JP6128501B2 - Time-series data analysis device, time-series data analysis method, and program - Google Patents

Description

  The present invention relates to a time-series data analysis device for analyzing time-series data.

  Conventionally, an apparatus for acquiring regularity from time-series data has been developed (see, for example, Patent Document 1).

JP 2006-338373 A

  However, in a conventional device for analyzing time-series data, only one type of feature amount data is acquired from one type of time-series data, and time-series data cannot be effectively used. There was a problem.

  In view of the above problems, an object of the present invention is to effectively use time-series data.

  One aspect of the present invention is an observation data acquisition unit that acquires one or more types of time-series observation data that is observation data for an observation target, and one or more types of time-series observation data acquired by the observation data acquisition unit. Is divided into a first period and the second period shorter than the first period, and a time series of characteristic values in each of the divided first periods First feature quantity data that is data, a feature quantity data acquisition unit that acquires second feature quantity data that is time-series data of characteristic values in each of the divided second periods, and the feature data Using the first feature amount data acquired by the acquisition unit, the first rule acquisition unit that acquires the first rule related to the object, and the second feature amount data acquired by the feature data acquisition unit The pair A second rule acquisition unit that acquires a second rule related to the first rule, a first rule acquired by the first rule acquisition unit, and an output unit that outputs the second rule acquired by the second rule acquisition unit This is a series data analysis device.

  According to one embodiment of the present invention, time-series data can be used effectively.

Block diagram of time-series data analysis apparatus according to Embodiment 1 The figure which shows an example of the observation data stored in the observation data storage means in the embodiment The figure for demonstrating the method in which the feature-value data acquisition means in the embodiment acquires a feature-value The figure for demonstrating the method in which the state label setting means in the same embodiment and the action label setting means set a label The figure for demonstrating the method in which the state rule acquisition means and action rule acquisition means in the embodiment acquire a rule The flowchart which shows operation | movement of the time series data analysis apparatus in the embodiment The figure which shows an example of the external appearance of the computer system in the embodiment The figure which shows an example of a structure of the computer system in the embodiment

  Hereinafter, embodiments of a time-series data analysis device and the like will be described with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.

(Embodiment 1)
In the present embodiment, a time-series data analysis device 1 that acquires state-related rules and behavior-related rules from time-series observation data will be described. FIG. 1 is a block diagram of a time-series data analysis apparatus 1 in the present embodiment.

  The time-series data analysis apparatus 1 includes an observation data storage unit 101, a feature data acquisition unit 102, a state rule (first rule) acquisition unit 103, an action rule (second rule) acquisition unit 104, and an output unit 105. The state rule acquisition unit 103 includes a state label (first label) setting unit 21 and a state rule specifying unit 22. The behavior rule acquisition unit 104 includes a behavior label (second label) setting unit 23 and a behavior rule specifying unit 24.

  The observation data storage means 101 stores one or more types of time-series observation data that are observation data for the observation target. Hereinafter, the observation data for the observation target may be referred to as target observation data. The observation target is preferably one that acts on its own, but may be other than one that acts on its own. For example, an object that acts on its own is preferably an animal, but may be a plant, a human, or not a human. For example, a vehicle other than the one that acts on its own, a tool having a movable part, or the like may be used. The time-series observation data is data in which data acquired by various sensors at a predetermined interval is continuous. The predetermined interval may be constant or indefinite. The predetermined interval may be, for example, once every second or once every 10 seconds. The predetermined interval may be equal to or less than an interval necessary for acquiring the state rule or the behavior rule.

  The time-series target observation data is preferably a video obtained by photographing the observation target with a camera, or a sound obtained by collecting sound generated by the observation target with a microphone, etc., but is data acquired using various sensors. Anything is fine. For example, the time-series target observation data may be time-series data indicating the body temperature measured by thermography, or may be time-series data of the position acquired by the GPS attached to the target, and attached to the target. It may be time-series data indicating the angle of a target acquired from a gyro sensor, or may be time-series data of a pulse acquired by a pulse meter attached to an observation target that is an animal. Note that the sound emitted from the observation target may be a sound such as a cry generated from the observation target, or may be a sound generated when the observation target operates. The sound generated by the movement of the observation target may be, for example, the sound of a squirrel breaking a sunflower seed, or the sound of a raccoon washing food. Sounds that rub against the road surface. In addition, the target observation data acquired using various sensors may be the data acquired using the various sensors as they are, or data calculated from the data acquired using the various sensors. There may be. Observation target data that is data calculated from data acquired using a sensor is, for example, time-series three-dimensional coordinate data indicating the movement history of an observation target acquired using two or more video cameras. May be.

  The observation data storage unit 101 may store time-series observation data related to the external environment. In addition, observation data related to the external environment is sometimes referred to as external environment observation data. The target observation data and the external environment observation data may be collectively referred to as observation data. The external environment is an event related to the environment around the observation target. The surrounding environment may be anything as long as it is in an area that affects the observation target. The surrounding environment may be around the target cage, or a room where the cage is placed. The external environment observation data may be any data related to the external environment acquired using various sensors. For example, the time-series external environment observation data may be a moving image captured by a camera, time-series data indicating sound collected by a microphone, or time-series indicating temperature acquired by a thermometer. The time series data indicating the precipitation obtained by the rain gauge or the time series data indicating the atmospheric pressure obtained by the barometer may be used. It may be time-series data indicating the strength of the wind. In addition, external environment observation data acquired using various sensors may be data acquired using various sensors as they are, and data calculated from data acquired using the various sensors. It may be. In addition, external environment observation data that is data calculated from data acquired using a sensor is, for example, a temperature difference calculated from a temperature history acquired by a thermometer and a temperature difference several hours ago. The data may also be data indicating time-series temperature differences calculated from indoor and outdoor temperatures acquired by two or more thermometers. Since each sensor used for acquiring the target observation data and the external environment observation data is a known technique, detailed description thereof is omitted.

