JP2013036887A - Observed value reliability assessing apparatus, observed value reliability assessing method, and observed value reliability assessing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accomplish more realistic threshold determination, even when the plausibility of virtual sensor data and/or threshold determination varies, by assessing the plausibility of the virtual sensor data and/or the threshold determination.SOLUTION: A virtual sensor data composing apparatus 100 comprises: an observed value statistical processing function unit 108 that calculates a representative value of observed values obtained by a plurality of observation devices; a meta-data statistical processing function unit 107 that calculates a representative value, regarding observation conditions in the plurality of observation devices and calculates fluctuations of the observation conditions from the representative value or the deviation of the representative value from a prescribed value; and a reliability assigning function unit 110 that calculates, on the basis of the fluctuations or the deviation, the reliability of the representative value regarding the observed values obtained by the plurality of observation devices.

Description

  The present invention relates to a technique that uses an index called reliability when threshold determination is performed using virtual sensor data calculated from sensor data with various accuracy and error distribution.

  For event detection processing in a general ubiquitous application, threshold value determination processing for an observation value included in sensor data is often used.

  Conventionally, it has been assumed that the sensor device used for threshold determination is predetermined. Estimating the accuracy of the observed value and the error range can be easily obtained, and the threshold value can be optimized by using them, and the probability that the observed value is within the error range can be calculated to realize stable threshold judgment. (See Patent Document 1).

  However, in an environment where the sensor device used for threshold determination and its area / position are not determined in advance, the accuracy of the observed value and the error range change every moment, and there is a problem that stable threshold determination is difficult.

  The environment targeted by the present invention is participatory sensing that allows a plurality of mobile users to collect sensor data. Consider a virtual sensor device that represents a certain region in a participatory sensing environment and perform threshold determination processing (see Non-Patent Document 1). Virtual sensor data representing a certain area can use statistical values such as a median value and an average value of sensor data for a plurality of sensor devices existing in that area at each time.

  Conventionally, as a method for estimating an observation value and its error range, it has been effective to perform sampling a plurality of times for one observation and use those statistical values. If there are correlations between a plurality of neighboring observation values, it is also effective to use these statistical values (see Non-Patent Document 2).

  However, virtual sensor data obtained in a participatory sensing environment in which a sensor device to be measured is not determined in advance has a feature that accuracy and an error range thereof dynamically change according to time. Further, the virtual sensor data representing a certain area has a feature that the area itself is variable according to a user's request.

JP-A-8-202970

Masanori Takeyama, 1 other, “Possibility of value co-creation model by participatory city sensing”, Journal “Information processing”, Vol. 51, no. 9, Information Processing Society of Japan, 2010 Yoshizawa Yasukazu, “New Error Theory”, Kyoritsu Publishing Co., Ltd., 1989 “Introductory Guide to Uncertainty”, 2004, Authorized Center for Product Evaluation Technology Foundation, 2004

  In the participatory sensing environment having the characteristics as described above, there is a problem described below in operating the threshold determination stably.

  Since the virtual sensor data is uncertain data, when performing threshold determination, not only whether the virtual sensor data exceeds the threshold, but also the error range of the virtual sensor data and the virtual sensor data is included in the section. It is desirable to consider the probability (see Non-Patent Document 3).

  However, since the sensor data set for calculating the virtual sensor data is not determined in advance, there is a problem that the probability of the calculated probability varies due to differences in sensor devices, observation positions, and observation time variations. Further, there is a problem that the probability of the virtual sensor data varies due to expansion and contraction of the area for calculating the virtual sensor data according to a user request.

  Therefore, in order to solve the above-described problem, the present invention evaluates the likelihood of virtual sensor data and threshold determination, and thus is more realistic even when the probability of virtual sensor data and threshold determination varies. The purpose is to make a proper threshold determination.

  In order to achieve the above objectives, the reliability of virtual sensor data is taken into account by introducing an index called reliability that quantifies differences in sensor devices and observation position and observation time while constructing virtual sensor data. It was decided to perform the probabilistic threshold judgment.

  Specifically, the present invention calculates a representative value for the observation values in a plurality of observation devices, calculates a representative value for the observation conditions in the plurality of observation devices, and varies the observation conditions from the representative values or A statistical processing unit that calculates the representative value from a predetermined value, and a representative value reliability calculation unit that calculates the reliability of the representative value for the observed values in the plurality of observation devices based on the variation or the method And an observation value reliability evaluation apparatus.

