JP7149499B2 - Data analysis method, data analysis device and data analysis program - Google Patents

Description

The present disclosure relates to a data analysis method, a data analysis device, and a data analysis program that perform data analysis processing for electrical appliances.

Conventionally, an autocorrelation function is often used to confirm the periodicity of time-series data or the relationship between time-series data and past data. However, when the time-series data is discrete time-series data that is not obtained at regular intervals, it is difficult to calculate the autocorrelation function of the discrete time-series data with high accuracy.

As a conventional technology to solve such a problem, sensor value interpolation that interpolates missing sensor values using a Gaussian process for discrete time series data in which sensor values are missing and data is not acquired at equal intervals A program has been proposed (see Patent Document 1, for example).

JP 2016-212066 A

However, when interpolating the missing sensor values as in the above-described conventional technique, there is a problem such as a decrease in interpolation accuracy if the number of data acquisitions is insufficient.

The present disclosure has been made to solve the above problems. An object of the present invention is to provide a data analysis method, a data analysis device, and a data analysis program capable of calculating an autocorrelation function from series data with high accuracy.

A data analysis method according to an aspect of the present disclosure is a data analysis method in a data analysis device that performs data analysis processing for an electrical appliance, comprising first information relating to an operation of the electrical appliance and temporal information of the first information. Acquiring time-series data information including M (M is a natural number of 3 or more) or more third information associated with second information representing an observation point, and acquiring N (N is using the second information included in each of the third information (a natural number of 3 or more), the absolute value of the time difference between each of the observation points is calculated, a set of the calculated time differences is generated as a first set, and the The value of the first information at a first time and the value of the first information at a second time after a predetermined time has passed from the first time for each set of the third information for which the absolute value of the time difference is calculated. calculating an autocorrelation coefficient representing the correlation of, generating a set of the calculated autocorrelation coefficients as a second set, and based on the first set and the second set, the set of time differences and the An autocorrelation function representing a relationship with a set of autocorrelation coefficients is calculated, the first information is analyzed in time series based on the autocorrelation function, and fourth information regarding the analysis result is output.

According to the present disclosure, even if the time series data is discrete time series data that is not acquired at equal intervals, the autocorrelation function is calculated with high accuracy from the discrete time series data using more data than the number of acquired data. can do.

1 is a diagram showing the overall configuration of an information processing system according to an embodiment of the present disclosure; FIG. 1 is a block diagram showing the configuration of a data analysis device according to an embodiment of the present disclosure; FIG. 1 is a block diagram showing the configuration of an electrical appliance according to an embodiment of the present disclosure; FIG. FIG. 5 is a sequence diagram showing time-series data storage processing according to the embodiment of the present disclosure; 4 is a flowchart showing abnormality determination processing of the data analysis device according to the embodiment of the present disclosure; FIG. 6 is a flowchart showing autocorrelation function calculation processing in step S13 of FIG. 5. FIG. It is a figure which shows an example of the database structure of the time series data storage part in embodiment of this indication. FIG. 5 is a diagram showing an example of a display screen displayed when an abnormality determination request is input in the data analysis device according to the embodiment of the present disclosure; FIG. 5 is a diagram showing an example of time-series data to be subjected to abnormality determination, which is extracted from the time-series data storage unit by the time-series data extraction process according to the embodiment of the present disclosure; FIG. 5 is a diagram showing an example of autocorrelation coefficients corresponding to time differences between observation points calculated by an autocorrelation coefficient set generator according to the embodiment of the present disclosure; FIG. 5 is a diagram showing an example of an abnormality determination result display screen generated by a data analysis unit according to the embodiment of the present disclosure; FIG.

(Findings on which this disclosure is based)
As described above, in analyzing time-series data, the autocorrelation function is used to know the data structure of the time-series data. By using the autocorrelation function, it is possible to grasp the correlation between data before and after, and data analysis such as future prediction becomes possible. A conventional autocorrelation function is calculated from time-series data acquired at a constant sampling period. However, for example, time-series data obtained from home appliances is often not obtained at regular sampling intervals, and it is difficult to calculate with high accuracy the autocorrelation function of time-series data in which data is not obtained at regular intervals. .

In the prior art, when a missing sensor value is detected within a predetermined period of time measured by the measurement unit, the missing sensor value is interpolated using a Gaussian process. Interpolation accuracy may decrease.

In order to solve the above problems, a data analysis method according to one aspect of the present disclosure is a data analysis method in a data analysis device that performs data analysis processing of an electrical appliance, wherein first information about the operation of the electrical appliance and , obtaining time-series data information including M (M is a natural number of 3 or more) third information associated with second information representing a temporal observation point of the first information, and obtaining time-series data information, Using the second information included in each of N pieces of the third information (N is a natural number of 3 or more) included in the information, calculating the absolute value of the time difference at each of the observation points, and calculating the set of the calculated time differences is generated as a first set, and for each set of the third information for which the absolute value of the time difference is calculated, the value of the first information at the first time and the second calculating an autocorrelation coefficient representing a correlation with the value of the first information at time, generating a set of the calculated autocorrelation coefficients as a second set, and dividing the first set and the second set into based on, calculating an autocorrelation function representing the relationship between the set of time lags and the set of autocorrelation coefficients, analyzing the first information in time series based on the autocorrelation function, and analyzing the result output the fourth information about

According to this configuration, using the second information included in each of N (N is a natural number of 3 or more) third information included in the time-series data information, the absolute value of the time difference between each observation point is calculated, A set of calculated time differences is generated, an autocorrelation coefficient representing a correlation between values of each first information is calculated for each set of third information for which the absolute values of the time differences are calculated, and the calculated autocorrelation coefficient A set of correlation coefficients is generated and an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients is calculated.

Therefore, while the number of variables used in the conventional method of interpolating the loss of discrete time series data by approximation is N, the number of variables used to calculate the autocorrelation function of the present disclosure is N Since the number of combinations of any two pieces of second information extracted from the third information is _N C ₂ , even if the time-series data is discrete time-series data that is not acquired at equal intervals, acquisition The autocorrelation function can be calculated with high accuracy from the discrete time-series data using more data than the number of data calculated. In addition, the autocorrelation function calculated with high accuracy can be used to accurately perform data analysis processing for electrical appliances.

Further, in the above data analysis method, a third set is generated by removing the overlapping time difference from the first set, and the generation of the second set is based on the third set. A set of the autocorrelation coefficients corresponding to the set is generated as the second set, and the calculation of the autocorrelation function includes calculating the autocorrelation function based on the third set and the second set. good too.

According to this configuration, in the first set, the third set is generated by removing overlapping time differences, the set of autocorrelation coefficients corresponding to the third set is generated as the second set, and the third set and the second set are generated. Since the autocorrelation function is calculated based on the set and the autocorrelation function is calculated based on the first set and the second set having overlapping time differences, the calculation process of the autocorrelation function requires time can be shortened.

Further, in the above data analysis method, further, based on the first set and the third set, a fourth A set may be generated, and the calculation of the autocorrelation function may be based on the second set, the third set and the fourth set to calculate the autocorrelation function.

According to this configuration, based on the first set and the third set, the fourth set, which is a set of duplicate deletion numbers corresponding to the set of duplicate time lags deleted from the first set, is generated, and the second set and the Since the autocorrelation function is calculated based on the third set and the fourth set, the autocorrelation function can be calculated by weighted approximation according to the number of overlaps of the deleted time differences.

Further, in the above data analysis method, the generation of the third set may add 0 to the elements of the third set.

According to this configuration, since 0 is added to the elements of the third set, it is possible to specify the value of the autocorrelation coefficient when the time difference is 0. For example, when the time difference is 0, the autocorrelation coefficient The value of the number can be specified as one.

Further, in the above data analysis method, the calculation of the autocorrelation function is based on the second set, the third set, and the fourth set, and the output value of the autocorrelation function is -1 or more and 1 or less. The autocorrelation function may be calculated as follows.

According to this configuration, the autocorrelation function is calculated so that the output value of the autocorrelation function is -1 or more and 1 or less based on the second set, the third set, and the fourth set. Output values of the autocorrelation function within the range of 1 or less can be used for data analysis.