  The observation data storage unit 101 may store two or more types of time-series target observation data. When two or more types of time-series target observation data are stored in the observation data storage unit 101, the observation data storage unit 101 stores, for example, observation data as shown in FIG. Each observation data stored in the observation data storage means 101 is data including a common period. In other words, two or more types of time-series observation data observed during an arbitrary period, for example, from January 1, 2013, 00:00:00 to 00: 00: 5 are stored. Is preferred. The observation data storage means 101 is preferably stored so that information of each observation data at a specific time point included in a period common to each time-series observation data can be synchronized. In order to synchronize the observation data of each time series, the observation data of each time series may have information necessary for synchronization such as a time code, for example. Hereinafter, the observation data storage unit 101 will mainly be described assuming that two types of target observation data and one type of external environment observation data are stored. The observation data storage means 101 is preferably a non-volatile recording medium, but can also be realized by a volatile recording medium. The process in which the observation data is stored in the observation data storage means 101 does not matter. For example, observation data may be stored in the observation data storage means 101 via a recording medium, and observation data transmitted via a communication line or the like is stored in the observation data storage means 101. Alternatively, the observation data input via the input device may be stored in the observation data storage means 101. Note that “two or more types of observation data” indicates observation data acquired by two or more types of different sensors.

  The feature amount data acquisition unit 102 acquires two or more types of feature amount data that are time-series data of characteristic values from one type of target observation data stored in the observation data storage unit 101. The feature data acquisition unit 102 may acquire external environment feature data that is time-series data of characteristic values also from time-series observation data related to the external environment. For example, as shown in FIG. 3, the feature amount data is data in which observation data is divided every predetermined period and characteristic values in the divided periods are arranged in time series. The value of the feature amount data may be a maximum value in a predetermined period, a minimum value in a predetermined period, an average value in a predetermined period, or a slope of a waveform in a predetermined period. It may be a characteristic value obtained by Fourier-transforming a waveform in a predetermined period, a value obtained by differentiating the movement amount in a predetermined period with time, and in a predetermined period. It may be a value obtained by differentiating the movement amount twice with respect to time, or may be a value obtained by another algorithm. The feature amount data may be a value associated with the value acquired as described above. For example, the feature amount data acquisition unit 102 may acquire the value of feature amount data corresponding to each value acquired as described above, using a correspondence table stored in a storage unit (not shown). Specifically, the feature amount data acquisition unit 102 may acquire feature amount data related to the volume of the call using a correspondence table that divides the value of the feature amount data into five levels according to the volume level. Note that the value of the feature data may be a numerical value obtained by rounding an arbitrary digit from the value acquired as described above. The method of rounding the value may be rounding off, rounding down, or rounding up.

  When two or more types of time-series observation data for the observation target are stored in the observation data storage unit 101, the feature data acquisition unit 102 stores the observation data stored in the observation data storage unit 101. You may acquire three or more types of feature-value data from two or more types of object observation data. When the feature amount data acquisition unit 102 acquires feature amount data from two or more types of time-series target observation data, two or more types of feature amount data are acquired from one type of target observation data. One or more types of feature data may be acquired from each target observation data other than the target observation data. That is, the feature data acquisition unit 102 may acquire (M + 1) or more types of feature data using M types of target observation data from M types of target observation data. M is a natural number of 1 or more. Therefore, it may be considered that the target observation data used for acquiring the feature data that is not used for acquiring the state rule or the behavior rule is not included in the M types of target observation data. Hereinafter, a case where two or more types of time-series target observation data are stored in the observation data storage unit 101 will be mainly described.

  The feature amount data acquisition unit 102 obtains state feature amount data (first feature amount data) that is feature amount data related to a state and action feature amount data (second feature amount) that is feature amount data related to an action from target observation data. Data) may or may not be acquired. When two or more types of time-series observation data for the observation target are stored in the observation data storage unit 101, the feature amount data acquisition unit 102 obtains N or more types of state feature amount data from the target observation data. And (3-N) or more types of action feature data may be acquired. N is 1 or 2. The state feature amount data may be data used to acquire a state rule. The state rule is information including a sequence of feature value data values, labels, and the like regarding the target state. The state rule may be, for example, information indicating a state during a meal, information indicating a state during drinking, information indicating a state during urination, and a state during defecation. It may be information indicating, information indicating a sleeping state, or information indicating other states. Further, the behavior feature amount data may be data used for acquiring a behavior rule. The behavior rule is information including a sequence of feature value data, labels, and the like regarding the target behavior. For example, the behavior rule may be information indicating an action whose position is periodically changed, may be information indicating a running action, or may be information indicating an action that makes a cry. For example, it may be information indicating a jumping action. The difference between the state feature quantity data and the behavior feature quantity data may be that the predetermined period for dividing the target observation data is different in the process for obtaining the feature quantity data, and the process for obtaining the feature quantity data. May be different, the target observation data from which the feature data is acquired may be different, or a combination of two or more of the above differences. The difference in the case where the predetermined period for dividing the target observation data is different is that the predetermined period in the process for acquiring the behavior feature quantity data is shorter than the predetermined period in the process for acquiring the state feature quantity data. Also good. For example, when the predetermined period in the process of acquiring state feature data is 10 seconds, the predetermined period in the process of acquiring behavior feature data may be 1 second or the like. For example, when the process for acquiring feature data is different, for example, when the process of acquiring state feature data is a process of acquiring an integral value, the process of acquiring action feature data is a differential value. It may be the process of acquiring. The difference when the target observation data from which feature data is acquired is different is that the target observation data used to acquire state feature data is different from the target observation data used to acquire behavior feature data in advance. It may be decided. For example, the processing in the feature amount data acquisition unit 102 may be determined in advance, such as acquiring state feature amount data and behavior feature amount data from a camera, and acquiring behavior feature amount data from a microphone.

