihood L1, the likelihood L2, and the likelihood L3 output from these physical condition estimation models may be compared to determine the adoption of the physical condition estimation model having the highest likelihood. Alternatively, it is also possible to compare the prediction errors obtained with respect to the sampled prediction points in these physical condition estimation models and decide to adopt the physical condition estimation model with the smallest prediction error.

Further, when the "physical condition estimation model related to training", the "physical condition estimation model related to training and non-training", and the "physical condition estimation model related to non-training" are all neural network estimation models, for example, from these physical condition estimation models. The output prediction error E1, prediction error E2, and prediction error E3 may be compared to determine the adoption of the physical condition estimation model with the smallest prediction error.

Next, the physical condition information determination unit 115
(C) The physical condition value of the estimated target (the athlete) determined for a certain point in time or period (estimated day), and
(D) When the "physical condition estimation model for training" or the "physical condition estimation model for training and non-training" is adopted, the training generated for the estimation target (the athlete) at the certain time point or period (estimation day). Sexual stress features and
(E) When the "physical condition estimation model for training and non-training" or the "physical condition estimation model for non-training" is adopted, it was generated for the estimation target (the athlete) at the certain time point or period (estimation day). Using the physical condition estimation model decided to be adopted based on the non-training stress feature amount, the estimation target at the time or period (when waking up the next day) that is in the future from the one time or period (estimated day) It is also preferable to determine the physical condition value of the athlete).

In this way, the <fourth form> takes into consideration the "information on the likelihood or error" of all the physical condition estimation models that may be adopted, so that a more appropriate physical condition estimation model can be adopted. As a result, it is possible to more accurately estimate the "physical condition" of the estimation target (the athlete) at a future time point or period (when waking up the next day).

By the way, although it is a guess based on experience, among the above three candidates for physical condition estimation, the "physical condition estimation model related to training" estimates, for example, elite athletes who are carrying out rigorous training on a daily basis. It is also considered that it is easy to be adopted in the target case, and the "training and non-training physical condition estimation model" is likely to be adopted in the case where the above-mentioned athletes other than the elite athletes, for example, standard student athletes are targeted for estimation. Furthermore, it is also considered that the "non-training physical condition estimation model" is likely to be adopted in a case where a standard sports enthusiast who enjoys sports as a hobby is targeted for estimation. As described above, according to the <fourth form>, it is possible to select an appropriate physical condition estimation model for a wide range of estimation targets and perform more accurate physical condition estimation.

The four physical condition information determination forms have been described above. Of these, the <third form> and the <fourth form> are two physical condition estimation models (third form) or three physical condition estimation models that are candidates for selection. For each of the (fourth form), for example, each model is constructed by using it as training data while shifting the physical condition information and feature amount for the current day and the past several days, and for example, the physical condition information and feature amount for the current day are used. Therefore, it is also preferable to derive "information on the likelihood or error" of each constructed model, and to select the model (determine the adopted model) by this. In addition, such daily learning / judgment processing may be carried out for the "training-related physical condition estimation model" of the <second form>.

Further, in any of the constructed physical condition estimation models in the above four physical condition information determination forms, the determined / estimated physical condition value (for example, Ln_RMSSD value) is equal to or higher than the predetermined physical condition value (for example, Ln_RMSSD value). If there is a difference, the items of trainable stress information and non-trainable stress information that should be used to generate the feature amount are, for example, based on the correlation between these information measured in advance and the HRV value (Ln_RMSSD value). , It is also preferable to reselect.

Here, as a specific example of the physical condition estimation model, a “physical condition estimation model related to training and non-training” based on a mixed normal distribution will be described. Incidentally, the "physical condition estimation model related to training" and the "physical condition estimation model related to non-training" can be configured in the same manner as follows. In any case, in the physical condition estimation model based on the mixed normal distribution, the variance of the prediction can be determined at the time of estimation, which is preferable in handling the estimated value.