Further, in the above data analysis method, the generation of the third set may add 0 to the elements of the third set.

According to this configuration, since 0 is added to the elements of the third set, it is possible to specify the value of the autocorrelation coefficient when the time difference is 0. For example, when the time difference is 0, the autocorrelation coefficient The value of the number can be specified as one.

Further, in the above data analysis method, the calculation of the autocorrelation function is performed based on the second set and the third set so that the output value of the autocorrelation function is -1 or more and 1 or less. An autocorrelation function may be calculated.

According to this configuration, the autocorrelation function is calculated based on the second set and the third set so that the output value of the autocorrelation function is -1 or more and 1 or less. Output values of the autocorrelation function within the range can be used for data analysis.

Further, in the above data analysis method, the calculation of the autocorrelation function is performed based on the second set and the third set so that the output value of the autocorrelation function is -1 or more and 1 or less. An autocorrelation function may be calculated.

According to this configuration, the autocorrelation function is calculated based on the second set and the third set so that the output value of the autocorrelation function is -1 or more and 1 or less. Output values of the autocorrelation function within the range can be used for data analysis.

Further, in the above data analysis method, instruction information for instructing the data analysis device to perform the data analysis processing is further received, and the generation of the first set is performed after the instruction information is received. , using the second information included in each of N pieces of the third information (N is a natural number of 3 or more) included in the time-series data information, calculating the absolute value of the time difference of each of the observation points, and calculating A set of the time differences obtained by the calculation may be generated as a first set.

According to this configuration, a set of time differences can be generated as the first set by triggering reception of instruction information instructing the data analysis device to perform data analysis processing.

Further, in the data analysis method described above, the instruction information may be input by a user at a terminal connected to the data analysis device via a network.

According to this configuration, since the instruction information is input by the user at the terminal connected to the data analysis device via the network, the terminal remotely instructs the data analysis device to perform the data analysis processing. be able to.

Further, in the above data analysis method, the analysis of the first information determines whether the operation of the electrical appliance is normal, and the fourth information determines whether the operation of the electrical appliance is normal. It may contain information about whether According to this configuration, it is possible to determine whether or not the operation of the electrical appliance is normal.

A data analysis device according to another aspect of the present disclosure is a data analysis device that performs data analysis processing of an electrical appliance, and includes first information about the operation of the electrical appliance and a temporal observation point of the first information. an acquisition unit that acquires time-series data information including M (M is a natural number of 3 or more) third information associated with the second information represented, and a control unit that performs data analysis processing on the time-series data information and the control unit uses the second information included in each of N (N is a natural number of 3 or more) third information included in the time-series data information to calculate the time difference between each of the observation points to generate a set of the calculated time differences as a first set, and for each set of the third information for which the absolute values of the time differences have been calculated, the value of the first information at a first time and the value of the first information at a second time after a predetermined time has elapsed from the first time, and a set of the calculated autocorrelation coefficients is generated as a second set. and calculating an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients based on the first set and the second set, and calculating the first set of autocorrelation coefficients based on the autocorrelation function. 1 information is analyzed chronologically, and fourth information relating to the analysis results is output.

According to this configuration, using the second information included in each of N (N is a natural number of 3 or more) third information included in the time-series data information, the absolute value of the time difference between each observation point is calculated, A set of calculated time differences is generated, an autocorrelation coefficient representing a correlation between values of each first information is calculated for each set of third information for which the absolute values of the time differences are calculated, and the calculated autocorrelation coefficient A set of correlation coefficients is generated and an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients is calculated.

Therefore, while the number of variables used in the conventional method of interpolating the loss of discrete time series data by approximation is N, the number of variables used to calculate the autocorrelation function of the present disclosure is N Since the number of combinations of any two pieces of second information extracted from the third information is _N C ₂ , even if the time-series data is discrete time-series data that is not acquired at equal intervals, acquisition The autocorrelation function can be calculated with high accuracy from the discrete time-series data using more data than the number of data calculated. In addition, the autocorrelation function calculated with high accuracy can be used to accurately perform data analysis processing for electrical appliances.

A data analysis program according to another aspect of the present disclosure is a data analysis program for performing data analysis processing of an electrical appliance, wherein a computer stores first information regarding the operation of the electrical appliance and the time of the first information. acquire time-series data information including M pieces (M is a natural number of 3 or more) or more of third information associated with second information representing a specific observation point, and obtain N pieces ( N is a natural number of 3 or more), using the second information included in each of the third information, calculating the absolute value of the time difference between each of the observation points, and generating a set of the calculated time differences as a first set. , for each set of the third information for which the absolute value of the time difference is calculated, the value of the first information at a first time and the first information at a second time after a predetermined time has passed from the first time calculating an autocorrelation coefficient representing a correlation with a value, generating a set of the calculated autocorrelation coefficients as a second set, and generating the set of time differences based on the first set and the second set and the set of autocorrelation coefficients, analyzes the first information in time series based on the autocorrelation function, and outputs fourth information about the analysis result. Let the process run.

According to this configuration, using the second information included in each of N (N is a natural number of 3 or more) third information included in the time-series data information, the absolute value of the time difference between each observation point is calculated, A set of calculated time differences is generated, an autocorrelation coefficient representing a correlation between values of each first information is calculated for each set of third information for which the absolute values of the time differences are calculated, and the calculated autocorrelation coefficient A set of correlation coefficients is generated and an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients is calculated.

Therefore, while the number of variables used in the conventional method of interpolating the loss of discrete time series data by approximation is N, the number of variables used to calculate the autocorrelation function of the present disclosure is N Since the number of combinations of any two pieces of second information extracted from the third information is _N C ₂ , even if the time-series data is discrete time-series data that is not acquired at equal intervals, acquisition The autocorrelation function can be calculated with high accuracy from the discrete time-series data using more data than the number of data calculated. In addition, the autocorrelation function calculated with high accuracy can be used to accurately perform data analysis processing for electrical appliances.

The embodiment described below shows one specific example of the present disclosure. Numerical values, shapes, components, steps, order of steps, and the like shown in the following embodiments are examples and are not intended to limit the present disclosure. In addition, among the constituent elements in the following embodiments, constituent elements that are not described in independent claims representing the highest concept will be described as arbitrary constituent elements.

(Embodiment)
FIG. 1 is a diagram showing the overall configuration of an information processing system according to an embodiment of the present disclosure.

The information processing system 1 includes a data analysis device 100 and electrical appliances 200 . Here, data analysis device 100 is connected to electrical appliance 200 via network 300 so as to be able to communicate with each other. Network 300 may be, for example, the Internet or an intranet. The connection between the data analysis device 100 and the electrical appliance 200 may be a wireless connection such as Wi-Fi (registered trademark), Bluetooth (registered trademark), ECHONET (registered trademark), or ECHONET LIGHT (registered trademark). However, a wired connection such as Ethernet (registered trademark) may be used, and any connection may be used as long as necessary information can be communicated.

The electrical appliance 200 is an electrical appliance capable of acquiring first information (hereinafter referred to as operation information) regarding the operation of the electrical appliance 200 . In the present embodiment, a case in which electrical appliance 200 is an air conditioner will be described as an example, but electrical appliance 200 may be any electrical appliance capable of acquiring time-series data. Specifically, the electrical appliance 200 is, for example, a television, an air conditioner, a washing machine, a heart rate monitor, an automatic weather data collection system, or the like.

FIG. 2 is a block diagram showing the configuration of the data analysis device according to the embodiment of the present disclosure. Data analysis device 100 is, for example, a computer. The data analysis device 100 includes a processor 11 , a memory 12 , a communication section 13 , an input section 14 and an output section 15 . The memory 12 is implemented, for example, by an auxiliary storage device such as a HDD (hard disk drive), RAM (random access memory), or semiconductor memory. The memory 12 has a time-series data storage unit 121 .

Time-series data storage unit 121 stores M (M is a natural number of 3 or more) in which operation information of electrical appliance 200 is associated with second information (hereinafter referred to as time information) representing temporal observation points of the operation information. Time-series data information (hereinafter referred to as time-series data) including the above third information is stored.

