JP5665534B2 - Level crossing evaluation system, level crossing evaluation method and program - Google Patents

Description

  The present invention relates to a level crossing evaluation system, a level crossing evaluation method, and a program for identifying a level crossing characteristic that has a strong influence on the occurrence of a level crossing accident among a plurality of level crossing characteristics indicating characteristics of a level crossing.

  In analyzing the background factors of level crossing accidents, how much influence each explanatory variable (background factor) has on the target variable (level crossing accident) by analyzing with quantification theory class I (class 1) Patent Document 1 discloses a technique for analyzing whether the data is given.

"Human Factors for Level Crossers", "online", "Search on 02/06/2010", Internet <URL: http://www.rtri.or.jp/infoce/rrr/2009/07/200907_04. pdf>

  However, the above-mentioned technology focuses on which level crossing characteristic variable <explanatory variable (background factor)> affects the level crossing accident, and which variable range (information content) of that variable is the level crossing accident. We are not analyzing whether it affects For example, when the level crossing characteristic variable affecting the level crossing accident is the level crossing length, no analysis is made as to how long the level crossing length affects the level crossing accident. As a result, it has not been possible to accurately identify the level crossing characteristic variable affecting the level crossing accident and its range (information amount).

  Accordingly, an object of the present invention is to provide a level crossing evaluation system, a level crossing evaluation method, and a program thereof that can accurately specify a level crossing characteristic variable affecting the level crossing accident and its range (information amount). .

In order to achieve the above-described object, the present invention provides a cross-sectional feature variable indicating the characteristics of a plurality of crossings for each of a plurality of different level crossings, an explanatory variable including the amount of information, and an objective variable including the number of accidents at each of the crossings. Based on the division of the domain in which the information amount is distributed, the domain division unit that divides the domain in which the information amount of the crossing feature variable in the explanatory variable is distributed, and the division of the domain in which the information amount is distributed. An information amount criterion calculation unit that converts each information amount in each of the converted regions into a dummy variable, and calculates an information amount criterion by multiple regression analysis using the dummy variable and the objective variable including the number of accident occurrences, The best information amount criterion is calculated among the plurality of information amount references obtained by repeatedly calculating the dividing position of the region where the information amount of the crossing feature variable is distributed every time the range dividing unit is changed. Each information amount in each divided region according to the division position is converted into a dummy variable, and multiple regression analysis using the dummy variable for each of the plurality of crossing feature variables is performed, and the crossing feature variable The degree of contribution to each level crossing accident, (1) the degree of contribution indicated by each partial regression coefficient as a result of the multiple regression analysis using each of the dummy variables, or (2) each change of the level crossing characteristic variable. the contribution of each crossing feature variables, either one of which is indicated by the difference between the maximum value and the minimum value of partial regression coefficient which is the result of the multiple regression analysis using a dummy variable each for each information amount in the range of the multivariate regression processing unit for calculating a contribution degree, wherein the level crossing characteristic variable indicating a large contribution of value out of the contribution to the railway crossing accident is the result of multiple regression analysis of multivariate And a output unit for outputting as a factor resulting from the crossing accident, a railroad crossing evaluation system, characterized in that it comprises a.

Further, the present invention is the crossing evaluation system described above, wherein the domain dividing unit performs a process of dividing a domain in which the information amount of the level crossing feature variable is distributed by a plurality of different dividing methods , and calculates the information criterion The domain division unit converts the information amount of each domain based on the division when the domain is divided by one of the plurality of different division methods into a dummy variable, and the dummy variable And processing for calculating the information criterion by multiple regression analysis using the objective variable consisting of the number of accident occurrences is performed for each of the cases where the domain is divided by all the division methods of the plurality of different division methods. The multivariate multiple regression processing unit repeatedly calculates the division position of the domain where the information amount of the crossing feature variable is distributed and the division method every time the domain division unit changes. Among the information amount criteria, the information amount in each divided region at the division position where the best information amount criterion is calculated is converted into a dummy variable, and the dummy variable for each of the plurality of crossing feature variables is A multiple regression analysis with the used multivariate is performed, and the degree of contribution of each of the crossing feature variables to the crossing accident is calculated .