  The “two or more types of feature data” may be considered as two or more feature data acquired by different processes. When two pieces of feature amount data acquired by different processes are the same feature amount data, they may be considered as two types of feature amount data or may be considered as one type of feature amount data. The feature amount data acquisition unit 102 can be usually realized by an MPU, a memory, or the like. The processing procedure of the feature data acquisition unit 102 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The state rule acquisition unit 103 uses the one or more types of feature amount data acquired by the feature amount data acquisition unit 102 to acquire a state rule that is a rule regarding the target state. When the feature amount data acquisition unit 102 acquires two or more types of feature amount data acquired from one type of target observation data, the state rule acquisition unit 103 selects at least one type of the two or more types of feature amount data. The state rule may be acquired using the above. When acquiring a state rule from one type of feature value data, the state rule acquisition unit 103 causes the continuous feature to appear when two or more consecutive values in the feature value data repeatedly appear in a predetermined period. You may acquire the state rule which is the value of quantity data. When the feature data acquisition unit 102 acquires three or more types of feature data from two or more types of time-series target observation data, the state rule acquisition unit 103 stores three or more types of feature data. The state rule may be acquired using any N or more types of feature amount data. N is 1 or 2 as described above. When acquiring a state rule from two or more types of feature amount data, the state rule acquisition unit 103 repeats a combination of one or two or more consecutive values in the two or more types of feature amount data within a predetermined period. When appearing, a state rule that is a combination of consecutive values may be acquired. Further, when the feature amount data acquisition unit 102 acquires state feature amount data, the state rule acquisition unit 103 may acquire the state rule from the state feature amount data as described above.

  Further, the state rule acquisition unit 103 may acquire a state rule for each value of the external environment feature value data or for each classification of the value of the external environment feature value data. “Acquisition of a state rule for each value of external environment feature value data” means acquisition of a state rule for each value acquired by the feature value data acquisition means 102. For example, when the external environment observation data is observation data relating to temperature, the state rule acquisition unit 103 may acquire a state rule when the temperature is 30 degrees and a state rule when the temperature is 31 degrees, respectively. good. “Acquisition of state rules for each classification of values of external environment feature data” classifies the values acquired by the feature data acquisition unit 102 into two or more classifications, and acquires the state rules for each classification. is there. For example, when the external environment observation data is observation data relating to the temperature, the state rule acquisition unit 103 obtains a state rule when the temperature is 30 degrees or higher and a state rule when the temperature is 20 degrees or higher and lower than 30 degrees. Each may be acquired. In addition to the above, the classification of the value of the external environment feature value data may be a classification of whether it is raining or not when the external environment observation data is observation data relating to precipitation. When the observation data is observation data related to voice, it may be classified as a library level, a living sound level, a construction site level, or the like, or may be another classification. The state rule acquisition unit 103 can be usually realized by an MPU, a memory, or the like. The processing procedure of the state rule acquisition unit 103 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The state rule acquisition unit 103 may acquire a state rule by adding a label to the feature amount data by the processing of the state label setting unit 21 and the state rule specifying unit 22. Note that the feature amount data referred to here may be state feature amount data. As shown in FIG. 4, the state label setting means 21 classifies the feature value data values into a plurality of groups, and sets the same state label to the feature value data values belonging to the same group. The state label setting means 21 may set a state label for any N or more types of feature amount data among three or more types of feature amount data. N is 1 or 2 as described above. There is no limitation on the criteria by which the state label setting means 21 classifies the value of the feature data. For example, the state label setting unit 21 may classify each feature value data value into a group, or may classify feature value data values into groups for each predetermined value range. The value of the feature amount data may be classified into groups according to the above. The rule determined in advance may be a rule for classifying into groups of even feature value data values and odd feature value data values. Or a rule for classifying them into groups. The status label is information that can identify the classified group. That is, the status label may be any information as long as it is information that can uniquely identify the group. By using the state label, a plurality of feature value data values can be set as one label, and information can be rounded rather than using feature value data values. That is, setting the state label facilitates discovery of the state rule.

  Note that the state label setting unit 21 may group two or more continuous feature value data values into one group. The state label setting means 21 may set one state label for two or more consecutive feature value data values when two or more consecutive feature value data values are grouped into a group. Specifically, the state label setting means 21 reads “value of feature quantity data indicating activity / value of feature quantity data indicating suspension / value of feature quantity data indicating suspension / feature data indicating activity”. Feature values such as “value of feature data indicating activity / value of feature data indicating stop / value of feature data indicating activity” Data values may be summarized. When the values of the feature amount data are collected, for example, in the above-described case, the state label setting unit 21 can set a label so as not to stick to the stopped period.