First, the type of training is i (i ∈ I, I is a set of all types of training), the type of non-training item is j (j ∈ J, J is a set of all types of non-training items), and then ^{The trainable stress features are represented by S t} _{i, d} , and the non-trainable stress features are S ^u , where each time of the day related to the time-series feature data as an explanatory variable is d (τ_0 ≤ d ≤ τ_1). _{It is represented by j and d} , and the physical condition value (Ln_RMSSD value) is H _d . In this case, the probability density function of the next day's physical condition value (Ln_RMSSD value) H _{τ_1 + 1} , which is the objective variable, is calculated by the following equation (3) p (H _{τ_1 + 1} ) = Σ _{i, d} π ^t _{i, d} · N ( H _d , S ^t _{i, d} | μ ^t _{i, d} , σ ^t _{i, d} )
＋ Σ _{j, d} π ^u _{j, d}・ N (H _d , S ^u _{j, d} | μ ^u _{j, d} , σ ^u _{j, d} )
Is represented by. Here, Σ _{i, d} is the summation of i (i ∈ I) and d (τ_0 ≦ d ≦ τ_1), and Σ _{j, d} are j (j ∈ J) and d (τ_0 ≦). d ≤ τ_1) is the summation, and N () is the probability density function of the normal distribution with the contents in parentheses. In addition, {π} is a mixing ratio of each normal distribution and is a parameter that satisfies the constraint condition of _{Σ i, d} π ^t _{i, d} + Σ _{j, d} π ^u _{j, d = 1, and further, {μ} } And {σ} are parameters that represent the mean and standard deviation of each normal distribution, respectively.

The above equation (3) is an estimated distribution equation that expresses the “physical condition estimation model for training and non-training”, and specifically, the parameters {π, μ, σ} in the equation are well known. The model is constructed by optimizing it with the EM algorithm, which is the method of.

In addition, for the above equation (3) in which the parameters {π, μ, σ} were determined in this way, the generated trainable / non-trainable stress features and physical condition values (Ln_RMSSD values) were time-series (τ_0 to τ_1). By inputting the data, the physical condition value (Ln_RMSSD value) (probability density) of the next day (τ_1 + 1) and its likelihood L can be determined.

In the functional block diagram of FIG. 1, the advice information generation unit 116
(A) Estimated physical condition value (for example, Ln_RMSSD value) and
(B) When the training stress information is used, a predetermined time or period (for example, the physical condition value of (a) above is the next day) that is past from the time point or period related to the physical condition value of (a) above. Training stress information on the day (if the value is related to the time of waking up) and
(C) When non-training stress information is used, a predetermined time or period (for example, the physical condition value in (a) above is the next day) which is a past time or period in view of the time or period related to the physical condition value in (a) above. Based on the non-training stress information on the day of waking up), advice information on reduction, maintenance, or increase of training stress and / or non-training stress is output.

Specifically, for example, when the Ln_RMSSD value, which is the estimated physical condition value, falls below the baseline (reference value) of the Ln_RMSSD calculated for the estimation target as the physical condition judgment threshold value, "I am in poor physical condition." Can be determined.

Here, as described above, the baseline as the physical condition judgment threshold is, for example, from "the average value of Ln_RMSSD in the last 2 weeks" to "the average value of Ln_RMSSD in the last 2 weeks" and "the coefficient of variation of Ln_RMSSD in the last 2 weeks". = (Standard deviation) / (average value)) (0.5 times) "may be subtracted from the multiplication value. For example, if the mean Ln_RMSSD for the last two weeks is 5 and the coefficient of variation is 5%, the baseline will be 4.875. Regarding the handling of such baselines, for example, non-patent literature: Daniel J. Plews et al., “Heart rate variability in elite triathletes, is variation in variability the key to effective training? A case comparison”, European Journal It is described in of Applied Physiology, 112, pp.3729-3741, 2012.

In addition, it is known that the baseline as the physical condition determination threshold is usually affected by the physical strength of the estimation target (for example, which can be quantified from the endurance test result), and the temperature or season at the time of calculating the baseline. Therefore, it is also preferable that the baseline calculation is carried out in consideration of this information.

Here, the advice information output by the advice information generation unit 116 can be, for example, the following information.
<Advice to reduce training stress>
The advice information generation unit 116 may output advice for reducing the training stress when it is determined that the training stress has an adverse effect on the physical condition. for example,
(A) The Ln_RMSSD value, which is the estimated next day's physical condition value, is below the baseline and is judged to be "unwell".
(B) When a specific training stress is higher than a predetermined value on the current day and in the last few days (for example, when a specific training stress feature exceeds a predetermined threshold value), the specific training stress decreases. You can output such advice.

