CN109641603B

CN109641603B - Abnormality detection device, abnormality detection method, and computer program

Info

Publication number: CN109641603B
Application number: CN201880003324.7A
Authority: CN
Inventors: 菊池元太; 丸地康平; 服部阳平
Original assignee: Toshiba Corp
Current assignee: Toshiba Corp
Priority date: 2017-07-19
Filing date: 2018-02-26
Publication date: 2021-10-19
Anticipated expiration: 2038-02-26
Also published as: CN109641603A; TWI688492B; JP6889059B2; JP2019018755A; WO2019016996A1; TW201908163A

Abstract

Embodiments of the present invention enable highly accurate anomaly detection. According to one embodiment, an abnormality detection apparatus includes: a condition generator, a threshold setter and an anomaly detector. The generator generates a plurality of conditions for classifying a difference between a predicted value of a state of the vehicle, which is based on the control command value and the prediction model, and a measured value of the state of the vehicle, based on the travel information of the vehicle. The threshold setter sets a plurality of thresholds for the condition. The abnormality detector performs abnormality detection on the vehicle based on the prediction model, the threshold value, and the condition.

Description

Abnormality detection device, abnormality detection method, and computer program

Technical Field

Embodiments of the present invention relate to an abnormality detection apparatus, an abnormality detection method, and a computer program.

Background

Maintenance management and inspection of railway vehicles needs to be performed daily to maintain safety and reliability of railway operations. Conventional maintenance management has been focused on regular inspections of railway vehicles. However, in order to achieve higher railway safety, a technology has recently been developed to achieve early abnormality detection by performing diagnosis and condition monitoring using vehicle information including sensor values and control values acquired from a railway vehicle.

In the related art, abnormality detection is performed by comparing a value measured by a sensor on a railway vehicle with a threshold value. In another prior art technique, a model that reconstructs the normal operation of the rail vehicle is prepared and used to detect anomalies or signs of anomalies. However, with a railway vehicle, the running condition dynamically changes in a time-series manner due to, for example, a gradient of a route, a change in weather, passengers getting on and off, and an operation by a driver. Therefore, it is difficult to accurately perform abnormality detection for such a vehicle based on a single threshold.

Embodiments of the present invention provide an abnormality detection device, an abnormality detection method, and a computer program that realize highly accurate abnormality detection.

Disclosure of Invention

According to one embodiment, an abnormality detection apparatus includes: a condition generator, a threshold setter and an anomaly detector. The generator generates a plurality of conditions for classifying a difference between a predicted value of a state of the vehicle, which is based on the control command value and the prediction model, and a measured value of the state of the vehicle, based on the travel information of the vehicle. The threshold setter sets a plurality of thresholds for the condition. The abnormality detector performs abnormality detection on the vehicle based on the prediction model, the threshold value, and the condition.

Drawings

FIG. 1 is a block diagram of an anomaly detection system according to an embodiment of the present invention.

Fig. 2 is a diagram showing an exemplary configuration of a brake groove, a brake, and an air spring of a railway vehicle.

Fig. 3 is a diagram showing an exemplary configuration of a power generation brake and a regenerative brake of a railway vehicle.

Fig. 4 is a diagram illustrating an exemplary table related to measurement information and environment information.

Fig. 5 is a diagram illustrating an exemplary table relating to braking information.

Fig. 6 is a diagram illustrating an exemplary conversion table.

FIG. 7 is a diagram illustrating an exemplary model database.

Fig. 8 is a diagram illustrating an exemplary method of determining a threshold value by using a normal distribution.

Fig. 9 is a diagram showing an exemplary detection result database.

Fig. 10 is a diagram showing an exemplary data set for generating a running condition.

FIG. 11 is a diagram illustrating an exemplary decision tree.

FIG. 12 is a diagram illustrating another exemplary model database.

Fig. 13 is a diagram illustrating an exemplary operation of the abnormality detection model.

Fig. 14 is a diagram illustrating an exemplary main screen output from the abnormality detection apparatus.

Fig. 15 is a diagram showing an exemplary abnormality detailed screen output from the abnormality detecting device.

Fig. 16 is a diagram showing an exemplary inspection screen for a driver.

Fig. 17 is a diagram showing a hardware configuration of the abnormality detection apparatus according to the present embodiment of the invention.

FIG. 18 is a flow diagram of an anomaly detection process according to an embodiment of the present invention.

Fig. 19 is a flowchart of the processing of the abnormality detection apparatus in the learning mode.

FIG. 20 is a diagram illustrating another exemplary model database.

Detailed Description

Embodiments of the present invention will be described below with reference to the accompanying drawings. Any identical components between the drawings are denoted by the same reference numerals, and the description thereof will be omitted as appropriate.

Fig. 1 is a block diagram showing an exemplary abnormality detection system according to a first embodiment of the present invention.

The abnormality detection system shown in fig. 1 includes an abnormality detection device 100, a vehicle system 500, an environmental information system 600, a terminal 700, an input device 800, and a screen display device 900. An outline of the abnormality detection system shown in fig. 1 will be described below.

The abnormality detection apparatus 100 operates in a learning mode and an operation mode. The abnormality detection apparatus 100 includes a function (abnormality detection model generator 200) that creates an abnormality detection model related to a brake system of a railway vehicle (hereinafter, vehicle) based on at least one of measurement information of the railway vehicle and environmental information of the vehicle in a learning mode. The measurement information is acquired from the vehicle system 500. The environment information is acquired from the environment information system 600. The running information of the vehicle according to the present embodiment includes at least one of the measurement information of the vehicle acquired from the vehicle system 500 and the environmental information of the vehicle acquired from the environmental information system 600. The abnormality detection model includes a prediction model for predicting the state of the vehicle, and a threshold value related to a residual (difference) of a prediction value from the prediction model. As an example, the state of the vehicle is a deceleration of the vehicle.

The abnormality detection apparatus 100 includes a function (abnormality detector 110) of performing abnormality detection of the vehicle by using the prediction model and the threshold value in the operation mode. Anomaly detection is the determination of whether an anomaly exists. The abnormality detection is also referred to as abnormality determination. Abnormality detection is performed by comparing a threshold value with a residual that is a difference between a predicted value of a prediction model and an actual value of a state of the vehicle acquired from the vehicle.

The abnormality detection apparatus 100 includes a function (threshold value setter 220) of setting a threshold value for each of a plurality of conditions (hereinafter, running conditions) based on running information of the vehicle. The abnormality detection apparatus 100 generates a prediction model and an abnormality detection model including threshold values corresponding to respective running conditions. In the abnormality detection using the abnormality detection model, a threshold value corresponding to a running condition in which the vehicle exists, in other words, a running condition satisfying running information (current running information) that is a target of the abnormality detection is selected from threshold values included in the abnormality detection model and used together with the prediction model.

The creation of a plurality of running conditions and the setting of a plurality of thresholds are performed in the learning mode. In the creation of the running condition, the running information of the vehicle and the result of abnormality detection (for example, a residual between the predicted value of the prediction model and the actual value of the state value) are used. The learning mode and the operation mode may be switched automatically or by an instruction from, for example, a maintenance person, or may be performed simultaneously.