  In addition, the present invention calculates a representative value for the observation values in a plurality of observation devices, calculates a representative value for the observation conditions in the plurality of observation devices, and determines from a variation in observation conditions from the representative values or a predetermined value. A statistical processing procedure for calculating the representative value of the representative value, and a representative value reliability calculating procedure for calculating the reliability of the representative value for the observed values in the plurality of observation devices based on the variation or the method. It is an observation value reliability evaluation method characterized by providing in order.

  According to this configuration, the virtual sensor data is used to calculate the reliability of the representative values for the observation values in the plurality of observation devices based on the variations or methods even when the observation conditions in the plurality of observation devices are inconsistent. Can be considered.

  In addition, the present invention further includes a threshold determination reliability calculation unit that calculates a reliability of a representative value and a threshold value determination for the observation values in the plurality of observation devices based on the variation or the method. Is an observation value reliability evaluation apparatus.

  Further, the present invention provides a threshold value determination reliability calculation procedure for calculating a reliability value of a representative value and a threshold value determination of the observed value in the plurality of observation devices based on the variation or the method. An observation value reliability evaluation method characterized by further comprising a reliability calculation procedure.

  According to this configuration, even when the observation conditions in the plurality of observation apparatuses have variations and methods, the reliability of the determination of the representative value and the threshold value for the observation values in the plurality of observation apparatuses is calculated based on the variations and methods. Therefore, probabilistic threshold determination can be performed in consideration of the reliability of the virtual sensor data.

  Further, according to the present invention, the threshold determination reliability calculation unit is configured to represent a representative value for the observation value in the plurality of observation devices based on a variation in the observation value from the representative value for the observation value in the plurality of observation devices, and It is an observation value reliability evaluation apparatus characterized by calculating the reliability of threshold size determination.

  Further, according to the present invention, the threshold determination reliability calculation procedure is based on a variation in the observed value from the representative value for the observed value in the plurality of observation devices, and a representative value for the observed value in the plurality of observation devices, and This is an observation value reliability evaluation method characterized by calculating the reliability of the threshold size determination.

  According to this configuration, it is possible to consider not only whether the virtual sensor data exceeds the threshold value but also the probability that the virtual sensor data is included in the error range of the virtual sensor data.

  The present invention further includes a data grouping unit that classifies data observed by the plurality of observation devices into groups according to observation conditions in the plurality of observation devices, and the statistical processing unit and the representative value reliability The degree calculation unit is an observation value reliability evaluation apparatus characterized in that the data grouping unit executes processing for each data classified into each group.

  According to this configuration, virtual sensor data can be calculated for data observed by a plurality of observation devices in detail for each observation condition in the plurality of observation devices.

  Further, the present invention is an observation value reliability evaluation program for causing a computer to execute the observation value reliability evaluation method.

  According to this configuration, the observation value reliability evaluation program can be recorded on the recording medium or provided through the network.

  The present invention can perform more realistic threshold determination even when the probability of virtual sensor data and threshold determination varies by evaluating the likelihood of virtual sensor data and threshold determination.

It is a figure which shows the structure of a virtual sensor data structure apparatus. It is a figure which illustrates the contents of sensor data. It is a figure which illustrates the conditions of grouping. It is a figure which shows the process of a sensor data grouping function part. It is a figure which shows the process of the virtual sensor data structure function part with reliability. It is a figure which illustrates the contents of a sensor data group. It is a figure which illustrates the content of the statistical information of metadata. It is a figure which illustrates the content of the sensor data group with the statistical information of metadata. It is a figure which illustrates the content of the statistical information of an observation value. It is a figure which illustrates the content of the sensor data group with the statistical information of an observed value. It is a figure which illustrates the contents of virtual sensor data. It is a figure which illustrates the contents of virtual sensor data with reliability. It is a figure which shows the process of the threshold value determination function part with reliability. It is a figure which illustrates the process of the threshold determination function part with reliability. It is a figure which illustrates the contents of virtual sensor data with reliability.

  Embodiments of the present invention will be described with reference to the accompanying drawings. The embodiments described below are examples of the present invention, and the present invention is not limited to the following embodiments.