More specifically, for example, the specific training stress feature in the last 7 days exceeds 150% of the average value of the specific training stress feature in the last month. If so, advice may be output to reduce the specific training stress (feature amount).

<Advice to reduce non-training stress>
On the other hand, the advice information generation unit 116 may output advice for reducing the non-training stress when it is determined that the non-training stress has an adverse effect on the physical condition. for example,
(A) The Ln_RMSSD value, which is the estimated next day's physical condition value, is below the baseline and is judged to be "unwell".
(B) When a specific non-training stress is higher than a predetermined value on the current day and in the last few days (for example, when a specific non-training stress feature exceeds a predetermined threshold value), the specific non-training stress Can output advice that reduces.

When it is practically difficult to reduce the specific non-training stress (for example, when the specific non-training stress feature must be less than a preset possible value range). , Instead of the advice to promote such reduction, the advice to perform a preset activity (for example, meditation, alternate bathing, etc.), which is generally considered to lead the physical condition in a preferable direction, may be output.

Further, it is also preferable that the advice information as described above is stored in a database in advance after setting a predetermined If-then rule by hand, for example. Further, the feature amount input value for guiding the physical condition value (Ln_RMSSD value) output from the physical condition estimation model used in the physical condition estimation in a more preferable direction is obtained in advance, and the feature amount for the trainable stress / non-trainable stress is obtained. It is also possible to give advice that encourages changes toward the input value.

As described above, by using the advice information output from the advice information generation unit 116, for example, the estimation target itself and the advisor who is in a position to give advice to the estimation target can know the future physical condition of the estimation target. Not only that, it is also possible to take appropriate measures and measures to improve physical condition.

Similarly, in the functional block diagram of FIG. 1, the physical condition information (physical condition value) determined by the physical condition information determination unit 115 and the advice information generated and output by the advice information generation unit 116 are instructed by the user via, for example, the keyboard 105. Therefore, it may be displayed on the display 106 via the input / output control unit 122. Further, for example, when an information provision request is received from a smartphone carried by a player who is an estimation target of physical condition via the communication interface unit 101, the physical condition information (physical condition value) and advice information are transmitted to the communication control unit 121 and the communication control unit 121. It is also preferable that the information is transmitted to the smartphone via the communication interface unit 101 and used for various purposes such as physical condition management and training menu update.

[Physical condition estimation method]
FIG. 3 is a flowchart showing an outline of one embodiment in the physical condition estimation method according to the present invention.

<When waking up on the day of estimated physical condition>
(S101) From the heart rate measurement information storage unit 104, the heart rate measurement information of the estimation target (for example, a certain player) at the time of waking up on the day of physical condition estimation is acquired.
(S102) Based on the acquired heart rate measurement information, the physical condition value of the estimation target (the athlete) is calculated.
(S103) Notifies the physical condition information including the calculated physical condition value (for example, Ln_RMSSD value). For example, information such as "The physical condition value is *** and the physical condition is good" is displayed on the display 106.

<Predetermined time on the day of physical condition estimation>
(S201) The training stress information on the day of physical condition estimation for the estimation target (the athlete) is acquired from the training stress information storage unit 102, and the estimation target (the athlete) is estimated from the non-training stress information storage unit 103. ), Get non-training stress information on the day of physical condition estimation.
(S202) The training stress feature amount of the estimation target (the athlete) is calculated from the acquired training stress information, and the non-training stress feature of the estimation target (the athlete) is calculated from the acquired non-training stress information. Calculate the amount.

(S203) One selected (by the method already described) from the constructed "physical condition estimation model for training", "physical condition estimation model for training and non-training", and "physical condition estimation model for non-training". On the other hand, the physical condition value of the day and the calculated trainable stress feature amount and / or non-trainable stress feature amount are input, and the physical condition value of the next day for the estimation target (the athlete) is output (as an estimated value). Determine your physical condition).
(S204) From the determined / estimated next day's physical condition value, it is determined whether or not the next day's physical condition of the estimated target (the athlete) is bad based on the baseline of the estimation target (the athlete) calculated in advance. Here, when it is determined that "it is not bad" (it is not "bad"), it is considered that no advice is required in this embodiment, and the physical condition estimation process on the day of physical condition estimation is terminated.

