JP5461224B2 - Interpolation apparatus, interpolation method and program - Google Patents

Description

  The present invention relates to an interpolation device, an interpolation method, and a program.

In a ubiquitous environment, various device groups such as sensors and actuators and a computer group transmit and receive data to and from each other via a network. In the ubiquitous environment, it is considered that the position of the device group and the activation state frequently change, and in order to reliably transmit and receive sensor data, operation through a shared sensor data storage device is effective ( Patent Document 1).
When using the shared sensor data storage device, for example, the device group writes sensor data including a measurement time and a measurement position in the shared sensor data storage device. The computer group can acquire sensor data necessary for an application by designating sensor data conditions in the range of measurement time and measurement position and reading the sensor data from the shared sensor data storage device. Further, by using the shared sensor data storage device, a plurality of computers can share sensor data and execute different applications.

By sharing the general-purpose sensor data among a plurality of applications using the shared sensor data storage device, it is not necessary to provide a dedicated sensor for each application and perform measurement, and the application can be executed more easily. On the other hand, since the general-purpose sensor data is used, the sensor data required by the application is not always stored in the shared sensor data storage device. For example, when an application requests data with an acquisition cycle shorter than the acquisition cycle of data stored in the shared sensor data storage device, sensor data other than the stored time is missing (the state in which no sensor data exists) )
As with normal applications, it is considered that applications in the ubiquitous environment are usually designed on the premise of acquiring sensor data without any defects. Therefore, the sensor data including the defect hinders the execution of the application. Therefore, it is conceivable to interpolate the missing data by applying the interpolation technique, which has conventionally been studied mainly to the data loss due to the failure of the network or the device, to the data loss in the ubiquitous environment.

  The loss interpolation technology for loss due to network or device failure includes spatial interpolation technology that uses sensor data (hereinafter referred to as “spatial data”) acquired at the same time as the missing data, or the same position or There has been a time-series interpolation technique using past or future sensor data (hereinafter referred to as “time-series data”) transmitted from the same sensor. As a method for performing spatial interpolation, for example, a method using a decision tree model, which is a pattern learning model, using a pattern of spatial data as teacher data (see Non-Patent Document 1), or calculating the degree of correlation between time series data In addition, there is a k-NN method (see Non-Patent Document 2) in which interpolation is performed using time-series data having a high degree of correlation. In addition, as a method for performing time series interpolation, for example, there is a method using an autoregressive model that calculates time series data at a certain time point from continuous past time series data.

JP 2009-93213 A

Kinaki Satoshi, "Data Science by R", Morikita Publishing, 2007, p. 230-233 Kazuya Murao, 3 others, “Sensing data interpolation method considering wearable sensor failure for wearable system”, IEICE Data Engineering Workshop (DEW2007) Proceedings, 2007 Kinaki Satoshi, "Data Science by R", Morikita Publishing, 2007, p. 211-213

However, while defects due to network or device failures are sporadic, the above-described defects when using a shared sensor data storage device in a ubiquitous environment are periodic. Furthermore, since defects occur periodically, the defect rate is higher than defects caused by network or device failures. For these reasons, the following problems arise when trying to interpolate the above-described deficiency when using a shared sensor data storage device in a ubiquitous environment using a conventional interpolation method.
First, in the interpolation method using a decision tree model, if past spatial data is missing, effective teacher data cannot be obtained and interpolation cannot be performed. When the shared sensor data storage device is used in the ubiquitous environment, the past spatial data may be periodically lost due to the difference between the data acquisition cycle by the sensor and the data acquisition cycle required by the application as described above. In this case, the interpolation method using the decision tree model cannot be applied as it is.

Further, in the interpolation method based on the k-NN method, when the time series data for calculating the correlation degree is missing, the correlation degree cannot be calculated and interpolation cannot be performed. When a shared sensor data storage device is used in a ubiquitous environment, time-series data may be periodically lost due to a difference between a data acquisition cycle by a sensor and a data acquisition cycle required by an application. In this case, the interpolation method by the k-NN method cannot be applied as it is.
In addition, in the interpolation method using the autoregressive model, interpolation is not possible if there is even one missing piece in the past continuous time series data. When a shared sensor data storage device is used in a ubiquitous environment, time-series data may be periodically lost due to a difference between a data acquisition cycle by a sensor and a data acquisition cycle required by an application. In this case, the interpolation method using the autoregressive model cannot be applied as it is.

  In addition, when these interpolation methods cannot be applied, as a method for performing spatial interpolation or time series interpolation, a method using average interpolation, linear interpolation, or the like of spatial data or time series data can be considered. However, in these interpolation methods, when the data loss rate is high, the interpolation accuracy is lowered. When using a shared sensor data storage device in a ubiquitous environment, the data loss rate can be high due to the difference between the data acquisition cycle by the sensor and the data acquisition cycle required by the application. In this case, the accuracy of interpolation is low in a method that uses average interpolation or linear interpolation of spatial data or time-series data.

  The present invention has been made in consideration of such circumstances, and the purpose of the present invention is to calculate the data acquisition cycle required by the application and the data acquisition cycle required by the application when reading sensor data from the shared sensor data storage device. An object of the present invention is to provide an interpolation device, an interpolation method, and a program capable of performing more accurate interpolation for data loss caused by a difference.

  [1] The present invention has been made to solve the above-described problems. An interpolation device according to an aspect of the present invention includes an actual measurement value obtained by a sensor, information indicating a measurement position of the actual measurement value, and the actual measurement value. The actual sensor data acquisition unit that acquires actual sensor data including information indicating the measurement time, and the acquisition request data that indicates the sensor and the measurement time and requests the measurement value of the sensor at the measurement time An actual sensor data matrix is generated for each sensor including all the sensors indicated by the request data and for each measurement time including all the measurement times indicated by the acquisition request data, and the measurement position of the actual measurement value is indicated. Based on the information and the information indicating the measurement time of the actual measurement value, a time series alignment unit that assigns the actual measurement value to an element of the actual sensor data matrix, and the element of the actual sensor data matrix Time series interpolation for interpolating the measured values of the elements that do not have the actual measurement value based on the actual measurement values measured by the same sensor as the sensor corresponding to the element that does not have the actual measurement value. Among the elements of the functional unit and the actual sensor data matrix, for the element that does not have the actual measurement value, the actual measurement value by a sensor different from the sensor corresponding to the element that does not have the actual measurement value, or the element that does not have the actual measurement value The acquisition request data requests by updating the measurement value in the element that does not have the actual measurement value using the measurement value interpolated by the time series interpolation function unit for the element corresponding to a sensor different from the sensor corresponding to A spatial interpolation function unit that generates a measurement value.

  [2] An interpolation device according to an aspect of the present invention is the above-described interpolation device, an acquisition request reception function unit that receives the acquisition request data, and a response function unit that transmits response data for the acquisition request data The time series aligning unit corresponds to all of the sensors selected by the information for selecting the sensors and all of the measurement times indicated by the measurement time range and the measurement time interval. The real sensor data matrix to be generated is generated, and the response function unit transmits the response data including the measurement values updated by the spatial interpolation function unit.

  [3] Further, an interpolation device according to an aspect of the present invention is the above-described interpolation device, wherein the spatial interpolation function unit is an actual measurement value obtained by a sensor different from a sensor corresponding to an element not having the actual measurement value, or A decision tree that determines a measurement value in an element that does not have the actual measurement value according to a conditional expression regarding the measurement value that is interpolated by the time-series interpolation function unit for an element that corresponds to a sensor different from the sensor that corresponds to the element that does not have the actual measurement value And a decision tree application unit that determines a measurement value of an element that does not have the actual measurement value using the decision tree.

  [4] An interpolation apparatus according to an aspect of the present invention is the above-described interpolation apparatus, wherein the spatial interpolation function unit includes a sensor corresponding to an element not having the actual measurement value and an element not having the actual measurement value. A distance calculation function unit that determines a distance between a sensor corresponding to the sensor and a different sensor, and a sensor is selected based on the distance, and a measured value in an element that does not have the actual measured value is calculated based on the measured value of the selected sensor. And a k-NN method application unit to be determined.

  [5] Further, an interpolation method according to an aspect of the present invention is an interpolation method of an interpolating apparatus that interpolates measurement values of a sensor, and includes an actual measurement value by a sensor, information indicating a measurement position of the actual measurement value, and the actual measurement. Based on the acquisition request data for acquiring the actual sensor data including the information indicating the measurement time of the value, and the acquisition request data for requesting the measurement value of the sensor at the measurement time by indicating the sensor and the measurement time. A real sensor data matrix is generated for each sensor including all the sensors indicated by the acquisition request data and each measurement time including all the measurement times indicated by the acquisition request data, and the measurement position of the actual measurement value A time series alignment step of assigning the actual measurement values to elements of the actual sensor data matrix based on the information indicating the measurement time and the information indicating the measurement time of the actual measurement values; Among the elements of the matrix, for the elements that do not have the actual measurement values, the measurement values for the elements that do not have the actual measurement values are calculated based on the actual measurement values measured by the same sensor as the sensor corresponding to the element that does not have the actual measurement values. Of the elements of the actual sensor data matrix, the time series interpolation step to interpolate and the element that does not have the actual measurement value, the actual measurement value by a sensor different from the sensor corresponding to the element that does not have the actual measurement value, or the actual measurement value Using the measurement values interpolated in the time-series interpolation step for the elements corresponding to the sensors different from the elements corresponding to the elements not having the measurement value, and updating the measurement values in the elements not having the actual measurement value, And a spatial interpolation step for generating a measurement value required by the data.