  A diagram showing the configuration of the virtual sensor data configuration apparatus is shown in FIG. A virtual sensor data composition device with reliability 100 that performs threshold determination with reliability includes a sensor data reading function unit 101, a virtual sensor data storage function unit 102, a setting reception function unit 103, a response function unit 104, and a sensor data grouping function unit. 105, a virtual sensor data configuration function unit 106 with reliability and a threshold determination function unit 111 with reliability are provided. The virtual sensor data configuration function unit 106 with reliability includes a metadata statistical processing function unit 107, an observation value statistical processing function unit 108, a virtual sensor data configuration function unit 109, and a reliability provision function unit 110.

  The virtual sensor data composing device 100 with reliability for performing threshold determination with reliability is connected to the sensor data storage device 112 in the sensor data reading function unit 101 and the virtual sensor data storage function unit 102, and the setting reception function unit 103 and the response The function unit 104 is connected to the application device 113. The sensor data storage device 112 can be easily created using a relational database management system (RDBMS) or the like.

  An overview of the virtual sensor data configuration apparatus 100 with reliability will be described. As a data grouping unit, the sensor data grouping function unit 105 classifies data observed by a plurality of observation devices into groups according to observation conditions of the plurality of observation devices. An observation device includes a temperature sensor and the like, and observation conditions include an observation position and an observation time.

  As a statistical processing unit, the metadata statistical processing function unit 107 calculates a representative value for observation conditions in a plurality of observation apparatuses, and calculates a variation in the observation conditions from the representative value or a representative value from a predetermined value. The representative value of the observation condition includes an average value of observation positions and observation times, and the predetermined value includes a center value in the group area. As the statistical processing unit, the observed value statistical processing function unit 108 calculates a representative value for the observed values in a plurality of observation apparatuses. The average value of temperature etc. are mentioned as a representative value of an observed value.

  As the representative value reliability calculation unit, the reliability providing function unit 110 calculates the reliability of the representative value for the observation values in the plurality of observation devices based on the above-described variation or the above. For example, the reliability assignment function unit 110 may evaluate that the reliability of the representative value of the observed value is higher as the variation in the observation condition is more uniform in the group, and the representative value of the observation condition in the group is the observation condition. It may be evaluated that the reliability of the representative value of the observed value is higher as it is closer to the center value of.

  The threshold determination function unit with reliability 111 calculates, as a threshold determination reliability calculation unit, the reliability of representative values and threshold determinations for observation values in a plurality of observation devices based on the above-described variation or the above. . For example, the threshold determination function unit with reliability 111 may evaluate that the reliability of the threshold determination is higher as the variation in the observation condition is more uniform in the group, and the representative value of the observation condition in the group is the observation condition. It may be evaluated that the closer to the center value, the higher the reliability of threshold determination.

  The threshold determination function unit with reliability 111 is a threshold determination reliability calculation unit, which is a representative value for observation values in a plurality of observation devices, based on variations in observation values from representative values for observation values in a plurality of observation devices. And the reliability of the threshold size determination is calculated. That is, the threshold value determination function unit 111 with reliability considers not only whether the representative value exceeds the threshold value but also the probability that the representative value is included in the error range of the representative value.

  Each configuration described in FIG. 1 can be realized by a computer and a program, and the program can be recorded on a recording medium or provided through a network.

  A diagram illustrating the contents of the sensor data is shown in FIG. The sensor data 200 is data described in pairs of attributes and values, and the attributes include observation positions 201 and 202, an observation time 203, and an observation value 204. The observation position includes at least one attribute, and there may be a plurality of observation positions such as latitude and longitude. Further, it is not limited to latitude and longitude, and may be, for example, an address section.

  The observation condition of the sensor data 200 is not limited to this, and both the observation position and the observation time may be included, or one of them may not be included. The other metadata may include, for example, a sensor type, a sensor model number, a network address, and the like.

  A diagram illustrating the grouping conditions is shown in FIG. The grouping condition 300 includes a sensor type 301 to be grouped, an observation start position 302 to be grouped, an observation end position 303 to be grouped, an observation start time 304 to be grouped, and an observation end time 305 to be grouped. The area represented by is defined.

  In the present embodiment, the start position and end position representing the latitude and longitude rectangles are used to specify the position range to be grouped. For example, the address section, the section in the building, the center and radius of the latitude and longitude, etc. Can be used. In this embodiment, the start time and end time are used to specify the time range to be grouped. For example, the start time and end time, or the start time and end time, and the grouping interval are specified. May be.