  Further, the present invention is the above-described crossing evaluation system, wherein after the multiple regression analysis with the multivariate, the best combination of the crossing feature variables is selected from among the plurality of crossing feature variables using a stepwise method. And

Further, the present invention is a level crossing evaluation method in a level crossing evaluation system, wherein the input unit of the level crossing evaluation system includes a level crossing feature variable indicating a characteristic of a plurality of level crossings for each of a plurality of different level crossings and an explanatory variable comprising the information amount thereof. And an objective variable consisting of the number of accident occurrences at each level crossing, and the domain division unit of the level crossing evaluation system divides a domain in which the information amount of the level crossing feature variable in the explanatory variable is distributed, The information amount standard calculation unit of the level crossing evaluation system converts each information amount in each divided region into a dummy variable based on the division of the region where the information amount is distributed, and the dummy variable and the accident An information criterion is calculated by multiple regression analysis using an objective variable consisting of the number of occurrences, and the multivariate multiple regression processing unit of the crossing evaluation system uses the information on the crossing feature variable Among the plurality of information amount criteria obtained by repeatedly calculating the division position of the domain where the distribution is changed, each time it is divided by the division position where the best information amount criterion is calculated. Each information amount in each domain is converted into a dummy variable, and a multiple regression analysis using the dummy variable for each of the plurality of level crossing feature variables is performed, and each level crossing feature variable contributes to the level crossing accident. (1) the contribution indicated by each partial regression coefficient that is the result of the multiple regression analysis using each dummy variable, or (2) each information amount in each domain of the crossing feature variable contribution of each of the crossings characteristic variables indicated by the difference between the maximum value and the minimum value of partial regression coefficient which is the result of the multiple regression analysis using dummy variables, respectively, calculated either contribution of Then, the output unit of the level crossing evaluation system causes the level crossing characteristic variable indicating the contribution level having a large value among the contributions to the level crossing accident as a result of the multiple regression analysis with the multivariate to be attributed to the level crossing accident. It is a level crossing evaluation method characterized by outputting as a factor.

Further, the present invention is the crossing evaluation method described above, wherein the domain dividing unit performs processing for dividing a domain in which the information amount of the level crossing feature variable is distributed by a plurality of different dividing methods , and calculates the information criterion The domain division unit converts the information amount of each domain based on the division when the domain is divided by one of the plurality of different division methods into a dummy variable, and the dummy variable And processing for calculating the information criterion by multiple regression analysis using the objective variable consisting of the number of accident occurrences is performed for each of the cases where the domain is divided by all the division methods of the plurality of different division methods. The multivariate multiple regression processing unit is configured to repeatedly calculate a domain division position where the information amount of the level crossing feature variable is distributed and the division method every time the domain division unit changes. Of the recorded information criteria, each information amount in each divided region at the division position for which the best information criterion is calculated is converted into a dummy variable, and the dummy variable for each of the plurality of crossing feature variables is used. A multi-variate multiple regression analysis is performed, and the degree of contribution of each level crossing feature variable to the level crossing accident is calculated .

  According to the present invention, in the above level crossing evaluation method, after the multivariate multiple regression analysis, a best combination of the level crossing feature variables is selected from among the plurality of level crossing feature variables using a stepwise method. And