  For example, the state rule specifying unit 22 obtains a state rule from an array of N or more types of state labels set by the state label setting unit 21 as shown in FIG. When acquiring a state rule from a sequence of one type of state label, the state rule specifying unit 22 determines that when two or more continuous state labels in the state label sequence repeatedly appear in a predetermined period, A state rule that is a state label may be acquired. When acquiring a state rule from a sequence of two or more types of state labels, the state rule specifying unit 22 determines that a combination of one or two or more continuous state labels in the sequence of two or more types of state labels is predetermined. When it repeatedly appears in the period, a state rule that is a combination of the continuous state labels may be acquired. For example, the state rule specifying unit 22 may acquire the state rule by dividing the arrangement of the state labels into a plurality of periods and performing frequent pattern mining. The state rule may be acquired by changing multiple times and performing frequent pattern mining in the same manner. In addition, since frequent pattern mining is a well-known technique, it abbreviate | omits about the detailed description. The state rule specifying means 22 can usually be realized by an MPU, a memory, or the like.

  The behavior rule acquisition unit 104 acquires a behavior rule that is a rule regarding the target behavior, using one or more types of feature amount data acquired by the feature amount data acquisition unit 102. When the feature data acquisition unit 102 acquires two or more types of feature data acquired from one type of target observation data, the behavior rule acquisition unit 104 selects at least one type of the two or more types of feature data. You may acquire an action rule using the above. When acquiring an action rule from one type of feature value data, the action rule acquiring unit 104 determines that the continuous feature is obtained when two or more consecutive values in the feature value data repeatedly appear in a predetermined period. You may acquire the action rule which is the value of quantity data. When the feature data acquisition unit 102 acquires three or more types of feature data from two or more types of time-series target observation data, the behavior rule acquisition unit 104 stores three or more types of feature data. You may acquire an action rule using any (3-N) or more types of feature-value data. N is 1 or 2 as described above. When acquiring a behavior rule from two or more types of feature data, the behavior rule acquisition unit 104 repeats a combination of one or more consecutive values in the two or more types of feature data in a predetermined period. When appearing, you may acquire the action rule which is the combination of the continuous value. When the feature data acquisition unit 102 acquires behavior feature data, the behavior rule acquisition unit 104 may acquire a behavior rule from the behavior feature data. The behavior rule acquisition unit 104 may acquire the behavior rule by using all the feature amount data not used by the state rule acquisition unit 103 among the feature amount data acquired by the feature amount data acquisition unit 102. The action rule may be acquired using one or more feature quantity data at least partially different from the type of feature quantity data used by the means 103, and the action rule using one or more feature quantity data satisfying both of the above You may get That is, the state rule acquisition unit 103 and the behavior rule acquisition unit 104 perform processing so that all feature amount data acquired by the feature amount data acquisition unit 102 is used to acquire either the state rule or the behavior rule. May be performed. Further, the behavior rule may be a rule within a period in which the state rule is acquired. That is, if the state rule acquisition unit 103 has acquired the state rule of the period x from the time point t to the time point (t + x), the behavior rule acquisition unit 104 determines that the period x from the time point t to the time point (t + x). You may obtain less action rules. Note that x is an arbitrary period.

  Further, the behavior rule acquisition unit 104 may acquire a behavior rule for each value of the external environment feature value data or for each classification of the value of the external environment feature value data. “Acquiring an action rule for each value of the external environment feature value data” means acquiring an action rule for each value acquired by the feature value data acquisition unit 102. For example, when the external environment observation data is observation data related to temperature, the behavior rule acquisition unit 104 may acquire a behavior rule when the temperature is 30 degrees and a behavior rule when the temperature is 31 degrees. good. “Acquire action rules for each classification of values of external environment feature data” classifies the values acquired by the feature data acquisition unit 102 into two or more classifications, and acquires the behavior rules for each classification. is there. For example, when the external environment observation data is temperature-related observation data, the behavior rule acquisition unit 104 obtains a behavior rule when the temperature is 30 degrees or more and a behavior rule when the temperature is 20 degrees or more and less than 30 degrees. Each may be acquired. In addition to the above, the classification of the value of the external environment feature value data may be a classification of whether it is raining or not when the external environment observation data is observation data relating to precipitation. When the observation data is observation data related to voice, it may be classified as a library level, a living sound level, a construction site level, or the like, or may be another classification. The behavior rule acquisition unit 104 can be usually realized by an MPU, a memory, or the like. The processing procedure of the behavior rule acquisition unit 104 is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The behavior rule acquisition unit 104 may acquire a behavior rule by assigning a label to the feature amount data by the processing of the behavior label setting unit 23 and the behavior rule specifying unit 24. Note that the feature amount data referred to here may be behavior feature amount data. As shown in FIG. 4, the action label setting unit 23 classifies the feature value data values into a plurality of groups, and sets the same action label to the feature value data values belonging to the same group. The action label setting means 23 may set an action label for any (3-N) or more types of feature quantity data among three or more types of feature quantity data. N is 1 or 2 as described above. There is no limitation on the criteria by which the action label setting unit 23 classifies the value of the feature amount data. For example, the action label setting unit 23 may classify the feature value data into groups, or may classify the feature value data values into groups for each predetermined value range. The value of the feature amount data may be classified into groups according to the above. The rule determined in advance may be a rule for classifying into groups of even feature value data values and odd feature value data values. Or a rule for classifying them into groups. The action label is information that can identify the classified group. That is, the action label may be any information as long as the information can uniquely identify the group. By using the action label, a plurality of feature value data values can be set as one label, and information can be rounded rather than using the feature value data value. In other words, setting action labels facilitates discovery of action rules.