Communication unit 13 communicates with electrical appliance 200 via network 300 . The communication unit 13 provides M (M is a natural number of 3 or more) or more third information in which the first information about the operation of the electrical appliance 200 is associated with the second information representing the temporal observation points of the first information. Get time series data containing Communication unit 13 receives the time-series data of electrical appliance 200 and outputs the data to processor 11 .

The processor 11 acquires information from the time-series data storage unit 121, the communication unit 13, and the input unit 14, and executes processing according to the acquired information. Also, the processor 11 transmits information to the time series data storage unit 121 , the communication unit 13 and the output unit 15 . The processor 11 includes an information processing section 111 , a time series data extraction section 112 , a calculation processing section 113 and a data analysis section 117 .

When information processing section 111 acquires the time series data of electrical appliance 200 from communication section 13 , information processing section 111 stores the time series data in time series data storage section 121 . Details of the time-series data storage processing will be described later. The information processing section 111 also outputs the abnormality determination request input by the input section 14 to the time-series data extraction section 112 and the calculation processing section 113 . The information processing section 111 also outputs the abnormality determination result display screen output by the data analysis section 117 to the output section 15 . The details of the abnormality determination result display screen will be described later.

When acquiring the abnormality determination request, the time-series data extraction unit 112 extracts from the time-series data storage unit 121 time-series data required for abnormality determination in response to the abnormality determination request.

When time-series data is extracted by the time-series data extraction unit 112, the calculation processing unit 113 executes autocorrelation function calculation processing. Calculation processing section 113 includes time difference set generation section 114 , autocorrelation coefficient set generation section 115 , and autocorrelation function calculation section 116 .

The time difference set generation unit 114 uses the second information included in each of the N (N is a natural number of 3 or more) third information included in the time series data to calculate the absolute value of the time difference at each observation point, A set of calculated time differences is generated as a first set. Also, the time difference set generation unit 114 generates a third set by removing redundant time differences from the first set.

The autocorrelation coefficient set generation unit 115 calculates the value of the first information at the first time and the An autocorrelation coefficient representing the correlation with the value of the first information is calculated, and a set of calculated autocorrelation coefficients is generated as a second set. Also, based on the third set, autocorrelation coefficient set generation section 115 generates a set of autocorrelation coefficients corresponding to the third set as the second set.

Autocorrelation function calculator 116 calculates an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients based on the second set and the third set.

In the present embodiment, time difference set generation section 114 generates a third set by removing overlapping time differences from the first set, and autocorrelation coefficient set generation section 115 generates, based on the third set, Although the set of autocorrelation coefficients corresponding to the third set is generated as the second set, the present disclosure is not particularly limited to this, and the time difference set generator 114 does not need to generate the third set. , the autocorrelation coefficient set generator 115 may generate a set of autocorrelation coefficients corresponding to the first set as the second set based on the first set. In this case, the autocorrelation function calculator 116 calculates an autocorrelation function representing the relationship between the set of time lags and the set of autocorrelation coefficients based on the first set and the second set.

Data analysis section 117 time-series analyzes the first information based on the autocorrelation function, and outputs fourth information regarding the analysis result. Data analysis unit 117 determines whether electrical appliance 200 operates normally. The fourth information includes information regarding whether the operation of electrical appliance 200 is normal.

The input unit 14 receives a user's operation as an input and outputs it to the processor 11 . The contents of the user's operation include an abnormality determination request. The input unit 14 is, for example, a remote controller, a touch panel, a voice input device, a keyboard, or the like, and may be realized by any means.

Note that the user's operation may be input from an external device via the communication unit 13, and in this case, the data analysis device 100 does not have to include the input unit 14. FIG. That is, the communication unit 13 may receive an abnormality determination request (instruction information) that instructs the data analysis device 100 to perform data analysis processing. After the abnormality determination request (instruction information) is received, the time difference set generation unit 114 uses the second information included in each of the N (N is a natural number of 3 or more) third information included in the time series data. , the absolute value of the time difference of each observation point may be calculated, and a set of calculated time differences may be generated as the first set. The abnormality determination request (instruction information) may be input by a user at a terminal connected to data analysis apparatus 100 via network 300 .

The output unit 15 can display various types of information, and specifically displays an abnormality determination result display screen acquired from the processor 11 and the like. The abnormality determination result display screen will be described later. The output unit 15 is, for example, a display or a projector, and may be realized by any means as long as it can display various information. Various information may be displayed on an external device via the communication unit 13 , and in this case, the data analysis device 100 does not have to include the output unit 15 .

Also, the data analysis device 100 may be a server. When the data analysis device 100 is a server, the data analysis device 100 does not have to include the input unit 14 and the output unit 15, and a user terminal communicatively connected to the data analysis device 100 via the network 300 An input unit 14 and an output unit 15 may be provided. The user terminal is, for example, a smartphone or a tablet computer, and may transmit an abnormality determination request to the data analysis device 100 and receive an abnormality determination result or an abnormality determination result display screen from the data analysis device 100 . Then, the user terminal may generate an abnormality determination result display screen from the received abnormality determination result and display the generated abnormality determination result display screen, or may display the received abnormality determination result display screen. good.

FIG. 3 is a block diagram showing the configuration of the electrical appliance according to the embodiment of the present disclosure. Electrical appliance 200 includes operation unit 21 , processor 22 , memory 23 , sensor 24 and communication unit 25 .

The operation unit 21 receives user's operations as input and issues instructions to the processor 22 . The operation unit 21 outputs an input instruction to the processor 22 according to a user's operation. The contents of the user's operation include, for example, power ON/OFF switching, operation state switching, and setting parameter change for the electrical appliance 200 . Also, the operation unit 21 is, for example, a remote controller, a touch panel, a voice input device, or the like. Any operation means may be used as the operation unit 21 .

Sensors 24 include various sensors provided on appliance 200 . If the electrical appliance 200 is an air conditioner, the sensor 24 measures, for example, indoor pipe temperature and air volume. The sensor 24 outputs the measured sensor value to the motion information acquisition unit 222 . Note that the sensor values measured by the sensor 24 are not limited to the indoor pipe temperature and the air volume, and differ according to the type of the electrical appliance 200 .

Processor 22 executes operation control for electrical appliance 200 . The processor 22 includes a device control section 221 , an operation information acquisition section 222 and a communication control section 223 .

Device control unit 221 acquires an input instruction according to a user's operation from operation unit 21 and causes electrical appliance 200 to perform an operation corresponding to the input instruction. Device control section 221 may acquire operation information of electrical appliance 200 from memory 23 and control the operation of electrical appliance 200 according to the state of electrical appliance 200 indicated by the operation information.

The memory 23 is implemented, for example, by an auxiliary storage device such as HDD, RAM, or semiconductor memory. The memory 23 has an operation information storage unit 231 . The motion information storage unit 231 stores motion information acquired by the motion information acquisition unit 222 . The motion information storage unit 231 outputs the stored motion information to the communication control unit 223 .

Operation information acquisition unit 222 acquires operation information from electrical appliance 200 and stores it in memory 23 . At this time, the time interval at which the operation information is acquired from electrical appliance 200 may not be constant. The operation information acquired from the electric appliance 200 includes, for example, identification information of the electric appliance 200, operation state information indicating the operation state of the electric appliance 200, user setting information indicating input by user operation, and sensor data measured by the sensor 24. It contains various information about the electrical appliance 200 such as value information.

The communication control section 223 reads the motion information from the motion information storage section 231 and outputs it to the communication section 25 .

The communication unit 25 communicates with the data analysis device 100 via the network 300 . Communication unit 25 acquires the operation information stored in operation information storage unit 231 and transmits the operation information of electrical appliance 200 to data analysis device 100 .

The processing of the information processing system according to the present embodiment will be specifically described below with reference to the drawings.

FIG. 4 is a sequence diagram showing time series data storage processing according to the embodiment of the present disclosure.