  [6] An interpolation method according to an aspect of the present invention is the above-described interpolation method, and includes information for selecting the sensor, information indicating a measurement time range, and information indicating a measurement time interval. An acquisition request reception step for receiving acquisition request data for requesting a measurement value of the sensor; and a response step for transmitting response data to the acquisition request data. The time series alignment step includes: The actual sensor data matrix corresponding to all of the sensors selected by the information to be selected and all of the measurement times indicated by the measurement time range and the measurement time interval is generated and updated in the spatial interpolation step The answer data including the measured value is transmitted in the answering step.

  [7] Further, an interpolation method according to an aspect of the present invention is the above-described interpolation method, wherein the spatial interpolation step includes an actual measurement value obtained by a sensor different from a sensor corresponding to an element not having the actual measurement value, or A decision tree for determining a measurement value in an element not having the actual measurement value according to a conditional expression relating to the measurement value interpolated in the time series interpolation step for an element corresponding to a sensor different from the sensor corresponding to the element not having the actual measurement value A decision tree generating step for generating; and a decision tree applying step for determining a measurement value in an element that does not have the actual measurement value using the decision tree.

  [8] Further, an interpolation method according to an aspect of the present invention is the above-described interpolation method, wherein the spatial interpolation step applies a sensor corresponding to an element not having the actual measurement value and an element not having the actual measurement value. A distance calculation step for determining a distance between a corresponding sensor and a different sensor; a sensor is selected based on the distance; and a measured value for an element that does not have the actual measured value is determined based on a measured value of the selected sensor. and a k-NN method application step.

  [9] A program according to an embodiment of the present invention is a program that causes a computer to execute one of the interpolation methods described above.

  According to this invention, when retrieving sensor data from the shared sensor data storage device, more accurate interpolation is performed for data loss caused by the difference between the data acquisition cycle by the sensor and the data acquisition cycle required by the application. Yes.

It is a block diagram which shows schematic structure of the interpolation apparatus in the 1st Embodiment of this invention. It is a figure which shows the example of real sensor data in the same embodiment. It is a data block diagram which shows the example of the acquisition request data in the same embodiment. It is a figure which shows the example of the search request data in the same embodiment. It is a data block diagram which shows the example of the real sensor data matrix in the same embodiment. It is a figure which shows the example of the time series interpolation matrix in the embodiment. It is a figure which shows the example of the decision tree of the unit area | region A in the embodiment. It is a figure which shows the example of the decision tree data which the decision tree storage part 111 memorize | stores in the same embodiment. It is a figure which shows the example of the spatial interpolation matrix in the same embodiment. It is a figure which shows the example of the reply data in the same embodiment. In the embodiment, the time series interpolation function unit 105 is a flowchart showing a processing procedure for time series interpolation of an actual sensor data matrix. 4 is a flowchart illustrating a processing procedure in which the first-stage interpolation function unit 106 performs time-series interpolation in the embodiment. 5 is a flowchart illustrating a processing procedure in which the spatial interpolation function unit 108 performs spatial interpolation in the embodiment. 5 is a flowchart illustrating a processing procedure for the decision tree generation unit 110 to generate a decision tree in the embodiment. 5 is a flowchart illustrating a processing procedure in which the decision tree application unit 112 performs spatial interpolation on a point of interest in the embodiment. FIG. 11 is a sequence diagram illustrating a processing procedure in which the interpolation device 100 generates and transmits (answers) response data according to acquisition request data transmitted from the ubiquitous application device 116 in the embodiment. FIG. 11 is a sequence diagram illustrating a processing procedure in which the interpolation device 100 generates and transmits (answers) response data according to acquisition request data transmitted from the ubiquitous application device 116 in the embodiment. It is a block diagram which shows schematic structure of the interpolation apparatus in the 2nd Embodiment of this invention. It is a figure which shows the example of the distance data matrix in the same embodiment. It is a figure which shows the example of the spatial interpolation matrix which the spatial interpolation matrix memory | storage part 113 memorize | stores in the same embodiment. 5 is a flowchart showing a processing procedure in which the spatial interpolation function unit 508 performs spatial interpolation in the embodiment. In the embodiment, a distance calculation function unit 510 is a flowchart showing a processing procedure for generating a distance data matrix. 5 is a flowchart illustrating a processing procedure in which the k-NN method application unit 512 performs spatial interpolation of a point of interest in the embodiment.

<First Embodiment>
Embodiments of the present invention will be described below with reference to the drawings.
FIG. 1 is a configuration diagram showing a schematic configuration of an interpolation apparatus according to the first embodiment of the present invention.
In the figure, an interpolation device 100 includes an acquisition request reception function unit 101, a shared sensor data storage device reading function unit 102, a time series alignment unit 103, an actual sensor data matrix storage unit 104, and a time series interpolation function unit 105. A time series interpolation matrix storage unit 107, a spatial interpolation function unit 108, a spatial interpolation matrix storage unit 113, and an answer function unit 114. The time series interpolation function unit 105 includes a first stage interpolation function unit 106. The spatial interpolation function unit 108 includes a second-stage interpolation function unit 109. The second-stage interpolation function unit 109 includes a decision tree generation unit 110, a decision tree storage unit 111, and a decision tree application unit 112.
The interpolation device 100 is connected to the shared sensor data storage device 115 and the ubiquitous application device 116.

The interpolation device 100 reads the sensor data from the shared sensor data storage device 115 according to the acquisition request data indicating the request for the sensor data from the ubiquitous application device 116, interpolates the missing data, and transfers the data to the ubiquitous application device 116. Send.
The acquisition request reception function unit 101 is connected to the ubiquitous application device 116, and when the ubiquitous application device 116 receives the acquisition request data requesting the sensor data, outputs the received acquisition request data to the time series sorting unit 103.
The shared sensor data storage device reading function unit 102 reads sensor data from the shared sensor data storage device 115.

The time series arranging unit 103 arranges the sensor data read from the shared sensor data storage device 115 by the shared sensor data storage device reading function unit 102 in time series, and generates an actual sensor data matrix.
The actual sensor data matrix storage unit 104 stores the actual sensor data matrix generated by the time series alignment unit 103.
The time series interpolation function unit 105 performs time series interpolation on the actual sensor data matrix to generate a time series interpolation matrix. The first stage interpolation function unit 106 performs time series interpolation on each element of the actual sensor data matrix.
The time series interpolation matrix storage unit 107 stores the time series interpolation matrix generated by the time series interpolation function unit 105.

The spatial interpolation function unit 108 performs spatial interpolation on the time series interpolation matrix to generate a spatial interpolation matrix. The second stage interpolation function unit 109 performs spatial interpolation on each element of the time series interpolation matrix. The decision tree generation unit 110 generates a decision tree used for spatial interpolation. The decision tree storage unit 111 stores the decision tree generated by the decision tree generation unit 110. The decision tree application unit 112 performs spatial interpolation on each element of the time series interpolation matrix using the decision tree.
The spatial interpolation matrix storage unit 113 stores the spatial interpolation matrix generated by the spatial interpolation function unit 108.
The response function unit 114 reads out sensor data requested by the ubiquitous application device 116 from the spatial interpolation matrix, generates response data that is a response to the acquired data, and transmits the generated response data to the ubiquitous application device 116.

The shared sensor data storage device 115 stores actual sensor data that is data measured by a sensor (not shown). The shared sensor data storage device 115 can be realized by using, for example, a relational database management system (RDBMS).
The ubiquitous application device 116 executes the application and transmits acquisition request data for requesting sensor data necessary for executing the application to the interpolation device 100.

Next, the data configuration of actual sensor data that is individual sensor data stored in the shared sensor data storage device reading function unit 102 will be described.
FIG. 2 is a diagram illustrating an example of actual sensor data.
As shown in the figure, the actual sensor data includes a line of actual measurement values and a line of metadata, and each line is composed of a combination of an attribute, a delimiter “=”, and a value. The actual sensor data includes metadata (Meta Data) indicating the measurement position of the actual measurement value and metadata indicating the measurement time of the actual measurement value. The actual sensor data includes metadata other than the metadata indicating the measurement position of the actual measurement value and the metadata indicating the measurement time of the actual measurement value, such as metadata indicating the sensor type or metadata indicating the sensor address. You may make it.
In the example of FIG. 2, the actual sensor data D120 includes an actual measurement value (data having the attribute “sensor value”), metadata indicating the measurement position of the actual measurement value (metadata having the attribute “position”), It includes metadata indicating the measurement time (metadata having the attribute “time”) and metadata indicating the sensor type (metadata having the attribute “sensor type”).