For example,
Start time = {2011-05-01 00:00:00}
End time = {2011-05-01 00:35:00}
Grouping interval = {00:10:00}
Is specified, grouping is performed at intervals of 10 minutes from 2011-05-01 00:00:00. as a result,
2011-05-01 00: 00: 00-00: 09: 59,
2011-05-01 00: 10: 00-00: 19: 59,
2011-05-01 00: 20: 00-00: 29: 59,
The remaining time groups may or may not be grouped. When a plurality of groups are specified, a plurality of grouping conditions may be listed.

  FIG. 4 is a diagram showing the processing of the sensor data grouping function unit. The sensor data grouping function unit 105 receives the grouping condition via the setting reception function unit 103, and generates a read request (step S401). The sensor data grouping function unit 105 transmits a read request to the sensor data read function unit 101 and receives the sensor data read from the sensor data storage device 112 (step S402).

  The sensor data grouping function unit 105 creates a group based on the sensor data grouping condition received in step S401 (step S403). Through steps S401 to S403, a sensor data group 600 as shown in FIG. 6 is created based on the sensor data grouping condition 300 shown in FIG.

  The sensor data grouping function unit 105 repeats steps S405 to S408 for each group (step S404). If the group is empty (“Yes” in step S405), steps S406 to S408 are omitted. If the group is not empty (“No” in step S405), steps S406 to S408 are executed.

  Details of the execution (step S406) of the virtual sensor data configuration function unit 106 with reliability will be described later. When step S406 is completed, the virtual sensor data configuration function unit with reliability 106 transmits virtual sensor data to the threshold determination function unit with reliability 111 (step S407), and the virtual sensor data is stored in the virtual sensor data storage function unit 102. Transmit (step S408). The virtual sensor data storage function unit 102 stores virtual sensor data in the sensor data storage device 112.

  FIG. 5 is a diagram showing processing of the virtual sensor data configuration function unit with reliability. When receiving the sensor data group 600, the virtual sensor data configuration function unit with reliability 106 corrects the missing data (step S501). Correction of missing data is performed by deleting data in which one or more of the observed value and metadata attributes are missing. The missing value 605 in FIG. 6 is an example in which only the observed value is missing. This line is deleted by correcting the missing data.

  The correction method is not limited to this. For example, data including values outside the range of values that can be measured by the sensor device is deleted, or missing data is compensated by using means such as Kalman filter, linear interpolation, spline interpolation, and kriging. It is possible to use means such as

  The metadata statistical processing function unit 107 receives the sensor data group 600 created in step S501, and calculates each statistical value of the observation position and the observation time. Then, the statistical data 700 of the metadata of the sensor data group 600 composed of each set of statistical values, and the sensor data group 800 that gives the statistical processing result of the metadata to each sensor data are displayed as an observed value statistical processing function unit. (Step S502).

  FIG. 7 shows an example of the contents of the statistical information of the metadata. In the metadata statistical information 700, an average value 701, a standard deviation 702, a skewness 703, a kurtosis 704, a normalized average value 705, and a normalized standard deviation 706 are used.

  The normalized average value 705 and the normalized standard deviation 706 are obtained in the unit region in which normalization is performed with the observation start point being −1 and the observation end point being 1 in each axis of the region defined by the spatial range and the time range. , Metadata mean and standard deviation. In addition, the statistical value 709 of the observation time is calculated as the number of elapsed seconds with the grouping start time 2011/5/1 01:55:00 (304) set to 0. The metadata statistical information 700 is not limited to these values, and may include, for example, a median value, unbiased variance, a centroid of values, and the like.

  FIG. 8 shows a diagram illustrating the contents of a sensor data group with metadata statistical information. A latitude deviation 801, a longitude deviation 802, and a time deviation 803 are given to the sensor data group 600 of FIG. The statistical processing result given to each sensor data is not limited to these attributes. For example, the distance from the center of gravity may be used.

  The observed value statistical processing function unit 108 receives the sensor data group 800 and the metadata statistical information 700 created in step S502, and calculates the statistical value of the observed value. Then, the statistical information 700 of the metadata created in step S502, the statistical information 900 of the observation value of the sensor data group 800 composed of a set of the statistical values of the observation values, and each of the metadata statistical processing results given in step S502 Further, the sensor data group 1000 to which the observation data statistical processing result is given to the sensor data is transmitted to the virtual sensor data configuration function unit 109 (step S503).