First, the operation information acquisition unit 222 of the electrical appliance 200 periodically acquires the operation information of the electrical appliance 200 (step S1).

Next, the motion information acquisition unit 222 stores the acquired motion information in the motion information storage unit 231 (step S2).

Next, the communication control unit 223 reads motion information from the motion information storage unit 231 (step S3).

Next, the communication unit 25 periodically transmits the operation information read by the communication control unit 223 to the data analysis device 100 (step S4).

Next, communication unit 13 of data analysis device 100 receives the operation information transmitted by electrical appliance 200 (step S5). The communication unit 13 outputs the received motion information to the information processing unit 111 .

Next, when the motion information is acquired, the information processing section 111 generates time-series data in which the motion information and the time information are associated (step S6). The time information is the time when the information processing section 111 generates the time series data. Note that the time information may be the time when the communication unit 13 receives the operation information.

Next, the information processing section 111 stores the generated time-series data in the time-series data storage section 121 .

Note that the time-series data in which the motion information and the time information are associated may be generated by the communication control unit 223 of the electrical appliance 200 . In this case, the communication control unit 223 reads the motion information from the motion information storage unit 231 and generates time-series data in which the read motion information and the time information are associated with each other. At this time, the time information may be the time when the communication control unit 223 generates the time-series data. Also, the time information may be the time when the motion information acquisition unit 222 acquires the motion information. Then, the communication unit 25 periodically transmits the generated time-series data to the data analysis device 100 . The communication unit 13 of the data analysis device 100 receives the time-series data transmitted by the electrical appliance 200 . The communication unit 13 outputs the received time series data to the information processing unit 111 . The information processing section 111 stores the acquired time-series data in the time-series data storage section 121 .

Further, when motion information is not acquired at regular time intervals, the time-series data stored in the time-series data storage unit 121 by the information processing unit 111 is time-series data at irregular intervals.

FIG. 5 is a flowchart showing abnormality determination processing of the data analysis device according to the embodiment of the present disclosure.

First, the information processing unit 111 of the data analysis device 100 acquires an abnormality determination request input by the input unit 14 (step S11). The abnormality determination request includes, for example, all information necessary for abnormality determination of the electrical appliance 200, such as identification information of the electrical appliance 200 to be subjected to abnormality determination, sensor information to be subjected to abnormality determination, and abnormality determination period information.

Next, the time-series data extraction unit 112 extracts corresponding time-series data for abnormality determination from the time-series data storage unit 121 in response to the abnormality determination request (step S12). That is, the time-series data extraction unit 112 extracts time-series data corresponding to the identification information, the sensor information, and the abnormality determination target period information included in the abnormality determination request from the time-series data stored in the time-series data storage unit 121. Extract data.

Next, the calculation processing unit 113 executes autocorrelation function calculation processing for calculating an autocorrelation function based on the time series data for abnormality determination extracted by the time series data extraction unit 112 (step S13). Details of the autocorrelation function calculation processing by the calculation processing unit 113 will be described later using FIG.

Next, the data analysis unit 117 determines whether or not the autocorrelation function has been calculated by the calculation processing unit 113 (step S14). Here, if it is determined that the autocorrelation function has not been calculated (NO in step S14), the data analysis unit 117 performs determination for notifying the user that the abnormality determination instructed by the abnormality determination request could not be performed. A disapproval notification screen is generated (step S15).

Next, the output unit 15 displays the judgment failure notification screen generated by the data analysis unit 117 (step S16). By confirming the determination failure notification screen displayed on the output unit 15, the user can know that the electrical appliance 200 has not been determined to be abnormal.

Note that in the present embodiment, when it is determined that the autocorrelation function has not been calculated, a determination failure notification screen is generated for notifying the user that the abnormality determination instructed by the abnormality determination request could not be performed. However, the present disclosure is not particularly limited to this, and when it is determined that the autocorrelation function has not been calculated, the abnormality determination process may end without performing the processes of steps S15 and S16.

On the other hand, if it is determined that the autocorrelation function has been calculated (YES in step S14), the data analysis unit 117 determines whether the time-series data subject to abnormality determination is abnormal based on the autocorrelation function calculated by the calculation processing unit 113. (step S17). The data analysis unit 117, for example, calculates the autocorrelation coefficient based on the autocorrelation function calculated by the calculation processing unit 113, and determines whether the calculated autocorrelation coefficient falls within a predetermined normal range. judge. When the data analysis unit 117 determines that the calculated autocorrelation coefficient falls within a predetermined normal range, the data analysis unit 117 determines that the time-series data to be subjected to abnormality determination is normal without abnormality. If the data analysis unit 117 determines that the calculated autocorrelation coefficient does not fall within the predetermined normal range, the data analysis unit 117 determines that the time-series data subject to abnormality determination has an abnormality.

The data analysis unit 117 uses the autocorrelation function calculated by the calculation processing unit 113 as an input variable in machine learning, thereby outputting an abnormality determination result indicating whether or not there is an abnormality in the time-series data to be subjected to abnormality determination. You may The data analysis unit 117 uses the autocorrelation function calculated by the calculation processing unit 113 to determine whether or not there is an abnormality in the time-series data subject to abnormality determination. good too.

Next, the data analysis unit 117 generates an abnormality determination result display screen based on the abnormality determination result indicating whether or not there is an abnormality in the time-series data subject to abnormality determination (step S18). The abnormality determination result display screen will be described later.

Next, the output unit 15 displays the abnormality determination result display screen generated by the data analysis unit 117 (step S19). The user can know whether or not electrical appliance 200 has an abnormality by checking the abnormality determination result display screen displayed on output unit 15 .

FIG. 6 is a flow chart showing the autocorrelation function calculation process in step S13 of FIG. The process leading to the derivation of the autocorrelation function calculation process will be described later.

First, the time difference set generation unit 114 determines whether or not the number of observation points of the time series data for abnormality determination extracted by the time series data extraction unit 112 is 3 or more (step S21). Here, if it is determined that the number of observation points of the time-series data of the electrical appliance 200 is not three or more, that is, if it is determined that the number of observation points of the time-series data of the electrical appliance 200 is two or less (step NO in S21), the autocorrelation function calculation process ends.

On the other hand, if it is determined that the number of observation points of the time series data of electrical appliance 200 is three or more (YES in step S21), time difference set generation unit 114 generates , the absolute values of the time differences between all observation points are calculated, and a first set, which is a set of the calculated time differences, is generated (step S22).

Next, the time difference set generation unit 114 generates a third set by removing redundant time differences from the first set (step S23).

Next, autocorrelation coefficient set generation section 115 generates autocorrelation coefficient set for all time differences included in the third set based on the time series data subject to abnormality determination and the third set generated by time difference set generation section 114 . A correlation coefficient is calculated, and a second set, which is a set of calculated autocorrelation coefficients, is generated (step S24). Note that the autocorrelation coefficient is calculated by Equation (4), which will be described later. Also, based on the third set, the autocorrelation coefficient set generator 115 calculates autocorrelation coefficients for all the time differences included in the third set.

Next, autocorrelation function calculation section 116 calculates the third set and the second set based on the third set generated by time difference set generation section 114 and the second set generated by autocorrelation coefficient set generation section 115. An autocorrelation function representing the relationship between the two sets is calculated (step S25). The autocorrelation function representing the relationship between the third set and the second set is calculated, for example, by an approximation method such as polynomial approximation or Gaussian approximation using the third set as an explanatory variable and the second set as an objective variable. At this time, the autocorrelation coefficient can only take values from -1 to 1, and the autocorrelation coefficient becomes 1 when the time difference is 0. Therefore, when calculating the autocorrelation function, these constraints should be added. That is, autocorrelation function calculation section 116 calculates the autocorrelation function so that the output value of the autocorrelation function is -1 or more and 1 or less based on the second set and the third set.

Moreover, in the present embodiment, after the process of step S23, the time difference set generation unit 114 may add 0 to the elements of the third set. In this case, autocorrelation function calculation section 116 may calculate an autocorrelation function representing the relationship between the third set and the second set based on the third set and the second set in which 0 is added to the elements. good. At this time, autocorrelation function calculation section 116 calculates autocorrelation function may be calculated.