Further, the present invention provides a computer of a level crossing evaluation system, comprising a crossing feature variable indicating characteristics of a plurality of level crossings for each of a plurality of different level crossings, an explanatory variable including the amount of information, and an objective variable including the number of accidents at each level crossing. And an input means for inputting, a domain dividing means for dividing a domain in which the information amount of the level crossing feature variable in the explanatory variable is distributed, and a division of the domain in which the information amount is distributed. Information amount criterion calculation means for converting each information amount in each domain into a dummy variable, and calculating an information amount criterion by multiple regression analysis using the dummy variable and the objective variable consisting of the number of accident occurrences, the crossing feature The best information amount among the plurality of information criterion obtained by repeatedly calculating the division position of the domain where the information amount of the variable is distributed every time the domain division unit changes Converting each information amount in each divided domain according to the division position where the quasi is calculated into a dummy variable, performing a multiple regression analysis with the multivariate using the dummy variable for each of the plurality of crossing feature variables, The degree of contribution of each level crossing feature variable to a level crossing accident, (1) the degree of contribution indicated by each partial regression coefficient as a result of the multiple regression analysis using each of the dummy variables, or (2) the level crossing feature The contribution for each level crossing feature variable indicated by the difference between the maximum value and the minimum value of the partial regression coefficients that are the results of the multiple regression analysis using each dummy variable for each information amount in each variable region , multivariate regression processing means for calculating either the contribution of, before showing a large contribution of value out of the contribution to the railway crossing accident is the result of multiple regression analysis with the multivariate The crossing characteristic variable is a program for causing to function as an output unit, which outputs as a factor resulting from the crossing accident.

Further, in the present invention, the domain dividing unit performs a process of dividing a domain in which the information amount of the crossing feature variable is distributed by a plurality of different dividing methods , and the information amount criterion calculating unit is configured to perform the domain dividing unit. When the means divides the domain by one of the plurality of different division methods, the information amount of each domain based on the division is converted into a dummy variable, and the dummy variable and the number of accidents are calculated. The multivariate multiple regression process is performed for each case where the domain is divided by all the division methods of the plurality of different division methods. A plurality of information criterion values obtained by the means repeatedly calculating the dividing position and the dividing method of the domain where the information amount of the crossing feature variable is distributed every time the domain dividing unit changes. Converting each information amount in each divided region according to the division position for which the best information amount criterion is calculated into a dummy variable, and multiplying each of the plurality of level crossing feature variables by a multivariate using the dummy variable. A program that performs regression analysis and calculates a degree of contribution of each level crossing characteristic variable to the level crossing accident .

  In the present invention, after the multiple regression analysis with the multivariate, the computer is caused to execute a process of selecting a best crossing feature variable combination among the plurality of crossing feature variables using a stepwise method. It is a program.

  According to the present invention, since a statistical model evaluated to be the best calculates a level crossing feature variable and a degree of contribution to the level crossing accident for each region of a certain amount of information in the level crossing feature variable, the user can Based on the degree of contribution, it is possible to accurately grasp which level crossing feature variable and the range of the information amount in the level crossing feature variable contribute to the level crossing accident.

It is a block diagram which shows the structure of a level crossing evaluation system. It is a figure which shows the processing flow of a level crossing evaluation system. It is a figure which shows the example of data of an analysis object memory | storage table. It is a figure which shows the outline | summary of the division process of a domain division part. It is a figure which shows the list of the contribution with respect to generation | occurrence | production of the accident of each level crossing characteristic variable.

Hereinafter, a crossing evaluation system according to an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is a block diagram showing a configuration of a level crossing evaluation system according to the embodiment.
In this figure, reference numeral 1 denotes a crossing evaluation system.