(S205) On the other hand, when the determination of "bad" is made in step S204, advice information, for example, "tomorrow" is obtained based on the heart rate variability information, the trainable stress information, and the non-training information on the day of physical condition estimation and in the predetermined past period. It is expected that your physical condition will deteriorate. *** Please shorten the training time. "

(S206) Notify the determined advice information. For example, the advice information is displayed on the display 106. Further, in the present embodiment, the physical condition estimation process on the day of physical condition estimation ends here.

As described in detail above, according to the present invention, at least in consideration of the training (training) stress information, the object at the time or period that is in the future in view of the time or period related to training (training). It is possible to estimate the physical condition.

In addition, since the physical condition estimation process according to the present invention handles training stress information as the main information, good estimation results can be obtained, for example, especially when an elite athlete including a first-class para-athlete is targeted for estimation. It is possible to output. However, of course, the present invention is not intended only for such elite athletes, but for improving performance and physical condition itself, for example, for student athletes, adult sports enthusiasts, and those who receive rehabilitation or long-term care. A variety of objects can be presumed by the present invention as long as they carry out training.

Therefore, the physical condition estimation device, program and method according to the present invention not only contribute to the training support of strengthening athletes / strengthening teams for a predetermined competition, for example, but also, for example, athletes and general patients admitted to a rehabilitation hospital. It is thought that it will also contribute to the improvement of the rehabilitation effect in Japan.

With respect to the various embodiments of the present invention described above, various changes, modifications and omissions within the scope of the technical idea and viewpoint of the present invention can be easily made by those skilled in the art. The above description is merely an example and does not attempt to limit anything. The present invention is limited only to the scope of claims and their equivalents.

1 Physical condition estimation device 101 Communication interface unit 102 Trainable stress information storage unit 103 Non-trainable stress information storage unit 104 Heart rate measurement information storage unit 105 Keyboard (KB)
106 Display (DP)
111 Trainable stress feature generation unit 112 Non-trainable stress feature generation unit 113 Physical condition information generation unit 114 Physical condition estimation model construction unit 115 Physical condition information determination unit 115a Model selection unit 116 Advice information generation unit 121 Communication control unit 122 Input / output control Part 2 Training stress information database (DB)
3 Non-training stress information DB
4 Heart rate measurement information DB

Claims (11)