Also, in the present embodiment, the third set is generated in step S23, but the present disclosure is not particularly limited to this, and the third set may not be generated. In this case, autocorrelation coefficient set generation section 115 may calculate autocorrelation coefficients corresponding to the first set based on the first set generated by time difference set generation section 114, and the calculated autocorrelation coefficient A second set, which is a set of relational coefficients, may be generated. Then, autocorrelation function calculation section 116 may calculate an autocorrelation function representing the relationship between the first set and the second set based on the first set and the second set.

Further, when calculating the autocorrelation function representing the relationship between the third set and the second set, the autocorrelation function calculation unit 116 uses weighted approximation to calculate the autocorrelation function may be calculated. Specifically, a time difference with an overlap number of 1 is given twice as much weight in approximation as a time difference with an overlap number of 0. Based on the first set and the third set, the time difference set generation unit 114 may generate a fourth set, which is a set of duplicate deletion numbers corresponding to sets of duplicate time differences deleted from the first set. Then, the autocorrelation function calculator 116 may calculate the autocorrelation function based on the second set, the third set, and the fourth set. At this time, the time difference set generator 114 may add 0 to the elements of the third set, or may not add 0 to the elements of the third set. Further, autocorrelation function calculation section 116 calculates the autocorrelation function based on the second set, the third set, and the fourth set so that the output value of the autocorrelation function is -1 or more and 1 or less. good too.

FIG. 7 is a diagram illustrating an example of a database configuration of a time-series data storage unit according to the embodiment of the present disclosure; The time-series data storage unit 121 includes an electrical appliance ID column indicating identification information for identifying electrical appliances, a time stamp column indicating time information, an indoor pipe temperature column indicating sensor value information, and an air volume column indicating sensor value information. have

Time information corresponding to the operation information of electrical appliance 200 is stored in the timestamp string. The information stored in the time stamp string may be any information indicating time or time, such as a character string or serial number indicating time. In the first row of the time stamp column of the database shown in FIG. 7, as an example, information "2017-12-22 15:27:11" indicating time information corresponding to the motion information is stored.

The electrical appliance ID column, the indoor pipe temperature column, and the air volume column store operation information of the electrical appliance 200 . When other operation information is acquired from electrical appliance 200 , time-series data storage unit 121 may increase columns for storing other operation information acquired from electrical appliance 200 . The column indicating the operation information acquired from electrical appliance 200 changes depending on the type of electrical appliance 200 . It is desirable to store the operation information acquired from the electrical appliance 200 in different time-series data storage units 121 for each type of electrical appliance 200 .

As an example, the electrical appliance ID column of the database shown in FIG. 7 stores information “XXX” indicating the electrical appliance ID of electrical appliance 200 . This information indicates that the information for identifying electrical appliance 200 is "XXX". The first row of the indoor pipe temperature column in the database shown in FIG. 7 stores information “7” indicating the value of the sensor that measures the indoor pipe temperature of the electrical appliance 200 . This information indicates that the indoor pipe temperature of electrical appliance 200 was 7 degrees. In addition, in the first row of the air volume column of the database shown in FIG. 7, information "0.5" indicating the value of the sensor that measures the air volume of the electrical appliance 200 is stored. This information indicates that the amount of blown air from electrical appliance 200 was 0.5 m ³ /h.

The process leading to the derivation of the autocorrelation function calculation process will be described below.

Suppose that N (N>2) samples are observed from t ₁ to t _N (t is time or information indicating time) at time intervals that are not always constant from stationary process time series data. Let this observed discrete time series data be X _t and its elements be _{{ x t1} , x _t2 , . At this time, the autocorrelation function R(h) of the discrete time-series data _Xt is defined by Equation (1) below. Note that the variable h in equation (1) indicates the time difference in calculating the autocorrelation coefficient.

Here, since the discrete time-series data _Xt is a sample obtained from the stationary process time-series data, the following equations (2) and (3) hold. Note that μ indicates the sample mean, and σ ² indicates the sample variance.

Equation (1) is expressed as Equation (4) shown below from Equations (2) and (3).

From equation (4), a desired autocorrelation coefficient can be obtained by substituting the value of the time difference for which the autocorrelation coefficient is to be calculated for the variable h. That is, the autocorrelation coefficient R(a) of the discrete time-series data _Xt at the time difference a is calculated by the following equation (5). However, for each element of the discrete time series data X _t , if there is no combination of k and l that satisfies x _tk = x _{tl + a} (k and l are natural numbers from 1 to N), the autocorrelation coefficient at the time difference a cannot be calculated.

Here, assuming that the acquisition time interval between the element x _ti and the element x _tj is represented by Δt _i,j (where i and j are natural numbers from 1 to N and i≠j), the time difference Δt _i,j is Since there always exists a combination of k and l that satisfies x _tk =x _{tl +Δti,j} , the autocorrelation coefficient can be calculated from Equation (4).

That is, irrespective of whether the sample acquisition time interval is constant or not, for two arbitrary combinations _N C of _two elements extracted from N samples, the autocorrelation in the time interval of the two elements The number can be calculated from equation (4).

Here, the set H of time differences for which autocorrelation coefficients can be calculated is H={Δt _i,j }. When the set H includes the time difference of the autocorrelation coefficient of the discrete time-series data _Xt to be calculated, the autocorrelation coefficient can be obtained from Equation (4). However, if the set H does not include the time difference of the autocorrelation coefficient of the discrete time series data X _t to be calculated, or if the set H includes the time difference of the autocorrelation coefficient of the discrete time series data X _t to be calculated Even if the number of elements of Δt _i,j corresponding to the time difference is not sufficient, the sample mean (expected value) of the product calculated by the numerator on the right side of Eq. descend. That is, when the number of elements of _{Δt i,j} corresponding to the time difference of the autocorrelation coefficient of the discrete time-series data Xt to be calculated is small, the accuracy of estimating the mother autocorrelation coefficient by the autocorrelation coefficient decreases. In particular, when the discrete time series data X _t is time series data with uneven intervals, the number of elements of Δt _i,j corresponding to the time difference of the autocorrelation coefficient of the discrete time series data X _t to be calculated is singular. is also assumed.

In order to solve the above problems, in the present disclosure, the obtained sample is set so that an arbitrary real number can be input to the variable h in Equation (4) and the population autocorrelation coefficient can be estimated with high accuracy. The autocorrelation function is reconstructed by approximation. The procedure will be described below.

Regarding the set H of time differences for which autocorrelation coefficients can be calculated, if duplicate values are included in the elements of the set H, the duplicate values are deleted. Assuming that a set obtained by removing duplicate values from the set H is H' and the number of removed elements is ε, the number of elements in the set H' is _N C ₂ -ε. When i=1, Δt _1,2 ≠Δt _1,3 ≠Δt _1,4 ≠…≠Δt _1,N . There is more than one element, ie _N C ₂ -ε≧N−1. Autocorrelation coefficients are calculated from equation (4) for all the elements of the set H', and let A be the set of autocorrelation coefficients calculated from all the elements of the set H'. Here, since the autocorrelation coefficient always indicates 1 when the time difference is 0, 0 is added to the elements of the set H' and 1 is added to the elements of the set A. FIG. Then, with the set H' as an explanatory variable and the set A as an objective variable, an autocorrelation function R'(h) representing the relationship between the set H' and the set A is calculated.

The approximation method used to calculate the autocorrelation function representing the relationship between the set H′ and the set A, with the set H′ as the explanatory variable and the set A as the objective variable, may be linear approximation, polynomial approximation, or Gaussian approximation. However, it is desirable to select the optimum approximation method according to the properties of the discrete time series data _Xt . Also, since the autocorrelation coefficient can only take values between -1 and 1, an approximation method that takes this constraint into account is effective. Regardless of which approximation method is selected, the number of variables used to calculate the autocorrelation function R'(h) is _N C ₂ -ε+1, and _N C ₂ -ε+1≧N. In other words, the accuracy of the approximation is improved compared to the number of variables used in the conventional method of interpolating the missing of the discrete time-series data _Xt by approximation.

