Information processing methods

Abstract

To provide an information processing method that can find out a setting in which extracted data capturing a feature of the entire original data can be obtained.SOLUTION: An information processing method includes: a first step (S120) of calculating a relative frequency distribution in original data for each feature amount; a second step (S125) of setting a plurality of time windows to cut out data corresponding to part of a period from the original data; a third step (S130) of cutting out the data from the original data; a fourth step (S140) of calculating a relative frequency distribution in extracted data; and a fifth step (S145) of calculating an error between the relative frequency distribution in the original data and the relative frequency distribution in the extracted data. An information processor repeatedly executes trials from the second step to the fifth step while changing the setting of the plurality of time windows after executing the first step, and selects and outputs settings of the plurality of time windows in which the error is less than a threshold value.SELECTED DRAWING: Figure 3

Description

Technical Field

[0004] , , , , ,

[0006] , , , , , ,

[0005] , , , , , ,

[0001] This invention relates to an information processing method.

Background Art

[0002] Patent Document 1 discloses an information processing apparatus that reduces the data size by compressing original data for analysis. The original data for analysis is data collected over a predetermined period using sensors mounted on a vehicle.

[0003] The information processing apparatus disclosed in Patent Document 1 compresses the data by extracting, from the original data, the data acquired when a certain vehicle speed is reached and the data acquired at the inflection point of the vehicle speed.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The above information processing apparatus extracts data focusing only on the vehicle speed. Therefore, the above information processing apparatus cannot extract data according to the characteristics of data other than the vehicle speed. There is a need for an information processing method capable of obtaining extraction data that captures the characteristics of the entire original data.

Means for Solving the Problems

[0006] Hereinafter, means for solving the above problems and their effects will be described. The information processing method to solve the above problem is one that reduces the amount of data used for analysis by extracting some data from the original data collected over a predetermined period using multiple sensors mounted on the vehicle.

[0007] This information processing method includes a first step in which the information processing device calculates the relative frequency distribution in the original data for each of the multiple features contained in the original data.

[0008] This information processing method includes a second step in which the information processing device sets up multiple time windows that extract data for a portion of the original data such that the sum of the periods of all time windows is shorter than the predetermined period.

[0009] This information processing method includes a third step in which the information processing device extracts data from the original data using the plurality of time windows. This information processing method includes a fourth step in which the information processing device calculates the relative frequency distribution for each feature in the extracted data obtained by combining all the data extracted by the plurality of time windows.

[0010] This information processing method includes a fifth step in which the information processing device calculates the error between the relative frequency distribution in the original data and the relative frequency distribution in the extracted data.

[0011] This information processing method involves the information processing device, after executing the first step, repeatedly performing the trials from the second to the fifth step by changing the settings of the multiple time windows, selecting the setting of the multiple time windows in which the error is less than a threshold, and outputting it.

[0012] In one embodiment of the information processing method, the information processing device further includes a step of performing clustering, which is a machine learning method, in which the data in each interval, obtained by dividing the original data into fixed period intervals, is classified into a predetermined number of clusters. In this information processing method, the information processing device sets the plurality of time windows in the second step such that the ratio of each cluster in the extracted data is equal to the ratio of each cluster in the original data as a whole. [Effects of the Invention]

[0013] According to the above information processing method, it is possible to find settings using an information processing device that allow for the acquisition of extracted data that captures the characteristics of the entire original data. [Brief explanation of the drawing]

[0014] [Figure 1] Figure 1 is a schematic diagram showing the relationship between a data center that executes information processing methods, a vehicle, and an information processing terminal. [Figure 2] Figure 2 is a graph showing the original data, where (a) shows the change in vehicle speed, (b) shows the change in tilt angle, and (c) shows the change in acceleration. [Figure 3] Figure 3 is a flowchart showing the processing flow performed by the processing unit in the data center. [Figure 4] Figure 4 is a graph showing an example of clustering the original data using two features. [Figure 5] Figure 5 is a graph showing an example of the relative frequency distribution for vehicle speed in the original data. [Figure 6] Figure 6 is a graph showing an example of the relative frequency distribution of slope angles in the original data. [Figure 7] Figure 7 is a flowchart showing a portion of the processing flow executed by a processing unit that implements the modified information processing method. [Modes for carrying out the invention]

[0015] Hereinafter, an embodiment of an information processing method and an information processing apparatus that executes the information processing method will be described with reference to FIGS. 1 to 6. <Configuration of Information Processing System> FIG. 1 shows the configuration of an information processing system including a data center 500 including an information processing apparatus according to an embodiment. As shown in FIG. 1, the data center 500 communicates with a plurality of vehicles 10 via a communication network 400. The data center 500 also communicates with an information processing terminal 600 via the communication network 400.