When an abnormality is detected, the abnormality detection apparatus 100 displays, for example, a place where the abnormality is detected, an abnormality detection model for abnormality detection, travel information for abnormality detection, and a prediction value of a prediction model on the screen display apparatus 900. This supports monitoring by the railway operator.

The following briefly describes a brake system of a vehicle according to the present embodiment. Fig. 2 shows an exemplary configuration of a brake groove, and a brake and air spring for a specific wheel of a vehicle. The brake groove is actually located in the cab of the consist. A brake system of a vehicle as an abnormality detection target of the abnormality detection apparatus 100 is described below with reference to fig. 2.

Brake lever 10, as an exemplary controller, provides a device for a driver's braking operation. The driver brakes the vehicle by moving the brake lever upwards. The numbers 1 to 8 indicated on the brake lever 10 indicate brake grooves (brake stages), and the larger the number means the stronger the braking force applied to the vehicle. This number of slots is exemplary and does not limit the vehicle to use a greater or lesser number of slots. Each brake slot is an exemplary control command value to the vehicle or brake.

The braking operation on the vehicle is not limited to the operation performed by the driver. For example, in a vehicle mounted with an automatic train unit stop (ATS), an automatic train unit control (ATC), or an automatic train unit operation (ATO), a braking operation is performed by the device instead of the driver in some cases. In this case, for example, the brake command output from the device corresponds to the control command value.

Fig. 2 shows a wheel 30 of a vehicle travelling on a railway 20. One device for braking a vehicle with a brake is a tread brake 42. In this example, only one wheel is shown for simplicity of description, but actually, a plurality of pairs of left and right wheels are provided.

The tread brake 42 uses a cylinder as a power source. When the brake cylinder pressure, which is the pressure inside the cylinder 43, increases, the brake pad 41 is pressed against the tread, which is the surface of the wheel 30 that contacts the rail. The frictional force between the wheel 30 and the brake pad 41 serves as the braking force of the tread brake 42.

Since the tread brake uses the frictional force of the brake pad in this way, the brake pad is worn due to continuous use, which may reduce the braking force. A tread brake is an exemplary mechanical brake for a vehicle, and another proposal uses a disc brake that obtains a braking force by pressing a disc fixed to a wheel shaft against a wheel with a brake pad or the like. The braking force of the brake varies with, for example, the state of wear of the brake pads or shoes. When an abnormality in the brake system is detected by the present abnormality detection apparatus, a worker or the like can check, for example, a brake pad or a disc of the brake system and check the actual presence of the abnormality.

In addition to wear of the components of the brake, the braking force of the brake also varies with the load on the vehicle. The load response device 50 is mounted on the vehicle shown in fig. 2. The load response device 50 includes an air spring 51, and measures the load on the vehicle by sensing the air spring pressure of the air spring 51. To control the braking of the vehicle, in addition to the operation of the brake lever 10, the braking force of the brake may be adjusted in response to the air spring pressure detected by the load-responsive device 50. Thus, a desired deceleration can be achieved regardless of changes in the load on the vehicle.

In order to supplement the braking force of the mechanical brake, an electric brake may additionally be used in the brake system of the vehicle. The electric brake will be described with reference to fig. 3. Fig. 3 shows an exemplary configuration of a power generation brake and a regenerative brake of a vehicle.

The

main motors

60a and 60b are mounted on the vehicle shown in fig. 3. When the power-generating brake is used, the

main motors

60a and 60b form a closed circuit with the resistor 70 to convert the electric power of the main motors into thermal energy.

When the regenerative brakes are applied, the electric power generated by the

main motors

60a and 60b is transmitted to the line 90 through the pantograph 80. When the secondary battery is mounted on a vehicle, the generated electric power may be used to charge the secondary battery. In this way, the regenerative brake obtains braking force by converting kinetic energy into electric power using the

main motors

60a and 60b as generators.

The mechanical brake and the power generation brake are exemplary, and the abnormality detection apparatus 100 may perform abnormality detection on any other scheme of brake used in the brake system.

Since the brake system has a relatively complicated configuration and the characteristics of the brake and the brake system vary depending on a plurality of factors and conditions, it is difficult to perform accurate abnormality detection on the brake system of the vehicle.

For example, a plurality of brake schemes having different characteristics are used in a brake system of a vehicle. Further, as described above, the braking force of the brake system of the vehicle varies with the load. For example, with a passenger vehicle, the number of passengers varies greatly with time slots and operating intervals, and therefore the braking force of the brake system varies greatly over a short duration. For freight vehicles, the load varies greatly with the amount of freight. Further, the deceleration when the braking operation is performed may vary between the travel route and the interval of the vehicle having different tendencies of gradient and inclination. Furthermore, any differences in rainfall weather conditions, atmospheric temperature, etc. may change the physical properties of the components of the brake system, thereby affecting the characteristics of the brake system. As other factors, the driver performs the braking operation in different ways, and the vehicle is manufactured with different braking characteristics.

In the present embodiment, highly accurate abnormality detection is easily performed by switching the threshold value according to the running condition in which the vehicle exists to use the threshold value suitable for the situation inside and outside the vehicle. In this way, the risk of false anomaly detection is reduced, thereby enabling early anomaly detection and safe and reliable railway operation.

The abnormality detection apparatus 100 shown in fig. 1 will be described in further detail below. In the following description, the brake system that is the target of abnormality detection by the abnormality detection apparatus 100 may be a brake apparatus for a specific wheel of an optional railway vehicle, all of a plurality of brake apparatuses in the entire vehicle, or a plurality of vehicles in a consist or the entire group of brake apparatuses in a car. The abnormality detection target is not limited to the brake system, but may be an electric power system, an air conditioning system, or a power generation system. The abnormality detection target is not limited to a railway vehicle, but may be an alternative vehicle including wheels, such as an automobile, a construction machine, and an airplane. The abnormality detection target also includes any device or system other than the vehicle.

The abnormality detection apparatus 100 includes a vehicle information collector 101, an environmental information collector 102, a data processor 103, an abnormality detection model generator 200, a condition generator 230, an abnormality detector 110, an alarm 120, and a screen generator 130.

Anomaly detection model generator 200 includes a model generator 210 and a threshold setter 220.

The vehicle information collector 101 acquires measurement information (also referred to as measurement data) related to the vehicle from various sensors of the vehicle system 500 inside the vehicle. Examples of sensors include: a sensor configured to detect a brake operation or the like of the vehicle as a control command value; a sensor configured to detect a deceleration of the vehicle; a sensor configured to detect a running speed; and a sensor configured to measure a load applied to the vehicle. Various other sensors may be included. The measurement information includes a detection value of the sensor (e.g., a control command value) and a measurement value of the sensor (e.g., a driving speed, a load, or a deceleration). When the vehicle system 500 calculates the deceleration from the value of the speed sensor, the calculated deceleration may be acquired as part of the measurement information.