A method of calculating an autocorrelation function representing the relationship between the set H' and the set A by weighting each element of the set H' in view of the number of overlapping elements of the set H is also effective.

Since any real number can be substituted for the variable h in the autocorrelation function R′(h) obtained according to the above procedure, the time difference of the autocorrelation coefficient of the discrete time series data X _t to be calculated is included in the set H. Even if not, the autocorrelation coefficient can be calculated from the autocorrelation function R'(h).

Also, even if the set H contains the time difference of the _{autocorrelation} coefficient of the discrete time series data X Since it is possible to obtain a value considering the autocorrelation coefficient of the time difference close to the time difference of the autocorrelation coefficient of _t , the accuracy of estimating the mother autocorrelation coefficient of the autocorrelation coefficient is improved.

In addition, since the autocorrelation function R′(h) is calculated by approximation using _N C ₂ −ε+1 variables, compared with conventional approximation using N observed variables, the autocorrelation function High accuracy of function estimation.

A specific process when electrical appliance 200 transmits operation information to data analysis device 100 will be described below.

At 15:27:11 on December 22, 2017, the electrical appliance 200 having the electrical appliance ID of "XXX" sent the operation information indicating that the room temperature piping temperature and the air volume were "7" and "0.5", respectively. Suppose that it is transmitted to the data analysis device 100 . As shown in FIG. 4, when the data analysis apparatus 100 receives motion information, it generates and stores time-series data including motion information and time information. In the present embodiment, the time at which information processing section 111 of data analysis device 100 generates time-series data is the same as the time at which electrical appliance 200 transmits the operation information.

At this time, the information processing unit 111 of the data analysis device 100 stores "XXX" in the electrical appliance ID column and "2017-12-22 15 2017-12-22 15 :27:11”, the indoor pipe temperature column indicates “7”, and the air volume column indicates “0.5”.

Specific processing performed by the time-series data extraction unit 112, the calculation processing unit 113, and the data analysis unit 117 in the data analysis device 100 when an abnormality determination request is acquired will be described below.

Regarding the electrical appliance 200 whose electrical appliance ID is "XXX", the indoor pipe temperature during the period from 15:28:30 on December 22, 2017 to 15:30:37 on December 22, 2017 is abnormal. Assume that the user requests determination of whether or not.

FIG. 8 is a diagram showing an example of a display screen displayed when an abnormality determination request is input in the data analysis device according to the embodiment of the present disclosure. The output unit 15 displays a display screen for accepting input of an abnormality determination request, and the input unit 14 accepts an input of an abnormality determination request by the user. As shown in FIG. 8, the user specifies, through the input unit 14, the electrical appliance ID of the electrical appliance 200 for which the abnormality determination is to be performed, the sensor for which the abnormality determination is to be performed, and the period during which the abnormality determination is to be performed. When the user inputs an abnormality determination request, the time-series data extraction unit 112 of the data analysis device 100 extracts time-series data corresponding to the specified electrical appliance ID, the sensor to be subjected to abnormality determination, and the period for performing the abnormality determination. Extract.

The time-series data extraction unit 112 stores "XXX" in the electrical appliance ID column from the database shown in FIG. The time stamp column and the indoor pipe temperature column in the row storing the time indicating the period up to 15:30:37 on December 22nd are extracted as time-series data for abnormality determination.

FIG. 9 is a diagram illustrating an example of time-series data for abnormality determination extracted from the time-series data storage unit by the time-series data extraction unit according to the embodiment of the present disclosure.

In the present embodiment, for an electrical appliance with an electrical appliance ID of "XXX", the indoor piping during the period from 15:28:30 on December 22, 2017 to 15:30:37 on December 22, 2017 A determination of whether the temperature is abnormal has been requested by the user. Therefore, as shown in FIG. 9, for the period from 15:28:30 on December 22, 2017 to 15:30:37 on December 22, 2017, the time stamp of the electrical appliance and the indoor pipe temperature are It is extracted as time-series data for abnormality determination used in the autocorrelation function calculation process.

When extraction of the time-series data ends, the calculation processing unit 113 executes autocorrelation function calculation processing. When the autocorrelation function calculation process is executed, the processes of steps S21 to S25 shown in FIG. 6 are executed in order.

First, in step S21, the time-difference set generating unit 114 determines whether the number of observation points of the time-series data for abnormality determination extracted by the time-series data extracting unit 112 is 3 or more. It is determined whether or not the number of rows of the time-series data extracted from the database of is three or more. Here, as shown in FIG. 9, the number of rows of the time-series data for abnormality determination extracted by the time-series data extraction unit 112 is 5, so the process proceeds to step S22.

Next, in step S22, the time difference set generation unit 114 calculates the absolute value of the time difference between all observation points based on the time series data for abnormality determination extracted by the time series data extraction unit 112. A first set is generated which is a set of the time differences. For example, the time difference set generation unit 114 generates data observed at 15:28:30 on December 22, 2017 and data observed at 15:29:25 on December 22, 2017. is calculated as "55 seconds". Here, in the time-series data for abnormality determination extracted from the database shown in FIG. Yes, 10 time differences are obtained. Specifically, the time differences between all observation points are "20 seconds", "35 seconds", "35 seconds", "37 seconds", "55 seconds", "57 seconds", "72 seconds", "92 seconds", "92 seconds" and "127 seconds" are calculated.

Next, in step S23, the time difference set generation unit 114 generates a third set by removing redundant time differences from the first set. The time difference set generation unit 114 generates the first sets of “20 seconds”, “35 seconds”, “35 seconds”, “37 seconds”, “55 seconds”, “57 seconds”, “72 seconds”, and “92 seconds”. , 92 seconds, and 127 seconds. As a result, finally, as the time difference between the observation points, "20 seconds", "35 seconds", "37 seconds", "55 seconds", "57 seconds", "72 seconds", "92 seconds" and " 127 seconds" is generated as the third set.

Next, in step S24, the autocorrelation coefficient set generation unit 115 extracts the time series data for abnormality determination extracted by the time series data extraction unit 112 and the time difference between the observation points generated by the time difference set generation unit 114. , the autocorrelation coefficients in the third set of time differences between observation points are calculated, and a second set, which is a set of the calculated autocorrelation coefficients, is generated. Here, the sample mean μ of the indoor pipe temperature of the time series data subject to abnormality determination is calculated as “3” using Equation (2), and the sample variance σ of the indoor pipe temperature of the time series data subject to abnormality determination ² is calculated as "3.2" using equation (3). For example, the autocorrelation coefficient of the time difference “35 seconds” between the observation points is obtained by the following formula (6) by substituting μ=3, σ ² =3.2 and h=35 into the formula (4). Calculated by

Here, when t is "December 22, 2017 15:28:30", t+35 is "December 22, 2017 15:29:05", and in the time-series data for abnormality determination , the observation point of “December 22, 2017 15:28:30” and the observation point of “December 22, 2017 15:29:05” both exist. Similarly, when t is "December 22, 2017 15:30:02", t+35 is "December 22, 2017 15:30:37", and in the time-series data for abnormality determination , the observation point of “December 22, 2017 15:30:02” and the observation point of “December 22, 2017 15:30:37” both exist. In this way, in the time-series data to be anomaly determination target, t such that there are both t and t+35 observation points is "December 22, 2017 15:28:30" and "December 22, 2017 15:30:02", and t+35 at this time is "December 22, 2017 15:29:05" and "December 22, 2017 15:30:37" respectively. . In the time-series data for abnormality determination, the observed value at the time "December 22, 2017 15:28:30" is "2", and at the time "December 22, 2017 15:29:05" The observed value is "4", the observed value at the time "December 22, 2017 15:30:02" is "6", and the time "December 22, 2017 15:30:37" The observed value is "1". Therefore, when the expected value calculation part in formula (6) is expanded, it is expressed by the following formula (7), and the autocorrelation coefficient of the time difference "35 seconds" between the observation points is calculated as "-0.781". be.