[0016] <Configuration of Data Center 500> As shown in FIG. 1, the data center 500 includes a processing apparatus 510 as an information processing apparatus. The data center 500 includes a storage device 520 and a communication device 530. The processing apparatus 510 includes a CPU that executes processing according to a program and a ROM in which the program is stored. The storage device 520 stores a large amount of data. The communication device 530 is implemented as hardware such as a network adapter, various communication software, or a combination thereof. The communication device 530 realizes wired or wireless communication via the communication network 400.

[0017] The data center 500 can be configured using a plurality of computers. For example, the data center 500 can be configured by a plurality of server devices. <Configuration of Vehicle 10> Each of the plurality of vehicles 10 includes a communication device 80. These communication devices 80 are implemented as hardware such as a network adapter, various communication software, or a combination thereof. These communication devices 80 are configured to be able to realize wired or wireless communication via the communication network 400. Each vehicle 10 is equipped with various sensors that collect information on each part of the vehicle 10.

[0018] In each vehicle 10, driving data is collected from such various sensors. From each vehicle 10, the driving data is transmitted to the data center 500 by the communication device 80. For example, the driving data including the driving distance, position information, and vehicle speed of each vehicle 10 is transmitted to the data center 500. The identification information for identifying each vehicle 10 is also transmitted to the data center 500 together with the driving data.

[0019] The data center 500 stores the driving data in the storage device 520 together with the received identification information. Thus, the driving data of a plurality of vehicles 10 is accumulated in the storage device 520 of the data center 500.

[0020] <Configuration of the information processing terminal 600> The information processing terminal 600 includes a processing device 610, a storage device 620, and a communication device 630. The processing device 610 includes a CPU that executes processing according to a program and a ROM in which the program is stored. The storage device 620 stores a large amount of data. The communication device 630 is implemented as hardware such as a network adapter, various communication software, or a combination thereof. The communication device 630 realizes wired or wireless communication via the communication network 400.

[0021] The information processing terminal 600 is, for example, a personal computer. <Regarding the analysis of the driving data of the vehicle 10> The information processing terminal 600 is used for the task of analyzing the driving data. When analyzing the driving data, a part of the driving data is downloaded from the huge amount of driving data stored in the storage device 520 of the data center 500 to the storage device 620 of the information processing terminal 600. The driving data to be downloaded is selected according to the purpose of the analysis. The processing device 610 performs analysis using the downloaded driving data.

[0022] For example, the processing unit 610 calculates the load on a specific part of the vehicle 10 based on driving data. Based on the calculated load, the processing unit 610 estimates the damage accumulated in that part. For example, the processing unit 610 performs an analysis to estimate the required durability of a specific part of the vehicle 10 based on driving data from multiple vehicles 10.

[0023] To perform this analysis, the processing unit 610 analyzes a large amount of driving data collected over a long period of time. Because the processing unit 610 needs to perform a massive amount of calculations, the analysis takes a long time.

[0024] Therefore, it is conceivable to extract data that captures the characteristics of the entire original data from the large amount of original driving data. If such extracted data can be obtained, the processing unit 610 can perform analysis in a shorter time by using the extracted data. For example, when estimating the damage to a part after 100,000 hours of driving, the processing unit 610 estimates the damage using 20,000 hours of extracted data extracted from 100,000 hours of original data. Then, the processing unit 610 calculates an estimated value of the damage to the part after 100,000 hours of driving by multiplying the estimated value calculated from the 20,000 hours of extracted data by 5.

[0025] Figure 2 shows an example of the original data. The original data shown in Figure 2 represents 100,000 hours of driving data from a single vehicle 10. The original data shown in Figure 2 includes vehicle speed, incline angle, and acceleration as features.

[0026] Figure 2(a) shows the change in vehicle speed over 100,000 hours. Figure 2(b) shows the change in the inclination angle of vehicle 10 over 100,000 hours. The inclination angle is positive when going uphill. The inclination angle is negative when going downhill. Figure 2(c) shows the acceleration of vehicle 10. The acceleration is positive when vehicle 10 is accelerating. The acceleration is negative when vehicle 10 is decelerating.

