JP2021157544A - Information processing device, information processing method, and program - Google Patents

Abstract

To generate a feature vector about respective sensors of a plurality of facilities with the same reference.SOLUTION: A feature vector generation device 1 includes a data acquisition part 121 for acquiring a first time series data corresponding to a plurality of first sensors provided in a first facility and second time series data corresponding to a plurality of second sensors provided in a second facility obtained by associating sensor identification information with a sensor operation time, a combination specification part 123 for specifying a combination of a first sensor and a second sensor having a similar attribute, and a feature vector generation part 124 for generating a feature vector for showing a relationship between an operation state of a sensor shown by the first time series data and the second time series data, and an operation state of the other sensors, and reflecting a similarity between the first sensor and the second sensor corresponding to the specified combination for each of the plurality of first sensors and the plurality of second sensors.SELECTED DRAWING: Figure 2

Description

The present invention relates to an information processing device, an information processing method, and a program for generating a feature vector.

With the widespread use of IoT (Internet of Things) devices, various analyzes using IoT devices are being performed. For example, in Non-Patent Document 1, time-series data is generated in which the sensor IDs of home sensors, which are IoT devices, are arranged in the order of time when the home sensors are operated, and the time-series data is described as sentences and the sensor IDs included in the time-series data. A technique is disclosed in which a feature vector indicating a co-occurrence relationship between sensors is assigned to each sensor by using Word2Vec as a word.

A plurality of sensors that are activated by a user in the house performing a predetermined action or an action linked to the predetermined action are arranged at relatively close positions on the feature space. Therefore, by associating the user's behavior with the operating status of the sensor, it is possible to recognize the behavior of the user in the house and predict the behavior based on the operating status of the sensor.

Kushal Singla, and Joy Bose. "IoT2Vec: Identification of Similar IoT Devices via Activity Footprints." 2018 International Conference on Advances in Computing, Communications and Informatics (ICACCI). IEEE, 2018

By the way, in the conventional technique, the feature vector is assigned to the sensor of one facility (in-house), and the relevance of the sensor in a plurality of facilities is not considered. Since multiple facilities are independent of each other, when the sensor feature vector is generated based on the operating status of the sensors installed in each of the multiple facilities, the feature vector is generated based on each of the multiple facilities. Will be done. Therefore, even if the sensor is used for the same purpose in each of a plurality of facilities, the feature vector of the sensor may differ depending on the facility in which the sensor is installed. Therefore, there is a problem that the user's behavior recognition and behavior prediction cannot be shared in a plurality of facilities. From this, it is required to generate the feature vector of each sensor of a plurality of facilities based on the same standard so that the user's behavior recognition and behavior prediction can be shared among the plurality of facilities.

Therefore, the present invention has been made in view of these points, and provides an information processing device, an information processing method, and a program capable of generating feature vectors of sensors of a plurality of facilities based on the same reference. With the goal.

The information processing device according to the first aspect of the present invention is time-series data in which the sensor identification information for identifying the sensor and the time when the sensor is activated are associated with each other, and a plurality of information processing devices provided in the first facility. The first time-series data, which is the time-series data corresponding to the first sensor, and the second time-series data, which is the time-series data corresponding to a plurality of second sensors provided in the second facility. The data acquisition unit to be acquired and the first attribute information indicating the attributes of the first sensor associated with the sensor identification information of each of the plurality of first sensors are acquired, and each of the plurality of second sensors is acquired. The attribute information acquisition unit that acquires the second attribute information indicating the attribute of the second sensor associated with the sensor identification information, and the attribute information that has similar attributes based on the first attribute information and the second attribute information. The first time-series data and the second time-series data are provided for each of the combination identification unit that specifies the combination of the first sensor and the second sensor, the plurality of first sensors, and the plurality of second sensors. The relationship between the operating status of the indicated sensor and the operating status of other sensors is shown, and the similarity between the first sensor and the second sensor corresponding to the combination specified by the combination specifying unit is reflected. It is provided with a feature vector generation unit that generates a feature vector.

The feature vector generation unit learns the relationship between the operating states of the plurality of first sensors indicated by the first time series data, generates feature vectors for each of the plurality of first sensors, and generates the feature vectors of the plurality of first sensors. After generating each feature vector of one sensor, the relationship between the operating states of the plurality of second sensors indicated by the second time series data is learned, and the first combination corresponding to the combination specified by the combination specifying unit. By generating the feature vectors of each of the plurality of second sensors so that the feature vectors of the one sensor and the second sensor match or resemble each other, the first combination corresponding to the combination specified by the combination identification unit. A feature vector of each of the plurality of second sensors may be generated, which reflects the similarity between the one sensor and the second sensor.

