JP2004080838A - Automatic train operating apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To put the learning of properties forward automatically also after business travel by reducing the time and labor required for tuning. <P>SOLUTION: This on-line processes data acquired during travel of a train and automatically learns train properties during travel of the train, with the control parameters of train travel, the physical properties of the train itself during train operation, and the physical values showing the rout properties as the train properties, based on the data obtained hereby and the data obtained beforehand, and this performs the automatic operation of the train, using the train properties obtained by these automatic properties. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an automatic train operation device that learns the characteristics of a train and uses the learning result for automatic operation of the train.
[0002]
[Prior art]
An automatic train operation device (hereinafter, referred to as "ATO") has already been put to practical use as an attempt to make a train follow a speed limit of the train or a speed limit having a certain margin from the speed limit. . However, since error tracking control such as PI control is mainly used, it often depends on the characteristics of trains and routes. At present, it is necessary to adjust the characteristics or control parameters for each train or route. Requires considerable time and effort.
[0003]
In addition, an automatic train operation device that creates a travel plan and performs train travel based on the travel plan has been considered. When creating a travel plan, a simple train travel model may be used. At its simplest, the train operation in question is simply a simple physical formula,
F−Fr = M · α (1)
There is a method of expressing with. Here, F is a power running traction force or a braking force, Fr is a train running resistance, M is a train weight, and α is an acceleration (including negative acceleration, that is, deceleration). The train running resistance Fr is a resistance generated at the time of running the train, and is often considered to be configured based on the following resistances for convenience of calculation. That is,
Departure resistance: Resistance at departure
Air resistance: Resistance due to air on the train
Gradient resistance: Resistance due to route gradient
Curve resistance: Resistance due to the curve of the route
Tunnel resistance: Resistance generated when traveling in a tunnel
That is it. The air resistance is often treated as a quadratic expression of the speed in consideration of the resistance between the wheel tread surface and the rail surface.
[0004]
Generally, the train running resistance Fr is often considered to be based on a resistance including a gradient resistance, an air resistance, a curve resistance, a tunnel resistance, a departure resistance, and the like. Here, it is assumed that a train other than a tunnel is running, and the slope resistance, air resistance, and curve resistance are considered. In this case, the gradient resistance, the air resistance, and the curve resistance are obtained by the following equations (2), (3), and (4), respectively (for example, see the document “Operation theory (DC AC electric locomotive)”, edited by Koyusha) ). That is,
(A) Gradient resistance type
Frg = s (2)
Frg: gradient resistance [kg weight / ton],
s: slope [‰] (positive for uphill, negative for downhill)
(B) Air resistance type
Fra = A + Bv + Cv ² … (3)
Fra: air resistance [kg weight / ton],
A, B, C: coefficients,
v: speed [km / h],
(C) Curve resistance type
Frc = 800 / r (4)
Frc: Curve resistance [kg weight / ton],
r: radius of curvature [m].
[0005]
When using a model such as Equation (1) in automatic train operation, even in the automatic train operation method based on a travel plan, characteristics such as train characteristics and route characteristics greatly affect ride comfort and stopping accuracy. I do.
[0006]
[Problems to be solved by the invention]
As described above, the conventional method largely depends on the characteristics of trains and lines or control parameters, and a large amount of labor is required for adjusting the automatic train operation device. In the adjustment, it was necessary to actually run the train many times to acquire actual data and proceed with the adjustment offline.
[0007]
Therefore, an object of the present invention is to provide an automatic train driving device that can reduce the time and labor required for adjustment, further advance the learning of characteristics automatically after business operation, further improve ride comfort, and improve stop accuracy. To provide.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, an automatic train driving device according to the invention according to claim 1 includes a data processing unit that online processes data acquired during train running, data acquired during train running by the data processing unit, Based on the acquired data, the control parameters during train running, and the automatic characteristic learning means for automatically learning train characteristics and route characteristics during train running, and the train characteristics and route characteristics learned by the automatic characteristic learning means are used. And automatic train operation means for automatically operating the train.
[0009]
According to the first aspect of the invention, it is possible to automatically learn train characteristics and route characteristics online while traveling, and it is possible to perform automatic train operation using the results of the automatic learning.
[0010]
According to a second aspect of the present invention, there is provided the automatic train driving apparatus according to the first aspect, wherein a train characteristic and a line characteristic required for automatic train operation and an initial value of a control parameter are preliminarily estimated by a preliminary test run before the business. The automatic characteristic learning means includes a front characteristic estimating means, and the learning is advanced by running after business based on the initial value estimated by the pre-business characteristic estimating means.
[0011]
According to the second aspect of the present invention, in addition to the effect of the first aspect, it is possible to automatically learn the train characteristics and the characteristics relating to the route on-line by test running before the start of business, and thus employ an automatic train driving device. It is possible to reduce the time required for the adjustment beforehand and the labor required for the adjustment.