[0027] Vehicle speed, tilt angle, and acceleration are correlated with the load on the vehicle 10's powertrain. The processing unit 610 of the information processing terminal 600 estimates the load on the components constituting the powertrain from driving data that includes vehicle speed, tilt angle, and acceleration as features.

[0028] When the data center 500 transmits data for analysis to the information processing terminal 600, it transmits not only the original data but also information for extracting data from the original data.

[0029] Extracted data is created by cutting out data from the original data using multiple time windows. Figure 2 shows five time windows as an example: the first time window W_1, the second time window W_2, the third time window W_3, the fourth time window W_4, and the fifth time window W_5, each indicated by a dashed line. The start and end dates of each time window are set so that they do not overlap. In this example, 20,000 hours of driving data are extracted as data. Therefore, the start and end dates of each time window are set so that the total length of the periods of all time windows W_1 to W_5 is 20,000 hours.

[0030] The data center 500 searches for the start and end dates of each time window that represent a cutting pattern for extracting extracted data that captures the characteristics of the entire original data. The data center 500 transmits the information on the cutting pattern for extracting the extracted data, along with the original data, to the information processing terminal 600. In other words, the information on the cutting pattern for extracting the extracted data, which is transmitted along with the original data, is the information on the settings of each time window found through the search.

[0031] The information processing terminal 600, having received the original data along with information on extraction patterns for extracting data from the original data, sets the extraction patterns based on the received information. Then, the information processing terminal 600 extracts the data from the original data using the set extraction patterns. The information processing terminal 600 then performs analysis using the extracted data.

[0032] <Search process for extraction patterns> Figure 3 is a flowchart showing the sequence of processes related to the extraction pattern search process. This sequence of processes is performed by the processing unit 510 of the data center 500.

[0033] As shown in Figure 3, the processing unit 510 acquires original data in step S100. The original data is a portion of the driving data selected from the vast amount of driving data stored in the storage device 520 of the data center 500 according to the purpose of the analysis. For example, the original data for calculating the load on a specific part of a single vehicle 10 is the driving data of the target vehicle 10 over a predetermined period, selected from the vast amount of driving data of multiple vehicles 10.

[0034] In step S110, the processing unit 510 assigns labels to the original data by clustering. Specifically, the processing unit 510 divides the original data into fixed-period intervals. The length of the interval for dividing the original data is, for example, several minutes. Then, the processing unit 510 performs clustering, a machine learning method that classifies the data in each interval into a predetermined number of clusters. The clustering algorithm used is, for example, the k-means method. The k-means method is a clustering algorithm that classifies data into a predetermined number of clusters. The clustering algorithm is not limited to the k-means method.

[0035] The original data includes driving data collected under different conditions, such as driving in urban areas, driving in suburban areas, and driving on highways. By performing clustering, the driving data included in the original data can be classified into clusters of driving data with similar characteristics. The number of clusters to be classified can be arbitrarily set depending on the content of the analysis.

[0036] Figure 4 is a graph showing an example of clustering the original data into four clusters using the k-means method, with two features included in the original data as explanatory variables. In Figure 4, each data point in each section of the original data is represented by a single point. When performing clustering, the processing unit 510 uses representative values ​​of the explanatory variables in the data for each section. For example, the processing unit 510 uses the mean value of the features in the data for each section as the representative value. The processing unit 510 may also use the moving average of the features over multiple consecutive time-series intervals as the representative value.

[0037] In Figure 4, these points are shown on a two-dimensional space with the first feature FV_a and the second feature FV_b as the coordinate axes. Figure 4 is an example of the original data being clustered into four clusters: the first cluster M_1, the second cluster M_2, the third cluster M_3, and the fourth cluster M_4. In Figure 4, the boundaries of the four clusters are shown by solid lines. In Figure 4, the centroids of each cluster are indicated by white triangles. Centroid cgM_1 is the centroid of the first cluster M_1. Centroid cgM_2 is the centroid of the second cluster M_2. Centroid cgM_3 is the centroid of the third cluster M_3. Centroid cgM_4 is the centroid of the fourth cluster M_4.