The feature vector generation unit has a relationship between the operating status of the plurality of first sensors indicated by the first time series data and a relationship between the operating status of the plurality of second sensors indicated by the second time series data. At the same time, the first sensor corresponding to the combination specified by the combination specifying unit and the second sensor of the plurality of first sensors and the second sensor so that the feature vectors of the second sensor match or resemble each other. By generating each feature vector, the plurality of the first sensor and the plurality of the first sensor and the plurality of the first sensor corresponding to the combination specified by the combination identification unit and the similarity between the first sensor and the second sensor are reflected. The feature vector of each of the second sensors may be generated.

The first attribute information and the second attribute information indicate a plurality of attributes, and in the information processing apparatus, the first attribute identification unit specifies the combination identification unit after the feature vector generation unit generates the feature vector. Based on the similarity of the feature vector between the 1 sensor and the 2nd sensor, it is significant from a plurality of attributes corresponding to the attribute information of the 1st sensor and the 2nd sensor specified by the combination specifying unit. It may further include an attribute extraction unit that extracts one or more attributes.

The feature vector generation unit reflects the similarity based on the attributes extracted by the attribute extraction unit for each of the plurality of first sensors and the plurality of second sensors after the attribute extraction unit extracts the attributes. The feature vector may be regenerated.

The data acquisition unit acquires the first label information in which the action type indicating the action performed by the first user who operates the first sensor in the first facility and the time when the action was performed are associated with each other. However, the information processing apparatus further includes a label specifying unit that identifies the label corresponding to the second sensor to the label indicating the action type corresponding to the first sensor having a feature vector similar to that of the second sensor. You may.

The information processing method according to the second aspect of the present invention is time-series data in which the sensor identification information for identifying the sensor, which is executed by the computer, is associated with the time when the sensor is activated, and is provided in the first facility. The first time-series data, which is the time-series data corresponding to the plurality of first sensors provided, and the second time-series data, which is the time-series data corresponding to the plurality of second sensors provided in the second facility. The step of acquiring the time-series data, the first attribute information indicating the attribute of the first sensor associated with the sensor identification information of each of the plurality of first sensors, and the plurality of second sensors are acquired. The step of acquiring the second attribute information indicating the attribute of the second sensor associated with each of the sensor identification information, and the said that the attributes are similar based on the first attribute information and the second attribute information. The step of specifying the combination of the first sensor and the second sensor, and the sensors indicated by the first time series data and the second time series data for each of the plurality of the first sensor and the plurality of the second sensors. A feature that shows the relationship between the operating status of the sensor and the operating status of other sensors, and reflects the similarity between the first sensor and the second sensor corresponding to the combination specified by the combination specifying unit. It has a step of generating a vector.

The program according to the third aspect of the present invention is time-series data in which the computer associates the sensor identification information for identifying the sensor with the time when the sensor is activated, and is provided in the first facility. The first time-series data, which is the time-series data corresponding to the plurality of first sensors, and the second time-series data, which is the time-series data corresponding to the plurality of second sensors provided in the second facility. The data acquisition unit that acquires the first attribute information indicating the attribute of the first sensor associated with the sensor identification information of each of the plurality of first sensors, and each of the plurality of second sensors. The attribute information acquisition unit that acquires the second attribute information indicating the attribute of the second sensor associated with the sensor identification information, the first attribute information, and the second attribute information having similar attributes based on the first attribute information and the second attribute information. The first time-series data and the second time-series data are provided for each of the combination identification unit that specifies the combination of the one sensor and the second sensor, and the plurality of the first sensor and the plurality of the second sensors. The relationship between the operating status of the indicated sensor and the operating status of other sensors is shown, and the similarity between the first sensor and the second sensor corresponding to the combination specified by the combination specifying unit is reflected. It functions as a feature vector generator that generates a feature vector.

According to the present invention, it is possible to generate a feature vector based on the same reference for each sensor of a plurality of facilities.

It is a figure explaining the outline of the feature vector generation apparatus which concerns on 1st Embodiment. It is a figure which shows the structure of the feature vector generation apparatus which concerns on 1st Embodiment. It is a figure which shows the example which arranged the sensor in the feature space based on the feature vector generated by the feature vector generation part. It is a flowchart which shows an example of the processing flow in the feature vector generation apparatus 1 which concerns on 1st Embodiment. It is a figure which shows the structure of the feature vector generation apparatus which concerns on 2nd Embodiment.

<First Embodiment>
[Overview of feature vector generator]
FIG. 1 is a diagram illustrating an outline of a feature vector generator according to the first embodiment. The feature vector generator is a computer that generates a feature vector showing the co-occurrence relationship of a plurality of sensors, which is a kind of IoT device installed in the facility.