The kind of the measurement information to be acquired (the kind of the sensor or the kind of the control command value) may be optionally set. The measurement information may be acquired in an optionally set period. For example, the measurement information related to the running speed of the vehicle is acquired in a short sampling period in milliseconds. The value of a sensor configured to measure a load applied to the vehicle is acquired in a sampling period in units of minutes.

The environment information collector 102 acquires environment information (also referred to as running environment data) of the vehicle from the environment information system 600. Examples of the environmental information include information related to an operation route and information related to weather. Examples of the information related to the operation route include a gradient and a slope (a difference in height between inner and outer rails of a railway) at each interval. Examples of weather-related information include weather, atmospheric temperature, precipitation, wind speed, and atmospheric pressure. The acquisition of the environmental information may be acquisition of information accumulated in a database in the ground system or acquisition of information distributed from an external server. The kind of the environmental information to be acquired and the frequency of acquisition may be optionally set.

The abnormality detection device 100 may be installed as a ground device outside the vehicle, for example, at a facility or an operation command center of a railway operation management company, or may be installed as an in-vehicle device on the vehicle. The abnormality detection device 100 is not limited to a specific mounting manner.

When the abnormality detection apparatus 100 is installed as a ground apparatus outside the vehicle, measurement information and the like of the vehicle system 500 inside the vehicle is received through, for example, an on-vehicle component, a transponder ground component, and a ground information network. Specifically, the vehicle system 500 transmits data to the ground information network through a ground element or the like, and the abnormality detection system receives data through the ground information network. The terrestrial information network may use, for example, a metal cable, a coaxial cable, an optical cable, a telephone line, a wireless device, or ethernet (registered trademark), but is not limited to a specific scheme. The abnormality detection apparatus 100 receives data from the environmental information system 600 through the ground information network.

When the abnormality detection apparatus 100 is an in-vehicle apparatus, the abnormality detection apparatus 100 acquires data from the vehicle system 500 through an information network inside the vehicle. The information network inside the vehicle is, for example, an ethernet or a wireless Local Area Network (LAN), but may be implemented in any other scheme. The anomaly detection device 100 can use the on-board components and the transponder ground components to acquire data from the environmental information system 600 connected to the ground information network.

The input device 800 provides an interface for the operation of maintenance personnel. The input device 800 includes a mouse, a keyboard, a voice recognition system, an image recognition system, a touch pad, or a combination thereof. The maintenance person can input various commands or data to the abnormality detection apparatus 100 through the input device 800 to perform operations.

The screen display device 900 displays data or information output from the abnormality detection device 100 as a still image or a moving image. The screen display device 900 is, for example, a Liquid Crystal Display (LCD), an organic electroluminescence display, or a Vacuum Fluorescent Display (VFD), but may be a display device in any other scheme.

Each of the input device 800 and the screen display device 900 may be one of a plurality of installed devices. For example, the input device 800 and the screen display device 900 may be installed at each of the operation command center and the operation console of the vehicle.

The input device 800 and the screen display device 900 may be one integrated device. For example, when a display having a touch panel function is used, a single device may be used as the input device 800 and the screen display device 900.

The abnormality detection apparatus 100 includes an information database 310, a model database 320, and a detection result database 330.

The

databases

310, 320, and 330 are all disposed inside the abnormality detection apparatus 100 shown in fig. 1. However, the arrangement of the database is not limited to a specific method. For example, the partial database may be arranged in an external server or storage device. Each database may be implemented by a relational database management system and various NoSQL systems, but may be implemented in any other scheme. Each database may be in the storage format of XML, JSON, or CSV, or any other format, such as a binary format. Not all databases within the anomaly detection apparatus 100 need be implemented by the same database system and the same storage format, but the databases may be implemented in a mixture of schemes.

The information database 310 stores the measurement information acquired by the vehicle information collector 101 and the environmental information acquired by the environmental information collector 102. A storage medium such as a memory device storing measurement information and environmental information may be inserted into the abnormality detection apparatus 100 and used as the information database 310.

Fig. 4 and 5 illustrate an exemplary information database 310. The travel information (measurement information and environment information) is stored in the form of a table 310a shown in fig. 4 and a table 310b shown in fig. 5.

The "data ID" column of table 310a shown in fig. 4 stores the identification numbers of the entries stored in table 310 a. The data ID serves as a primary key. Each data ID is associated with a table 310b shown in fig. 5. Table 310b is stored in information database 310. The "time" column stores the time of generation of the entry. In this example, an entry is generated in each constant sample time. However, the entries may be generated at intervals preset to the railroad track or at any other criteria.

The "driver" column of table 310a stores the name of the driver who performed the braking operation. For example, when a device such as an ATS, ATC, or ATO performs a braking operation instead of the driver, the name of the device performing the operation or an identifier indicating the device may be stored instead.

The "weather" column of the table 310a stores information related to weather acquired from the environment information system 600.

The "atmospheric temperature" column of table 310a stores information related to the atmospheric temperature acquired from the environmental information system 600. The information related to the atmospheric temperature may be an actual value or a label classifying the actual value. In the example shown in fig. 4, the column "atmospheric temperature" stores a label of any one of categories T1, T2, T3, T4, T5, T6, and T7 into which the atmospheric temperature as a real number is converted by using the conversion table 310c shown in fig. 6. For example, an atmospheric temperature of-11 ℃ is converted into the category T1, an atmospheric temperature of 15 ℃ is converted into the category T4, and an atmospheric temperature of 33 ℃ is converted into the category T6. When creating a prediction model described later, data obtained by converting a category name into an optional integer, for example, category T1 into 1, category T2 into 2, and category T3 into 3, can be used as an explanatory variable.

As exemplarily described by the "atmospheric temperature" column, the information database 310 may store processing information obtained by performing calculation or conversion on measurement information or environmental information.

The "occupancy" column of table 310a stores the occupancy as an indicator of the load applied to the vehicle in percentage. Another indicator may be used to indicate load. The occupancy is defined by, for example, a ratio of the capacity of the passenger vehicle to the number of passengers in the passenger vehicle. Ride rate is typically estimated based on the air spring pressure of the load-responsive device. In this case, the air spring pressure can be directly used as an index.

The air spring pressure is the actual value of the sensor, unlike the ride rate, which is not a value indirectly estimated by using, for example, a conversion table or formula, and thus can be used to reduce the residual error in model generation. However, the value of the air spring pressure depends on the manufacturer and the model of the load response device mounted on the vehicle, and thus lacks versatility. Therefore, when a commonly used index such as a riding rate is used, in some cases, a difference between vehicles due to different load response devices thereof can be absorbed.

The "grade" column of table 310a stores the grade of the route in values expressed in units of thousandths (permils). The permillage (permil) is a value obtained by expressing a height difference of a horizontal distance of 1000 meters in meters. The "grade" column may store another unit of value, with the per mil (Permil) being exemplary.

The "slope" column of table 310a stores the slope in millimeters, but may store another unit of value.

Table 310a also includes a "wind speed" column and a "barometric pressure" column. Table 310a may include columns that store other information such as the current location and current spacing on the railroad track.