The level crossing evaluation system 1 includes an input unit 11, a domain dividing unit 12, an information criterion calculation unit 13, a multivariate multiple regression processing unit 14, an output unit 15, and a control unit 16.
Here, the input unit 11 inputs a level crossing feature variable indicating the characteristics of a plurality of level crossings for each of a plurality of different level crossings, an explanatory variable including the amount of information, and a target variable including the number of accidents at each level crossing. Part.
The domain dividing unit 12 is a processing unit that divides a domain in which the information amount of the crossing feature variable in the explanatory variable is distributed.
The information amount reference calculation unit 13 converts each information amount in each divided region into a dummy variable based on the division of the region where the information amount is distributed, and includes the dummy variable and the number of accidents. It is a processing unit that calculates an information criterion by multiple regression analysis using objective variables.
In addition, the multivariate multiple regression processing unit 14 selects the best information amount criterion from among a plurality of information amount criteria obtained by repeatedly calculating by changing the division position of the domain in which the information amount of the level crossing feature variable is distributed. It is a processing unit that converts each information amount in each divided region according to the calculated division position into a dummy variable, and performs multiple regression analysis with multivariate using the dummy variable for each of a plurality of level crossing feature variables. .
The output unit 15 is a processing unit that outputs a level crossing feature variable indicating a contribution level having a large value among the contributions to the level crossing accident as a result of the multivariate multiple regression analysis as a factor resulting from the level crossing accident.
The control unit 16 is a processing unit that controls other processing units.
The level crossing evaluation system 1 includes each processing unit as described above, and performs a process of accurately identifying a level crossing characteristic variable affecting the level crossing accident and its range (information amount).

FIG. 2 is a diagram showing a processing flow of the crossing evaluation system.
Hereinafter, the details of the processing flow of the crossing evaluation system will be described in order.
First, the input unit 11 of the level crossing evaluation system 1 uses a plurality of level crossing feature variables indicating the characteristics of level crossings and explanatory variables including the amount of information from a level crossing ledger database, for example, a plurality of railroad lines owned by a certain railway company. Each level crossing (with ID for identifying a level crossing) is input (step S101). Here, the level crossing characteristic variables are information such as level crossing length, level crossing gradient, level crossing road width, level crossing width, proximity of intersections, railway traffic volume, blocking time, pedestrian traffic volume, etc. If it is a width or a level crossing, the amount of information is length, if it is rail traffic, the amount of information is the number of passes per unit time (1 day), and if it is pedestrian traffic, the amount of information is unit time ( The number of people crossing a level crossing per day). In addition to this, the input unit 11 inputs various level crossing feature variables and explanatory variables including information amounts from the level crossing ledger database.
In addition, the input unit 11 inputs a combination of a level crossing ID of a plurality of level crossings existing on a railway line owned by a certain railway company and the number of accidents (including 0) as an objective variable at each level crossing (step S102). .
Then, the input unit 11 generates combination information of a plurality of crossing feature variables of the crossing indicated by the crossing ID, their respective information amounts, and the number of accidents (including 0) based on the input crossing ID. Then, the crossing ID, the plurality of crossing characteristic variables of the crossing indicated by the crossing ID, and the combination information of each information amount and the number of accidents are output to the domain division unit 12. Then, the regional dividing unit 12 determines the information amount of each level crossing feature variable and the number of accidents for each level crossing ID based on the level crossing ID, the plurality of level crossing characteristic variables, the information amount of each of those variables, and the combination information of the number of accident occurrences. Are stored in the analysis target storage table (FIG. 3). The analysis target storage table is stored in a memory or the like held in the domain division unit 12.

  Next, the domain dividing unit 12 has an accident for each different information amount (for example, different level crossing lengths) in the range (for example, 5 to 20 m) of the information amount (level crossing length) of a certain level crossing characteristic variable (for example, level crossing). The number of occurrences is specified from the information recorded in the analysis target storage table (FIG. 3) (step S103). For example, if the variable for the crossing feature variable 1 is the crossing length, specify the number of crossings with a crossing length of 5 m, the number of crossings with 6 m, the number of crossings with 7 m, and so on. If there is, 5m to 20m is specified as the range of the information amount (crossing length). Then, the domain dividing unit 12 divides the domain in which the information amount of a certain level crossing feature variable is distributed (step S104).