Next, the data structure of the acquisition request data that indicates the sensor data conditions that the ubiquitous application apparatus 116 requests from the interpolation apparatus 100 will be described.
FIG. 3 is a data configuration diagram illustrating an example of acquisition request data.
In the example shown in the figure, the acquisition request data D200 has a sensor position range of N35.00.00.000 (north latitude 35.00.00.000 degrees as a condition of sensor data requested by the ubiquitous application device 116. The same applies hereinafter. ) And E139.00.00.000 (East longitude 139.00.00.000 degrees; the same applies hereinafter) to N35.00.10.000 and E139.00.10.000, and the range of sensor data measurement time is 2006 / 03/01 00:00:00 (March 1, 2006 0: 0: 0. The same applies below) to 2006/3/1 03:00:00, and sensor data acquisition time interval Is 01:00:00 (one hour), and a condition C14 in which the sensor type is a temperature sensor is included.

Note that the acquisition request data handled by the interpolation apparatus 100 is not limited to that shown in FIG. 3, and may be any data that indicates the sensor and the measurement time and requests the measurement value of the sensor at the measurement time.
In particular, the acquisition request data D200 of FIG. 3 shows the range of the position of the sensor as a condition for selecting the sensor, but instead of this, the range of the sensor ID (Identifier, identification information) may be used as the condition. Alternatively, the sensor condition may be indicated using a network address range of the sensor. Or you may use the conditions which consist of these combination. Further, the method for indicating the measurement time is not limited to the method for indicating the time range and the period in FIG. 3. For example, all the measurement times may be listed.

Next, the data configuration of search request data, which is data indicating the conditions of actual sensor data read by the interpolation device 100 from the shared sensor data storage device 115, will be described.
FIG. 4 is a diagram illustrating an example of search request data.
In the example shown in the figure, the search request data D210 has a sensor position (sensor data measurement position) range of N34.50.00.000 and E138.50.00.000 to N35. Condition C21 for 10.00.000 and E139.10.00.000 and Condition C22 for the sensor data measurement time range from 2006/02/01 00:00:00 to 2006/03/01 03:00:00 Including.
As will be described later, the interpolation device 100 generates search request data indicating a wider range than the received acquisition request data. For example, the interpolation apparatus 100 sets a range including several times the sensor or a range of several times the sensor position indicated by the acquisition request data as the sensor position indicated by the search request data.
Also regarding the time direction, the interpolation apparatus 100 generates search request data indicating a time range wider than the acquisition request data. For example, if the sensor data requested by the acquisition request data changes daily, search request data for selecting sensor data for several days, or if it changes monthly, is generated.

The sensor type of the search request data is not limited to the sensor type of the acquisition request data. For example, when the acquisition request data requests temperature information, the interpolation apparatus 100 may generate search request data that requests humidity information in addition to the temperature information.
The search conditions for the search request data are not limited to those shown in FIG. 4, but may include a condition indicating the range of sensor data measurement time and a condition for selecting a sensor. For example, instead of the sensor data measurement position range shown in FIG. 4, the condition for selecting the sensor may include a sensor ID range, a sensor address, or a combination of these conditions. Further, other information such as the type of sensor may be included.

Next, the data configuration of the actual sensor data matrix generated by the time series sorting unit 103 and stored by the actual sensor data matrix storage unit 104 will be described.
FIG. 5 is a data configuration diagram illustrating an example of an actual sensor data matrix.
Each row of the actual sensor data matrix corresponds to one unit region, and each column corresponds to one measurement time.
Here, the “unit area” refers to each area obtained by dividing the space in which the sensor is installed into several areas. In the example of FIG. 5, “A”, [B], and “C” in each row indicate one section each of the space in which the sensor is installed divided into three regions according to latitude and longitude, One sensor is installed in each section. Hereinafter, the sensors in the unit area are indicated by unit areas “A”, [B], and “C” in which the sensor is installed. The position range C11 indicated by the acquisition request data in FIG. 4 includes three sensors A, B, and C, and the time-series aligning unit 103 uses an actual sensor data matrix corresponding to all three sensors. Is generated.
A single unit area may include a plurality of sensors. In this case, the time series generation unit 103 generates an actual sensor data matrix using an average value of measured values of each sensor in the unit area as an actual measurement value (actual sensor data value). Alternatively, a value other than the average value such as an intermediate value or a mode value may be used as the actual measurement value.

Note that the sensors A, B, and C of this embodiment are all temperature sensors, and measure the temperature at the installed position in degrees Celsius (° C.). However, the sensor data attribute included in the actual sensor data matrix is not limited to temperature and is not limited to one type. For example, the actual sensor data matrix may include sensor data having a plurality of attributes such as humidity and precipitation.
Moreover, in the example of FIG. 5, each column is 2006/02/01 00:00:00, 00:30:00, ..., 2006/02/28 22:30:00, 23:00: 00,..., Each of the measurement times at intervals of 30 minutes is shown. Hereinafter, each measurement time in the actual sensor data matrix is referred to as “unit time”. This unit time is determined such that the acquisition interval indicated by the acquisition request data is a multiple of the unit time interval. For example, the acquisition interval indicated by the acquisition request data D200 in FIG. 3 is one hour. On the other hand, in the actual sensor data in FIG. 5, half the acquisition interval is 30 minutes. In this way, the time series sorting unit 103 determines that the acquisition interval indicated by the acquisition request data is a multiple of the unit time interval, and generates an actual sensor data matrix corresponding to all the measurement times indicated by the acquisition request data. To do.

Each element of the actual sensor data matrix indicates the data value measured by the sensor corresponding to the row at the time corresponding to the column. In the example of FIG. 5, the value of the element Va102 is the data value “7.0” (° C.) measured by the sensor A at 2006/02/01 00:30:00.
Here, all elements of the actual sensor data matrix are not necessarily filled with data actually measured by the sensor. In the case of FIG. 5, the sensor A does not measure at 2006/02/01 00:00:00, and the actual measurement data Va101 at this time does not exist. In the figure, “−” indicates that there is no actual measurement data. Hereinafter, the absence of the actual measurement data is referred to as “missing”, and the sensor data value when the actual measurement data does not exist is referred to as “missing value”.

  For example, in the unit area A, actual measurement data Va102, Va103, and Va109 are obtained at intervals of 3 hours from 2006/02/01 00:30:00, and in the unit area C, 2006/02/01 0:00:00 Measured data Vc101, Vc104, Vc108, and Vc112 are obtained at intervals of 2 hours. In FIG. 5, the unit time interval is set to 30 minutes. In the range from 2006/03/01 00:00:00 to 03:00:00, the unit area A is 2006/03/01 00:00: The values Va106 to Va108 from 00 to 11:00 and the values Va110 to Va112 from 02:00:00 to 03:00:00 are missing values, and in the unit area C from 00:00:00 Values Vc106 and Vc107 from 0:30:00 and values Vc109 to Vc111 from 01:30:00 to 02:30:00 are missing values.

Next, the data structure of the time series interpolation matrix stored in the time series interpolation matrix storage unit 107 will be described. Here, “time series interpolation” is to calculate a value to be interpolated based on actual sensor data at another time in the same unit region. The time series interpolation matrix is a matrix obtained by interpolating each missing value of the actual sensor data matrix based on actual sensor data at other times in the same unit region.
FIG. 6 is a diagram illustrating an example of a time series interpolation matrix.
In the example shown in the figure, for example, the actual sensor data value Va102 of 2006/02/01 00:03:00 of the unit area A is the same as the value Va201 of 2006/02/01 00:00:00 of the unit area A. Is retained. Further, values Va204 to Va208 from 2006/02/28 23:00:00 to 2006/03/01 01:00:00 in the unit area A are also the same as those in the unit area A 2006/02/28 22:30:00. The actual sensor data value Va103 and the actual sensor data value Va109 of 2006/03/01 01:30:00 are linearly interpolated. In addition, values Va210 to Va212 from 2006/03/01 02:00:00 to 2006/03/01 03:00:00 in the unit area A are also the same as 2006/03/01 01:30:00 in the unit area A. Value Va109 is held.

Next, the data structure of the decision tree generated by the decision tree generation unit 110 will be described.
FIG. 7 is a diagram illustrating an example of a decision tree of the unit region A.
The decision tree is a binary tree used when spatial interpolation is performed on the elements of the time-series interpolation matrix, and candidate values of the elements after interpolation are assigned to each of the leaves N8 to N15. . Here, “spatial interpolation” means that the value to be interpolated is calculated based on the values of other unit areas at the same time. For example, the value Va206 of the unit region A in FIG. 6 is set to the value Vb206 (“10.6”) of the unit region B at the same time 2006/03/01 00:00:00 and the value Vc206 of the unit region C at the same time. ("10.1").