The table 310b shown in fig. 5 stores information such as time information, brake grooves, and actual values of deceleration of corresponding entries in the table 310a shown in fig. 4. In the example shown in FIG. 5, table 310b corresponds to the data ID of 2560 in FIG. 4. The entries in table 310b are generated at shorter time intervals than the time intervals of table 310 a. The generation interval (sampling interval) of the entries in table 310b may be the same as that of table 310 a. Tables 310b and 310a are provided separately in this example, but may be integrated with each other.

Data stored in the information database 310 may be processed. For example, the data processor 103 causes the screen display device 900 to display the contents of each table stored in the information database 310. The maintenance person or the driver performs the manufacturing operation on the data by using the input device 800. The data processor 103 performs data fabrication according to the fabrication operation.

The interval at which the vehicle information collector 101 or the environmental information collector 102 acquires information or data may be adjusted. For example, the data processor 103 receives an operation of designating an acquisition interval from a maintenance person or a driver through the input device 800, and adjusts the acquisition interval according to the content of the operation.

The abnormality detection model generator 200 creates an abnormality detection model of the brake system of the vehicle by using the data stored in the information database 310. The anomaly detection model includes a predictive model and one or more thresholds. The predictive model is generated by model generator 210 and each threshold is generated by threshold setter 220. The generated anomaly detection model is stored in the model database 320.

FIG. 7 illustrates an exemplary model database 320. The model database 320 may store one or more anomaly detection models. Each anomaly detection model is identified by a model ID. The column of the prediction model stores data indicating the prediction model or a memory address (pointer) storing the prediction model. The data indicative of the predictive model includes, for example, a deceleration model. Each cell in the column of thresholds stores one or more thresholds. When a plurality of threshold values are stored, the running condition corresponding to each threshold value is stored together (to be described later in detail).

The generation of the abnormality detection model is performed in the learning mode, for example, when the abnormality detection apparatus 100 is started or when a system is newly added as an abnormality detection target. When there are a plurality of systems as abnormality detection targets, an abnormality detection model is generated for each system.

The anomaly detection model is created by using data samples (feature vectors) extracted from the information database 310.

The data samples (feature vectors) include one or more interpretation variables. Examples of the explanatory variables include values of brake grooves (control command values) in the table 310 b. Further, any other type of value (such as speed) in the running information and specifications of the vehicle (such as the size and weight of the vehicle) may be used as the explanatory variable. The explanatory variable may be generated by calculation of a plurality of items included in the travel information. In this example, the target variable of the predictive model is deceleration. A data sample may be generated for each entry in the braking information table 310b or a data sample may be generated by collecting multiple consecutive entries with a lower time granularity.

The method of generating the predictive model is described below. The regression model is assumed to be used as the prediction model. The model generator 210 obtains a feature vector X ═ (X) having an explanatory variable as an element by using the information database 310₁,x₂,x₃,...,x_n)。

Subsequently, the model generator 210 performs multiple regression analysis to obtain formula (1) for predicting deceleration as a target variable.

[ expression 1]

y＝b₀+b₁x₁+b₂x₂+b₃x₃+...+b_nx_n (1)

In the above formula, y represents the target variable, x_nRepresents an explanatory variable, and b_nThe partial regression coefficients are represented. To absorb the difference in measurement units between the explanatory variables, a standard partial regression coefficient may be used as the partial regression coefficient b by normalizing the target variable and all explanatory variables to the average value 0 and the degree of dispersion 1_n. The number of interpretation variables may be one or more.

Model generation by multivariate regression analysis is exemplary, and a predictive model of the target variable may be created by any other method, such as support vector regression or autoregressive.

Cross-validation may be used when creating the predictive model. For example, the data samples may be divided into sets such that at least one set is used as test data for verification and the other sets are used for model creation. This allows the performance of the generated model to be checked.

In each embodiment of the present invention, the information database stores information acquired when the brake system that performs abnormality detection is in a normal state. Therefore, the generated prediction model is a model obtained by modeling the behavior of the brake system of the vehicle in a normal state. However, the information database may store information when a particular brake fails.

The number of interpretation variables used to create the predictive model may be reduced by adjusting the number of interpretation variables, for example, by a variable selection method or principal component analysis. Such adjustment is effective when there is a correlation between different explanatory variables or when it is necessary to reduce the calculation time and the processing load.

In the variable selection method, a model is generated by selecting a subset of the interpretation variables that are valid for prediction from a set of all interpretation variables. Useful explanatory variables can be selected by first generating a model using one or a small number of explanatory variables, and then generating a model to which the explanatory variables are added one by one. Alternatively, useful explanatory variables may be specified by first generating a model having a large number of explanatory variables, and then generating a model obtained by removing the explanatory variables one by one. The selection of the interpretation variables may be performed by using a genetic algorithm.

In principal component analysis, eigenvalue problems of correlation matrices or variance-covariance matrices for model-generated data are solved to generate new explanatory variables, thereby reducing dimensionality. Using the newly explained variable obtained by principal component analysis as the variable x of the formula (1)_nThe regression analysis of (2) is called principal component regression.

An example of using the variable selection method is described below. For example, when the brake system of the vehicle is an abnormality detection target and the deceleration of the brake is a target variable of the prediction model, a prediction model using only the brake groove as an explanatory variable is first created. It is assumed that the brake groove has the highest correlation with deceleration. Thereafter, any other explanatory variable (such as the traveling speed of the vehicle) that is considered to have a correlation with the deceleration is sequentially added to the prediction model, and the accuracy of the prediction is checked. An explanatory variable is employed when the desired prediction accuracy is obtained.

The threshold value set to the predictive model by the threshold value setter 220 is described below. The threshold value is used to determine that there is an abnormality when the difference between the predicted value of the target variable (in this example, the predicted value of the deceleration) and the measured value (actual value) of the deceleration, which are calculated based on the prediction model, exceeds the threshold value. Determining the presence of an anomaly is also referred to as detection of an anomaly. The difference between the actual and predicted values is also referred to as the residual. The actual value may be greater or less than the predicted value, and thus the value of the residual may be positive or negative. When the sign is not important, the residual may be defined as the absolute value of the difference, which is the absolute value of the distance from the predictor.

FIG. 8 is used to describe an exemplary threshold determination method using a normal distribution. Fig. 8 shows a graph of a normal distribution 400 of residuals. The horizontal axis represents the residual and the vertical axis represents the probability density. Assuming that these residuals follow a normal distribution, a number of residuals between the predicted and actual values of the prediction model are taken to create a normal distribution 400. The data used to obtain the residual may be data samples used to generate the prediction model, test data, driving information unrelated to the generation of the prediction model, or a selectable combination thereof. When the residuals have a large variance, the distribution has a wider skirt, as shown by

normal distributions

401 and 402 shown in the figure with dashed lines.