FIG. 4 is a diagram showing an outline of the dividing process of the domain dividing unit.
FIG. 4 shows the number of accidents occurring at each level crossing length (information amount) of a railroad crossing in which a certain railway company has. In FIG. 4, the three arrows a, b, and c show an example in which the domain dividing unit 12 divides the domain indicated by the railroad crossing length into three.
When the domain dividing unit 12 divides the domain in which the information amount of a certain level crossing feature variable is distributed, the information criterion calculation unit 13 then calculates each information amount in each divided domain. An information criterion is obtained by multiple regression analysis using the dummy variable (step S105) and the number of accidents (objective variable) corresponding to the dummy variable and the explanatory variable (crossing feature variable) corresponding to the dummy variable. Is calculated (step S106). The AIC is

It is represented by In the formula (1), K represents the number of free parameters of the statistical model, and N represents the number of data (number of crossings). In addition, the log likelihood l (θ) is

Can be represented by In f of this equation (2), x (information amount of the level crossing feature variable) indicates a probability function in the case of a discrete distribution and a probability density function in the case of a continuous distribution. Θ represents a statistical model.
Further, in the case of Equation (2), the maximum log likelihood is

It is.
One AIC is calculated based on one division process in the domain division unit 12 by such processing of the information amount reference calculation unit 13.
In addition, the domain division unit 12 determines whether to repeatedly calculate the AIC using another division method (step S107), and changes the division position of the domain where the information amount of the level crossing feature variable is distributed, The number is increased from 2 to n, and the division process is repeated. Each time the division process is performed, the information amount criterion calculation unit 13 converts each information amount in each divided domain into a dummy variable, and performs AIC (Akaike's Akaike) by multiple regression analysis using the dummy variable. Information criterion). Then, the information amount criterion calculation unit 13 determines the AIC having the lowest value as the best value among the calculated AICs as the AIC of the crossing feature variable (step S108). That is, the statistical model that is evaluated to be the best by selecting the AIC having the lowest value is selected. Then, the control unit 16 determines whether or not the AIC has been determined for all the level crossing feature variables (step S109), and if the AIC has not been determined for all the level crossing feature variables, the domain division unit 12 and the information amount are determined. The reference calculation unit 13 is controlled, and the above-described processing from step S103 to step S107 is repeated until AIC is determined for all crossing feature variables.

  Next, when AIC is determined for all crossing feature variables, the control unit 16 controls the multivariate multiple regression processing unit 14 to start processing, and the multivariate multiple regression processing unit 14 starts processing. Then, the multivariate multiple regression processing unit 14 is the highest value among a plurality of information criterion (AIC) obtained by repeatedly calculating by changing the division position of the domain in which the information amount of the crossing feature variable is distributed. By converting each information amount in each divided domain according to the division position where the low (best) information criterion is calculated into a dummy variable, and using the dummy variable for each of a plurality of level crossing feature variables Multiple regression analysis (quantification theory class I) is performed (step S110). As a result, the multivariate multiple regression processing unit 14 calculates the degree of contribution of each level crossing feature variable to the occurrence of an accident. And the output part 15 outputs the contribution with respect to generation | occurrence | production of each accident of each level crossing characteristic variable (step S111).

In addition, when the information amount of a plurality of level crossing feature variables is divided into arbitrary domains, converted into dummy variables, and multiple regression analysis is performed, the following two types of contributions exist. At this time, which definition is adopted depends on the purpose of utilization.
(I) When comparing contributions between dummy variables The contribution is defined by the partial regression coefficient of each dummy variable in the result of multiple regression analysis.
(Ii) When comparing contributions between level crossing feature variables The contribution is defined for each level crossing feature variable. In the result of the multivariate multiple regression analysis, the difference between the maximum value and the minimum value among the partial regression coefficients of the dummy variable divided from one level crossing feature variable is defined as the contribution of each level crossing feature value. This is the same as what is called category range in quantification class I, which is one of multiple regression analysis methods using dummy variables.

FIG. 5 is a diagram showing a list of contributions to the occurrence of each accident at each level crossing feature variable.
In the figure, it can be seen that the level crossing feature variable is output sorted in descending order of contribution. Thus, the user can grasp which level crossing characteristic variable contributes to the level crossing accident based on the magnitude of the numerical value.