As shown in FIG. 1, the first house A as the first facility _{is provided with at least the first sensors ia 1 to ia 5} , and the second house B as the second facility is provided with the second sensor. At least sensors ib _{1 to ib 5} are provided. The first sensor ia _{1 to ia 5} and the second sensor ib _{1 to ib 5} output an operation signal indicating that the operation has been performed in response to the operation. It is assumed that the first sensor ia _{1 to ia 5} and the second sensor ib _{1 to ib 5} are attached with attribute information indicating the type, installation location, and number of operations, respectively.

For example, the first sensor ia ₁ is a pressure sensor as shown in FIG. 1, and is provided with attribute information indicating that it is installed in a bedroom and operates 5 times per hour. Further, it is assumed that the first sensor ia ₂ is an illuminance sensor and is provided with attribute information indicating that it is installed in the bedroom and operates 0.2 times per hour.

Further, here, it is assumed that a sensor that constantly measures (for example, a temperature / humidity sensor or an illuminance sensor) is considered to have been activated if the measured value fluctuates beyond the threshold value set for the sensor. Time first terminal provided in the first housing A (not shown), which receives the actuation signals output from the first sensor _ia 1 _{~ia 5,} first sensor _ia 1 _{~ia 5} is actuated And the sensor ID that identifies the first sensor ia _{1 to ia 5} are associated with each other to generate the first time series data.

Similarly, the second sensors ib _{1 to ib 5} output an operation signal according to the operation. Time second terminal provided on the second housing B (not shown), which receives the actuation signals output from the second sensor _ib 1 _{~ib 5,} second sensor _ib 1 _{~ib 5} is actuated And the second time-series data in which the sensor IDs of the second sensors ib _{1 to ib 5 are associated with each other are generated.}

In the present embodiment, for convenience of explanation, the symbols (ia _{1 to ia 5} , ib _{1 to ib 5} ) provided on each sensor are used as the sensor ID. When the first sensors ia _{1 to ia 5} are not treated separately, these first sensors ia _{1 to ia 5} are referred to as the first sensor ia. Similarly, when the second sensors ib _{1 to ib 5} are not treated separately, these second sensors ib _{1 to ib 5} are referred to as the second sensor ib.

The feature vector generator acquires the first time-series data from the first terminal provided in the house A and the second time-series data from the second terminal provided in the house B. Further, the feature vector generator acquires the attribute information of the first sensor ia and the attribute information of the second sensor ib. The feature vector generator identifies the similarity of each of the plurality of second sensors ib with respect to each of the plurality of first sensors ia based on the acquired attribute information, and identifies a combination of similar sensors. In the example shown in FIG. 1, the feature vector generator _{identifies that the first sensor ia 3} and the second sensor ib ₃ are a combination of similar sensors, and the first sensor ia ₄ and the second sensor ib ₅ Is identified as a combination of similar sensors.

The feature vector generator generates a feature vector indicating the relationship between the operating states of the sensors based on the operating order of the sensors indicated by the first time series data and the second time series data. The feature vector generator generates a feature vector so that the feature vectors of the first sensor ia and the second sensor ib corresponding to the specified combination are substantially the same. As a result, the feature vector generator can generate a feature vector that reflects the similarity between the first sensor ia and the second sensor ib corresponding to the specified combination.

The lower part of FIG. 1 shows an example in which the first sensor ia and the second sensor ib are arranged in the feature space. As shown in the feature space of FIG. 1, the first sensor ia ₁ (pressure sensor _{installed in the bedroom), ia 2} (illuminance sensor installed in the bedroom), and the second sensor ib ₁ (bed). It can be confirmed that the acceleration sensors installed in the room) are located close to each other.

Further, as shown in the feature space of FIG. 1, the first sensor ia ₃ and the second sensor ib ₃ corresponding to the specified combination are arranged at positions close to each other, and the first sensor corresponding to the specified combination is arranged. It _{can be confirmed that the ia 4} and the second sensor ib ₅ are arranged close to each other. In this way, the feature vector generator can generate the feature vectors of the sensors provided in the houses A and B according to the same reference, and can arrange the sensors showing the same properties at positions close to each other. ..

[Configuration example of feature vector generator 1]
Next, the configuration of the feature vector generator will be described. FIG. 2 is a diagram showing a configuration of the feature vector generation device 1 according to the first embodiment. The feature vector generation device 1 includes a storage unit 11 and a control unit 12.

The storage unit 11 is, for example, a ROM (Read Only Memory), a RAM (Random Access Memory), or the like. The storage unit 11 stores various programs for operating the feature vector generation device 1. For example, the storage unit 11 causes the control unit 12 of the feature vector generation device 1 to function as a data acquisition unit 121, an attribute information acquisition unit 122, a combination identification unit 123, a feature vector generation unit 124, and a label identification unit 125. Memorize the program.