[0012]
According to a third aspect of the present invention, in the automatic train driving device according to the first or second aspect, the automatic characteristic learning means learns when the assumed characteristic value is determined to be clearly different from the actual value during train running. Is performed, and the learning content is reflected in subsequent train running.
[0013]
According to the third aspect of the invention, in addition to the effect of the first aspect, efficient characteristic learning can be performed by performing automatic characteristic learning at an appropriate timing.
[0014]
According to a fourth aspect of the present invention, in the automatic train driving device according to the first or second aspect, the automatic characteristic learning means performs learning based on a running result between one station and trains the trained contents to the next station. It is characterized in that it is reflected in driving.
[0015]
According to the fourth aspect of the invention, in addition to the effect of the first aspect, it is possible to perform online automatic learning of characteristics from the result of traveling between stations.
[0016]
According to a fifth aspect of the present invention, in the automatic train driving device according to the first or second aspect, the automatic characteristic learning means performs learning based on a result of traveling on one route, and reflects the learning content at the time of traveling on the next route. It is characterized by the following.
[0017]
According to the fifth aspect of the present invention, in addition to the effect of the first aspect, at the time of acquiring the traveling results of all the routes, it is possible to advance the automatic online learning of the characteristics based on the specific phenomenon obtained from the results.
[0018]
According to a sixth aspect of the present invention, in the automatic train driving device according to the first or second aspect, the automatic characteristic learning means performs learning based on a running result for one day, and learns the learning content when the train runs on the next day. It is characterized by being reflected.
[0019]
According to the sixth aspect of the present invention, in addition to the effect of the first aspect, when a one-day driving result (one-day entire route driving result) is acquired, a characteristic based on a specific phenomenon obtained from the result is obtained. You can do online automatic learning.
[0020]
According to a seventh aspect of the present invention, in the automatic train driving device according to the first or second aspect, the automatic characteristic learning means performs learning based on a running result for at least several days, and trains the trained contents of the train after the next day. It is characterized in that it is reflected during running.
[0021]
According to the seventh aspect of the invention, in addition to the effect of the first aspect, when running results for several days or months are acquired, online automatic learning of characteristics based on a specific phenomenon obtained from the results is performed. be able to.
[0022]
According to an eighth aspect of the present invention, in the automatic train driving device according to the first or second aspect, at least two of the automatic characteristic learning means according to the third to seventh aspects are combined, and each of the provided automatic characteristic learning means is provided. It is characterized by further comprising learning result comparing means for comparing learning results of learning means, and learning result correcting means for correcting individual learning results according to the comparison results of the learning result comparing means.
[0023]
According to the eighth aspect of the present invention, it is possible to avoid an impossible learning situation and improve each learning effect.
[0024]
According to a ninth aspect of the present invention, in the automatic train driving device according to the second aspect, the result of the estimation by the pre-business characteristic estimating means is a characteristic value that cannot actually occur or a characteristic of a limit that can actually occur. If the value is out of range, the apparatus further comprises estimation result correction means for correcting the estimation result to within the limit characteristic value.
According to a tenth aspect of the present invention, in the automatic train driving device according to any one of the first to seventh aspects, the result of learning by the automatic characteristic learning means is a characteristic value that cannot actually occur, And a second learning result correcting means for correcting the learning result to within the limit characteristic value when the limit characteristic value is out of range.
[0025]
According to the ninth and tenth aspects of the present invention, by correcting the learning / estimation, it is possible to avoid a learning / estimation result that cannot be obtained by any chance.
[0026]
According to an eleventh aspect of the present invention, in the automatic train operating device according to any one of the first to tenth aspects, an automatic train that performs an automatic train operation by correcting a control command based on an error from a target travel plan value. In the driving device, the automatic learning characteristic means performs characteristic learning in accordance with a control command correction amount based on an error between the target traveling plan value and the characteristic learning at the time of business traveling.
[0027]
According to the eleventh aspect of the present invention, more detailed and quick learning can be performed by performing learning in accordance with a control command correction state based on an error from a target travel plan.
[0028]
According to a twelfth aspect of the present invention, in the automatic train operating device according to any one of the first to eleventh aspects, the automatic characteristic learning means performs characteristic learning using an adaptive observer.
[0029]
According to the twelfth aspect of the present invention, in addition to the effect of the first aspect, by learning characteristics using an adaptive observer, more accurate learning can be performed.
[0030]
According to a thirteenth aspect of the present invention, in the automatic train operating device according to any one of the first to eleventh aspects, the automatic characteristic learning means performs characteristic learning using a disturbance observer.
[0031]
According to the thirteenth aspect, in addition to the effect of the first aspect, by using the disturbance observer, it is possible to explicitly estimate the disturbance and perform the characteristic learning using the result.