In the example of FIG. 7, the candidate value “−3.7” after interpolation when interpolating any element of the row of the unit region A of the time series interpolation matrix M320 of FIG. Assigned. Similarly, candidate value “−0.9” after interpolation is assigned to leaf N9. Hereinafter, the variable indicating the sensor value of the unit area to be interpolated is referred to as “object variable”, and the variable indicating the sensor value of the other unit area is referred to as “explanatory variable”. In the example of FIG. 7, the objective variable A indicates the sensor value of the unit area A to be interpolated, and the explanatory variables B and C indicate the sensor values of the unit areas B and C, respectively.
A conditional expression using explanatory variables is assigned to each of the nodes N1 to N7 including the root N1. For example, the conditional expression “C <2.838” using the explanatory variable C is assigned to the node N1 in FIG.

In order to determine the value of the unit region A at a certain time (hereinafter referred to as “attention unit time”) using the decision tree of FIG. 7, first, each of the explanatory variables B and C is set to the interpolation target time. The values of the unit areas B and C are substituted. Then, in order from the root (node of node number N1), if the conditional expression of each node is satisfied, it proceeds to the leaf (nodes of node numbers N8 to N15) while branching to the left if it does not hold, and reaches the arrived leaf The assigned value is a value at the unit time of interest in the unit area A.
For example, when the value Va206 of the unit area A at 2006/03/01 00:00:00 in FIG. 6 is interpolated, the explanatory variable B of the decision tree in FIG. 7 includes the unit at the same time in the time series interpolation matrix. The value of region B (Vb 206 value “10.6” in FIG. 6) is substituted, and the value of unit region C at the same time (Vc 206 value “10.1” in FIG. 6) is substituted for explanatory variable C. To do.
Then, when the decision tree is traced from the route N1, the conditional expression C (= 10.1) <2.838 of the route N1 is not established, and therefore branches to the right and reaches the node N3. Since the conditional expression C (= 10.1) <8.913 of the node N3 is not established, it branches to the right and reaches the node N7. Since the conditional expression C (= 10.1) <12 of the node N3 is satisfied, the left branch branches to the leaf N14, and the value “9.6” of the leaf N14 is set as the value after interpolation.

Next, the data structure of decision tree data stored in the decision tree storage unit 111 will be described.
FIG. 8 is a diagram illustrating an example of decision tree data stored in the decision tree storage unit 111.
For each objective variable, the decision tree storage unit 111 stores the decision tree generated by the decision tree generation unit 110 as decision tree data in a format in which the structure of the decision tree can be understood.
In the example of FIG. 8, the leftmost value in each row indicates a node number. The node with node number 1 is the root. The second value from the left in the row of node numbers 2 to 15 indicates a branch condition that reaches the node. The third (rightmost) value from the left in the row of node numbers 8 to 15 indicates the value of the objective variable A.

In order to obtain the value of the objective variable A using the decision tree data shown in the figure, first, the node number among the data of the objective variable A stored in the decision tree storage unit 111 starting from the root (node number 1) is stored. Are searched for the node having the node number of 2 and the node having the node number of 3 to obtain the respective branch conditions. According to the obtained branching condition, if the value of the explanatory variable C is less than 2.838, the process proceeds to node 2, and if it is 2.838 or more, the process proceeds to node 3.
In this way, from the node of node number i (i is a positive integer of 1 ≦ i ≦ 7), the node of node number 2i and the node of node number 2i + 1 are searched, and any node that satisfies the branch condition Proceed to By repeating this, the same leaf as when the decision tree shown in FIG. 7 is used is reached, and the value of the objective variable A is obtained.

Next, the data configuration of the spatial interpolation matrix stored in the spatial interpolation matrix storage unit 113 will be described.
FIG. 9 is a diagram illustrating an example of a spatial interpolation matrix.
The spatial interpolation matrix is a matrix obtained by performing spatial interpolation on the time series interpolation matrix. For example, when obtaining the value Va306 of 2006/03/01 00:00:00 of the unit area A, the value Vb206 of 2006/03/01 00:00:00 of the unit area B of the time series interpolation matrix M320 of FIG. 7 and the value Vc206 “10.1” of 2006/03/01 00:00:00 of the unit region C are used to trace the decision tree of the unit region A shown in FIG. Then, since the value reaches the leaf N14 having a value of “9.6”, the value of the unit region A at “2006/03/01 00:00:00” in the spatial interpolation matrix is set to “9.6”.
Similarly to the unit area A, the unit trees B and C are values obtained by the decision tree generation unit 110 generating a decision tree and the spatial interpolation function unit 108 performing spatial interpolation using the generated decision tree. ing.

Next, the data structure of the answer data generated by the answer function unit 114 will be described.
FIG. 10 is a diagram illustrating an example of answer data.
The response data is data generated by the response function unit 114 in response to the acquisition request data. Among the sensor data stored in the spatial interpolation matrix storage unit 113, the range of the position and the time range requested by the acquisition request data Data matching the acquisition interval is shown in the same format as the actual sensor data.
In the example of FIG. 10, the sensors in the region from N35.00.00.00 and E139.00.00.000 to N35.00.10.00 and E139.00.10.000 in accordance with the position condition C11 of the acquisition data request of FIG. The sensor value of A is in the time range from 2006/03/01 00:00:00 to 2006/03/01 03:00:00 according to the time condition C12, and one hour interval according to the condition C13. Is shown.

Here, at the time when the actual sensor data is missing, not only the sensor value but also the metadata does not exist. Therefore, the response function unit 114 generates metadata at the time when the actual sensor data is lost based on the actual sensor data at other times in the same unit area, for example, by copying the metadata of the previous actual sensor data. To do. In the example of FIG. 10, the response function unit 114 sets the metadata of “sensor type”, “position”, and “time” of the data of 2006/03/01 02:00:00 as the actual sensor data immediately before. It is generated by copying the metadata of actual sensor data of 2006/03/01 01:30:00.
Note that the answer data generated by the answer function unit 114 is not limited to data of one unit area shown in FIG. For example, the answer function unit 114 may generate answer data including data of a plurality of unit areas, such as generating tabular data including a plurality of unit areas.

Next, the operation of the interpolation apparatus 100 will be described.
When the ubiquitous application device 116 transmits the acquisition request data, the acquisition request reception function unit 101 receives the acquired acquisition request data and outputs it to the time-series sorting unit 103. Based on the acquisition request data from the acquisition request reception function unit 101, the time series sorting unit 103 generates search request data in which the conditions of actual sensor data to be read from the shared sensor storage device 115 are described using a SQL search expression.
For example, upon receiving the output of the acquisition request data in FIG. 3, the time series sorting unit 103 obtains search request data indicating a position condition C21 and a time condition C22 wider than the position condition C11 and the time condition C12 indicated by the acquisition request data. Generate.

The time series sorting unit 103 outputs the generated search request data to the shared sensor data storage device reading function unit 102, and the shared sensor data storage device reading function unit 102 searches the shared sensor data storage device 115 to search. All the actual sensor data that matches the condition indicated by the request data is read, and the read actual sensor data is output to the time series alignment unit 103.
The time series sorting unit 103 generates a matrix divided by unit time and unit area, and writes the real sensor data read from the shared sensor storage device 115 to the unit area and unit time corresponding to each position and time. .
The time series aligning unit 103 writes the generated matrix as a real sensor data matrix in the real sensor data matrix storage unit 104 and outputs a signal instructing to perform time series interpolation to the time series interpolation function unit 105.

FIG. 11 is a flowchart illustrating a processing procedure in which the time-series interpolation function unit 105 performs time-series interpolation on the actual sensor data matrix. When the time series interpolation function unit 105 receives an output of a signal instructing to perform time series interpolation from the time series alignment unit 103, the time series interpolation function unit 105 starts the processing of FIG.
In step S300, the time-series interpolation function unit 105 first reads the actual sensor data matrix from the actual sensor data matrix storage unit 104.
In step S301, the time-series interpolation function unit 105 sequentially sets each unit region of the read actual sensor data matrix as a target unit region, and performs a loop for performing the following processing (procedure 302 and subsequent steps) in the set target unit region. Run.
In step S302, the time-series interpolation function unit 105 sets each unit time of the target unit area as a target point in order, and executes a loop for performing the following processing (from step 303) on the set target point.
In step S303, the time series interpolation function unit 105 determines whether data exists at the point of interest. If it is determined that there is data (step S303: Yes), the process proceeds to step 305, and if it is determined that there is no data (step S303: No), the process proceeds to step S304.
In step 304, the first stage interpolation function unit 106 performs time series interpolation.

FIG. 12 is a flowchart illustrating a processing procedure in which the first stage interpolation function unit 106 performs time-series interpolation.
In step S310, the first-stage interpolation function unit 106 determines whether there is past data from the target point in the same unit area as the target point of the matrix. If it is determined that there is past data from the point of interest (step S310: YES), the process proceeds to step S312. If it is determined that there is no data (step S310: NO), the process proceeds to step S311.
In step S311, the first stage interpolation function unit 106 determines whether there is future data from the point of interest. If it is determined that there is future data from the point of interest (step S311: YES), the process proceeds to step S315. If it is determined that there is no future data from the point of interest (step S311: No), there is no data in this unit area, so the first-stage interpolation function unit 106 gives up time-series interpolation and performs the processing shown in FIG. Exit.