  Note that the action label setting means 23 may group two or more consecutive feature value data values into one group. The action label setting means 23 may set one action label to the values of two or more consecutive feature quantity data when the values of two or more consecutive feature quantity data are grouped into a group. Specifically, the action label setting unit 23 determines that “the value of feature data indicating a running action, the value of feature data indicating a walking action, the value of feature data indicating a walking action, The value of the feature value data such as “value of feature value data indicating the action being performed” is the value of the feature value data indicating the action being performed, the value of the feature value data indicating the action being performed, and the running. The value of feature amount data may be collected as “value of feature amount data indicating behavior”. Summarizing the values of the feature amount data, for example, in the case described above, the action label setting unit 23 can set a label so as not to stick to the walking period.

  For example, the behavior rule specifying unit 24 acquires the behavior rule from the sequence of (3-N) types of behavior labels set by the behavior label setting unit 23 as shown in FIG. N is 1 or 2 as described above. When acquiring a behavior rule from a sequence of one type of behavior label, the behavior rule specifying means 24, when two or more consecutive behavior labels in the sequence of the behavior label repeatedly appear in a predetermined period, You may acquire the action rule which is an action label. In addition, when obtaining a behavior rule from a sequence of two or more types of behavior labels, the behavior rule specifying unit 24 determines that a combination of one or two or more consecutive behavior labels in the sequence of the two or more types of behavior labels is predetermined. When it repeatedly appears in the period, an action rule that is a combination of consecutive action labels may be acquired. For example, the behavior rule specifying unit 24 may acquire the behavior rule by dividing the sequence of behavior labels into a plurality of periods and performing frequent pattern mining. The behavior rule may be acquired by changing multiple times and performing frequent pattern mining in the same manner. The behavior rule specifying unit 24 can be usually realized by an MPU, a memory, or the like.

  The output unit 105 outputs the state rule acquired by the state rule acquisition unit 103 and the behavior rule acquired by the behavior rule acquisition unit 104. The output means 105 outputs a state rule and an action rule for each external environment feature value data or for each value classification of the external environment feature value data. Note that the output unit 105 may output a state rule including the same period and an action rule in association with each other. The output in association with each other may be to output the state rule and the action rule at the same time, or may output the state rule, the action rule, and the action rule including the same ID separately. The state rule and the behavior rule may be output in order. Further, the output means 105 may output the state rule and the behavior rule in association with the value of the external environment feature value data or the classification of the value of the external environment feature value data in the same manner as described above. Note that the output means 105 may output only when both the state rule and the action rule within the period in which the state rule is acquired, or output when only one of them is acquired. You may do it. Output refers to display on a display, projection using a projector, printing on a printer, sound output, transmission to an external device, storage on a recording medium, processing results to other processing devices or other programs, etc. It is a concept that includes delivery. The output unit 105 may or may not include an output device such as a display or a speaker. The output means 105 can be implemented by output device driver software, or output device driver software and an output device.

  The time-series data analysis device 1 acquires the state rule and the behavior rule as described above. Each rule acquired in this way is used in a device or the like that classifies data according to the state or action of the target using the rule. For example, a device that classifies data according to the state or behavior of a target using a rule stores, for example, data that associates each rule about an animal with information about a disease indicated by the rule, and performs target observation on an animal. It may be a device that searches for a rule corresponding to the received target observation data by receiving data, and transmits information related to a disease associated with the hit rule when it is hit.

  FIG. 6 is a flowchart showing an example of the operation of the time-series data analysis apparatus 1 in the present embodiment. The operation will be described below with reference to FIG. In this flowchart, a case where two types of observation data for the observation target are stored in the observation data storage unit 101 will be described.

  (Step S201) The feature data acquisition unit 102 acquires three or more types of feature data from the two types of target observation data stored in the observation data storage unit 101.

  (Step S202) The feature data acquisition unit 102 acquires external environment feature data from the external environment observation data stored in the observation data storage unit 101.

  (Step S203) The state label setting means 21 classifies the feature amount data acquired in step S201 into groups for each value of the feature amount data, and sets a state label for each classified group.

  (Step S204) The state rule specifying means 22 specifies a state rule for each value of the external environment feature value data from the state label sequence set in step S203.

  (Step S205) The action label setting unit 23 classifies the feature amount data acquired in step S201 into groups for each value of the feature amount data, and sets an action label for each classified group.

  (Step S206) The behavior rule specifying means 24 specifies a behavior rule for each value of the external environment feature value data from the sequence of behavior labels set in step S205.

  (Step S207) The state rule specified in step S204 and the action rule specified in step S206 are associated with each other and output for each value of the external environment feature data. Then, the process ends.

  Hereinafter, a specific description will be given of a case where the observation data storage unit 101 stores target observation data that is a moving image obtained by photographing a squirrel with a camera. In this specific example, it is assumed that the camera is arranged so that all the inside of the cage can be photographed from the upper part of the squirrel cage. The feature amount data acquisition means 102 performs background difference processing from the target observation data that is a moving image, thereby obtaining feature amount data “A, B, C, C, C, C, A... And feature amount data “Z, Z, Y, Z, Z, X, Y. In this specific example, “C, C, C, C” periodically appears in the feature amount data relating to the position of the squirrel, and in the same period as “C, C, C, C”, It is assumed that “Z, Z, X” appears periodically in the feature amount data regarding the posture. Here, “C” is a feature value data value indicating the position of the toilet, “Z” is a feature value data value indicating that it is lying down, and “X” is a look around. It is assumed that the value is feature value data indicating The state rule acquisition unit 103 acquires the state rule “C, C, C, C” from the feature amount data regarding the position of the squirrel. Then, the behavior rule acquisition unit 104 acquires a behavior rule “Z, Z, X” corresponding to the state rule “C, C, C, C”. And the output means 105 outputs the acquired state rule and action rule. Note that the state rule “C, C, C, C” output in this specific example indicates staying in the toilet, and the action rule “Z, Z, X” Suppose that you are looking around.