  As described above, the embodiment of the present invention has been described. However, according to the above-described processing, a plurality of level crossing feature variables and a certain amount of information in each level crossing feature variable according to the statistical model evaluated to be the best. Since the degree of contribution to the level crossing accident is calculated, the user can accurately determine which level crossing feature variable and the range of information in the level crossing feature variable contribute to the level crossing accident based on the degree of contribution. I can grasp it.

  In the above-described processing, the information amount criterion calculation unit 13 calculates the information amount criterion AIC for Akaike, but instead, the information amount criterion calculation unit 13 calculates the information amount criterion. ) Or MDL (minimum description length) may be used. The calculation formula for the Bayesian information criterion (BIC) is:

The calculation formula of MDL (longest description length) is

Is represented by
Further, in the above-described processing, the domain dividing unit 12 changes the division position of the domain in which the information amount of the level crossing feature variable is distributed, increases the number of divisions from 2 to n, and repeats the division process. An example in which the information amount criterion calculation unit 13 calculates an information amount criterion by multiple regression analysis using dummy variables based on a plurality of different divisions is described. However, as a division method, for example, a domain in which the information amount of a crossing feature variable is distributed is further divided using another division method, and an information criterion is used using a dummy variable based on the division performed by the division processing. The calculation unit 13 may calculate an information amount criterion by multiple regression analysis, and may determine the information amount criterion indicating the lowest value among all the calculated information amount criteria as the AIC of the crossing feature variable. . As another division method, division by one value indicated by the information amount of the domain (for example, a division method in the case of dividing each crossing length of 5 m, 6 m, 7 m, and 8 m for each information amount), There are 10 divisions of the domain, 10 divisions of the percentile of the domain, 9 divisions with the standard deviation of the domain.

Furthermore, in the above-described processing, multivariate multiple regression analysis (quantification theory class I) is performed using dummy variables for each of all crossing feature variables among a plurality of crossing feature variables. However, in order to select the best crossing feature variable and obtain the regression equation, after performing multiple regression analysis (quantification theory class I) with multivariate, perform Stepwize method <stepwise method (sequential method)> Thus, a crossing feature variable that performs multivariate multiple regression analysis among a plurality of crossing feature variables may be selected.
In this stepwise method, the variable selection is often based on the F value in the multiple regression analysis, but AIC, BIC, and MDL can also be used as the variable selection reference value. Here, the variable may be defined as a crossing feature variable composed of a plurality of dummy variables, or may be defined as each dummy variable. Which definition is adopted depends on the purpose of utilization.

  The above-mentioned crossing evaluation system has a computer system inside. Each process described above is stored in a computer-readable recording medium in the form of a program, and the above process is performed by the computer reading and executing the program. Here, the computer-readable recording medium means a magnetic disk, a magneto-optical disk, a CD-ROM, a DVD-ROM, a semiconductor memory, or the like. Alternatively, the computer program may be distributed to the computer via a communication line, and the computer that has received the distribution may execute the program.

  The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

DESCRIPTION OF SYMBOLS 1 ... Railroad crossing evaluation system 11 ... Input part 12 ... Domain division | segmentation part 13 ... Information amount reference | standard calculation part 14 ... Multivariate multiple regression process part 15 ... Output part 16 ... Control unit

Claims (9)