The control unit 12 is, for example, a CPU (Central Processing Unit). The control unit 12 controls the function related to the feature vector generation device 1 by executing various programs stored in the storage unit 11. The control unit 12 functions as a data acquisition unit 121, an attribute information acquisition unit 122, a combination identification unit 123, a feature vector generation unit 124, and a label identification unit 125 by executing a program stored in the storage unit 11.

[Generation of sensor feature vector]
In the first embodiment, the data acquisition unit 121, the attribute information acquisition unit 122, the combination identification unit 123, and the feature vector generation unit 124 cooperate with each other to correspond to each of the plurality of sensors provided in the facility. Generate a feature vector. Hereinafter, the functions of the data acquisition unit 121, the attribute information acquisition unit 122, the combination identification unit 123, and the feature vector generation unit 124 related to the generation of the feature vector will be described.

The data acquisition unit 121 acquires time-series data in which the sensor ID that identifies the sensor corresponding to each of the plurality of facilities and the time when the sensor is activated are associated with each other at predetermined time (for example, one day).

Specifically, the data acquisition unit 121 is a first time-series data corresponding to a plurality of first sensors ia provided in the first facility from the first terminal provided in the first facility. Get time series data. The data acquisition unit 121 acquires the first time-series data in which the sensor ID of the first sensor ia that has been operated for a predetermined period (for example, one day) and the operation time of the first sensor ia are associated with each other from the first terminal. .. When the same first sensor ia is operated a plurality of times, the first time series data may include a plurality of sensor IDs of the first sensor ia in response to the operation.

In addition, the data acquisition unit 121 associates the action type indicating the action performed by the first user who activates the first sensor ia in the first facility from the first terminal with the time when the action was performed. Acquire the first label information. Examples of actions performed by the first user include "getting up", "washing face", "cooking", and "meal". Note that the label information is generated, for example, by recording the user's own actions.

Further, the data acquisition unit 121 may acquire a plurality of data acquisition units 121 from the second terminal provided in the second facility to the second facility in the same manner as acquiring the first time series data from the first terminal. The second time series data which is the time series data corresponding to the second sensor ib is acquired. The data acquisition unit 121 acquires the second time-series data in which the sensor ID of the second sensor ib that has been activated in a predetermined period and the operation time of the second sensor ib are associated with each other from the second terminal.

The attribute information acquisition unit 122 acquires attribute information indicating the attributes of the first sensor ia associated with the sensor IDs of the plurality of first sensors ia corresponding to the first facility. The attribute information acquisition unit 122 acquires attribute information indicating the attributes of the second sensor ib associated with the sensor IDs of the plurality of second sensor ibs corresponding to the second facility. The attribute information of the first sensor ia is also referred to as the first attribute information, and the attribute information of the second sensor ib is also referred to as the second attribute information.

The attribute information acquisition unit 122 may acquire a part of the attribute information among the plurality of attribute information by analyzing the time series data. For example, the attribute information acquisition unit 122 may acquire the number of times the sensor is operated as attribute information by analyzing the time series data.

The combination specifying unit 123 identifies the combination of the first sensor ia and the second sensor ib having similar attributes based on the first attribute information and the second attribute information. For example, the combination specifying unit 123 performs normalization such as matching the minimum value and the maximum value for each of the plurality of attributes indicated by the first attribute information and the second attribute information. The combination specifying unit 123 uses each of the plurality of attributes indicated by the normalized attribute information as the attribute vector of the sensor, and calculates the degree of similarity between the attribute vector of the first sensor ia and the attribute vector of the second sensor ib. The combination identification unit 123 calculates the cosine similarity between the attribute vector of the first sensor ia and the attribute vector of the second sensor ib, so that the attribute vector of the first sensor ia and the attribute of the second sensor ib are calculated. Calculate the similarity with the vector.

The combination specifying unit 123 specifies the combination of the first sensor ia and the second sensor ib whose calculated similarity exceeds a predetermined threshold value as the combination of the first sensor ia and the second sensor ib having similar attributes. The combination specifying unit 123 calculated the similarity by calculating the cosine similarity, but the similarity is not limited to this, and the similarity may be calculated by using another method.

The feature vector generation unit 124 sets the operating status of the sensor indicated by the first time series data and the second time series data and the operating status of the other sensors for each of the plurality of first sensor ia and the plurality of second sensor ib. A feature vector that reflects the similarity between the first sensor ia and the second sensor ib corresponding to the combination specified by the combination specifying unit 123 is generated.

First, the feature vector generation unit 124 learns the relationship between the operating states of the plurality of first sensors ia indicated by the first time series data, and generates the feature vectors of the plurality of first sensors ia. For example, the feature vector generation unit 124 generates sequence data in which the sensor IDs included in the first time-series data are arranged in the order of time when the sensors of the sensor IDs are activated as the first sequence data.

