Downsampling method

Abstract

To provide a downsampling method which accurately performs downsampling of traveling data.SOLUTION: A traveling data downsampling method using machine learning includes steps of: acquiring traveling data; calculating a predetermined feature quantity of the traveling data; setting time length or a position of a window on the basis of distribution density of the traveling data; generating cutout data by using the window from the traveling data; calculating an error between appearance frequencies of the feature quantity of the traveling data and the feature quantity of the cutout data; and selecting the cutout data in a manner to make the error a predetermined value or less.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a downsampling method. [Background Art]

[0002] When a vehicle travels, if travel data is collected and accumulated, it can be used for various types of information processing. Since travel data accumulated over a long period of time amounts to an enormous volume, there is a demand for technology to compress travel data. For example, Patent Document 1 discloses an invention for compressing travel data by outputting vehicle speed data at a time when the vehicle speed becomes equal to a set vehicle speed and at a time when a change in vehicle speed becomes an inflection point. [Prior Art Documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2008-108247 [Summary of the Invention] [Problem to be Solved by the Invention]

[0004] However, with conventional technology, the accuracy of vehicle speed data after compression is not sufficient. In order to efficiently compress travel data, it is preferable to shorten the sampling time as much as possible, but if the sampling time is too short, there is a possibility that the accuracy may decrease.

[0005] In view of the above technical problems, an object of an aspect of the present disclosure is to accurately perform downsampling of travel data. [Means for Solving the Problem]

[0006] A downsampling method according to one aspect of the present disclosure includes the steps of: acquiring driving data; calculating predetermined features for the driving data; setting the time length or position of a window based on the distribution density of the driving data; generating extracted data from the driving data using the window; calculating the error between the frequency of occurrence of features in the driving data and the frequency of occurrence of features in the extracted data; and selecting extracted data such that the error is less than or equal to a predetermined value. [Effects of the Invention]

[0007] According to one aspect of this disclosure, it is possible to perform downsampling of driving data with high accuracy. [Brief explanation of the drawing]

[0008] [Figure 1] This flowchart shows an example of a downsampling method. [Figure 2] This figure shows an example of a feature. [Figure 3] This flowchart shows an example of a fault prediction detection method. [Modes for carrying out the invention]

[0009] Hereinafter, embodiments of this disclosure will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0010] [Embodiment] One embodiment of the present disclosure is a method for downsampling driving data using artificial intelligence (AI). In this embodiment, the downsampling method calculates predetermined features for the entire driving data, compares the frequency of occurrence of the features for the entire driving data with the frequency of occurrence of the features for the extracted data extracted from the driving data, and selects the extracted data such that the error in the frequency of occurrence is less than or equal to a predetermined value.

[0011] In the downsampling method, when extracting data from the driving data, a window is used in which at least one of the time length or position is set based on the distribution of the driving data. The window may be set by fixing the total time of the window, or it may be set after extending the total time of the window. In this case, the proportion of the time extended may be increased when the error in the frequency of occurrence is large. The window may be set so that the error in the sum of the features is less than or equal to a predetermined value.

[0012] Various metrics can be used to measure the error in frequency of occurrence. For example, the error in frequency of occurrence may be measured using the Root Mean Squared Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), etc.

[0013] <Driving Data> In this embodiment, the target of processing is driving data acquired from a drive system used in a hybrid vehicle. This type of drive system is disclosed, for example, in Reference 1. [Reference 1] Japanese Patent Publication No. 2008-232287

[0014] The drive device disclosed in Reference 1 is configured as follows: The drive device has a plurality of multi-plate clutches. Each multi-plate clutch has a cancellation chamber. The plurality of multi-plate clutches may share one cancellation chamber.

[0015] One of the multiple multi-plate clutches comprises a clutch drum, a clutch piston, a clutch hub, a return spring, and a cancellation plate. The clutch drum is a bottomed cylindrical member connected to rotate integrally with the first intermediate shaft, and multiple ring-shaped friction plates are spline-engaged to its inner circumferential surface so as to rotate integrally. The clutch piston is located inside the clutch drum and forms a hydraulic chamber between itself and the clutch drum. The clutch hub is a cylindrical member having a cylindrical portion facing the inner circumferential surface of the cylindrical portion of the clutch drum, and connected to rotate integrally with the second intermediate shaft, and multiple ring-shaped friction mating plates are spline-engaged to its outer circumferential surface so as to rotate integrally.

[0016] Driving data is output from the Electronic Control Unit (ECU), which controls the operation of the drivetrain. The ECU receives data such as vehicle speed, access pedal position, shift position, power transmission oil temperature, and mode select switch. Based on this input data, the ECU controls the operation of the drivetrain. The ECU outputs various parameters, including motor-generator torque, electric oil pump pressure, line pressure, and engagement pressure. Multiple engagement pressure values ​​are output, corresponding to each friction material within the drivetrain.