An input unit for inputting a level crossing characteristic variable indicating the characteristics of a plurality of level crossings for each of a plurality of different level crossings and an explanatory variable including the amount of information, and an objective variable including the number of accidents at each level crossing;
A domain dividing unit that divides a domain in which the information amount of the level crossing feature variable in the explanatory variable is distributed;
Based on the division of the region where the information amount is distributed, each information amount in each divided region is converted into a dummy variable, and the overlap using the dummy variable and the objective variable consisting of the number of accident occurrences. An information amount criterion calculation unit for calculating an information amount criterion by regression analysis;
The best information amount criterion was calculated among the plurality of information amount criteria obtained by repeatedly calculating the region dividing position where the information amount of the level crossing feature variable is distributed every time the region dividing unit is changed. Each information amount in each divided region according to the division position is converted into a dummy variable, and multiple regression analysis using the dummy variable for each of the plurality of crossing feature variables is performed, and the crossing feature variable The degree of contribution to each level crossing accident, (1) the degree of contribution indicated by each partial regression coefficient as a result of the multiple regression analysis using each of the dummy variables, or (2) each change of the level crossing characteristic variable. The contribution for each level crossing feature variable indicated by the difference between the maximum value and the minimum value of the partial regression coefficients that are the results of the multiple regression analysis using each dummy variable for each information amount in the region , A multivariate multiple regression processing unit for calculating one of the contributions ,
An output unit that outputs the level crossing characteristic variable indicating a large contribution value among the contributions to the level crossing accident as a result of the multiple regression analysis with the multivariate, as a factor resulting from the level crossing accident;
A crossing evaluation system characterized by comprising: The domain dividing unit performs a process of dividing the domain in which the information amount of the crossing feature variable is distributed by a plurality of different dividing methods,
The information criterion calculation unit converts the information amount of each domain based on the division into a dummy variable when the domain division unit divides the domain by one of the plurality of different division methods. Then, the process of calculating the information criterion by multiple regression analysis using the dummy variable and the objective variable consisting of the number of accident occurrences, the domain is divided by all the division methods of the plurality of different division methods Do for each of the cases,
The multivariate multiple regression processing unit is configured to repeatedly calculate the domain division position and the division method in which the information amount of the level crossing feature variable is distributed every time the domain division unit changes. Of the information criterion, each information amount in each divided region at the division position where the best information criterion is calculated is converted into a dummy variable, and the dummy variable for each of the plurality of crossing feature variables is used. The level crossing evaluation system according to claim 1, wherein a multi-variate multiple regression analysis is performed, and a degree of contribution of each level crossing feature variable to the level crossing accident is calculated. The multi-variable multiple regression analysis is performed, and the best crossing feature variable combination is selected from the plurality of crossing feature variables using a stepwise method. Railroad crossing evaluation system. A crossing evaluation method in a crossing evaluation system,
The input unit of the crossing evaluation system inputs a crossing feature variable indicating the characteristics of a plurality of crossings for each of a plurality of different crossings, an explanatory variable including the amount of information, and an objective variable including the number of accidents at each crossing. And
The domain dividing unit of the level crossing evaluation system divides a domain in which the information amount of the level crossing feature variable in the explanatory variable is distributed,
Based on the division of the domain in which the information amount is distributed, the information amount criterion calculation unit of the level crossing evaluation system converts each information amount in each divided region into a dummy variable, the dummy variable, An information criterion is calculated by multiple regression analysis using objective variables consisting of the number of accidents,
A plurality of pieces of information obtained by the multivariate multiple regression processing unit of the level crossing evaluation system repeatedly calculating each time the domain division unit changes the division position of the domain in which the information amount of the level crossing feature variable is distributed. Among the amount criteria, each information amount in each divided region at the division position where the best information amount criterion is calculated is converted into a dummy variable, and the dummy variable for each of the plurality of crossing feature variables is used. Performing multiple regression analysis with variables, and the degree of contribution of each level crossing feature variable to a level crossing accident, (1) Contribution indicated by each partial regression coefficient as a result of the multiple regression analysis using each of the dummy variables time, or (2) the difference between the maximum value and the minimum value of partial regression coefficient which is the result of the multiple regression analysis using a dummy variable each for each information amount at each variance range of the level crossing characteristic variable Calculating the degree of contribution for each of the crossing feature variables indicated by