  In addition, in the case where the observation data storage means 101 stores target observation data that is a moving image of a squirrel photographed by a camera and target observation data that is a sound collected by a microphone from a squeak generated by the squirrel. This will be specifically described. In addition, in this specific example, the description which overlaps with the specific example demonstrated previously may be abbreviate | omitted. In this specific example, it is assumed that the microphone is arranged in the immediate vicinity of the squirrel cage. The feature amount data acquisition means 102 performs background difference processing from the target observation data that is a moving image, thereby obtaining feature amount data “A, B, C, C, C, C, A... And the feature amount data “Z, Z, Y, Z, Z, X, Y...” Relating to the posture of the squirrel, and the feature amount data “ "0, 0, 0, 0, 0, 1, 0 ..." is acquired (step S201). In this specific example, it is assumed that “0, 0, 1” also appears periodically during the period in which “Z, Z, X” appears in the period in which “C, C, C, C” appears. . Here, “0” is a value of feature amount data indicating that no cry is raised, and “1” is a value of feature amount data indicating that a cry is raised. The state rule acquisition unit 103 acquires the state rule “C, C, C, C” from the feature amount data regarding the position of the squirrel (step S203 to step S204). Then, the behavior rule acquisition unit 104 acquires a behavior rule [“Z, Z, X” “0, 0, 1”] corresponding to the state rule “C, C, C, C” (from step S205 to step S206). ). And the output means 105 outputs the acquired state feature-value and action feature-value. In addition, the state rule “CCCC” output in this specific example indicates staying in the toilet, and the action rule [“Z, Z, X” “0, 0, 1”] Let ’s say you ’re looking around and making a cry.

  As described above, according to the present embodiment, the feature data acquisition unit 102 acquires two or more feature data from one type of time series data. Thereby, since the rules for the observation target can be acquired from a plurality of viewpoints, the rules can be acquired by effectively using time-series data. Further, according to the present embodiment, the feature data acquisition unit 102 can acquire external environment feature data. Thus, the state rule and the action rule can be acquired for each value of the external environment feature value data or the classification of the value of the external environment feature value data. For example, when the feature amount data acquisition unit 102 acquires external environment feature amount data, the search condition can be limited in an apparatus that classifies data according to the state or action of the target using the rule. It becomes possible to search. Further, according to the present embodiment, the feature amount data acquisition unit 102 can acquire two or more types of feature amount data from one type of observation data. Accordingly, it is possible to acquire feature amount data suitable for acquiring the state rule and feature amount data suitable for acquiring the action rule. When the feature quantity data acquisition unit 102 acquires two or more types of feature quantity data from one type of observation data, for example, in a device that classifies data according to the state or behavior of the target using rules, the search condition is set. Because it can be limited, rules can be searched at high speed. Moreover, according to this Embodiment, the state label setting means 21 can set a label to feature-value data. Thereby, when acquiring a state rule, it can classify | categorize into the feature-value group of observation data, and can acquire a state rule. Therefore, for example, a state rule obtained by rounding the value of the observation data can be acquired. For example, in a device that classifies data according to the state or action of a target using rules, a similar rule can be searched at high speed. Moreover, according to this Embodiment, the action label setting means 23 can set a label to feature-value data. Thereby, when acquiring an action rule, it can classify | categorize into the feature-value group of observation data, and can acquire an action rule. Therefore, for example, an action rule obtained by rounding the value of the observation data can be acquired. For example, in a device that classifies data according to the state or action of a target using rules, a similar rule can be searched at high speed. Further, according to the present embodiment, the state rule acquisition unit 103 and the behavior rule acquisition unit 104 obtain the state rule and the behavior rule from observation data obtained by photographing an animal or collecting sounds emitted by the animal. You can get it.

  Although the case where the state label setting unit 21 and the state rule specifying unit 22 are included has been described, the time-series data analysis device 1 may not include the state label setting unit 21 and the state rule specifying unit 22. When the state label setting unit 21 and the state rule specifying unit 22 are not included, the state rule acquisition unit 103 may acquire the state rule without setting a label from the feature amount data.

  Although the case where the action label setting unit 23 and the action rule specifying unit 24 are included has been described, the time-series data analysis device 1 may not include the action label setting unit 23 and the action rule specifying unit 24. When the behavior label setting unit 23 and the behavior rule specifying unit 24 are not included, the behavior rule acquisition unit 104 may acquire the behavior rule without setting a label from the feature amount data.

  Moreover, the software which implement | achieves the time series data analysis apparatus 1 in this Embodiment is the following programs. In other words, the program uses a computer that can access observation data storage means that stores one or more types of time-series observation data, which is observation data for the observation target, as one type of observation data stored in the observation data storage means. From the feature amount data acquisition means for acquiring two or more types of feature amount data, which is time-series data of characteristic values, and the state rule acquisition for acquiring a state rule that is a rule relating to the target state using the feature amount data Means, action rule acquisition means for acquiring an action rule that is a rule relating to a target action using the feature data, a state rule acquired by the state rule acquisition means, and an action rule acquired by the action rule acquisition means It is a program for functioning as output means.