[0038] Figure 4 shows an example with two explanatory variables, but the number of explanatory variables is not limited to two. For example, as shown in Figure 2, if the original data includes three features: vehicle speed, tilt angle, and acceleration, the processing unit 510 may perform clustering using these three features as explanatory variables. In that case, the processing unit 510 will cluster the original data in a three-dimensional coordinate space.

[0039] The processing unit 510 then assigns labels to the original data indicating the clustering results. Specifically, it assigns a label to each data point, which was represented as a point in coordinate space, to identify the classified cluster. In this way, the processing unit 510 creates the original data with labels.

[0040] Next, in step S120, the processing unit 510 calculates the relative frequency distribution of the original data. As mentioned above, the original data contains multiple features. The processing unit 510 calculates the relative frequency distribution of each feature in the original data.

[0041] A frequency distribution classifies data into multiple classes and represents the distribution of the number of data points in each class. Relative frequency indicates what percentage of the total sum of frequencies a particular class represents.

[0042] Figure 5 shows the relative frequency distribution for vehicle speed in the original data shown in Figure 2. In this relative frequency distribution, the vehicle speed in the original data is divided into m classes from 1 to m, and the relative frequency distribution is shown for each class.

[0043] Figure 6 shows the relative frequency distribution for the slope angle in the original data shown in Figure 2. In this relative frequency distribution, the slope angle in the original data is divided into m classes from 1 to m, and the relative frequency distribution is shown for each class.

[0044] In step S120, the processing unit 510 calculates this relative frequency distribution for each feature contained in the original data. The number of classes in the relative frequency distribution of each feature is the same.

[0045] For example, as shown in Figure 2, if the original data includes three features—vehicle speed, tilt angle, and acceleration—the processing unit 510 calculates the relative frequency distribution of each of these three features.

[0046] Next, in the processing of step S125, the processing unit 510 sets multiple time windows in order to extract extracted data from the original data. Figure 2 shows five time windows W_1 to W_5 as an example of multiple time windows: the first time window W_1, the second time window W_2, the third time window W_3, the fourth time window W_4, and the fifth time window W_5. In the example shown in Figure 2, the duration of each time window is equal. As shown in Figure 2, the data extracted by each extraction window is the data for each feature over the same period.

[0047] In step S125, the processing unit 510 randomly sets multiple time windows such that the sum of the periods of all time windows is shorter than the default period, which is the period of the entire original data. As will be described later, the processing unit 510 combines all the data extracted using the multiple time windows set here to create extracted data. The sum of the periods of all time windows is a value that determines the capacity of the extracted data. Therefore, the sum of the periods of all time windows is set in advance.

[0048] For example, each time the processing unit 510 executes the process in step S125, it randomly sets the number of time windows, the start date of each time window, and the end date of each time window. At this time, the processing unit 510 sets each time window so that they do not overlap. In this way, the processing unit 510 randomly sets multiple time windows so that the sum of the periods of all time windows equals a predetermined period. In the process in step S125, the processing unit 510 may set multiple time windows by fixing the period of each time window to a constant value, as shown in Figure 2. In the process in step S125, the processing unit 510 may set multiple time windows by fixing the number of time windows to a constant value.

[0049] In addition to the requirements described above, when the processing unit 510 sets multiple time windows through step S125, it sets multiple time windows such that the ratio of each cluster in the extracted data is equal to the ratio of each cluster in the original data as a whole.

[0050] In this way, by setting multiple time windows through the process in step S125, an extraction pattern for extracting data from the original data is determined. Once the extraction pattern is determined, the processing unit 510 proceeds to step S130.

[0051] In step S130, the processing unit 510 extracts data from the original data according to the determined extraction pattern. In other words, in step S130, the processing unit 510 extracts data from the original data according to multiple set time windows. Then, the processing unit 510 combines all the data extracted according to the multiple time windows to create extracted data.

[0052] In the next step, S140, the processing unit 510 calculates the relative frequency distribution of the extracted data. The processing unit 510 calculates the relative frequency distribution of the extracted data in the same way as the method used to calculate the relative frequency distribution in step S120. That is, in the processing of step S140, the processing unit 510 calculates the relative frequency distribution of the extracted data for each feature. At this time, the processing unit 510 makes the number of classes in the relative frequency distribution of each feature the same as the relative frequency distribution in step S120.