The feature vector generation unit 124 regards the first sequence data as one sentence, and the first sensor ia included in the first sequence data as one word included in the sentence. The feature vector generation unit 124 generates the feature vectors of the plurality of first sensors ia by using Word2vec for the first sequence data. The feature vector generation unit 124 generates each feature vector of the plurality of first sensors ia by learning so that the _{objective function T 1} shown in the following equation (1) is maximized.

ここで、目的関数Ｔ_１を示す式に出現するＰ（ｘ｜ｙ）は、条件付き確率（ｙが出現した場合にｘが出現する確率）である。また、Ｉ_Ａは、第１の施設としての第１住宅Ａで生活する世帯Ｓにおける第１センサーｉａのシーケンスデータ、Ｋはウィンドウサイズのパラメータである。

Here, P (x | y) that appears in the equation showing the objective function T ₁ is a conditional probability (probability that x appears when y appears). Also, I _A is the sequence data of the first sensor ia in households S living in the first housing A of the first facility, K is a parameter of the window size.

Subsequently, the feature vector generation unit 124 learns and combines the relationship between the operating states of the plurality of second sensors ib indicated by the second time series data after generating the feature vectors of the plurality of first sensors ia. The feature vectors of the plurality of second sensors ib are generated so that the feature vectors of the first sensor ia and the second sensor ib corresponding to the combination specified by the specific unit 123 match or are similar to each other. As a result, the feature vector generation unit 124 has the features of the plurality of second sensors ib reflecting the similarity between the first sensor ia corresponding to the combination specified by the combination identification unit 123 and the second sensor ib. Generate a vector.

For example, the feature vector generation unit 124 generates sequence data in which the sensor IDs included in the second time-series data are arranged in the order of time when the sensors of the sensor IDs are activated as the second sequence data. The feature vector generation unit 124 generates the feature vectors of the plurality of second sensor ibs by using Word2vec, assuming that the second sequence data is the data indicating the contextual relationship of the second sensor ib. Here, the feature vector generation unit 124 has the objective function T shown in the following equation (2) so that the feature vectors of the first sensor ia and the second sensor ib specified by the combination identification unit 123 are similar or match. _{By learning so that 2} is the maximum, the feature vectors of the plurality of second sensors ib are generated.

ここで、目的関数Ｔ_２を示す式に出現するＩ_Ｂは、第２の施設としての第２住宅Ｂで生活する世帯Ｔにおける第２センサーｉｂのシーケンスデータである。Ｃは、組合せ特定部１２３が特定した全ての組合せの集合を示す。α＜ｉａ_ｔ，ｉｂ_ｔ＞は、組合せ特定部１２３が特定した組合せに対応する第１センサーｉａの特徴ベクトルと、第２センサーｉｂの特徴ベクトルとの類似度である。β＜ｉａ_ｔ，ｉｂ_ｔ＞は、組合せ特定部１２３が特定した組合せに対応する第１センサーｉａの属性ベクトルと、第２センサーｉｂの属性ベクトルとの類似度である。また、式（２）における第２項の関数ｆ（ｘ，ｙ）は、ｘとｙとの差が小さければ小さいほど大きくなる関数であり、例えば、以下の式（３）に示される。

Here, I _B appearing in equation showing the objective function T _2, a sequence data of the second sensor ib in households T living on the second housing B as a second facility. C represents a set of all combinations specified by the combination specifying unit 123. α <ia _t , ib _t > is the degree of similarity between the feature vector of the first sensor ia corresponding to the combination specified by the combination specifying unit 123 and the feature vector of the second sensor ib. β <ia _t , ib _t > is the degree of similarity between the attribute vector of the first sensor ia corresponding to the combination specified by the combination specifying unit 123 and the attribute vector of the second sensor ib. Further, the function f (x, y) of the second term in the equation (2) is a function that increases as the difference between x and y becomes smaller, and is shown in the following equation (3), for example.

式（３）に示す、ｃはパラメータであり、適合する値（例えば十分に大きい値）が設定されるものとする。

It is assumed that c shown in the equation (3) is a parameter, and a matching value (for example, a sufficiently large value) is set.

The feature vector generation unit 124 generated the feature vector of the first sensor ia and then generated the feature vector of the second sensor ib, but the present invention is not limited to this. The feature vector generation unit 124 simultaneously obtains the relationship between the operating status of the plurality of first sensors ia indicated by the first time series data and the relationship between the operating status of the plurality of second sensors ib indicated by the second time series data. You may learn. The feature vector generation unit 124 has a plurality of first sensor ia and second sensor ib so that the feature vectors of the first sensor ia and the second sensor ib corresponding to the combination specified by the combination identification unit 123 match or are similar to each other. By generating each feature vector of the above, a plurality of first sensors ia and a plurality of first sensors ia and a plurality of sensors ia that reflect the similarity between the first sensor ia corresponding to the combination specified by the combination identification unit 123 and the second sensor ib. The feature vector of each of the second sensor ib may be generated.