The output unit of the level crossing evaluation system uses the level crossing characteristic variable indicating a large contribution among the contributions to the level crossing accident as a result of the multiple regression analysis with the multivariate as a factor resulting from the level crossing accident. A level crossing evaluation method characterized by output. The domain dividing unit performs a process of dividing the domain in which the information amount of the crossing feature variable is distributed by a plurality of different dividing methods,
The information criterion calculation unit converts the information amount of each domain based on the division into a dummy variable when the domain division unit divides the domain by one of the plurality of different division methods. Then, the process of calculating the information criterion by multiple regression analysis using the dummy variable and the objective variable consisting of the number of accident occurrences, the domain is divided by all the division methods of the plurality of different division methods Do for each of the cases,
The multivariate multiple regression processing unit is configured to repeatedly calculate the domain division position and the division method in which the information amount of the level crossing feature variable is distributed every time the domain division unit changes. Of the information criterion, each information amount in each divided region at the division position where the best information criterion is calculated is converted into a dummy variable, and the dummy variable for each of the plurality of crossing feature variables is used. The level crossing evaluation method according to claim 4, wherein a multivariate multiple regression analysis is performed, and a degree of contribution of each level crossing feature variable to the level crossing accident is calculated. The multi-variate multiple regression analysis is performed, and the best crossing feature variable combination is selected from among the plurality of crossing feature variables using a stepwise method. Crossing evaluation method. The computer of the level crossing evaluation system
Input means for inputting a crossing feature variable indicating the characteristics of a plurality of crossings for each of a plurality of different crossings and an explanatory variable including the amount of information, and an objective variable including the number of accidents at each of the crossings;
Domain dividing means for dividing a domain in which the information amount of the crossing feature variable in the explanatory variable is distributed;
Based on the division of the region where the information amount is distributed, each information amount in each divided region is converted into a dummy variable, and the overlap using the dummy variable and the objective variable consisting of the number of accident occurrences. An information criterion calculation means for calculating an information criterion by regression analysis;
The best information amount criterion was calculated among the plurality of information amount criteria obtained by repeatedly calculating the region dividing position where the information amount of the level crossing feature variable is distributed every time the region dividing unit is changed. Each information amount in each divided region according to the division position is converted into a dummy variable, and multiple regression analysis using the dummy variable for each of the plurality of crossing feature variables is performed, and the crossing feature variable The degree of contribution to each level crossing accident, (1) the degree of contribution indicated by each partial regression coefficient as a result of the multiple regression analysis using each of the dummy variables, or (2) each change of the level crossing characteristic variable. The contribution for each level crossing feature variable indicated by the difference between the maximum value and the minimum value of the partial regression coefficients that are the results of the multiple regression analysis using each dummy variable for each information amount in the region , Multivariate multiple regression processing means for calculating one of the contributions ,
An output means for outputting the level crossing characteristic variable indicating a large degree of contribution among the degrees of contribution to the level crossing accident as a result of the multiple regression analysis with the multivariate as a factor resulting from the level crossing accident;
A program characterized by functioning as The domain dividing means performs a process of dividing the domain in which the information amount of the crossing feature variable is distributed by a plurality of different dividing methods,
The information amount reference calculating unit converts the information amount of each domain based on the division into a dummy variable when the domain dividing unit divides the domain by one of the plurality of different division methods. Then, the process of calculating the information criterion by multiple regression analysis using the dummy variable and the objective variable consisting of the number of accident occurrences, the domain is divided by all the division methods of the plurality of different division methods Do for each of the cases,
The multivariate multiple regression processing means is configured to repeatedly calculate the domain division position and the division method in which the information amount of the level crossing feature variable is distributed each time the domain division means changes. Of the information criterion, each information amount in each divided region at the division position where the best information criterion is calculated is converted into a dummy variable, and the dummy variable for each of the plurality of crossing feature variables is used. The program according to claim 7, wherein multivariate multiple regression analysis is performed, and a degree of contribution of each level crossing feature variable to the level crossing accident is calculated. 8. The computer is caused to perform a process of selecting a best combination of crossing feature variables among the plurality of crossing feature variables using a stepwise method after the multiple regression analysis with the multivariate. Or the program of Claim 8.

Images