[0017] In this embodiment, driving data is uploaded to an external storage device via a wireless communication network such as a mobile phone network. The external storage device may be, for example, a cloud storage service provided by cloud computing. Driving data from various vehicles is uploaded to the external storage device, and big data is constructed.

[0018] <Downsampling method> The downsampling method in this embodiment will be described with reference to Figure 1. Figure 1(A) is a flowchart showing an example of a downsampling method.

[0019] The downsampling method in the present embodiment is executed by an information processing apparatus implemented by a computer or the like. The information processing apparatus includes a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), an input device, an output device, an external I / F (interface), a communication I / F, and the like, each of which is mutually connected via a system bus.

[0020] In step S101, the information processing apparatus acquires travel data acquired in each travel mode. As an example, the travel modes may be an urban area mode, a highway mode, a mountainous area mode, or the like.

[0021] In step S102, the information processing apparatus combines the travel data of each travel mode acquired in step S101 to generate one piece of travel data. Hereinafter, the combined travel data is referred to as "overall data". When evaluating a component, it is necessary to create an evaluation mode in consideration of all usage scenarios of the component, and thus it is necessary to combine the travel data of each travel mode. As an example, the purpose of the evaluation may be calculation of a damage rate.

[0022] In step S103, the information processing apparatus performs machine learning on the overall data generated in step S102 and performs clustering.

[0023] In step S104, the information processing apparatus calculates a predetermined feature quantity (explanatory variable) for the overall data clustered in step S103. As an example, the explanatory variable may be the number of engagements of a friction material for shifting, engagement torque, oil temperature, relative rotational speed during idling, or the like.

[0024] In step S105, the information processing apparatus cuts out N pieces of data having a time-series length TL from the overall data using a window with a variable time length. Hereinafter, the cut-out travel data is referred to as "cut-out data".

[0025] In step S106, the information processing device calculates the frequency of occurrence of the feature calculated in step S104 for each of the extracted data points obtained in step S105.

[0026] In step S107, the information processing device calculates the error between the frequency of occurrence of features in the overall data and the frequency of occurrence of features in the extracted data. Next, the information processing device determines whether the error is less than a predetermined value. If the error is less than the predetermined value (YES), the information processing device proceeds to step S108. On the other hand, if the error is greater than or equal to the predetermined value (NO), the information processing device returns to step S105.

[0027] After returning to step S105, the information processing device modifies at least one of the window length TL or the window position based on the distribution density of the overall data, and extracts N data again. In this way, the information processing device repeats the extraction of data while adjusting at least one of the window length TL or the window position until the error in the frequency of feature occurrences falls below a predetermined value.

[0028] In step S108, the information processing device selects the N extracted data points from step S105 as the compressed driving data.

[0029] Figure 1(B) is a flowchart showing an example of the data extraction process (step S105 in Figure 1(A)).

[0030] In step S111, the information processing device determines whether the counter Sd is greater than a predetermined value S1. If the counter Sd is greater than the predetermined value S1 (YES), the information processing device proceeds to step S115. On the other hand, if the counter Sd is less than or equal to the predetermined value S1 (NO), the information processing device proceeds to step S112.

[0031] In step S112, the information processing device fixes (maintains) the total time of the window.

[0032] In step S113, the information processing device changes at least one of the time length of each window or the position of each window based on the distribution density of the overall data. Specifically, it sets up one or more windows by determining the positions in order of the highest distribution density of the overall data. The time length of each window may be uniform, or the time length may be increased in order of increasing distribution density. As the counter Sd increases, the information processing device increases the number of windows N. For example, after N windows, the next number is N+1. Here, since the total time of the windows is fixed, the time length of each window becomes shorter. In step S114, 1 is added to the counter Sd.

[0033] In step S115, the information processing device determines whether the counter Sb is greater than a predetermined value S2. If the counter Sb is greater than the predetermined value S2 (YES), the information processing device proceeds to step S119. On the other hand, if the counter Sb is less than or equal to the predetermined value S2 (NO), the information processing device proceeds to step S116.

[0034] In step S116, the information processing device extends the total window time by X%, where X is a predetermined positive number. In this embodiment, the sampling time cannot be shortened, but this is an acceptable constraint in order to reduce the error. The percentage X% by which the total time is changed may be set according to the magnitude of the error. For example, if the error is large, X should be set to a large value.