_{For example, the feature vector generation unit 124 performs learning so that the objective function T 3} shown in the equation (4), which is a combination of the above equations (1) and (2), is maximized, thereby causing a plurality of first sensors. The feature vectors of each of the ia and the plurality of second sensors ib are generated.

By doing so, the feature vector generator 1 has the feature vectors of the plurality of first sensors ia and the plurality of second sensors ib, respectively, as in the case where the equations (1) and (2) are executed individually. Can be generated.

FIG. 3 is a diagram showing an example in which the sensor is arranged in the feature space based on the feature vector generated by the feature vector generation unit 124. Note that FIG. 3 shows an example in which the feature space is compressed two-dimensionally and the sensor is arranged in the compressed feature space for convenience of explanation. In FIG. 3, a plurality of elliptical marks M1 and a plurality of diamond-shaped marks M2 are arranged. These marks indicate the position of the sensor in the feature space. The mark M1 indicates the position of the first sensor ia, and the mark M2 indicates the position of the second sensor ib.

Further, in the feature space, an approximate range of action types corresponding to each sensor based on the operation time of each sensor and the time when the user's action occurs is shown by a region surrounded by a broken line. As shown in FIG. 3, the first sensor ia ₁ , ia ₂ and the second sensor ib ₁ are arranged in the vicinity on the feature space and are aggregated in one area corresponding to the action type of "wake up". Can be confirmed. Further, as shown in FIG. 3, the first sensor ia ₃ , the second sensor ib ₃ and the ib ₂ are integrated in one area corresponding to the action type of “face washing”, and the first sensor ia ₄ , Ia ₅ , and the second sensors ib ₄ , ib ₅ can be confirmed to be aggregated in one area corresponding to the action type of "cooking".

[Labeling the sensor]
In the first embodiment, the label specifying unit 125 functions as a predictor that predicts the behavior corresponding to the sensor that is not associated with the label information. In explaining this function, it is assumed that the second sensor ib is not associated with the action type indicating the action performed by the second user who operates the second sensor ib.

The label specifying unit 125 identifies the label corresponding to the second sensor ib corresponding to the second time series data to the label indicating the action type corresponding to the first sensor ia whose feature vector is similar to that of the second sensor ib. .. For example, the label specifying unit 125 has a first sensor ia having a feature vector whose distance (for example, Euclidean distance) from the feature vector of the second sensor ib is within a predetermined distance, and a second sensor ib having a feature vector similar to that of the second sensor ib. 1 Specify as sensor ia. Then, the label specifying unit 125 refers to the first label information and specifies the action type corresponding to the action performed by the user at the operation time of the specified first sensor ia as the action type corresponding to the second sensor ib. Thereby, the label corresponding to the second sensor ib is specified. By doing so, it is possible to give a label to the unlabeled second sensor ib and predict the behavior corresponding to the second sensor ib.

[Process flow in feature vector generator 1]
Subsequently, an example of the processing flow in the feature vector generator 1 will be described. FIG. 4 is a flowchart showing an example of the processing flow in the feature vector generation device 1 according to the first embodiment.

First, the data acquisition unit 121 acquires the first time series data and the second time series data (S1).
Subsequently, the attribute information acquisition unit 122 acquires the first attribute information and the second attribute information (S2).
Subsequently, the combination specifying unit 123 identifies the combination of the first sensor ia and the second sensor ib having similar attributes based on the first attribute information and the second attribute information (S3).

Subsequently, the feature vector generation unit 124 secondly arranges the first sequence data in which the sensor IDs included in the first time series data are arranged in the order of the operation time of the first sensor ia, and the sensor IDs included in the second time series data. The second sequence data arranged in the order of the operation time of the sensor ib is generated (S4).

Subsequently, the feature vector generation unit 124 generates the feature vectors of the first sensor ia and the second sensor ib based on the first sequence data and the second sequence data generated in S4 (S5).

Subsequently, the label specifying unit 125 sets the label corresponding to each of the plurality of second sensors ib corresponding to the second time series data to the action corresponding to the first sensor ia whose feature vector is similar to that of the second sensor ib. It is specified on the label indicating the type (S6).

[Effect in the first embodiment]
As described above, the feature vector generation device 1 according to the first embodiment includes the first time-series data in which the sensor ID of the first sensor corresponding to the first facility and the operating time of the sensor are associated with each other, and the second. The second time series data in which the sensor ID of the second sensor corresponding to the facility of the above and the operation time of the sensor are associated with each other is acquired, and a combination of similar sensors in the first facility and the second facility is specified. The feature vector generator 1 has a relationship between the operating status of the sensor indicated by the first time series data and the second time series data and the operating status of the other sensors for each of the plurality of first sensors and the plurality of second sensors. A feature vector that shows the characteristics and reflects the similarity between the first sensor and the second sensor corresponding to the specified combination is generated. By doing so, the feature vector generation device 1 can generate a feature vector based on the same reference for each sensor of the plurality of facilities.