In step S312, the first-stage interpolation function unit 106 determines whether there is future data from the point of interest. If it is determined that there is future data from the attention point (step S312: YES), the process proceeds to step S313, and if it is determined that there is no data (step S312: NO), the process proceeds to step S314.
In step S313, the first-stage interpolation function unit 106 substitutes a value obtained by linear interpolation using two points of past data and future data closest to the attention point, into the attention point of the matrix. Thereafter, the process of FIG.
In step S314, in this case, there is past data from the attention point, but no future data exists, so the first-stage interpolation function unit 106 sets the past data closest to the attention point as the attention point of the matrix. substitute. Thereafter, the process of FIG.
In step S315, in this case, there is future data from the attention point, but there is no past data. Therefore, the first-stage interpolation function unit 106 substitutes the future data closest to the attention point into the attention point of the matrix. To do. Thereafter, the process of FIG.
Note that the interpolation algorithm performed by the first-stage interpolation function unit 106 in step 313 is not limited to the linear interpolation described above. For example, an interpolation algorithm other than linear interpolation such as average value interpolation, previous value interpolation, or spline interpolation may be used.

Returning to FIG. 11, in step S <b> 305, the time-series interpolation function unit 105 determines whether or not the processing in step S <b> 302 and subsequent steps has been performed for all unit times of the unit area of interest. If it is determined that the unprocessed unit time remains, the process returns to step S302 and the unprocessed unit time is processed. If it is determined that the processing has been completed for all unit times, the process proceeds to step S306.
In step S306, the time series interpolation function unit 105 determines whether or not the processing in step S301 and subsequent steps has been performed for all unit regions of the actual sensor data matrix. If an unprocessed unit area remains, the process returns to step S301 to process the unprocessed unit area. If the processing has been completed for all the unit areas, the process proceeds to step S307.
In step S307, the time series interpolation function unit 105 writes a matrix (time series interpolation matrix) obtained by performing time series interpolation into the time series interpolation matrix storage unit 107. When the writing is completed, the time series interpolation function unit 105 outputs a signal instructing the spatial interpolation function unit 108 to perform spatial interpolation. Thereafter, the process of FIG.

FIG. 13 is a flowchart illustrating a processing procedure in which the spatial interpolation function unit 108 performs spatial interpolation. When the spatial interpolation function unit 108 receives an output of a signal instructing to perform spatial interpolation from the time-series interpolation function unit 105, the spatial interpolation function unit 108 starts the processing in FIG.
In step S400, the decision tree generation unit 110 of the spatial interpolation function unit 108 generates a decision tree.

FIG. 14 is a flowchart illustrating a processing procedure in which the decision tree generation unit 110 generates a decision tree. A decision tree is a binary tree that represents a classification algorithm in a tree structure as in the example of FIG. 7, and is widely used for classification and prediction. Classification and prediction using a decision tree are performed by following the decision tree according to the value of the explanatory variable and the conditional expression assigned to each node, as described with reference to FIG. It is known that the conditional expression assigned to each node can be easily generated by calculating the information gain, Gini coefficient, or entropy.
In step S410, the decision tree generation unit 110 reads, from the time series interpolation matrix storage unit 107, the time series interpolation matrix in the past time range from the acquisition request data as decision tree teacher data. For example, the decision tree generation unit 110 includes the time 2006/03/01 00:00:00 indicated in the acquisition request data in FIG. 3 among the time series interpolation matrices stored in the time series interpolation matrix storage unit 107 in FIG. The time series interpolation matrix in the range from 2006/02/01 00:00:00 to 2006/02/28 23:30:00 is read.

In step S411, the decision tree generation unit 110 sequentially sets each unit region of the read time-series interpolation matrix as a target unit region, and performs a loop for performing the following processing (from step 412) on the set target unit region. Run.
In step S <b> 412, the decision tree generation unit 110 determines the target unit region decision tree with the target unit region as the objective variable and the unit region other than the target unit region as the explanatory variable at the same unit time in the read teacher data matrix. Is generated.
The decision tree generation unit 110 generates a decision tree using a CART algorithm. Note that the method by which the decision tree generation unit 110 generates the decision tree is not limited to the method using the CART algorithm. For example, the decision tree may be generated using another algorithm such as ID3, C4.5, or C5.0.

In step S413, the decision tree generation unit 110 writes the decision tree data indicating the generated decision tree in the corresponding objective variable column of the table stored in the decision tree storage unit 111. In the example of FIG. 8, the decision tree generation unit 110 stores the decision tree data of the decision tree having the unit region A as the objective variable (the decision tree in FIG. 7) in the objective variable A of the table stored in the decision tree storage unit 111. It fills in the column.
In step S414, the decision tree generation unit 110 determines whether or not the processing after step S412 has been performed on all unit regions in the time-series interpolation matrix. If it is determined that the unprocessed unit area remains, the process returns to step S411 to process the unprocessed unit area. If it is determined that the processing has been completed for all the unit areas, the processing in FIG. 14 ends.

Returning to FIG. 13, in step S <b> 401, the spatial interpolation function unit 108 reads from the actual sensor data matrix storage unit 104 an actual sensor data matrix in a range that matches the time range of the acquisition request data.
In step S402, the spatial interpolation function unit 108 sets each unit region of the read actual sensor data matrix as a target unit region in order, and performs a loop for performing the following processing (from step 403) on the set target unit region. Run.
In step S403, the spatial interpolation function unit 108 sets each unit time of the target unit area as a target point in order, and executes a loop for performing the following processing (from step 404) on the set target point.
In step S404, the spatial interpolation function unit 108 determines whether data exists at the point of interest. If it is determined that data exists (step S404: YES), the process proceeds to step S406. If it is determined that it does not exist (step S404: NO), the process proceeds to step S405.
In step S405, the decision tree application unit 112 performs spatial interpolation on the point of interest.

FIG. 15 is a flowchart illustrating a processing procedure in which the decision tree application unit 112 performs spatial interpolation on the point of interest.
In step S430, the decision tree application unit 112 searches the decision tree storage unit 111 to determine whether or not there is decision tree data having a unit region including the target point of the matrix as an objective variable. If it is determined that there is the decision tree data (step S430: YES), the process proceeds to step S431. On the other hand, when it is determined that the decision tree data does not exist (step S430: No), the spatial interpolation is abandoned and the process of FIG.
In step S431, the decision tree application unit 112 reads the decision tree data from the decision tree storage unit 111.

In step S432, the decision tree application unit 112 converts the data of the unit area corresponding to the explanatory variable of the unit time including the attention point from the time series interpolation matrix storage unit based on the decision tree data of the unit area. Read as.
In step S433, the decision tree application unit 112 determines whether or not the corresponding explanation data is missing. If it is determined that the data is missing (step S433: Yes), the spatial interpolation is abandoned and the processing in FIG. On the other hand, if it is determined that there is no loss (step S433: NO), the process proceeds to step S434.
In step S434, the decision tree application unit 112 includes units of interest for the conditional expressions assigned to the nodes of the decision tree data read from the decision tree storage unit 111, as described with reference to FIG. Substituting time explanatory data, traces the decision tree according to the success or failure of the conditional expression, and writes the value assigned to the arrived leaf to the element of the point of interest of the spatial interpolation matrix as the value after spatial interpolation. Thereafter, the decision tree application unit 112 ends the process of FIG.

Returning to FIG. 13, in step S <b> 406, the spatial interpolation function unit 108 determines whether or not the processing in step S <b> 404 and subsequent steps has been performed for all unit times of the target unit area. If it is determined that an unprocessed unit time remains, the process returns to step S403 to perform an unprocessed unit time. If it is determined that the processing has been completed for all unit times, the process proceeds to step S407.
In step S407, the spatial interpolation function unit 108 determines whether or not the processing in step S403 and subsequent steps has been performed for all unit regions of the actual sensor data matrix. If an unprocessed unit area remains, the process returns to step S402 to process the unprocessed unit area. If the processing has been completed for all the unit areas, the process proceeds to step S408.
In step S <b> 408, the spatial interpolation function unit 108 stores the generated spatial interpolation matrix in the spatial interpolation matrix storage unit 113 and outputs a signal instructing the ubiquitous application device 116 to make a response to the response function unit 114. Thereafter, the spatial interpolation function unit 108 ends the process of FIG.