  In the present embodiment, each process (each function) may be realized by centralized processing by a single device (system), or by distributed processing by a plurality of devices. May be. In the present embodiment, it goes without saying that two or more communication means existing in one apparatus may be physically realized by one means.

  In the present embodiment, each component may be configured by dedicated hardware, or a component that can be realized by software may be realized by executing a program. For example, each component can be realized by a program execution unit such as a CPU reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory.

  In the program, the functions realized by the program do not include functions that can be realized only by hardware. For example, functions that can be realized only by hardware such as a modem and an interface card in an acquisition unit that acquires information, an output unit that outputs information, and the like are not included in the functions realized by the program.

  FIG. 7 is a schematic diagram showing an example of the appearance of a computer that executes the program and implements the present invention according to the embodiment. The embodiment described above can be realized by computer hardware and a computer program executed on the computer hardware.

  In FIG. 7, a computer system 1100 includes a computer 1101 including a CD-ROM drive 1105 and an FD drive 1106, a keyboard 1102, a mouse 1103, and a monitor 1104.

  FIG. 8 is a diagram showing an internal configuration of the computer system 1100. In FIG. 8, in addition to the CD-ROM drive 1105 and the FD drive 1106, a computer 1101 is connected to an MPU 1111, a ROM 1112 for storing a program such as a bootup program, and an MPU 1111. And a hard disk 1114 for accumulating application programs, system programs, and data, an MPU 1111, a ROM 1112, and a bus 1115 for interconnecting the ROM 1112 and the like. The computer 1101 may include a network card (not shown) that provides connection to the LAN.

  A program for causing the computer system 1100 to execute the functions of the present invention according to the above-described embodiment is stored in the CD-ROM 1121 or the FD 1122, inserted into the CD-ROM drive 1105 or the FD drive 1106, and transferred to the hard disk 1114. May be. Instead, the program may be transmitted to the computer 1101 via a network (not shown) and stored in the hard disk 1114. The program is loaded into the RAM 1113 when executed. The program may be loaded directly from the CD-ROM 1121, the FD 1122, or the network.

  The program does not necessarily include an operating system (OS) or a third-party program that causes the computer 1101 to execute the functions of the present invention according to the above-described embodiment. The program may include only a part of an instruction that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 1100 operates is well known and will not be described in detail.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention. The “means” of each means in the present invention may be read as “part” or “circuit”.

  The time-series data analysis apparatus of the first reference example is stored in observation data storage means for storing one or more types of time-series observation data that are observation data for an observation target, and in the observation data storage means 1 Feature quantity data acquisition means for acquiring two or more types of feature quantity data that is time-series data of characteristic values from types of observation data, and a state rule that is a rule relating to a target state using the feature quantity data State rule acquisition means to acquire, action rule acquisition means to acquire a behavior rule that is a rule related to the target action using the feature data, state rules acquired by the state rule acquisition means, and action rule acquisition means It is a time series data analysis apparatus provided with the output means which outputs the action rule which performed.

  In the time series data analysis apparatus of the second reference example, the observation data storage means stores two or more types of time series observation data, and the feature quantity data acquisition means is From the two or more types of observation data stored in the observation data storage unit, three or more types of feature data are acquired, and the state rule acquisition unit is any one of the three or more types of feature data. The state rule is acquired using the feature amount data (N is 1 or 2), and the action rule acquisition means is any one of the three or more types of feature amount data (3-N) or more of the feature amount. It is a time series data analysis device that acquires action rules using data.

  In the time-series data analysis apparatus according to the third reference example, in contrast to the first or second invention, the observation data storage means also stores time-series observation data related to the external environment, and the feature data acquisition means Obtains external environment feature value data that is characteristic time-series data from time-series observation data related to the external environment, and the state rule acquisition means obtains each value of the external environment feature value data or the external environment. The state rule is acquired for each classification of the feature data value, and the behavior rule acquisition unit acquires the behavior rule for each value of the external environment feature data or for each value of the external environment feature data. The output means is a time-series data analysis device that outputs a state rule and an action rule for each external environment feature quantity data or for each value classification of the external environment feature quantity data.

  In the time-series data analysis apparatus according to the fourth reference example, for any one of the first to third inventions, the state rule acquisition unit classifies the values of the feature data into a plurality of groups, and the same A time series comprising state label setting means for setting the same state label to the value of feature quantity data belonging to the group, and state rule specifying means for acquiring a state rule from the arrangement of the state labels set by the state label setting means Data analysis device.

  In the time-series data analysis apparatus according to the fifth reference example, the action rule acquisition unit classifies the values of the feature amount data into a plurality of groups, and the same for any one of the first to fourth inventions. A time series comprising action label setting means for setting the same action label for the value of feature quantity data belonging to a group, and action rule specifying means for acquiring action rules from the arrangement of action labels set by the action label setting means Data analysis device.

  The time-series data analysis apparatus according to the sixth reference example is directed to any one of the first to fifth inventions, wherein the object is an animal, the observation data is image data obtained by photographing the animal, and the animal This is a time-series data analysis device including sound data obtained by collecting sounds.

  According to the time-series data analysis device or the like according to the reference example, by acquiring two or more feature amount data from one type of time-series data, it is possible to acquire rules for the observation target from a plurality of viewpoints. The rules can be obtained by effectively utilizing.