It is a physical condition estimation device that estimates the physical condition determined by using the measured values related to the body.
Learned based on the physical condition information of the subject determined for one time point or period and the training stress information which is the stress information related to the physical training performed by the subject at the one time point or period. A physical condition estimation device comprising a physical condition information determining means for determining the physical condition information of the target at a time point or period that is in the future with respect to the one time point or period by using the physical condition estimation model according to the training of the above. The physical condition estimation device according to claim 1, wherein the physical condition information includes an amount determined by using a measured value related to heart rate measurement. The physical condition information determining means is
When it is determined that non-training stress information, which is information on stress other than the stress related to the physical training, is also necessary based on the information on the likelihood or error obtained for the determined physical condition information, a certain time point or period Based on the physical condition information of the estimation target determined for the above, the training stress information related to the time point or period, and the non-training stress information related to the time point or period, the trained and non-trained information has been learned. The physical condition estimation device according to claim 1 or 2, wherein the physical condition estimation model according to the training is used to determine the physical condition information of the estimation target at a time point or period that is in the future from the certain time point or period. The physical condition information determining means is
Non-information on physical condition related to one time point or period, relevant training stress information related to the one time point or period, and information on stress other than stress related to the physical training at the one time point or period. Based on the trainable stress information, the physical condition information of the target at the time or period that is in the future from the one time or period is determined by using the physical condition estimation model related to the trained and non-trained.
Information on the likelihood or error obtained for the physical condition information determined using the physical condition estimation model related to the training, and physical condition information determined using the physical condition estimation model related to the training and non-training were obtained. Based on comparison with information on likelihood or error, we decided to adopt one of these physical condition estimation models.
When adopting the physical condition information of the estimation target determined for a certain time point or period, the training stress information related to the certain time point or period, and the physical condition estimation model related to the training and non-training, the certain time point Or, based on the non-training stress information related to the period, the physical condition information of the estimation target at the time point or period that is in the future from the certain time point or period is determined by using the physical condition estimation model decided to be adopted. The physical condition estimation device according to any one of claims 1 to 3, wherein the physical condition estimation device is characterized. The physical condition information determining means is
Non-information on physical condition related to one time point or period, relevant training stress information related to the one time point or period, and information on stress other than stress related to the physical training at the one time point or period. Based on the trainable stress information, the physical condition information of the target at the time or period that is in the future from the one time or period is determined by using the physical condition estimation model related to the trained and non-trained.
Based on the physical condition information related to one time point or period and the non-training stress information related to the one time point or period, using the learned physical condition estimation model related to non-training, the one time point or period Determine the physical condition information of the target at the time or period in the future from the viewpoint,
Information on the likelihood or error obtained for the physical condition information determined using the physical condition estimation model related to the training, and physical condition information determined using the physical condition estimation model related to the training and non-training were obtained. Any one of these physical condition estimation models based on the comparison between the information related to the likelihood or error and the information related to the likelihood or error obtained for the physical condition information determined by using the physical condition estimation model related to the non-training. Decided to hire one
When adopting the physical condition information of the estimation target determined for a certain time point or period and at least the physical condition estimation model related to the training or the physical condition estimation model related to the training and non-training, the relevant time point or period. Adopted based on the training stress information and at least the non-training stress information related to the certain time point or period when adopting the physical condition estimation model related to the training and non-training or the physical condition estimation model related to the non-training. 3. The physical condition estimation device described in. For the non-training stress information, a candidate corresponding to an explanatory variable having at least a predetermined correlation with the measured value related to the body, which is the objective variable, is selected from a plurality of non-training stress information candidates. The physical condition estimation device according to any one of claims 3 to 5, wherein the physical condition is determined. The physical condition estimation model used or adopted is one that has been trained by learning data relating to a plurality of objects that are similar to each other in the measured values relating to the body, or after being trained by learning data relating to the plurality of objects. The physical condition estimation device according to any one of claims 1 to 6, wherein the physical condition is adjusted by learning data related to an estimation target. The physical condition estimation device according to any one of claims 1 to 7, wherein the physical condition estimation model used or adopted is an estimation model based on a mixed normal distribution, a Bayesian model, or a neural network estimation model. When the determined physical condition information and the trainable stress information are used, the trainable stress information at a predetermined time or period which is the past from the time or period related to the physical condition information and the physical training When non-training stress information, which is information on stress other than the stress concerned, is used, it is trainable based on the non-training stress information at a predetermined time or period that is past from the predetermined time or period. The physical condition estimation device according to any one of claims 1 to 8, further comprising an advice information generating means for outputting advice information relating to reduction, maintenance or increase of stress and / or non-training stress. It is a physical condition estimation program that makes a computer function to estimate the physical condition determined by using the measured values related to the body.
Learned based on the physical condition information of the subject determined for one time point or period and the training stress information which is the stress information related to the physical training performed by the subject at the one time point or period. Physical condition estimation characterized in that a computer functions as a physical condition information determining means for determining the physical condition information of the target at a time point or period that is in the future from the one time point or period by using the physical condition estimation model related to the training of the above. program. It is a physical condition estimation method in a computer that estimates the physical condition determined by using the measured values related to the body.
Physical condition information of the subject determined for one time point or period, training stress information which is information on stress related to physical training performed by the subject at the one time point or period, and corresponding correct answer data. The step of generating a physical condition estimation model related to training by performing learning processing using the physical condition information of the target at a time point or period that is in the future from the one time point or period.
Using the physical condition estimation model for the training generated based on the physical condition information of the estimation target determined for a certain time point or period and the training stress information of the estimation target for the certain time point or period. A method for estimating physical condition, which comprises a step of determining physical condition information of the estimation target at a time or period that is in the future with respect to the certain time or period.

Images