[0053] For example, as shown in Figure 2, if the original data includes three features: vehicle speed, tilt angle, and acceleration, the processing unit 510 calculates the relative frequency distribution of each of these three features in step S140.

[0054] Next, in step S145, the processing unit 510 calculates the error between the relative frequency distribution in the original data and the relative frequency distribution in the extracted data. For example, the processing unit 510 calculates the Mean Absolute Error (MAE). The Mean Absolute Error (MAE) is expressed by the following formula.

[0055]

number

[0056] In the above formula, "n" is the number of features. "m" is the number of classes in the relative frequency distribution. "Y" is the frequency of the corresponding feature in the corresponding class in the original data. "y" is the frequency of the corresponding feature in the corresponding class in the extracted data.

[0057] As shown in the formula above, the processing unit 510 calculates the error as the sum of the frequency errors in each class for each feature between the relative frequency distribution in the entire original data and the relative frequency distribution in the extracted data.

[0058] Once the error is calculated, the processing unit 510 proceeds to step S150. In step S150, the processing unit 510 determines whether the calculated error is less than or equal to a threshold. The threshold is a value used to determine whether extracted data having a relative frequency distribution similar to the relative frequency distribution in the original data has been extracted using the set extraction pattern. The magnitude of this threshold is pre-set so that, based on the error being less than or equal to the threshold, it can be determined that extracted data having a relative frequency distribution similar to the relative frequency distribution in the original data has been extracted.

[0059] In step S150, if it is determined that the error is below a threshold (step S150: YES), the processing unit 510 proceeds to step S160. In step S160, the processing unit 510 records the extraction pattern. Specifically, the processing unit 510 stores the start and end data of each time window in the extraction pattern in the storage device 520 as information that identifies the extraction pattern. After recording the extraction pattern in this way, the processing unit 510 proceeds to step S170.

[0060] On the other hand, if the processing in step S150 determines that the error is greater than the threshold (step S150: NO), the processing unit 510 proceeds to step S170 without executing the processing in step S160. In other words, the processing unit 510 proceeds to step S170 without recording the extraction pattern.

[0061] In step S170, the processing unit 510 determines whether the number of trials is equal to or greater than a predetermined number. The number of trials is the number of times the trials consisting of the processes in steps S125 to S145 have been executed. These trials are trials to search for extraction patterns for extracting data. The predetermined number is a threshold for determining whether a sufficient number of trials to search for extraction patterns for extracting data have been executed. This predetermined number is set in advance so that it can be determined that the extraction patterns have been sufficiently searched based on whether the number of trials is equal to or greater than a predetermined number.

[0062] If, during the process in step S170, it is determined that the number of trials is less than the predetermined number (step S170: NO), the processing unit 510 returns to step S125 and performs the trials again. That is, the processing unit 510 sets multiple new time windows and performs trials to extract the data.

[0063] On the other hand, if the processing in step S170 determines that the number of trials is equal to or greater than a predetermined number (step S170: YES), the processing unit 510 proceeds to step S180.

[0064] The processing unit 510 repeats the trial until the number of trials reaches a predetermined number. The extracted pattern whose error is below the threshold is then stored in the storage device 520. When the number of trials reaches the predetermined number, the processing unit 510 proceeds to step S180.

[0065] In step S180, the processing unit 510 outputs the extraction pattern stored in the storage device 520. Specifically, the processing unit 510 sends the time window setting in which the error is below a threshold, which is recorded in the storage device 520, to the information processing terminal 600 as a candidate for the extraction pattern. After outputting the extraction pattern in step S180, the processing unit 510 terminates this series of processes.

[0066] In this way, through this series of processes, the processing unit 510 selects and outputs all of the time window settings from the trialed time window settings in which the error is less than or equal to a threshold. <Operation of this embodiment> As described above, the processing unit 510 of the data center 500 executes an information processing method to search for settings that reduce the amount of data used for analysis.

[0067] The information processing method of this embodiment includes a first step (step S120) in which the processing unit 510 calculates the relative frequency distribution in the original data for each of the multiple features contained in the original data. The information processing method includes a second step (step S125) in which the processing unit 510 sets multiple time windows such that the sum of the periods of all time windows is shorter than a predetermined period. The information processing method includes a third step (step S130) in which the processing unit 510 extracts data from the original data using the multiple time windows. The information processing method includes a fourth step (step S140) in which the processing unit 510 calculates the relative frequency distribution in the extracted data, obtained by combining all the data extracted using the multiple time windows, for each of the features. The information processing method includes a fifth step (step S145) in which the processing unit 510 calculates the error between the relative frequency distribution in the original data and the relative frequency distribution in the extracted data.