The normal distribution 400 is used to set the threshold of the prediction model. For example, when the standard deviation is represented by σ, a multiple of the standard deviation, such as 2 σ or 3 σ, is set as the threshold value. When the residual exceeds a threshold set to 2 σ, an abnormality is detected in abnormality detection. When such a threshold value is set, about 95% of the actual value is determined as no abnormality (normal). As another exemplary threshold setting, a residual value corresponding to a predetermined probability (e.g., the highest X percentage point or the lowest X percentage point) or an absolute value thereof may be set as a threshold. The above threshold determination method is exemplary, and any other method may be used. For example, the threshold value may be determined by assuming a distribution other than the normal distribution, or may be determined by a person such as a maintenance person or a driver based on his experience.

The anomaly detector 110 performs anomaly detection on the target system by using an anomaly detection model (prediction model and threshold value) stored in the model database 320. A feature vector is generated from the running information for abnormality detection, and the deceleration is predicted by using the generated feature vector and a prediction model. The residual between the predicted deceleration and the actual deceleration is compared to a threshold. When the residual is equal to or less than the threshold, it is determined that the deceleration is normal. When the residual is greater than the threshold, it is determined that the deceleration is abnormal. The abnormality detector 110 stores information in the detection result database 330 based on the result of abnormality detection. The abnormality detector 110 notifies the screen generator 130 and the alarm 120 of information related to the result of the abnormality detection.

Fig. 9 shows an exemplary detection result database. The database stores the brake groove, the actual value of deceleration, the predicted value (ID0001) of deceleration based on the prediction model, and the abnormality detection result (presence or absence of abnormality) in a time series manner. In the example shown in fig. 9, no anomaly is detected at any time. Travel information (e.g., items of a driver, weather, atmospheric temperature, occupancy, gradient, inclination, wind speed, and atmospheric pressure) corresponding to time may be added to the detection result database.

The above description is an example regarding setting a threshold to a predictive model. However, in the present embodiment, the threshold setter 220 may set a plurality of thresholds to the prediction model according to the running condition. In this case, at the time of abnormality detection, the abnormality detector 110 specifies the running condition satisfied by the current running information. Then, abnormality detection is performed by using a threshold value corresponding to the specified running condition. The driving conditions required for the threshold value setter 220 to set a plurality of threshold values are generated by the condition generator 230.

The condition generator 230 generates a plurality of running conditions (a plurality of conditions) for classifying the difference values according to the difference values between the predicted values and the actual values of the prediction model by using the detection result database 330 and the running information for abnormality detection. The operation of the condition generator 230 is described in detail below.

The condition generator 230 creates a classifier (e.g., a decision tree) for predicting a category from the residual by using the detection result database 330 and the travel information.

Each residual between the predicted value and the actual value in the detection result database is classified into a plurality of categories (residual categories). For example, a residual equal to or less than a threshold a is classified as a residual class a for small residuals, a residual greater than the threshold a and less than the threshold B is classified as a residual class B for intermediate residuals, or a residual equal to or greater than the threshold B is classified as a residual class C for large residuals. The threshold a may be a threshold that is first set to the predictive model, but is not limited thereto. Another exemplary classification will be described next. The residuals are arranged in ascending or descending order. When the residuals are arranged in ascending order of size, the data set of the smallest 20% of the residuals is classified as a residual class a, the data set of the next smallest 60% of the residuals is classified as a residual class B, and the data set of the remaining 20% of the residuals is classified as a residual class C. The ratio of the categories may be optionally determined and is not limited to the above values.

The condition generator 230 selects any one of the residual categories a to C from the residuals of the entries for each entry in the detection result database 330 as shown in fig. 9, and assigns the selected residual category to the entry. Further, the condition generator 230 specifies the travel information corresponding to each entry in the table 310a shown in fig. 4, and associates the travel information with the entry. However, when the travel information is already included in the detection result database 330, this operation is unnecessary. In this way, as shown in fig. 10, a data set in which each entry in the detection result database is associated with the residual category and the travel information is generated. The data set may be stored in an internal buffer of the condition generator 230 or may be stored in a storage device or database not shown. In the data set shown in fig. 10, all entries are set to the residual category a. The decision tree is generated by using the data set as learning data, specifically, by setting the residual category as a target variable and the other items as explanatory variables. In the databases from which the data sets are created (e.g., those shown in fig. 9 and 4), there is no need to use any items that are not needed to produce the data sets. For example, when not needed, the items of the abnormality detection result need not be included in the data set shown in fig. 10.

The example assumes that the data set does not include data that detects anomalies. However, when there is data for which an anomaly is detected, the data may be excluded from the data set. When the abnormality detector 110 detects an abnormality, the maintenance person can check whether the detection result is correct. For example, when an abnormality is detected, the screen display device 900 displays a check screen (see fig. 16 described later) to check whether the detection result is correct. When it is determined that the detection result is erroneous, the maintenance person inputs an instruction for error detection. The condition generator 230 or another processor corrects the detection results in the detection result database 330 based on the instruction. In creating the decision tree, data that detects anomalies may be excluded from the dataset.

FIG. 11 illustrates an exemplary generated decision tree. The decision tree predicts a residual category corresponding to a target variable based on two explanatory variables related to precipitation and occupancy. Nodes 1001a, 1001b, and 1001c are end nodes corresponding to target variables. The top node is called the root node. Any node other than the end node and the root node is referred to as an intermediate node. The root node and the intermediate nodes are explanatory variable nodes. The end node is a residual category node (target variable node). Nodes 1001a, 1001B, and 1001C correspond to residual category a, residual category B, and residual category C, respectively. The decision tree classifies residuals into a residual category a for "sunny (no precipitation) and occupancy equal to or below 90%", a residual category B for "sunny (no precipitation) and occupancy above 90%", or a residual category C for "rainy (no precipitation)".

The condition generator 230 acquires a condition included in a path from the residual category node (end node) to the root node as a travel condition corresponding to each residual category. Specifically, the condition generator 230 acquires the condition "clear day (no precipitation) and occupancy equal to or lower than 90%" corresponding to the path from the residual category a to the root node, the condition "clear day (no precipitation) and occupancy higher than 90% corresponding to the path from the residual category B to the root node, and the condition" rainy day (no precipitation) "corresponding to the path from the residual category C to the root node as the driving conditions A, B and C, respectively.

The threshold value setter 220 sets a threshold value for each of the running conditions a to C. Specifically, the threshold setter 220 classifies the travel information used to generate the prediction model (or the travel information not used to generate the prediction model) into groups a to C satisfying the travel conditions a to C. Abnormality detection is performed on the traveling information classified into the group a, and a residual is calculated based on the detection result. A distribution (e.g., a normal distribution) of the residuals is calculated and used to set the threshold a (see description with respect to fig. 8). In this way, the threshold value a is set for the running condition a. Similarly, a threshold value B corresponding to the running condition B is set as a group B, and a threshold value C corresponding to the running condition C is set as a group C.

The distribution of residuals corresponding to group a has a small standard deviation, so the threshold a is small. The distribution of residuals corresponding to group C has a large standard deviation, so the threshold C is large. The threshold B corresponding to group B is between threshold a and threshold C.