  As described above, the time-series data analysis device according to the present invention can acquire rules for an observation target from a plurality of viewpoints by acquiring two or more feature amount data from one type of time-series data. Rules can be acquired by using time-series data effectively. This is useful as a data analysis device or the like used in a device or the like that classifies data according to the state or action of a target using rules.

DESCRIPTION OF SYMBOLS 1 Time series data analyzer 101 Observation data storage means 102 Feature-value data acquisition means 103 State rule acquisition means 104 Action rule acquisition means 105 Output means 21 State label setting means 22 State rule specification means 23 Action label setting means 24 Action rule specification means 24

Claims (9)

An observation data acquisition unit for acquiring one or more types of time-series observation data as observation data for the observation target;
One type of observation data among one or more types of time series observation data acquired by the observation data acquisition unit is divided into a first period and the second period shorter than the first period. , First feature amount data that is time-series data of characteristic values in each of the divided first periods and time-series data of characteristic values in each of the divided second periods. A feature amount data acquisition unit for acquiring the feature amount data of 2;
A first rule acquisition unit that acquires a first rule related to the object using the first feature amount data acquired by the feature data acquisition unit;
A second rule acquisition unit that acquires a second rule related to the object using the second feature amount data acquired by the feature data acquisition unit;
A time-series data analysis device comprising: an output unit that outputs a first rule acquired by the first rule acquisition unit and a second rule acquired by the second rule acquisition unit. The first rule acquisition unit acquires a value that continuously appears in the first feature amount data acquired by the feature data acquisition unit as a first rule related to the object,
The second rule acquisition unit acquires a value that continuously appears in the second feature amount data acquired by the feature data acquisition unit as a second rule related to the object.
The time-series data analysis apparatus according to claim 1. The observation data acquisition unit
Acquire two or more types of time series observation data,
The feature data acquisition unit
From the two or more types of observation data acquired by the acquisition unit, acquire three or more types of feature data,
The first rule acquisition unit includes:
The first rule is acquired using any N or more (N is 1 or 2) feature data among the three or more feature data.
The second rule acquisition unit
Obtaining the second rule using any (3-N) or more types of feature data among the three or more types of feature data;
The time-series data analysis apparatus according to claim 1 or 2. The observation data acquisition unit
Acquire time-series observation data about the external environment,
The feature data acquisition unit
From the time-series observation data related to the external environment, obtain external environment feature data that is time-series data of characteristic values,
The first rule acquisition unit includes:
Obtaining the first rule for each value of the external environment feature value data or for each classification of the value of the external environment feature value data;
The second rule acquisition unit
Obtaining the second rule for each value of the external environment feature value data or for each classification of the value of the external environment feature value data;
The output unit is
Outputting the first rule and the second rule for each external environment feature value data or for each classification of the value of the external environment feature value data;
The time series data analysis device according to any one of claims 1 to 3. The first rule acquisition unit includes:
A first label setting unit that classifies values of the feature amount data into a plurality of groups and sets the same first label to the values of the feature amount data belonging to the same group;
A first rule specifying unit for obtaining a first rule from the arrangement of the first labels set by the first label setting unit,
The time series data analysis device according to any one of claims 1 to 4. The second rule acquisition unit
A second label setting unit that classifies values of the feature amount data into a plurality of groups and sets the same second label to the values of the feature amount data belonging to the same group;
A second rule specifying unit for obtaining a second rule from the arrangement of the second labels set by the second label setting unit,
The time series data analysis device according to any one of claims 1 to 5. The subject is an animal;
The observation data is
Including image data obtained by photographing the animal, and sound data obtained by collecting sounds emitted by the animal,
The time series data analysis device according to any one of claims 1 to 6. An observation data acquisition unit that acquires one or more types of time-series observation data that is observation data for an observation target, a feature data acquisition unit, a first rule acquisition unit, a second rule acquisition unit, and an output unit A time-series data analysis method processed using
The feature amount data acquisition unit obtains one type of observation data among the one or more types of time series observation data acquired by the acquisition unit as a first period and the second period shorter than the first period. The first feature value data, which is time-series data of characteristic values in each of the divided first periods, and the characteristic values in each of the divided second periods A feature data acquisition step for acquiring second feature data that is time-series data;
A first rule acquisition step in which the first rule acquisition unit acquires a first rule using the first feature amount data acquired in the feature amount data acquisition step;
A second rule obtaining step in which the second rule obtaining unit obtains a second rule using the second feature quantity data obtained in the feature quantity data obtaining step;
An output step in which the output unit outputs the first rule acquired in the first rule acquisition step and the second rule acquired in the second rule acquisition step;
A time series data analysis method comprising: A computer that can acquire one or more types of time-series observation data, which is the observation data for the observation target,
One type of observation data among the acquired one or more types of time-series observation data is divided into a first period and the second period shorter than the first period, and the divided first data First feature amount data that is time-series data of characteristic values in each of one period and second feature amount data that is time-series data of characteristic values in each of the divided second periods A feature data acquisition unit for acquiring
A first rule acquisition unit that acquires a first rule related to the object using the first feature amount data acquired by the feature data acquisition unit;
A second rule acquisition unit that acquires a second rule related to the object using the second feature amount data acquired by the feature data acquisition unit; and
A program for functioning as an output unit that outputs the first rule acquired by the first rule acquisition unit and the second rule acquired by the second rule acquisition unit.

Images