  A diagram illustrating the contents of the statistical information of the observed values is shown in FIG. In the observed value statistical information 900, an average value 901, a standard deviation 902, a skewness 903, and a kurtosis 904 are used.

  The observed statistical information 900 is not limited to these values, and may include, for example, a median value, unbiased variance, and the like. For the calculation of these statistical values, the metadata statistical information 700 and the statistical information 801 to 803 of each sensor data created in step S502 may be used. For example, the statistical information 900 of observation values may include a load average value given a weight using a deviation of metadata, a load average value given a weight using a distance from the center of gravity of metadata, and the like.

  A diagram illustrating the contents of a sensor data group with observation statistical information is shown in FIG. An observed value deviation 1001 is given to the sensor data group 800 of FIG.

  In this embodiment, calculation is performed in the order of the metadata statistical processing function unit 107 and the observed value statistical processing function unit 108 (steps S502 and S503). However, these calculation orders may be interchanged. Then, the statistical information 700 of the metadata may be calculated using the statistical information 900 of the observation value and the statistical information 1001 of each sensor data.

  The virtual sensor data configuration function unit 109 selects the representative value of the observed value, the error range of the observed value, and the representative value of the metadata from the statistical information calculated in steps S502 and S503, and generates virtual sensor data obtained by combining these values. Configure (step S504).

  A diagram illustrating the contents of the virtual sensor data is shown in FIG. The virtual sensor data 1100 includes a group ID 1101, a metadata representative value 1102-1104, an observed value representative value 1105, and an observed value error range 1106. The virtual sensor data 1100 is not limited to this configuration. For example, the virtual sensor data 1100 may not include all the representative values 1102-1104 of the metadata.

  As the representative value 1105 and the error range 1106 of the observed value, a value (95% confidence interval) twice the average value 901 and the standard deviation 902 of the observed value is used. The statistical value used for the representative value of the observed value and the error range is not limited to the set of the average value of the observed value and the standard deviation, and for example, the set of the load average value of the observed value and the load standard deviation can be used. Further, the value is not limited to twice the standard deviation 902 (95% confidence interval), but may be a value one time standard deviation 902 (68% confidence interval).

  The metadata average value 701 is used as the metadata representative value 1102-1104. As a representative value of metadata, not only an average value of metadata but also a median value of metadata or a weighted average value of metadata can be used, for example.

  The reliability degree assigning function unit 110 assigns a reliability degree to the virtual sensor data 1100 configured in step S504 (step S505). The reliability is a value obtained by quantifying two indexes, the condition and the variation degree based on the observation position and the observation time.

  In the present embodiment, it is assumed that the more reliable the observation conditions are scattered in the normalized unit region, the higher the reliability. Note that it may be assumed that the more the observation conditions are concentrated at the center of the normalized unit area, the higher the reliability.

  For quantification of the two indexes, metadata statistical information 700 and observation statistical information 900 are used. In this embodiment, a value is calculated from the distance from the center point of the unit area to the normalized average value 705, and a variation value is calculated using the normalized standard deviation 706.

  First, using the distance r between the center point of the unit area and the normalized average value 705 in each axis of the unit area, a value is calculated from the distance from the center point. In the present embodiment, the range that the distance r can take is 0 ≦ r ≦ √3. The distance r calculated from the metadata statistical information 700 is 0.254, and the value is 14.7% by dividing this by the maximum value that r can take.

  Next, a variation value is calculated using the normalized standard deviation σ_x (x = i, j, k) of each axis in the unit region. A rectangular parallelepiped having one side of 2σ_x (x = i, j, k) is assumed, and the ratio of the rectangular parallelepiped to the unit area is set as a variation value. In the present embodiment, the range that the volume v can take is 0 ≦ v ≦ 8. The volume v of the rectangular parallelepiped calculated from the statistical information 700 of the metadata is 0.0619, and when this is divided by the maximum value that v can take, the variation value is 0.773%.

  FIG. 12 shows an example of the contents of virtual sensor data with reliability. Therefore, the reliability is represented by binary values 1201 and variation values 1202. In addition, the calculation of the value and the variation value is not limited to the normalized average value and the normalized standard deviation. For example, the moment around the center in the unit area of the sensor data group, the skewness and kurtosis of the metadata, Entropy or the like in the unit area of the sensor data group may be used.