The threshold a is used when the weather is clear and the occupancy is equal to or lower than 90%. Threshold B is used when the weather is clear and the occupancy is above 90%. The threshold C is used when the precipitation is greater than zero. The threshold value is switched according to such a running condition at the time of abnormality detection, thereby accurately realizing abnormality detection based on the braking characteristic. Although only the precipitation amount and the occupancy are used as the above-described running conditions, other items of atmospheric temperature and humidity may be used as the running conditions, for example.

Examples of algorithms for learning decision trees include ID3 and C4.5, but any algorithm may be used. Pruning of the decision tree may be performed to prevent noise and over-learning. Learning of the decision tree is exemplary and any other classifier may be used. Which interpretation variable of the plurality of travel information items is used in the decision tree depends on the algorithm and learning data to be used.

Although a decision tree is used to set a threshold value for each of the above running conditions, a maintenance person may employ expert knowledge to set a threshold value for each of the running conditions.

Each pair of the traveling condition generated by the condition generator 230 and the threshold value set by the threshold value setter 220 is stored in a corresponding cell in the threshold value column in the model database 320. Fig. 12 shows an example in which a plurality of threshold values and corresponding running conditions are stored for the model 0001.

When an abnormality detection model in which a plurality of threshold values are set for the prediction model in this way is used in abnormality detection, the abnormality detector 110 specifies a running condition that satisfies the current running information among the running conditions. Then, the abnormality detector 110 performs abnormality detection by using a threshold value corresponding to the specified running condition and a prediction model. A specific example of the operation of the abnormality detector 110 in this case is described below.

Fig. 13 is a diagram for describing an exemplary operation of the abnormality detector 110. The upper part of fig. 13 shows the detent groove. The middle part of which shows the braking deceleration. The lower part shows the residual between the actual value and the predicted value of the prediction model. In the predictive model, the brake groove corresponds to the explanatory variable, and the deceleration corresponds to the target variable.

At time t1, an operation is performed to set the brake groove to groove "4". Upon receiving the operation, the brake system applies a braking force to the vehicle. Therefore, the deceleration of the vehicle increases and thereafter is kept close to a constant value. In the interval between times t1 and t2, there is a slight difference between the predicted value of deceleration and its measured value (actual value), but essentially transitions at the same value, and the residual between the predicted value and the actual value is less than the threshold value α. In this interval, the running environment of the vehicle satisfies the running condition a. The threshold value α corresponds to the running condition a.

At time t2, the running condition that the running environment of the vehicle satisfies is changed from a to B. The abnormality detector 110 detects a change in the running condition and changes the threshold value used to β.

The running environment and its change can be detected from the measured value and the control command value included in the measurement information and the route data and the weather data included in the environment information. Alternatively, the driving environment and its changes may be detected based on explicit commands, e.g. from the driver or a command center, or radio signals received from ground elements.

In the interval between times t2 and t3, the predicted value of deceleration is constant, but the actual value varies greatly compared to the actual value between times t1 and t 2. Therefore, the residual exceeds the threshold β at three times, and the abnormality detector 110 detects an abnormality at each time.

At time t3, an operation to change the brake groove from groove "4" to groove "2" is performed. Upon receiving the operation, the brake system reduces the braking force applied to the vehicle, and accordingly, the deceleration of the vehicle is reduced. In the interval between times t3 and t4, the residual is within the threshold β, so no anomaly is detected.

At time t4, the running condition satisfied by the running environment of the vehicle is returned from B to a. Upon detecting the change in the running condition, the abnormality detector 110 changes the threshold value used from β to α at time t 4. In the interval between times t4 and t5, the residual is within the range of the threshold α, so no anomaly is detected.

At time t5, an operation to cancel braking is performed. Upon receiving this operation, the brake system further reduces the braking force applied to the vehicle, and therefore the deceleration of the vehicle further decreases.

After time t5, the residual is within the threshold α, so no anomaly is detected.

The alarm 120 notifies the terminal 700 used by a railway operator, a driver, or a maintenance person of the abnormality detected by the abnormality detector 110. The notification may be performed by sending an e-mail, displaying a pop-up message on an operation screen of the terminal 700, or performing the notification by a predetermined tool management protocol, or may be performed by any other means. The notification may include detailed information (e.g., a location on a map where the abnormality occurred (current value), or an identifier of the vehicle where the abnormality occurred). The operator or maintenance personnel can learn about the detection of an anomaly and its details by receiving a notification.

The screen generator 130 displays, for example, an abnormality detection result, the current position of the vehicle if an abnormality is detected, an abnormality detection model and a threshold value for abnormality detection, sensor data, and a prediction value based on the prediction model on the screen display device 900. The screen generator 130 may be included in the abnormality detection apparatus 100, or may be included in a vehicle information system connected to the abnormality detection apparatus 100, or in a terminal or a management server on an information network of a ground system.

Fig. 14 shows an exemplary home screen 901 displayed by the screen generator 130. The main screen 901 displays information about a plurality of consist. In this example, the screen display 900 is installed in a command room for managing and monitoring the consist.

Information related to the vehicle is displayed in a table form on the upper portion of the main screen 901. Examples of the display items include items of a consist, an abnormality detection result, a line number (consist identifier), a ride rate, and a current location, but any other information may be displayed. The column "abnormal" in the table shows the abnormal detection result. Exclamation mark! "indicates that an anomaly is detected. Thus, an anomaly is indicated as being detected at consist B. Such display of the anomaly detection results is exemplary and may be implemented in any other manner.

A map is displayed in the lower portion of the main screen 901 indicating the current location of each consist. The detection of an anomaly and the name of the model used are displayed in the word balloon of consist B where the anomaly was detected.

Clicking on the consist of interest on the main screen 901 shown in fig. 14 performs a transition to an exceptional detail screen. The scheme of screen transition is not limited thereto, but any other scheme such as a predetermined keyboard operation may be employed.

FIG. 15 shows an exemplary exception detail screen 902 for transitioning by clicking on consist B.

In fig. 15, the same graph as that shown in fig. 13 is displayed on the right side of the screen. Specifically, a graph of the brake groove, a graph of the predicted value of deceleration, a graph of the actual value of deceleration, and a graph of the residual are displayed. Each graph is displayed for a particular duration including the time when the anomaly was detected. A bar is displayed indicating that the threshold is exceeded to allow visual inspection of the time at which the anomaly was detected.

A check box is provided in the left part of the screen shown in fig. 15 to allow selection of an item whose graph is to be displayed. Another separate means may be provided to allow specification of the time range of the graphical display. When such an interface as described above is provided, the railway operator or the like can understand the details of the abnormality and take measures quickly.

When the abnormality detector 110 detects an abnormality, a check screen may be presented to a maintenance person to allow the maintenance person to check whether the detection result is correct. Fig. 16 shows an exemplary check screen 903. When the detection result is determined to be wrong, the maintenance personnel inputs an instruction for correcting the result. The condition generator 230 corrects the detection result in the detection result database based on the instruction.