[0035] In step S117, the information processing device changes at least one of the time length of each window or the position of each window based on the distribution density of the overall data. As the counter Sb increases, the information processing device increases the number of windows N. Here, since the total time of the windows is extended, a new window corresponding to the extended time is set. In step S118, the information processing device adds 1 to the counter Sb.

[0036] In step S119, the information processing device terminates the downsampling method.

[0037] An example of the feature quantities in this embodiment will be explained with reference to Figure 2. Figure 2(A) is a diagram showing an example of the overall data. In the overall data shown in Figure 2(A), the horizontal axis is time and the vertical axis is the engagement torque of the friction material. In this example, the extracted data is shown in windows W1 to W4 in Figure 2(A).

[0038] Figure 2(B) shows an example of features from the overall data and the extracted data. In Figure 2(B), the frequency of occurrence of features from the overall data and the extracted data are compared for engagement torque. For each feature, the horizontal axis represents the feature value and the vertical axis represents the frequency. The smaller the error in the distribution of feature occurrences between the overall data and the extracted data, the more similar the features of the overall data and the extracted data are.

[0039] Figure 2(C) shows the first example of the distribution of the overall data. Since the actual overall data consists of tens of thousands of data points, Figure 2(C) shows 100 data points in one plot. In the overall data shown in Figure 2(C), the vertical axis represents the required driving force (representative accelerator opening), and the horizontal axis represents the vehicle speed. When setting the position of each window, the center points of the regions D1 to D5 with high distribution density are selected in order, and the overall data traveling at those positions is used as the window.

[0040] Figure 2(D) shows a second example of the distribution of driving data. In the overall data shown in Figure 2(D), the vertical axis is acceleration and the horizontal axis is vehicle speed. When setting the position of each window, the center points of the regions D6 to D11 with high distribution density are selected in order, and the overall data driving at those positions is used as the window.

[0041] <Method for detecting signs of failure> Figure 3 is a flowchart showing an example of a fault prediction detection method in this embodiment. The fault prediction detection method is an example of information processing to which the downsampling method in this embodiment is applied.

[0042] In step S201, the information processing device acquires driving data of the vehicle to be detected from the big data. In step S202, the information processing device extracts the driving data acquired in step S201 using the downsampling method in this embodiment (see Figure 1).

[0043] In step S203, the information processing device calculates a predetermined damage rate Ss based on the extracted data obtained in step S202. Since the extracted data is more compact due to downsampling compared to the original driving data, the calculation of the damage rate Ss can be performed in a short time.

[0044] Next, the information processing device calculates the current damage rate Sr by multiplying the damage rate Ss by a predetermined multiplier. The predetermined multiplier can be calculated by (actual driving time or distance) / (driving time or distance of extracted data).

[0045] In step S204, the information processing device displays the current damage rate Sr calculated in step S203.

[0046] In step S205, the information processing device compares the current damage rate Sr with a specified value and determines whether the current damage rate Sr is significantly large relative to the mileage or mileage. If the damage rate Sr is equal to or greater than the specified value (YES), the information processing device proceeds to step S206. On the other hand, if the damage rate Sr is less than the specified value (NO), the information processing device skips steps S206 to S207 and terminates the fault prediction detection method.

[0047] In step S206, the information processing device determines that it has detected a sign of impending failure. In step S207, the information processing device notifies the vehicle user of the impending failure. Specifically, the information processing device notifies the vehicle user that there is a high probability that a failure will occur in the near future.

[0048] <Effects> The downsampling method in this embodiment selects extracted data such that the error between the frequency of occurrence of features in the entire driving data and the frequency of occurrence of features in the extracted data is less than or equal to a predetermined value. The extracted data selected by the downsampling method has features similar to the entire driving data. Therefore, according to this embodiment, downsampling of driving data can be performed with high accuracy. In one aspect, according to this embodiment, signs of failure can be detected with high accuracy.

[0049] Although embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments, and various modifications or changes are possible within the scope of the gist of the present invention as described in the claims.

Claims

[Claim 1] An information processing device, The process of acquiring driving data for multiple driving modes, A step of calculating predetermined feature quantities for the overall data obtained by combining the driving data of the aforementioned multiple driving modes in a time series, The process of clustering the aforementioned overall data, Based on the clustering results, the process involves identifying data from the overall data that corresponds to areas with high distribution density, The process of setting a window at a time-series position corresponding to the identified data within the overall data, A step of generating extracted data from the overall data using the window, in which the time series is maintained. A step of calculating the error between the frequency of occurrence of features in the overall data and the frequency of occurrence of features in the extracted data, If the error is greater than or equal to a predetermined value, the process involves changing the time duration or position of the window. A step of selecting the extracted data generated using the window in which the error is less than or equal to a predetermined value, A downsampling method that performs this operation.

Images