The response function unit 114 that has received the output of the signal for instructing the ubiquitous application device 116 to perform the response from the spatial interpolation function unit 108 reads the spatial interpolation matrix from the spatial interpolation matrix storage unit 113, and based on the read spatial interpolation matrix Response data is generated, and the generated response data is transmitted to the ubiquitous application device 116.
Specifically, the reply function unit 114 reads information indicating the sensor and information indicating the range of the measurement time from the acquisition request data, and calculates the sensor corresponding to the read information and the measurement value of the measurement time by a spatial interpolation matrix. Read from. For example, the response function unit 114 falls within the area from N35.00.00.00 and E139.00.00.000 to N35.00.10.00 and E139.00.10.000 in accordance with the position condition C11 of the acquisition data request in FIG. The sensor value of a certain sensor A is set to 1 within the time range from 2006/03/01 00:00:00 to 2006/03/01 03:00:00 according to the time condition C12 according to the condition C13. Measurement values are read from the spatial interpolation matrix of FIG. 9 at time intervals. In addition, as described with reference to FIG. 10, the answer function unit 114 sets the “sensor type”, “position”, and “time” of the data of 2006/03/01 02:00:00 and 03:00:00 on the same day. Is generated by copying the metadata of the actual sensor data of 2006/03/01 01:30:00 as the previous actual sensor data. In addition, the metadata of “sensor type”, “position”, and “time” of the data of 2006/03/01 00:00:00 and 01:00:00 on the same day is the actual sensor data immediately before 2006 / Copy the metadata of the actual sensor data at 02/28 22:30:00 to generate.
The response function unit generates the response data in FIG. 10 including the measurement values read from the spatial interpolation matrix and the metadata copied from the actual sensor data, and transmits the response data to the ubiquitous application device 116.

16 and 17 are sequence diagrams illustrating a processing procedure in which the interpolation device 100 generates and transmits (answers) response data according to the acquisition request data transmitted from the ubiquitous application device 116.
In FIG. 16, first, when the ubiquitous application device 116 transmits acquisition request data to the interpolation device 100 (sequence S801), the acquisition request reception function unit 101 of the interpolation device 100 receives the acquisition request data, and a time series alignment unit. The data is output (transferred) to 103 (sequence S802).
The time series sorting unit 103 generates search request data based on the transferred acquisition request data, and outputs the generated search request data to the shared sensor data storage device reading function unit 102 (sequence S803).

The shared sensor data storage device reading function unit 102 searches the shared sensor data storage device 115 based on the search request data from the time series sorting unit 103 and reads the actual sensor data. Specifically, shared sensor data storage device reading function unit 102 transmits search request data to shared sensor data storage device 115 (sequence S804). The shared sensor data storage device 115 searches the internal storage unit, reads actual sensor data corresponding to the search request data (sequence S085), and transmits the read actual sensor data to the interpolation device 100 (sequence S806). The shared sensor data storage device reading function unit 102 of the interpolation device 100 receives the actual sensor data, and outputs (transfers) it to the time series alignment unit 103 (sequence S807).
The time series aligning unit 103 generates an actual sensor data matrix based on the actual sensor data from the shared sensor data storage device reading function unit 102 (sequence S808), and the generated actual sensor data matrix is converted into the actual sensor data matrix storage unit 104. (Sequence S809). Thereafter, the time series alignment unit 103 outputs a signal instructing to perform time series interpolation to the time series interpolation function unit 105 (sequence S810).

Moving to FIG. 17, the time series interpolation function unit 105 that has received the signal output from the time series alignment unit 103 reads the actual sensor data matrix from the actual sensor data matrix storage unit 104 (sequence S821), and performs time series interpolation. Then, a time series interpolation matrix is generated (sequence S822), and the generated time series interpolation matrix is written in the time series interpolation matrix storage unit 107 (sequence S823). Thereafter, the time series interpolation function unit 105 outputs a signal instructing the spatial interpolation function unit 108 to perform spatial interpolation (sequence S824).
Upon receiving the signal output from the time series interpolation function unit 105, the spatial interpolation function unit 108 reads the actual sensor data from the actual sensor data matrix storage unit 104 (sequence S825), and generates a decision tree (sequence S826). Then, the spatial interpolation function unit 108 reads the actual sensor data matrix (sequence S827), reads the explanation data for each unit time from the time series interpolation matrix storage unit 107 (sequence S828), performs spatial interpolation, and performs the spatial interpolation matrix. Is generated (sequence S829), and the generated spatial interpolation matrix is written in the spatial interpolation matrix storage unit 113 (sequence S830). Thereafter, the spatial interpolation function unit 108 outputs a signal instructing the ubiquitous application device 116 to make a reply to the reply function unit 114 (sequence S831).
The response function unit 114 reads the spatial interpolation matrix from the spatial interpolation matrix storage unit 113 (sequence S832), generates response data according to the acquisition request data (sequence S833), and transmits the generated response data to the ubiquitous application device 116. Is performed (sequence S834).

As described above, in the interpolation device 100, the spatial interpolation function unit 108 performs spatial interpolation on the measured data obtained from the shared sensor data storage device after the time-series interpolation function unit 105 performs time-series interpolation. Therefore, it is possible to perform interpolation for data loss caused by the difference between the data acquisition cycle by the sensor and the data acquisition cycle required by the application, particularly, a periodic loss or a loss at a high loss rate.
Further, by using both time series interpolation and spatial interpolation, it can be expected that a more accurate interpolation value reflecting the temporal correlation and the spatial correlation of the sensor data can be obtained.
For example, when time-series interpolation is performed simply by linear interpolation, the measured data may fluctuate significantly due to temporary sensor mismeasurements, noise contamination during sensor data transmission, or local and instantaneous fluctuations in measured values. In this case, the sensor data at the peripheral time that is interpolated using the actual measurement data becomes an inaccurate value including a large error. On the other hand, in the interpolation performed by the interpolation device 100, the interpolation is performed by correlating with a plurality of sensor data, so that the influence of the error included in one sensor data is reduced and a more accurate value can be obtained. Can be expected.
In addition, when the sensor data requested from the ubiquitous application device 116 is data of a unit region having a large acquisition cycle (the number of actual measurement values is small), the unit having a small acquisition cycle (a large number of actual measurement values). It can be expected that a more accurate value can be obtained by performing interpolation using the sensor data of the region.
Furthermore, the decision tree can be generated by using the value of the result of time series interpolation, and the data of the unit area highly correlated with the target unit area is selected as the explanatory variable, and the value of the highly correlated data is more strongly interpolated. Can be reflected. Thereby, it can be expected that a more accurate value can be obtained than when linear interpolation or average value interpolation is simply performed.

Note that the spatial interpolation function unit 108 directly reflects the interpolation value generated by the spatial interpolation generated by the second stage interpolation function unit and the interpolation value generated by the time series interpolation generated by the first stage interpolation function unit 106. A value may be generated. For example, when the decision tree application unit 112 calculates an interpolation value by spatial interpolation of the point of interest in step S405 of FIG. 13, the decision tree application unit 112 writes the calculated interpolation value in the spatial interpolation matrix storage unit 113. First, the data is output to the spatial interpolation function unit 108. The spatial interpolation function unit 108 reads the interpolation value of the target point by time series interpolation from the time series interpolation matrix storage unit 107, and by the read interpolation value by time series interpolation and the spatial interpolation output from the decision tree application unit 112. An average value with the interpolation value is calculated, and the calculated average value is written in the spatial interpolation matrix storage unit 113 as the interpolation value of the point of interest.
In this way, when calculating the value of the target point by taking the average of the interpolated value by time series interpolation and the interpolated value by spatial interpolation, other sensors at the same time and the time correlation with the measured data by the same sensor It is possible to directly reflect the spatial correlation with the actual measurement data obtained by, and it can be expected that a more accurate value can be obtained.

  Further, when the spatial interpolation function unit 108 calculates the average value of the interpolation value by time series interpolation and the interpolation value by spatial interpolation, the spatial interpolation function unit 108 performs weighted averaging based on the temporal relationship or spatial relationship with the point of interest. May be. For example, by determining the weight based on the time difference between the unit time of the target point and the data measurement time, or the distance between the sensor position of the target point and the data measurement position, the interpolation value by time series interpolation or the interpolation by spatial interpolation It can be expected that the correlation between the value and the sensor data at the point of interest can be reflected more accurately, and a more accurate value can be obtained.

<Second Embodiment>
Spatial interpolation performed by the interpolation device is not limited to using the decision tree described in the first embodiment. In the present embodiment, an interpolation apparatus that performs spatial interpolation by the k-NN method as spatial interpolation other than a decision tree will be described.
FIG. 18 is a configuration diagram illustrating a schematic configuration of the interpolation device according to the second embodiment of the present invention.
In the figure, an interpolation device 500 includes an acquisition request reception function unit 101, a shared sensor data storage device reading function unit 102, a time series alignment unit 103, an actual sensor data matrix storage unit 104, and a time series interpolation function unit 105. A time series interpolation matrix storage unit 107, a spatial interpolation function unit 108, a spatial interpolation matrix storage unit 113, and an answer function unit 114. The time series interpolation function unit 105 includes a first stage interpolation function unit 106. The spatial interpolation function unit 508 includes a second-stage interpolation function unit 509. The second-stage interpolation function unit 509 includes a distance calculation function unit 510, a distance storage unit 511, and a k-NN method application unit 512.
The interpolation device 500 is connected to the shared sensor data storage device 115 and the ubiquitous application device 116.
In the figure, the same reference numerals (101 to 107, 113 to 116) are assigned to portions having the same functions as those in FIG.