Fig. 17 shows a hardware configuration of the abnormality detection apparatus according to the present embodiment. The abnormality detection apparatus according to the present embodiment is realized by a computer apparatus 100. The computer device 100 includes a CPU 151, an input interface 152, a display device 153, a communication device 154, a main storage device 155, and an external storage device 156. These components are interconnected by a bus 157.

A Central Processing Unit (CPU)151 executes an abnormality detection program as a computer program on a main storage 155. The abnormality detection program is a computer program that realizes each of the above-described functional components of the abnormality detection apparatus. Each functional component is realized by the CPU 151 executing an abnormality detection program.

The input interface 152 is a circuit for inputting an operation signal from an input device such as a keyboard, a mouse, or a touch panel to the abnormality detection device.

The display device 153 displays data or information output from the abnormality detection device. The display device 153 is, for example, a Liquid Crystal Display (LCD), a Cathode Ray Tube (CRT), or a Plasma Display Panel (PDP), but is not limited thereto. Data or information output from the computer device 100 may be displayed by the display device 153.

The communication device 154 is a circuit that allows the abnormality detection device to communicate with an external device in a wireless or wired manner. The measurement information may be input from an external device through the communication device 154. The measurement information input from the external device may be stored in the information database 310.

The main storage 155 stores, for example, an abnormality detection program, data necessary for executing the abnormality detection program, and data generated by executing the abnormality detection program. The exception detection program is loaded onto main storage 155 and executed. The main memory device 155 is, for example, a RAM, a DRAM, or an SRAM, but is not limited thereto. The information database 310, the model database 320, and the test result database 330 may be built on the main storage 155.

The external storage device 156 stores, for example, an abnormality detection program, data necessary for executing the abnormality detection program, and data generated by executing the abnormality detection program. When the abnormality detection program is executed, the program and data are read into the main storage device 155. The external storage device 156 is, for example, a hard disk, an optical disk, a flash memory, or a magnetic tape, but is not limited thereto. The information database 310, the model database 320, and the detection result database 330 may be built on the external storage device 156.

The abnormality detection program may be installed in advance on the computer apparatus 100, or may be stored in a storage medium such as a CD-ROM. Alternatively, the anomaly detection program may be uploaded on the internet.

The computer apparatus 100 may include one or more CPUs 151, one or more input interfaces 152, one or more display devices 153, one or more communication devices 154, and one or more main storage devices 155, and may be connected to a peripheral device such as a printer or a scanner.

The abnormality detection apparatus may be implemented by a single computer apparatus 100, or may be configured as a system including a plurality of computer apparatuses 100 connected to each other.

Fig. 18 is a flowchart of an abnormality detection process performed in the operation mode according to an embodiment of the present invention. The processing of the flowchart shown in fig. 18 may be executed at a certain operation of the abnormality detection target system, may be executed in a constant cycle, may be executed upon receiving an instruction from a user such as a maintenance person, or may be executed at any other time.

In step S101, the abnormality detector 110 acquires the travel information as the abnormality detection target from the information database 310.

In step S102, the abnormality detector 110 selects a prediction model corresponding to an abnormality detection target system (in this example, a brake system of the vehicle) from the model database 320. The abnormality detector 110 also selects a threshold value corresponding to a running condition that is satisfied by the acquired running information among the plurality of running conditions. For example, the prediction model is a model for predicting a target variable representing a state (e.g., deceleration) of the vehicle from an explanatory variable representing a control command value (e.g., brake groove) for the vehicle. In other words, the predictive model is a model that associates an explanatory variable representing a control command value for the vehicle with a target variable representing a state of the vehicle.

In step S103, the abnormality detector 110 generates a feature vector from the acquired travel information. For example, the anomaly detector 110 generates a feature vector that includes control command values. The number of elements of the feature vector may be one or more. The anomaly detector 110 predicts a target variable (deceleration in this example) based on the feature vector and the prediction model. In other words, the abnormality detector 110 calculates a predicted value of the state of the vehicle based on the control command value and the prediction model.

In step S104, the abnormality detector 110 calculates a residual that is a difference between the predicted deceleration and the deceleration included in the running information, and compares the calculated residual with a threshold value.

When the residual is larger than the threshold (yes), the abnormality detector 110 detects an abnormality and outputs information notifying that the abnormality is detected to the screen display device 900 or the like (S105).

When the residual is equal to or smaller than the threshold (no), the abnormality detector 110 detects no abnormality (S106). In other words, the abnormality detector 110 determines that the brake system of the vehicle is normal. When no abnormality is detected, information notifying that the brake system of the vehicle is normal may be output to the screen display device 900 or the like.

Fig. 19 is a flowchart of the threshold setting processing at the abnormality detecting device in the learning mode. This process may be performed at a constant cycle, may be performed at a time instructed by a maintenance person, or may be performed at any other time. An exemplary operation when a plurality of threshold values according to the running condition are set to the prediction model is described below. It is assumed that the abnormality detector 110 performs abnormality detection based on a prediction model generated in advance and a threshold value, and the detection result database 330 stores data relating to abnormality detection.

In step S201, the condition generator 230 assigns a residual category to a residual between the predicted value and the actual value according to a value of the residual based on the detection result database 330. The condition generator 230 generates a data set in which the residual category is associated with the travel information (see fig. 10).

In step S202, the condition generator 230 sets each item of the data set as an explanatory variable and sets the residual category as a target variable, and performs, for example, machine learning to generate a classifier that predicts the target variable from at least one of the plurality of explanatory variables. In particular, condition generator 230 generates a classifier that associates a plurality of conditions related to at least one explanatory variable with a plurality of residual categories. In this example, a decision tree (refer to fig. 11) is generated as the classifier.

In step S203, the condition generator 230 acquires the conditions included in the classifier as a plurality of running conditions. When the decision tree is used, a condition included in a path from each residual category node (end node) to the root node is acquired as a traveling condition corresponding to the residual category.

In step S204, the threshold setter 220 sets a plurality of thresholds to the running condition. For example, the threshold setter determines the threshold based on the distribution of the residuals classified as the residual category satisfying each of the traveling conditions of the traveling information. For example, the travel information used to generate the decision tree (or the travel information not used to generate the decision tree) is classified into a group satisfying the travel condition. Abnormality detection is performed for each group, and a residual is calculated based on the detection result. Then, a probability distribution of the residuals is generated (refer to fig. 8). A residual value corresponding to a predetermined probability (such as a higher level X percentile) in the probability distribution, or a value based on a value of two or three times the standard deviation σ, is determined as the threshold value.

In step S205, the threshold setter 220 stores pairs of thresholds and running conditions in the model database 320 in association with the corresponding prediction models.

Although the present embodiment describes an example in which the target variable of the prediction model is the deceleration of the brake, a prediction model for predicting another state of the vehicle (for example, the braking distance of the brake) may be used instead. The braking distance may be measured by calculating the distance, for example, since braking was initiated until the vehicle stopped or until a desired deceleration or speed was reached. Alternatively, a prediction model for predicting both the deceleration and the braking distance of the brake may be used. In this case, for example, equation (1) is prepared for each of the deceleration and the braking distance. Accordingly, the number of target variables of the predictive model is two. In this way, the predictive model may have multiple target variables instead of a single target variable. In this case, an anomaly may be detected when the residual error for each target variable or any one of the target variables exceeds a threshold.