As the time difference between observation points, "20 seconds", "35 seconds", "37 seconds", "55 seconds", "57 seconds", "72 seconds", "92 seconds" and "127 seconds" are calculated. , and the autocorrelation coefficients for the time difference between each observation point are "-0.313", "-0.781", "-0.938", "0.313", "0.938", "0.625", "-0.781" and "0.625" are calculated.

FIG. 10 is a diagram illustrating an example of autocorrelation coefficients corresponding to time differences between observation points calculated by the autocorrelation coefficient set generation unit according to the embodiment of the present disclosure.

Autocorrelation function calculation section 116 calculates time differences based on the third set of time differences generated by time difference set generation section 114 and the second set of autocorrelation coefficients generated by autocorrelation coefficient set generation section 115. and the second set of autocorrelation coefficients. For example, "20 seconds", "35 seconds", "37 seconds", "55 seconds", "57 seconds", "72 seconds", "92 seconds" and "127 seconds" are calculated as the time difference which is an explanatory variable. The autocorrelation coefficients, which are objective variables corresponding to the respective time differences, are "-0.313", "-0.781", "-0.938", "0.313", "0. 938”, “0.625”, “−0.781” and “0.625” are calculated. Therefore, the autocorrelation function representing the relationship between the third set of time lags and the second set of autocorrelation coefficients uses the least-squares method and is constrained so that the autocorrelation coefficient indicates 1 when the time lag is 0. is expressed by the following fourth-order polynomial equation (8). Variable h in equation (8) indicates the time difference, and R(h) indicates the autocorrelation function.

R(h)=4.24*10 ⁻⁷ h ⁴ −1.08*10 ⁻⁴ h ³ +8.69*10 ⁻³ h ² −2.32*10 ⁻¹ h+1 (8)

After completing the autocorrelation function calculation process, the data analysis unit 117 executes the abnormality determination process.

The data analysis unit 117 determines abnormality of the time-series data for abnormality determination based on the autocorrelation function calculated in the autocorrelation function calculation process. For example, if the autocorrelation coefficient corresponding to the time difference of 40 seconds for the indoor pipe temperature of the electrical appliance 200 is 0.5 or less, and it is known that there is an abnormality in the time series data of the indoor pipe temperature, the data Based on the autocorrelation function calculated by the autocorrelation function calculation unit 116, the analysis unit 117 determines whether the autocorrelation coefficient corresponding to the time difference of 40 seconds between the indoor pipe temperatures of the electrical appliance 200 is 0.5 or less. judge.

Here, when the data analysis unit 117 determines that the autocorrelation coefficient corresponding to the time difference of 40 seconds is 0.5 or less, it determines that the time-series data of the indoor pipe temperature is abnormal. Further, when the data analysis unit 117 determines that the autocorrelation coefficient corresponding to the time difference of 40 seconds is greater than 0.5, there is no abnormality in the time-series data of the indoor pipe temperature, that is, the time-series data of the indoor pipe temperature Determined to be normal. The autocorrelation coefficient corresponding to the time difference of 40 seconds between the indoor pipe temperatures of the electrical appliance 200 is calculated as "-0.203" by substituting h=40 into the equation (8), and the autocorrelation coefficient is 0.203. Since it is 5 or less, the data analysis unit 117 determines that there is an abnormality in the time-series data of the indoor pipe temperature.

Data analysis section 117 calculates an autocorrelation coefficient corresponding to the predetermined time difference by substituting the predetermined time difference into the autocorrelation function calculated by autocorrelation function calculation section 116, and the calculated autocorrelation coefficient is It is determined whether or not it is equal to or less than a predetermined threshold. In the present embodiment, in the abnormality determination of the time-series data of the indoor pipe temperature, if the autocorrelation coefficient when the predetermined time difference is 40 seconds is 0.5 or less, which is a predetermined threshold, the indoor pipe temperature It is known in advance that there is an anomaly in the time-series data of . Therefore, the data analysis unit 117 substitutes the predetermined time difference of 40 seconds into the autocorrelation function calculated by the autocorrelation function calculation unit 116, thereby calculating the autocorrelation coefficient corresponding to the predetermined time difference. It is determined whether or not the obtained autocorrelation coefficient is equal to or less than 0.5, which is a predetermined threshold. Note that the predetermined time difference and the predetermined threshold value described above are examples, and are not limited to the above values. Moreover, it is preferable that the predetermined time difference and the predetermined threshold value are set according to the abnormality determination target. The memory 12 stores in advance a predetermined time difference and a predetermined threshold in association with an abnormality determination target. The data analysis unit 117 uses a predetermined time difference and a predetermined threshold stored in the memory 12 to determine whether or not there is an abnormality in the time-series data subject to abnormality determination.

The output unit 15 displays the result of abnormality determination by the data analysis unit 117 . The data analysis unit 117 generates an abnormality determination result display screen based on the abnormality determination result indicating whether or not there is an abnormality in the time-series data subject to abnormality determination, and outputs the generated abnormality determination result display screen to the output unit 15. Output.

11 is a diagram illustrating an example of an abnormality determination result display screen generated by the data analysis unit according to the embodiment of the present disclosure; FIG.

On the abnormality determination result display screen shown in FIG. An anomaly judgment result indicating that there is an anomaly in the time-series data is displayed. The data analysis unit 117 generates an abnormality determination result display screen according to an abnormality determination result indicating whether or not there is an abnormality in the time-series data to be subjected to abnormality determination. Then, the output unit 15 displays the abnormality determination result display screen generated by the data analysis unit 117 .

As described above, using the second information included in each of the N (N is a natural number of 3 or more) third information included in the time-series data, the absolute value of the time difference between each observation point is calculated and calculated. A set of time differences is generated, and an autocorrelation coefficient representing the correlation between the values of the first information is calculated for each set of third information for which the absolute value of the time difference is calculated, and the calculated autocorrelation A set of numbers is generated and an autocorrelation function is calculated representing the relationship between the set of time differences and the set of autocorrelation coefficients.

Therefore, while the number of variables used in the conventional method of interpolating the loss of discrete time series data by approximation is N, the number of variables used to calculate the autocorrelation function of the present disclosure is N Since the number of combinations of any two pieces of second information extracted from the third information is _N C ₂ , even if the time-series data is discrete time-series data that is not acquired at equal intervals, acquisition The autocorrelation function can be calculated with high accuracy from the discrete time-series data using more data than the number of data calculated. In addition, the autocorrelation function calculated with high accuracy can be used to accurately perform data analysis processing for electrical appliances.

Note that, in the present embodiment, data analysis section 117 uses the autocorrelation function calculated in the autocorrelation function calculation process to perform abnormality determination processing for determining whether or not there is an abnormality in the time-series data to be subjected to abnormality determination. However, the present disclosure is not particularly limited to this, and the autocorrelation function calculated by the autocorrelation function calculation process may be used for any data analysis such as drawing a correlogram, future prediction, or clustering. . When the autocorrelation function calculated in the autocorrelation function calculation process is used for any data analysis, the abnormality determination request in the present embodiment may be replaced with another request, and the abnormality determination process and the abnormality determination result display The processing may be replaced with another processing depending on the intended use of the autocorrelation function.

A correlogram is a graph of an autocorrelation function in which the horizontal axis is the time difference and the vertical axis is the autocorrelation coefficient. When the autocorrelation function calculated from the time-series data is used for drawing a correlogram, for example, the input unit 14 draws a correlogram including a correlogram drawing period, a sensor for correlogram drawing, a correlogram bin width, and the like. It accepts request input from the user. Using the autocorrelation function calculated in the autocorrelation function calculation process, the data analysis unit 117 generates a correlogram showing the autocorrelation coefficient for each time difference separated by the designated bin width, and generates the generated correlogram Generates a display screen that draws Then, the output unit 15 displays a display screen on which the correlogram is drawn.

Further, when the autocorrelation function calculated from the time-series data is used for future prediction, for example, the input unit 14 inputs the past period used for future prediction, the future prediction target sensor, and the future period output for future prediction. , etc., from the user. Using the autocorrelation function calculated in the autocorrelation function calculation process, the data analysis unit 117 predicts changes in sensor values for a specified future period, and generates a prediction result display screen showing the prediction result. Then, the output unit 15 displays a prediction result display screen showing the prediction result.