  Further, the scope of the present invention is not limited to the above, and for example, a four-dimensional distribution including the latitude, longitude, time, and observed value of metadata may be assumed, and the probability of entering the confidence interval may be used as the reliability. Also, weighting may be performed by calculating the density of the number of sensors. Further, the reliability may be synthesized into one value, or may be a vector expression composed of a plurality of values.

  The virtual sensor data configuration function unit 106 with reliability transmits the virtual sensor data 1200 to which the reliability is given to the threshold determination function unit 111 with reliability (step S506). Then, the virtual sensor data 1200 to which the reliability is given is stored in the sensor data storage device 112 via the virtual sensor data storage function unit 102 (step S507).

  FIG. 13 is a diagram showing the processing of the threshold determination function unit with reliability. FIG. 14 is a diagram illustrating the processing of the threshold determination function unit with reliability. FIG. 15 is a diagram illustrating the contents of virtual sensor data with reliability. FIG. 14 illustrates virtual sensor data values, virtual sensor data error ranges, virtual sensor data reliability, and application thresholds at times t1, t2, and t3. FIG. 15 illustrates the virtual sensor data 1400 of FIG.

  The application device 113 requests a determination result as to whether or not the virtual sensor data is equal to or greater than a threshold value. At time t1, it intersects with the threshold value of the application at ¼ point from the lower limit of the error range. At time t2, it intersects with the application threshold at 3/4 point from the lower limit of the error range. At time t3, no intersection has occurred.

  The threshold determination processing function unit with reliability 111 calculates threshold determination from the error range and threshold as the first-stage determination (step S1301). The error distribution of observations within a group is assumed to follow a normal distribution. The width of the error range of the measurement value x uses a 95% confidence interval. That is, when the average value μ and the standard deviation σ, the section (μ−2σ ≦ x ≦ μ + 2σ) is used.

  When it is determined that the threshold value a is exceeded, the probability that the measured value x exists between the threshold value and the error range upper limit (a ≦ x ≦ μ + 2σ) is calculated. At this time, 0% is assumed when (μ + 2σ ≦ a), and 100% is assumed when (a ≦ μ−2σ).

  When it is determined that the value falls below the threshold value a, the probability that the measured value x exists below the threshold value and below the error range lower limit (μ−2σ ≦ x ≦ a) is calculated. At this time, the case of (μ + 2σ ≦ a) is regarded as 100%, and the case of (a ≦ μ−2σ) is regarded as 0%.

  In the present embodiment, it is determined that the threshold value is exceeded. At time t1, the probability of being above the threshold and below the upper limit of the error range is 81.85%. At time t2, the probability of being above the threshold and below the upper limit of the error range is 13.59%. At time t3, the probability of being above the threshold and below the upper limit of the error range is 0%.

  The threshold value determination processing function unit with reliability 111 evaluates the correctness of the first-stage threshold determination result by using the reliability (value and variation value) as the second-stage determination (step S1302). That is, when the reliability of the representative value is low, the reliability of the error range is also considered to be low, so the first stage threshold determination result may not be correct.

  In the present embodiment, it is assumed that the reliability of the representative value of the observed value is high when the observation conditions are uniformly scattered in the normalized unit region. In other words, when the observation conditions are uniformly distributed in the unit area, the reliability of the representative value of the observation value is 100%, and when the sensor placement is local, the reliability of the representative value of the observation value is Consider it low.

  In the present embodiment, the variation value × (1−value) is calculated using the reliability assigned to the virtual sensor data. By multiplying by (1-value from), the observation condition is uniformly scattered in the unit region, and the closer the representative value of the observation condition is to the central value of the unit region, the more reliable the representative value of the observation value is If it is high, it can be regarded more reliably.

  The variation value × (1−value) is evaluated. The first-stage threshold determination result at time t1 is 0.02% reliability. The first-stage threshold determination result at time t2 has a reliability of 51.2%. The first stage threshold determination result at time t3 has a reliability of 0.59%.

  The threshold determination processing function unit with reliability 111 integrates the threshold determination using the first-stage determination result and the second-stage determination result (step S1303).