[0068] In this information processing method, after the processing unit 510 executes the first step, it repeatedly performs the trials from the second to the fifth step, changing the settings of multiple time windows. Then, the processing unit 510 selects and outputs the settings of multiple time windows that result in an error below a threshold.

[0069] In this information processing method, the processing unit 510 calculates the error between the relative frequency distribution of each feature in the entire original data and the relative frequency distribution of each feature in the extracted data. The processing unit 510 repeats this error calculation while changing the settings of multiple time windows used to extract data from the original data. The processing unit 510 then selects and outputs the setting that results in an error below a threshold.

[0070] This information processing method allows the processing unit 510 to search for a time window setting. The setting output by the processing unit 510 using this information processing method is a setting that allows extracting data in which the relative frequency distribution of each feature is similar to that of the entire original data.

[0071] <Effects of this embodiment> (1) According to the above information processing method, it is possible to find a setting that can obtain extracted data that captures the characteristics of the entire original data.

[0072] (2) In the above example, the original data includes vehicle speed, tilt angle, and acceleration as features. In this case, the processing unit 510 calculates the relative frequency distribution for vehicle speed in the original data in the first step above. In this case, the processing unit 510 calculates the relative frequency distribution for tilt angle in the original data in the first step above. In this case, the processing unit 510 calculates the relative frequency distribution for acceleration in the original data in the first step above.

[0073] In this case, the processing unit 510 calculates the relative frequency distribution for vehicle speed in the extracted data in the fourth step. In this case, the processing unit 510 calculates the relative frequency distribution for tilt angle in the extracted data in the fourth step. In this case, the processing unit 510 calculates the relative frequency distribution for acceleration in the extracted data in the fourth step.

[0074] Vehicle speed, tilt angle, and acceleration are correlated with the load on the vehicle 10's powertrain. By obtaining extracted data including vehicle speed, tilt angle, and acceleration, the load on the vehicle 10's powertrain can be estimated with less data.

[0075] This allows the processing unit 510 to find a setting that can obtain extracted data used to estimate the load of the vehicle 10's powertrain. (3) The above information processing method includes a step (step S110) in which the processing unit 510 performs clustering, which is a machine learning method that classifies the data in each interval, obtained by dividing the original data into fixed period intervals, into a predetermined number of clusters. The above information processing method also includes a step (step S125) in which the processing unit 510 sets multiple time windows such that the ratio of each cluster in the extracted data is equal to the ratio of each cluster in the original data as a whole.

[0076] Multiple intervals classified into the same cluster are intervals with similar characteristics. The settings output from the processing unit 510 by the above information processing method are settings that allow for the extraction of data where the ratio of each cluster to the entire original data is equal, and the relative frequency distribution of each feature is similar.

[0077] Therefore, using the information processing method described above, it is possible to find settings that allow us to obtain extracted data that more closely resembles the characteristics of the entire original data. (4) In the above information processing method, the processing unit 510 selects and outputs all of the time window settings in which the error is less than or equal to a threshold value from among the time window settings in which the trial was performed.

[0078] The above information processing method outputs all extraction patterns whose error is below a threshold. This allows for comparison of the multiple outputted extraction patterns and selection of which extraction pattern to ultimately adopt.

[0079] <Example of changes> This embodiment can be implemented with the following modifications. This embodiment and the following modifications can be combined with each other to the extent that they do not contradict each other technically.

[0080] The example shown illustrates how the processing unit 510 sets multiple time windows such that the ratio of each cluster in the original data is equal to the ratio of each cluster in the extracted data. However, the processing unit 510 may set multiple time windows without such constraints. In this case, the clustering step S110 may be omitted.

[0081] In the above information processing method, in the fifth step, the processing unit 510 calculates the interval errors, which are the differences in relative frequencies for each class between the relative frequency distribution of the original data and the relative frequency distribution of the extracted data, and simply calculates their sum. Alternatively, the calculated interval errors may be adjusted by multiplying them by weights, and then the sum of the interval errors may be calculated as the error.