Furthermore, when the autocorrelation function calculated from the time-series data is used for clustering, for example, the input unit 14 receives an input of a clustering request including a clustering target period and clustering target sensors from the user. The data analysis unit 117 uses the autocorrelation function calculated in the autocorrelation function calculation process to identify the cluster name to which the time series data of the clustering target sensor in the specified clustering target period belongs, and expresses the identified cluster name. Generate a clustering result display screen. Then, the output unit 15 displays a clustering result display screen showing cluster names to which the time-series data of the clustering target sensors belong.

Further, the data analysis apparatus 100 according to the present embodiment can be used for abnormality determination or future prediction of home appliances, factory equipment, vehicle-mounted equipment, or the like. Further, the data analysis device 100 according to the present embodiment can be used for stock price prediction based on market information. Furthermore, the data analysis device 100 according to the present embodiment can be used for diagnosing or predicting a health condition based on biological information. In other words, the data analysis device 100 can be used for diagnosing or predicting various states correlated with the log information based on various chronological log information.

In each of the above-described embodiments, each component may be configured by dedicated hardware, or realized by executing a software program suitable for each component. Each component may be realized by reading and executing a software program recorded in a recording medium such as a hard disk or a semiconductor memory by a program execution unit such as a CPU or processor.

Some or all of the functions of the device according to the embodiment of the present disclosure are typically implemented as an LSI (Large Scale Integration), which is an integrated circuit. These may be made into one chip individually, or may be made into one chip so as to include part or all of them. Further, circuit integration is not limited to LSIs, and may be realized by dedicated circuits or general-purpose processors. An FPGA (Field Programmable Gate Array) that can be programmed after the LSI is manufactured, or a reconfigurable processor that can reconfigure connections and settings of circuit cells inside the LSI may be used.

Also, some or all of the functions of the device according to the embodiment of the present disclosure may be implemented by a processor such as a CPU executing a program.

In addition, the numbers used above are all examples for specifically describing the present disclosure, and the present disclosure is not limited to the illustrated numbers.

In addition, the order in which each step shown in the above flowchart is executed is for illustrative purposes in order to specifically describe the present disclosure, and may be an order other than the above as long as the same effect can be obtained. . Also, some of the above steps may be executed concurrently (in parallel) with other steps.

The data analysis method, data analysis device, and data analysis program according to the present disclosure are discrete time series data that is not acquired at equal intervals, and are discrete time series using more data than the number of acquired data. Since the autocorrelation function can be calculated from series data with high accuracy, it is useful for data analysis methods, data analysis devices, and data analysis programs for performing data analysis processing of electric appliances.

1 information processing system 11 processor 12 memory 13 communication unit 14 input unit 15 output unit 21 operation unit 22 processor 23 memory 24 sensor 25 communication unit 100 data analysis device 111 information processing unit 112 time series data extraction unit 113 calculation processing unit 114 time difference set Generation unit 115 Autocorrelation coefficient set generation unit 116 Autocorrelation function calculation unit 117 Data analysis unit 121 Time-series data storage unit 200 Electrical appliance 221 Device control unit 222 Operation information acquisition unit 223 Communication control unit 231 Operation information storage unit 300 Network

Claims (13)

A data analysis method in a data analysis device that performs data analysis processing for electrical appliances,
A time series including M or more (M is a natural number of 3 or more) third information in which first information about the operation of the electrical appliance is associated with second information representing temporal observation points of the first information. get data information,
Using the second information included in each of the N pieces of the third information (N is a natural number of 3 or more) included in the time series data information, the absolute value of the time difference between each of the observation points is calculated and calculated generating the set of time differences as a first set;
The value of the first information at a first time and the value of the first information at a second time after a predetermined time has passed from the first time for each set of the third information for which the absolute value of the time difference is calculated. Calculating an autocorrelation coefficient representing the correlation with, generating a set of the calculated autocorrelation coefficients as a second set,
calculating an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients based on the first set and the second set;
Based on the autocorrelation function, chronologically analyze the first information,
outputting fourth information about the results of the analysis;
Data analysis method. Further, in the first set, generating a third set by removing the overlapping time difference,
generating the second set based on the third set, generating a set of the autocorrelation coefficients corresponding to the third set as the second set;
The calculation of the autocorrelation function calculates the autocorrelation function based on the third set and the second set.
The data analysis method according to claim 1. Further, based on the first set and the third set, generating a fourth set that is a set of duplicate deletion numbers corresponding to the set of the duplicate time differences deleted from the first set,
Calculation of the autocorrelation function includes calculating the autocorrelation function based on the second set, the third set, and the fourth set.
The data analysis method according to claim 2. generating the third set adding 0 to the elements of the third set;
The data analysis method according to claim 3. The calculation of the autocorrelation function is based on the second set, the third set, and the fourth set, and the autocorrelation function is calculated so that the output value of the autocorrelation function is -1 or more and 1 or less. calculate,
The data analysis method according to claim 4. generating the third set adding 0 to the elements of the third set;
The data analysis method according to claim 2. The calculation of the autocorrelation function is based on the second set and the third set, and calculates the autocorrelation function so that the output value of the autocorrelation function is -1 or more and 1 or less.
The data analysis method according to claim 6. The calculation of the autocorrelation function is based on the second set and the third set, and calculates the autocorrelation function so that the output value of the autocorrelation function is -1 or more and 1 or less.
The data analysis method according to claim 2. Further, receiving instruction information instructing the data analysis device to perform the data analysis process,
The generation of the first set uses the second information included in each of the N pieces of the third information (N is a natural number of 3 or more) included in the time-series data information after the instruction information is received. calculating the absolute value of the time difference at each of the observation points, and generating a set of the calculated time differences as a first set;
The data analysis method according to any one of claims 1-8. the instruction information is input by a user at a terminal connected to the data analysis device via a network;
The data analysis method according to claim 9. The analysis of the first information determines whether the operation of the electrical appliance is normal,
The fourth information includes information regarding whether or not the electrical appliance operates normally.
The data analysis method according to any one of claims 1-10. A data analysis device for performing data analysis processing of electrical appliances,
A time series including M or more (M is a natural number of 3 or more) third information in which first information about the operation of the electrical appliance is associated with second information representing temporal observation points of the first information. an acquisition unit that acquires data information;
A control unit that performs data analysis processing on the time-series data information,
The control unit
Using the second information included in each of the N pieces of the third information (N is a natural number of 3 or more) included in the time series data information, the absolute value of the time difference between each of the observation points is calculated and calculated generating the set of time differences as a first set;
The value of the first information at a first time and the value of the first information at a second time after a predetermined time has passed from the first time for each set of the third information for which the absolute value of the time difference is calculated. Calculating an autocorrelation coefficient representing the correlation with, generating a set of the calculated autocorrelation coefficients as a second set,
calculating an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients based on the first set and the second set;
Based on the autocorrelation function, chronologically analyze the first information,
outputting fourth information about the results of the analysis;
Data analysis equipment. A data analysis program for performing data analysis processing of electrical appliances,
to the computer,
A time series including M or more (M is a natural number of 3 or more) third information in which first information about the operation of the electrical appliance is associated with second information representing temporal observation points of the first information. get data information,
Using the second information included in each of the N pieces of the third information (N is a natural number of 3 or more) included in the time series data information, the absolute value of the time difference between each of the observation points is calculated and calculated generating the set of time differences as a first set;
The value of the first information at a first time and the value of the first information at a second time after a predetermined time has passed from the first time for each set of the third information for which the absolute value of the time difference is calculated. Calculating an autocorrelation coefficient representing the correlation with, generating a set of the calculated autocorrelation coefficients as a second set,
calculating an autocorrelation function representing the relationship between the set of time differences and the set of autocorrelation coefficients based on the first set and the second set;
Based on the autocorrelation function, chronologically analyze the first information,
outputting fourth information about the results of the analysis;
A data analysis program that causes processing.

Images