  In the present embodiment, if the probability of the first-stage determination result is greater than 0%, it is considered that “there is a possibility that the threshold is exceeded”, and the uncertainty of the first-stage determination result is expressed by the probability. Further, the correctness of the first stage determination result is expressed by the second stage determination result.

  If “the probability of the first stage determination result is 0% and its reliability is low”, it is considered that the threshold may be exceeded. Only when the “probability of exceeding the threshold is 0% and the reliability is high”, it is determined that the threshold is not exceeded.

  There is a possibility that the threshold is exceeded at time t1. The probability is 81.85% as judged in the first stage, and the reliability of the first stage is low as judged in the second stage. There is a possibility that the threshold value is exceeded at time t2. The probability is 13.59% as determined in the first stage, and the reliability of the first stage is high as determined in the second stage. At time t3, the threshold may be exceeded. The probability is 0% as determined in the first stage, but the reliability in the first stage is low as determined in the second stage, so it is considered that the threshold may be exceeded.

  The threshold value determination processing function unit with reliability 111 adds reliability to the determination result obtained in step S1303, and transmits the result to the application apparatus 113 via the response function unit 104 (step S1304). In this embodiment, the threshold value determination processing function unit 111 with reliability assigns binary values 1201 and a variation value 1202 as reliability.

  The observed value reliability evaluation apparatus, the observed value reliability evaluation method, and the observed value reliability evaluation program according to the present invention include, for example, a plurality of sensor devices used for threshold determination and an environment in which the region and position thereof are not determined in advance. It can be applied when performing a stable threshold determination in a participatory sensing environment that allows mobile users to collect sensor data.

100: Virtual sensor data configuration device 101: Sensor data read function unit 102: Virtual sensor data storage function unit 103: Setting reception function unit 104: Response function unit 105: Sensor data grouping function unit 106: Virtual sensor data configuration with reliability Function unit 107: Metadata statistical processing function unit 108: Observation value statistical processing function unit 109: Virtual sensor data configuration function unit 110: Reliability assignment function unit 111: Threshold determination function unit with reliability 112: Sensor data storage device 113: Application device

Claims (8)

A representative value is calculated for observation values in a plurality of observation devices, a representative value is calculated for observation conditions in the plurality of observation devices, and a variation in observation conditions from the representative value or a representative value from a predetermined value is calculated. A statistical processing unit to calculate,
Based on the variation or the method, a representative value reliability calculation unit that calculates the reliability of the representative value for the observation values in the plurality of observation devices;
An observation value reliability evaluation apparatus comprising: Based on the variation or the method, a threshold determination reliability calculation unit that calculates the reliability of the determination of the representative value and the threshold value for the observation values in the plurality of observation devices,
The observation value reliability evaluation apparatus according to claim 1, further comprising:   The threshold determination reliability calculation unit is configured to determine the reliability of the representative value and the threshold value for the observation value in the plurality of observation devices based on the variation in the observation value from the representative value for the observation value in the plurality of observation devices. The observation value reliability evaluation apparatus according to claim 2, wherein the degree is calculated. According to the observation conditions in the plurality of observation devices, further comprising a data grouping unit that classifies the data observed by the plurality of observation devices into groups,
4. The method according to claim 1, wherein the statistical processing unit and the representative value reliability calculation unit execute processing on each data classified into each group by the data grouping unit. 5. Observation value reliability evaluation equipment. A representative value is calculated for observation values in a plurality of observation devices, a representative value is calculated for observation conditions in the plurality of observation devices, and a variation in observation conditions from the representative value or a representative value from a predetermined value is calculated. Statistical processing procedure to calculate,
A representative value reliability calculation procedure for calculating the reliability of the representative value for the observed value in the plurality of observation devices based on the variation or the method;
In order, the observation value reliability evaluation method. Based on the variation or the method, a threshold determination reliability calculation procedure for calculating the reliability of the representative value and the threshold determination of the observation value in the plurality of observation devices,
The observation value reliability evaluation method according to claim 5, further comprising: after the representative value reliability calculation procedure.   The threshold determination reliability calculation procedure is based on the variation in the observed value from the representative value for the observed value in the plurality of observation devices, and the reliability of the determination of the representative value and the threshold value for the observation value in the plurality of observation devices. The observed value reliability evaluation method according to claim 6, wherein the degree is calculated.   An observation value reliability evaluation program for causing a computer to execute the observation value reliability evaluation method according to any one of claims 5 to 7.

Images