<Second embodiment>
Subsequently, the second embodiment will be described. The attribute information acquired by the attribute information acquisition unit 122 may include attributes that are not related to the generation of the feature vector. Then, if the feature vector is generated based on the attribute information including the unrelated attributes, there is a problem that the accuracy of the feature vector deteriorates. Therefore, the feature vector generation device 1 according to the second embodiment extracts attributes that are highly likely to be related to the generation of the feature vector as significant attributes, and regenerates the feature vector based on the extracted attributes. Hereinafter, the feature vector generation device 1 according to the second embodiment will be described.

FIG. 5 is a diagram showing a configuration of the feature vector generation device 1 according to the second embodiment. As shown in FIG. 5, the feature vector generation device 1 according to the second embodiment further includes an attribute extraction unit 126.

The attribute extraction unit 126 generates the feature vector after the feature vector generation unit 124 generates the feature vector, and then the combination identification unit 123 is based on the similarity between the feature vectors of the first sensor ia and the second sensor ib specified by the combination identification unit 123. Extracts one or more significant attributes from the plurality of attributes corresponding to the attribute information of the first sensor ia and the second sensor ib specified by.

For example, when all the attributes included in the attribute information are {M1, M2, ..., MN} and the combination of some of all the attributes is {m1, m2, ..., mK}, the attribute is extracted. Of all the combinations of attributes that can be taken from all the attributes, the unit 126 selects one or more significant combinations of attributes {m1, m2, ..., MK} that maximize the value of the following equation (5). Extract as a combination corresponding to the attribute.

The feature vector generation unit 124 includes the first sensor ia based on the attributes extracted by the attribute extraction unit 126 for each of the plurality of first sensors ia and the plurality of second sensors ib after the attribute extraction unit 126 extracts the attributes. A feature vector that reflects the similarity with the second sensor ib is regenerated.

Specifically, the feature vector generation unit 124 calculates the similarity of the attribute vector based on the attribute extracted by the attribute extraction unit 126, and applies the similarity to the equation (2) or the equation (4). Regenerates the feature vectors of each of the plurality of first sensor ia and the plurality of second sensor ib.

[Effect in the second embodiment]
As described above, after the feature vector generation device 1 according to the second embodiment generates the feature vector, the similarity of the feature vectors of the first sensor ia and the second sensor ib corresponding to the combination having similar attributes is obtained. Based on this, one or more significant attributes are extracted from the plurality of attributes corresponding to the attribute information of the first sensor ia and the second sensor ib corresponding to the combination. By doing so, the feature vector generation device 1 can generate a feature vector based on one or more significant attributes, so that the accuracy of the feature vector can be improved.

Although the present invention has been described above using the above-described embodiment, the technical scope of the present invention is not limited to the scope described in the above-described embodiment. It will be apparent to those skilled in the art that various changes or improvements can be made to the above embodiments. For example, in the above-described embodiment, the feature vector generator 1 generates a feature vector based on the same reference for each sensor of the two facilities, but the present invention is not limited to this. The feature vector generator 1 may generate a feature vector based on the same reference for each sensor of three or more facilities. Further, in particular, the specific embodiment of the distribution / integration of the apparatus is not limited to those shown above, and all or a part thereof may be arbitrarily added according to various additions or functional loads. It can be functionally or physically distributed / integrated in units.

1 ... Feature vector generator, 11 ... Storage unit, 12 ... Control unit, 121 ... Data acquisition unit, 122 ... Attribute information acquisition unit, 123 ... Combination identification unit, 124.・ ・ Feature vector generation part, 125 ・ ・ ・ Label identification part, 126 ・ ・ ・ Attribute extraction part

Claims (8)