[0082] In this case, as shown in Figure 7, the processing unit 510 calculates the relative frequency distribution of the extracted data in step S140, and then proceeds to step S142. In step S142, the processing unit 510 calculates the interval error, which is the difference in relative frequencies for each class between the relative frequency distribution of the original data and the relative frequency distribution of the extracted data.

[0083] Next, in step S144, the processing unit 510 adjusts the interval error for each class by multiplying the calculated interval error by a weight for each class. Then, in step S146, the processing unit 510 calculates the sum of the adjusted interval errors as the error. The processing unit 510 then uses the error calculated in this way to execute the processes from step S150 onward.

[0084] Depending on the features, there may be a significant bias in the frequency of occurrence for each class. As described above, by using an information processing method that adjusts each interval error by multiplying it by a weight, the influence of the bias in frequency of occurrence can be suppressed and the error can be calculated by adjusting the weight of the interval error.

[0085] The method for determining the time window setting in the cropping pattern does not have to be random. The setting of the time window in the cropping pattern can be changed according to a pre-defined rule, and the trial can be repeated.

[0086] The error calculated in step S145 is not limited to the mean absolute error (MAE). For example, the processing unit 510 may calculate the mean squared error as the error. The processing unit 510 may also calculate the root mean squared error as the error.

[0087] In the information processing method of the above embodiment, all extraction patterns stored in the storage device 520 were output during the processing of step S180. In contrast, the information processing method does not have to output all extraction patterns. For example, the information processing method may select and output the extraction pattern with the smallest error from among the extraction patterns stored in the storage device 520.

[0088] In the above embodiment, an example was shown in which the information processing device is implemented as the processing device 510 of the data center 500. Alternatively, the information processing device may be implemented as the processing device 610 of the information processing terminal 600. In this case, the processing device 610 downloads the original data from the data center 500 and executes the series of processes shown in Figures 3 and 7. [Explanation of symbols]

[0089] 10...Vehicle, 80...Communication equipment, 400...Communication network, 500...Data center, 510...Processing device, 520...Storage device, 530...Communication equipment, 600...Information processing terminal, 610...Processing device, 620...Storage device, 630...Communication equipment

Claims

1. This is an information processing method that reduces the amount of data used for analysis by extracting some data from the original data collected over a predetermined period using multiple sensors mounted on the vehicle. The information processing device performs a first step of calculating the relative frequency distribution in the original data for each of the multiple features contained in the original data, The information processing device performs a second step of setting up multiple time windows that extract data for a portion of the original data such that the sum of the periods of all time windows is shorter than the predetermined period. The information processing device performs a third step of extracting data from the original data using the multiple time windows, The information processing device performs a fourth step in which it calculates the relative frequency distribution for each feature in the extracted data obtained by combining all the data extracted by the plurality of time windows, The information processing device includes a fifth step of calculating the error between the relative frequency distribution in the original data and the relative frequency distribution in the extracted data, An information processing method in which, after the information processing device has performed the first step, it repeatedly performs the trials from the second step to the fifth step by changing the settings of the plurality of time windows, selects a setting of the plurality of time windows in which the error is less than a threshold, and outputs it.

2. The aforementioned multiple feature quantities include vehicle speed, tilt angle, and acceleration. In the first step, the information processing device calculates the relative frequency distribution for vehicle speed in the original data, the relative frequency distribution for tilt angle in the original data, and the relative frequency distribution for acceleration in the original data. In the fourth step, the information processing device calculates the relative frequency distribution for vehicle speed in the extracted data, the relative frequency distribution for tilt angle in the extracted data, and the relative frequency distribution for acceleration in the extracted data. The information processing method according to claim 1.

3. The information processing device further includes the step of performing clustering, which is a machine learning method that classifies the data in each interval obtained by dividing the original data into fixed period intervals, into a predetermined number of clusters. In the second step, the information processing device sets the multiple time windows such that the ratio of each cluster in the extracted data is equal to the ratio of each cluster in the original data as a whole. The information processing method according to claim 1.

4. The information processing device selects and outputs all of the time window settings used in the trial in which the error is less than or equal to the threshold. The information processing method according to claim 1.

5. In the fifth step, the information processing device calculates interval errors, which are the differences in relative frequencies for each class between the relative frequency distribution of the original data and the relative frequency distribution of the extracted data. After adjusting each calculated interval error by multiplying it by a weight, the device calculates the sum of the interval errors as the error. The information processing method according to claim 1.

Images