According to the present embodiment, a large number of abnormality detection models suitable for various conditions can be generated by setting thresholds according to the running conditions. Thresholds suitable for, for example, a plurality of time slots such as morning, afternoon and evening, routes in a plurality of areas such as urban, suburban and mountain areas, all seasons such as spring, summer, fall and winter, and detailed conditions of various weather such as rainy, snowy, sunny weather may be set.

In the first embodiment described above, a plurality of threshold values are set to the same prediction model according to the running condition. However, in the second embodiment, a plurality of abnormality detection models (a plurality of pairs of prediction models and threshold values) may be generated in accordance with the running condition. In this case, when abnormality detection is performed, a travel condition satisfying the current travel information is specified, and an abnormality detection model (prediction model and threshold value) corresponding to the specified travel condition is used.

Model generator 210 generates a predictive model for each of a plurality of driving conditions. The threshold value setter 220 sets a threshold value corresponding to each predictive model (in other words, a threshold value corresponding to each driving condition).

Specifically, the model generator 210 generates a plurality of running conditions, similarly to the first embodiment. The model generator 210 extracts data satisfying each of the driving conditions from the driving information, and generates a prediction model by using the extracted data. The prediction model is generated by the same method as that in the above-described embodiment. The threshold value setter 220 sets a threshold value corresponding to each predictive model (in other words, a threshold value corresponding to each driving condition) in the same manner as in the above-described embodiment. The generated prediction model, the set threshold value, and the corresponding running condition are stored in the model database 320. Fig. 20 shows an exemplary model database 320 according to a second embodiment. Models 0001_ a, 0001_ B, and 0001_ C are generated instead of the model 0001 illustrated in fig. 7. In other words, three abnormality detection models are newly generated instead of one abnormality detection model. The generation of multiple models in place of one model is referred to as model partitioning. A travel condition column is additionally provided to store travel conditions corresponding to each model.

When the above-described decision tree shown in fig. 11 is generated, the model 0001_ a is used when the driving condition "sunny day (no precipitation) is satisfied and the occupancy is equal to or lower than 90%". The model 0001_ B is used when the driving condition "clear day (no precipitation) and riding rate higher than 90%" is satisfied. The model 0001_ C is used when the driving condition "rainy weather (no rainfall)" is satisfied.

By recursively repeating the model division on the abnormality detection models generated in the present embodiment, a large number of abnormality detection models suitable for various conditions can be generated. An abnormality detection model suitable for, for example, detailed conditions such as a plurality of time periods such as morning, afternoon, and evening, routes in a plurality of areas such as urban areas, suburban areas, mountain areas, all seasons such as spring, summer, fall, and winter, and various weather such as rainy, snowy, sunny weather can be generated.

This embodiment may be combined with the first embodiment. Specifically, a plurality of thresholds according to the running condition may be set for each of a plurality of abnormality detection models generated by model division. This allows generation of an abnormality detection model corresponding to more detailed conditions.

While certain embodiments have been described, these embodiments have been presented by way of example only, and are not intended to limit the scope of the invention. Indeed, the novel embodiments described herein may be embodied in various other forms; furthermore, various omissions, substitutions and changes in the form of the embodiments described herein may be made without departing from the spirit of the inventions.

[ description of reference numerals ]

10 brake lever

20 railway

30 wheel

41 brake pad

42 tread brake

43 air cylinder

50 load response device

51 air spring

60a, 60b main motor

70 resistor

80 pantograph

90 line

100 abnormality diagnosis device

101 vehicle information collector

102 environment information collector

103 data processor

110 anomaly detector

120 alarm

130 screen generator

151 CPU

152 input interface

153 display device

154 communication device

155 main storage device

156 external storage device

157 bus

200 anomaly detection model generator

210 model generator

220 threshold value setter

230 condition generator

310 database of information

310a, 310b, 310c table

320 model database

330 database of test results

500 vehicle system

600 environment information system

700 terminal

800 input device

900 screen display device

901 main screen

902 Exception detail screen

903 check screen

Claims

1. An abnormality detection device comprising:

a condition generator configured to generate a classifier based on:

a difference between the predicted value of the state of the vehicle and the measured value of the state of the vehicle,

depending on the class of the difference, an

An explanatory variable of the vehicle travel information is,

the classifier is configured to classify the difference values into classes according to an interpretation variable;

a threshold setter configured to set a plurality of thresholds for abnormality detection for the categories based on the difference values classified into the categories by the classifier; and

an anomaly detector configured to:

calculating a predicted value of the state of the vehicle based on the control command value corresponding to the first timing and the prediction model,

classifying, by a classifier, a difference between the calculated predicted value and a measured value corresponding to the state of the vehicle at the first time as one of the classes based on an explanatory variable corresponding to the vehicle information at the first time; and

the difference between the calculated predicted value and the measured value corresponding to the state of the vehicle at the first time is compared with a threshold value corresponding to the classified category, thereby detecting whether the vehicle has an abnormality.

2. The abnormality detection apparatus according to claim 1, wherein the threshold setter determines the threshold corresponding to the category based on a distribution of the difference values classified into the category by the classifier.

3. The abnormality detection apparatus according to claim 2, wherein the threshold setter generates a probability distribution of the difference values, and sets the threshold to a value based on a standard deviation of the probability distribution or a value of the difference value corresponding to a predetermined probability in the probability distribution.

4. The abnormality detection apparatus according to any one of claims 1 to 3, wherein the threshold setter receives an instruction to set the plurality of thresholds through the user interface, and sets the thresholds based on the setting instruction.

5. The abnormality detection device according to any one of claims 1 to 3, further comprising: a model generator configured to generate a plurality of prediction models for a category, in each of which a control command value for a vehicle is associated with a state of the vehicle;

wherein the abnormality detector specifies a category based on an explanatory variable corresponding to the vehicle information at the first time and the classifier,

calculating a predicted value of the state of the vehicle based on the prediction model corresponding to the specified category and the control command value corresponding to the first time, and

the difference between the calculated predicted value and the measured value corresponding to the state of the vehicle at the first time is compared with a threshold value corresponding to the predicted category, thereby detecting whether the vehicle has an abnormality.

6. The abnormality detection device according to any one of claims 1 to 3,

the control command value is a command value related to the magnitude of braking of the vehicle, an

The state includes deceleration of the vehicle or air brake pressure.

7. The abnormality detection device according to any one of claims 1 to 3, wherein the running information includes at least one of measurement information of at least one of sensors of the vehicle and environmental information of the vehicle.

8. An anomaly detection method comprising:

generating a classifier based on:

depending on the class of the difference, an

An explanatory variable of the vehicle travel information is,

setting a plurality of thresholds for abnormality detection for the categories based on the difference values classified into the categories by the classifier; and

9. A computer program for causing a computer to execute processing comprising:

generating a classifier based on:

depending on the class of the difference, an

An explanatory variable of the vehicle travel information is,