 The spatial interpolation function unit 508 performs spatial interpolation on the time series interpolation matrix to generate a spatial interpolation matrix. The second-stage interpolation function unit 509 performs spatial interpolation on each element of the time series interpolation matrix. The distance calculation function unit 510 calculates the distance between each unit region used for spatial interpolation by the k-NN method. The distance storage unit 511 stores the distance between the unit areas calculated by the distance calculation function unit 510. The k-NN method application unit 512 performs spatial interpolation on individual elements of the time-series interpolation matrix based on the distance between the unit regions.

Next, the data configuration of the distance data matrix stored in the distance storage unit 511 will be described.
FIG. 19 is a diagram illustrating an example of a distance data matrix.
Each distance data in the distance data matrix indicates a distance between two unit row regions. For example, the distance data D171 in the first row and the second column in the figure indicates that the distance between the unit area A and the unit area B is 5.0.

Next, the spatial interpolation matrix in this embodiment will be described.
FIG. 20 is a diagram illustrating an example of a spatial interpolation matrix stored in the spatial interpolation matrix storage unit 113.
In the spatial interpolation matrix of the present embodiment, the value of the element to be interpolated (hereinafter referred to as “attention point”) is the distance from the unit area including the attention point among the unit areas not including the attention point. This is the average value of the values on the time-series interpolation matrix at the time including the point of interest in the unit region equal to or less than the average value of the distances between the unit region including the point and each unit region not including the point of interest.
For example, the distance between the unit area A and the unit area B stored in the distance storage unit 511 in FIG. 19 is 5.0, and the distance between the unit area A and the unit area C is 5. 0. Therefore, the average value of the distance between the unit area A and the other unit areas (B, C) is 5.0, and the distance between the unit area A and the unit area A is equal to or less than the average value between the unit area B and the unit area C. It is.
Therefore, the value Va406 (“10.4”) in the unit region A in FIG. 20 at 2006/03/01 00:00:00 is the value Vb206 (at the same time in the unit region B of the time-series interpolation matrix in FIG. “10.6”) and the average value (the value obtained by rounding the second decimal place) of the value Vc206 (“10.1”) at the same time in the unit region C of the time series interpolation matrix of FIG. .
Similarly to the unit region A, the values of the unit regions B and C in FIG.

Next, the operation of the spatial interpolation function unit 508 will be described.
FIG. 21 is a flowchart illustrating a processing procedure in which the spatial interpolation function unit 508 performs spatial interpolation.
In step S <b> 600, the distance calculation function unit 510 calculates the distance between the unit areas, and writes the calculated distance in the distance data matrix of the distance storage unit 511.

FIG. 22 is a flowchart illustrating a processing procedure in which the distance calculation function unit 510 generates a distance data matrix.
In step S610, the distance calculation function unit 510 reads a time series interpolation matrix in a past time range from the acquisition request data from the time series interpolation matrix storage unit 107, as in the decision tree generation unit 110 of the first embodiment. . For example, the distance calculation function unit 510 includes the time 2006/03/01 00:00:00 indicated in the acquisition request data in FIG. 3 among the time series interpolation matrix stored in the time series interpolation matrix storage unit 107 in FIG. The time series interpolation matrix in the range from 2006/02/01 00:00:00 to 2006/02/28 23:30:00 is read.
In step S611, the distance calculation function unit 510 sets each unit region of the read time-series interpolation matrix as a target unit region in order, and performs a loop for performing the following processing (from step 612) on the set target unit region. Run.

In step S612, the distance calculation function unit 510 sequentially sets each of all unit areas except the target unit area as a distance calculation unit area, and performs the following processing (from step 613) on the set distance calculation unit area. Execute.
In step S613, the distance calculation function unit 510 calculates the Euclidean distance between the two based on the “position” metadata of the target unit area and the “position” metadata of the distance calculation unit area. When the sensors A to C move, the distance calculation function unit 510 may calculate the distance between the target unit area and the distance calculation unit area for each unit time. In this case, the distance storage unit 511 stores the distance between the unit areas for each unit time.
In step S614, the distance calculation function unit 510 writes the calculated distance in the distance storage unit 511.

In step S615, the distance calculation function unit 510 determines whether or not the processing in step S613 and subsequent steps has been performed for all unit regions other than the target unit region. If it is determined that the unprocessed unit area remains, the process returns to step S612 to process the unprocessed unit area. If it is determined that the processing has been completed for all the unit areas, the process proceeds to step S616.
In step S616, the distance calculation function unit 510 determines whether or not the processing after step S612 has been performed for all unit regions of the time-series interpolation matrix. If it is determined that the unprocessed unit area remains, the process returns to step S611 to process the unprocessed unit area. If it is determined that the processing has been completed for all the unit areas, the processing in FIG.

Returning to FIG. 21, in step S <b> 601, the spatial interpolation function unit 508 reads the actual sensor data matrix from the actual sensor data matrix storage unit 104.
In step S602, the spatial interpolation function unit 508 sequentially sets each unit region of the read actual sensor data matrix as a target unit region, and performs a loop for performing the following processing (from step 603) on the set target unit region. Run.
In step S603, the spatial interpolation function unit 508 sets each unit time of the target unit area as a target point in order, and executes a loop for performing the following processing (from step 604) on the set target point.
In step S604, the spatial interpolation function unit 508 determines whether data exists at the point of interest. If it is determined that data exists (procedure S604: Yes), the process proceeds to step S606. If it is determined that data does not exist (procedure S604: No), the process proceeds to step S605.
In step S605, the k-NN method application unit 512 performs spatial interpolation on the point of interest.

FIG. 23 is a flowchart illustrating a processing procedure in which the k-NN method application unit 512 performs spatial interpolation of a point of interest.
The k-NN method is a widely used algorithm for determining a value of a target point by calculating an average value, an intermediate value, a mode value, or the like of k pieces of data having a short distance from the target point. .
In step S620, the k-NN method application unit 512 reads all distance data between the unit region including the target point and other unit regions from the distance storage unit 511.
In step S621, the k-NN method application unit 512 calculates an average value of the read distances, and a unit area sensor in which the distance from the unit area including the attention point is equal to or less than the average value at the unit time including the attention point. Data is read from the time-series interpolation matrix storage unit 107.

In step S622, the k-NN method application unit 512 determines whether the data read in step 621 is missing (sensor data of some unit areas does not exist). If it is determined that the data is missing (step S622: Yes), the spatial interpolation is abandoned and the process of FIG. On the other hand, if it is determined that there is no loss (step S622: No), the process proceeds to step S623.
In step S623, the k-NN method applying unit 512 writes the average value of the data read in step 621 in the spatial interpolation matrix of the spatial interpolation matrix storage unit 113 as a value after spatial interpolation at the point of interest. Thereafter, the process of FIG.

Returning to FIG. 21, in step S <b> 606, the spatial interpolation function unit 508 determines whether or not the processing in step S <b> 604 and subsequent steps has been performed for all unit times of the unit area of interest. If it is determined that an unprocessed unit time remains, the process returns to step S603 to perform an unprocessed unit time. If it is determined that the processing has been completed for all unit times, the process proceeds to step S607.
In step S607, the spatial interpolation function unit 108 determines whether or not the processing in step S603 and subsequent steps has been performed for all unit regions of the actual sensor data matrix. If an unprocessed unit area remains, the process returns to step S602 to process the unprocessed unit area. If the processing has been completed for all the unit areas, the process proceeds to step S608.
In step S <b> 608, the spatial interpolation function unit 508 stores the generated spatial interpolation matrix in the spatial interpolation matrix storage unit 113 and outputs a signal instructing the ubiquitous application apparatus 116 to make a response to the response function unit 114. Thereafter, the spatial interpolation function unit 108 ends the process of FIG.

Note that the distance used by the spatial interpolation function unit 508 is not limited to the Euclidean distance of the position data. For example, the distance calculation function unit 510 calculates a distance other than the Euclidean distance, such as the number of co-occurrence of metadata, the correlation coefficient of the data of the time series interpolation matrix in the unit area, or a distance calculated from a combination thereof. Then, the k-NN method application unit 512 may perform spatial interpolation of the attention point using the calculated distance.
When the k-NN method application unit 512 calculates a mean value after multiplying the value of the unit region B by a coefficient of 0.8, the value relationship between the unit regions has a certain tendency. The k-NN method application unit 512 may calculate the average value after performing conversion according to this tendency. Thereby, the tendency between unit areas is reflected, and a more accurate interpolation value can be obtained.

  Note that the value after spatial interpolation at the point of interest calculated by the k-NN method application unit 512 is not limited to the average value described above. For example, the k-NN method application unit 512 may calculate a value other than the average value, such as an intermediate value or a mode value. Alternatively, for each of the unit areas to be averaged, a weight that decreases as the distance from the unit area including the target point becomes longer may be calculated, and a weighted average may be calculated using the calculated weight. When the correlation between the values of the unit regions is higher as the distance between the unit regions is closer, a more accurate interpolation value can be obtained by calculating the weighted average.