The time-series data in which the sensor identification information for identifying the sensor and the time when the sensor is activated are associated with each other, and the time-series data corresponding to a plurality of first sensors provided in the first facility. A data acquisition unit that acquires one time-series data and a second time-series data that is the time-series data corresponding to a plurality of second sensors provided in the second facility.
The first attribute information indicating the attribute of the first sensor associated with the sensor identification information of each of the plurality of first sensors is acquired, and the first attribute information is associated with the sensor identification information of each of the plurality of second sensors. The attribute information acquisition unit that acquires the second attribute information indicating the attributes of the second sensor, and
A combination specifying unit that specifies a combination of the first sensor and the second sensor having similar attributes based on the first attribute information and the second attribute information.
For each of the plurality of first sensors and the plurality of second sensors, the relationship between the operating status of the sensors indicated by the first time series data and the second time series data and the operating status of the other sensors is shown. In addition, a feature vector generation unit that generates a feature vector that reflects the similarity between the first sensor corresponding to the combination specified by the combination identification unit and the second sensor.
Information processing device equipped with. The feature vector generation unit learns the relationship between the operating states of the plurality of first sensors indicated by the first time series data, generates feature vectors for each of the plurality of first sensors, and generates the feature vectors of the plurality of first sensors. After generating each feature vector of one sensor, the relationship between the operating states of the plurality of second sensors indicated by the second time series data is learned, and the first combination corresponding to the combination specified by the combination specifying unit. By generating the feature vectors of each of the plurality of second sensors so that the feature vectors of the one sensor and the second sensor match or resemble each other, the first combination corresponding to the combination specified by the combination identification unit. A feature vector of each of the plurality of second sensors reflecting the similarity between the one sensor and the second sensor is generated.
The information processing device according to claim 1. The feature vector generation unit has a relationship between the operating status of the plurality of first sensors indicated by the first time series data and a relationship between the operating status of the plurality of second sensors indicated by the second time series data. At the same time, the first sensor corresponding to the combination specified by the combination specifying unit and the second sensor of the plurality of first sensors and the second sensor so that the feature vectors of the second sensor match or resemble each other. By generating each feature vector, the plurality of the first sensor and the plurality of the first sensor and the plurality of the first sensor corresponding to the combination specified by the combination identification unit and the similarity between the first sensor and the second sensor are reflected. Generate a feature vector for each of the second sensors.
The information processing device according to claim 1. The first attribute information and the second attribute information indicate a plurality of attributes.
After the feature vector generation unit generates the feature vector, the combination identification unit identifies the feature vector based on the similarity between the first sensor and the second sensor specified by the combination identification unit. It further includes an attribute extraction unit that extracts one or more significant attributes from a plurality of attributes corresponding to the attribute information of the first sensor and the second sensor.
The information processing device according to any one of claims 1 to 3. The feature vector generation unit reflects the similarity based on the attributes extracted by the attribute extraction unit for each of the plurality of first sensors and the plurality of second sensors after the attribute extraction unit extracts the attributes. Regenerate the feature vector
The information processing device according to claim 4. The data acquisition unit acquires the first label information in which the action type indicating the action performed by the first user who operates the first sensor in the first facility and the time when the action is performed are associated with each other. death,
A label specifying unit for specifying a label corresponding to the second sensor as a label indicating the action type corresponding to the first sensor having a feature vector similar to that of the second sensor is further provided.
The information processing device according to any one of claims 1 to 5. Computer runs,
The time-series data in which the sensor identification information for identifying the sensor and the time when the sensor is activated are associated with each other, and the time-series data corresponding to a plurality of first sensors provided in the first facility. A step of acquiring one time-series data and a second time-series data which is the time-series data corresponding to a plurality of second sensors provided in the second facility.
The first attribute information indicating the attribute of the first sensor associated with the sensor identification information of each of the plurality of first sensors is acquired, and the first attribute information is associated with the sensor identification information of each of the plurality of second sensors. The step of acquiring the second attribute information indicating the attribute of the second sensor, and
A step of identifying a combination of the first sensor and the second sensor having similar attributes based on the first attribute information and the second attribute information.
For each of the plurality of first sensors and the plurality of second sensors, the relationship between the operating status of the sensors indicated by the first time series data and the second time series data and the operating status of the other sensors is shown. At the same time, a step of generating a feature vector reflecting the similarity between the first sensor corresponding to the combination specified by the combination specifying unit and the second sensor, and
Information processing method having. Computer,
The time-series data in which the sensor identification information for identifying the sensor and the time when the sensor is activated are associated with each other, and the time-series data corresponding to a plurality of first sensors provided in the first facility. A data acquisition unit that acquires one time-series data and second time-series data, which is the time-series data corresponding to a plurality of second sensors provided in the second facility.
The first attribute information indicating the attribute of the first sensor associated with the sensor identification information of each of the plurality of first sensors is acquired, and the first attribute information is associated with the sensor identification information of each of the plurality of second sensors. Attribute information acquisition unit that acquires the second attribute information indicating the attributes of the second sensor.
A combination specifying unit that specifies a combination of the first sensor and the second sensor having similar attributes based on the first attribute information and the second attribute information, and
For each of the plurality of first sensors and the plurality of second sensors, the relationship between the operating status of the sensors indicated by the first time series data and the second time series data and the operating status of the other sensors is shown. In addition, a feature vector generation unit that generates a feature vector that reflects the similarity between the first sensor corresponding to the combination specified by the combination identification unit and the second sensor.
A program that functions as.

Images