As described above, in the interpolation device 500, after the time series interpolation function unit 105 performs time series interpolation on the measured data obtained from the shared sensor data storage device, the space interpolation function unit 508 performs space interpolation. Therefore, as in the case of the first embodiment, the data loss caused by the difference between the data acquisition cycle by the sensor and the data acquisition cycle required by the application, particularly the periodic loss and the loss at a high loss rate. Interpolation can also be performed.
Further, by using time series interpolation and spatial interpolation together, it is possible to obtain a more accurate interpolation value reflecting the time correlation and spatial correlation of sensor data, as in the first embodiment. I can expect. Thereby, it can be expected that a more accurate value can be obtained than when linear interpolation or average value interpolation is simply performed.

As in the case of the first embodiment, the spatial interpolation function unit 508 performs the interpolation value generated by the spatial interpolation generated by the second-stage interpolation function unit 509 and the time-series interpolation generated by the first-stage interpolation function unit 106. An interpolation value that directly reflects the interpolation value may be generated. For example, as in the case of the first embodiment, when calculating the value of the target point by taking the average of the interpolated value by time series interpolation and the interpolated value by spatial interpolation, the time with the measured data by the same sensor is calculated. The correlation and the spatial correlation with the measured data by other sensors at the same time can be directly reflected, and it can be expected that a more accurate value can be obtained.
Furthermore, as in the case of the first embodiment, when the spatial interpolation function unit 508 calculates the average value of the interpolation value obtained by time series interpolation and the interpolation value obtained by spatial interpolation, You may make it perform a weighted average based on a spatial relationship. For example, by determining the weight based on the time difference between the unit time of the target point and the data measurement time, or the distance between the sensor position of the target point and the data measurement position, the interpolation value by time series interpolation or the interpolation by spatial interpolation It can be expected that the correlation between the value and the sensor data at the point of interest can be reflected more accurately, and a more accurate value can be obtained.

The apparatus of the present invention can be realized by a computer and a program, and can be recorded on a recording medium or provided through a network.
For example, a program for realizing all or part of the functions of the interpolation apparatus 100 or 500 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed. You may process each part by. Here, the “computer system” includes an OS and hardware such as peripheral devices.
Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used.
The “computer-readable recording medium” refers to a holding device such as a portable medium such as a flexible disk, a magneto-optical disk, a ROM, and a CD-ROM, and a hard disk built in the computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory in a computer system serving as a server or a client in that case, and a program that holds a program for a certain period of time are also included. The program may be a program for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in a computer system.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes design changes and the like within a scope not departing from the gist of the present invention.

  The present invention is suitable for use in an interpolation apparatus, an interpolation method, and a program.

DESCRIPTION OF SYMBOLS 100, 500 Interpolation apparatus 101 Acquisition request reception function part 102 Shared sensor data memory reading function part 103 Time series alignment part 104 Real sensor data matrix memory part 105 Time series interpolation function part 106 First stage interpolation function part 107 Time series interpolation matrix Storage unit 108, 508 Spatial interpolation function unit 109, 509 Second stage interpolation function unit 110 Decision tree generation unit 111 Decision tree storage unit 112 Decision tree application unit 510 Distance calculation function unit 511 Distance storage unit 512 k-NN method application unit 113 Spatial interpolation matrix storage unit 114 Reply function unit 115 Shared sensor data storage device 116 Ubiquitous application device

Claims (9)

An actual sensor data acquisition unit for acquiring actual sensor data including an actual measurement value by the sensor, information indicating a measurement position of the actual measurement value, and information indicating a measurement time of the actual measurement value;
Based on the acquisition request data indicating the sensor and the measurement time and requesting the measurement value of the sensor at the measurement time, for each sensor including all the sensors indicated by the acquisition request data, and the measurement time indicated by the acquisition request data The actual sensor data matrix having the measurement values for each measurement time including all the elements as elements is generated, and the actual measurement values are calculated based on the information indicating the measurement positions of the actual measurement values and the information indicating the measurement times of the actual measurement values. A time series alignment unit assigned to elements of the real sensor data matrix;
Among the elements of the actual sensor data matrix, for elements that do not have the actual measurement values, elements that do not have the actual measurement values based on the actual measurement values measured by the same sensor as the sensor corresponding to the element that does not have the actual measurement values A time-series interpolation function unit for interpolating the measured values at
Of the elements of the actual sensor data matrix, for the elements that do not have the actual measurement values, the actual measurement values obtained by a sensor different from the sensor corresponding to the element that does not have the actual measurement values, or the sensors that correspond to the elements that do not have the actual measurement values The measurement value required by the acquisition request data is generated by updating the measurement value of the element that does not have the actual measurement value using the measurement value interpolated by the time series interpolation function unit for the element corresponding to the sensor different from A spatial interpolation function unit,
An interpolation device comprising: An acquisition request reception function unit for receiving the acquisition request data;
An answer function unit for sending answer data for the acquisition request data;
Further comprising
The time series alignment unit calculates the real sensor data matrix corresponding to all of the sensors selected by the information for selecting the sensors and all of the measurement times indicated by the measurement time range and the measurement time interval. Generate and
The answer function unit transmits the answer data including the measurement value updated by the spatial interpolation function unit.
The interpolating apparatus according to claim 1. The spatial interpolation function unit
The measured value obtained by the time series interpolation function unit interpolating the measured value by a sensor different from the sensor corresponding to the element not having the actual measured value or the element corresponding to the sensor different from the sensor corresponding to the element not having the actual measured value A decision tree generation unit that generates a decision tree for determining a measurement value in an element that does not have the actual measurement value, according to a conditional expression for:
A decision tree application unit that determines a measurement value in an element that does not have the actual measurement value using the decision tree;
The interpolating apparatus according to claim 1, further comprising: The spatial interpolation function unit
A distance calculation function unit for determining a distance between a sensor corresponding to an element not having the actual measurement value and a sensor different from the sensor corresponding to the element not having the actual measurement value;
A k-NN method application unit that selects a sensor based on the distance and determines a measurement value in an element that does not have the actual measurement value based on a measurement value of the selected sensor;
The interpolating apparatus according to claim 1, further comprising: An interpolation method of an interpolation device for interpolating sensor measurement values,
An actual sensor data acquisition step for acquiring actual sensor data including an actual measurement value by the sensor, information indicating a measurement position of the actual measurement value, and information indicating a measurement time of the actual measurement value;
Based on the acquisition request data indicating the sensor and the measurement time and requesting the measurement value of the sensor at the measurement time, for each sensor including all the sensors indicated by the acquisition request data, and the measurement time indicated by the acquisition request data The actual sensor data matrix having the measurement values for each measurement time including all the elements as elements is generated, and the actual measurement values are calculated based on the information indicating the measurement positions of the actual measurement values and the information indicating the measurement times of the actual measurement values. Time series alignment step assigned to elements of the real sensor data matrix;
Among the elements of the actual sensor data matrix, for elements that do not have the actual measurement values, elements that do not have the actual measurement values based on the actual measurement values measured by the same sensor as the sensor corresponding to the element that does not have the actual measurement values A time series interpolation step for interpolating the measurement values at
Of the elements of the actual sensor data matrix, for the elements that do not have the actual measurement values, the actual measurement values obtained by a sensor different from the sensor corresponding to the element that does not have the actual measurement values, or the sensors that correspond to the elements that do not have the actual measurement values The measurement value requested by the acquisition request data is generated by updating the measurement value of the element not having the actual measurement value using the measurement value interpolated in the time series interpolation step for the element corresponding to the sensor different from Spatial interpolation step to
An interpolation method characterized by comprising: An acquisition request receiving step for receiving the acquisition request data;
A reply step of sending reply data to the acquisition request data;
Further comprising
The actual sensor data matrix corresponding to all of the sensors selected by the information for selecting the sensors in the time series alignment step and all of the measurement times indicated by the range of the measurement time and the measurement time interval Produces
Transmitting the answer data including the measurement value updated in the spatial interpolation step in the answer step;
The interpolation method according to claim 5, wherein: The spatial interpolation step includes
Measured value obtained by interpolating in the time-series interpolation step for an element corresponding to a sensor different from a sensor corresponding to an element not having the actual measurement value or a sensor different from the sensor corresponding to the element not having the actual measurement value A decision tree generating step for generating a decision tree for determining a measurement value in an element that does not have the actual measurement value according to a conditional expression for:
A decision tree application step of determining a measurement value in an element not having the actual measurement value using the decision tree;
The interpolation method according to claim 5 or 6, further comprising: The spatial interpolation step includes
A distance calculating step for determining a distance between a sensor corresponding to the element not having the actual measurement value and a sensor different from the sensor corresponding to the element not having the actual measurement value;
A k-NN method application step of selecting a sensor based on the distance, and determining a measured value in an element that does not have the measured value based on a measured value of the selected sensor;
The interpolation method according to claim 5 or 6, further comprising: A program for causing a computer to execute the interpolation method according to any one of claims 5 to 8.

Images