[0032]
BEST MODE FOR CARRYING OUT THE INVENTION
<Embodiment 1>
1 and 2 show an embodiment of an automatic train driving device according to the present invention. An automatic train operation device (ATO) 8 mounted on the illustrated train 1 acquires speed limit data from an automatic train control device (ATC) 2 which is a ground system, and also has a route from a database (DB) 3 within the train 1. Data such as conditions (inclination angle, curve radius of curvature, etc.), vehicle conditions (train formation, weight, etc.), operating conditions, etc. are acquired, and a departure signal from the cab 4 and a variable load signal from the variable load device 5 are detected. A train speed signal is obtained from the detector 6, and a train position signal is obtained from the ground detector 7 which responds to the ground positioned appropriately on the route. The grounds that are appropriately arranged on the route are used for confirming the train position. Here, the DB 3 is shown as being mounted in the train 1, but may be provided as a ground system outside the train 1 in some cases, and in some cases, distributed and arranged in the train 1 and on the ground. It may be.
[0033]
The ATO 8 includes, in addition to the data processing means 80 for performing online data processing and the automatic train driving means 81, an estimating means and a learning means typified by a pre-business characteristic estimating means 24 and a post-business characteristic learning means 34 described later. . The data processing means 80 processes the train speed signal, and in addition to the train speed, calculates the train position (time integrated value of speed), train acceleration (differential value of speed), and train travel distance (time integrated value of absolute speed value). Operate continuously. The train position or the train travel distance is appropriately corrected by the train position signal from the ground detector 7. The data processing unit 80 performs a predetermined calculation based on each input signal as described above, and provides measurement data necessary for learning and automatic train operation described later. Measurement data required for automatic train operation is provided to the automatic train operation means 81. The automatic train driving means 81 issues a powering command to the driving device 9 and a deceleration command to the speed reduction device 10 according to the result calculated based on each input data. Drive device 9 includes a main motor for towing the train and a power converter for controlling the main motor. Further, the reduction gear transmission 10 usually has both a mechanical brake and an electric brake.
[0034]
The ATO 8 is mounted on the train 1, and a part of the pre-business characteristic estimating means 24 and the post-business characteristic learning means 34 related to learning of the present invention includes a pre-business running determination means as shown in detail in FIG. 20, pre-operation characteristic initial value setting means 21, pre-operation test running train automatic driving means 22, running result accumulation means 23, pre-operation characteristic estimation means 24, estimation result correction means 25, characteristic estimation value accumulation means 26, learning characteristic It comprises a database (learning characteristic DB) 30, characteristic initial value setting means 31, automatic train driving means 32, post-operation running result accumulation means 33, post-operation characteristic learning means 34, and learning result correction means 35. The means 21 to 26 are processing means for test driving before business travel, the means 31 to 35 are processing means for after business driving, and the pre-business running determining means 20 and the learning characteristic DB 30 are used before and after business driving. Regardless, it is provided in common for both. FIG. 2 omits the data processing means 80 and the automatic train operation means 81 which the ATO 8 originally has as an automatic train operation device.
[0035]
Next, the operation of the apparatus shown in FIGS. 1 and 2 will be described.
[0036]
In FIG. 1, the ATO 8 acquires speed limit data from the ATC 2 and information that can be acquired in advance such as route conditions, vehicle conditions, and operation conditions from the DB 3, respectively.
As described above, a predetermined operation is performed based on the command, a control command including a powering command or a deceleration command is created, and the automatic operation of the train 1 is realized.
[0037]
The ATO 8 receives the departure signal from the cab 4 and starts the automatic operation by the automatic train operation means. After departure, the adaptive load information obtained from the adaptive load device 5, the speed data obtained from the speed detector 6, and the ground child detection information obtained from the ground child detector 7 are used. The adaptive load information is used as information on the weight of the train, and the ground detection information is used for correcting position information. Using these information, the ATO 8 creates a train control command (powering command / deceleration command). When a powering command is created as a control command, the powering command is output and the train is driven by the drive device 9. Examples of the powering command include a powering torque (powering traction force) command, and a powering notch command in the case of notch traveling. When a deceleration command is created as a control command, the deceleration command is output, and the train is decelerated by the speed reduction device 10. Examples of the deceleration command include a brake force command and a brake notch command in the case of notch traveling.
[0038]
Next, the operation of the ATO 8 will be described in more detail with reference to FIG.
[0039]
Upon receiving the departure signal from the driver's cab 4, the pre-business travel determination means 20 first determines whether the test travel is before the business or after the business. As a judgment method at this time, a soft flag is created, and a soft flag such as "test driving when the flag is not set" and "business driving when the flag is set" is used. And a method of using the setting result of the hard switch.
[0040]
When the pre-business running determination means 20 determines that the test run is before the business, the pre-business characteristic initial value setting means 21 sets the initial characteristic parameters during the pre-business test run. As a setting method, it is conceivable to use a man-machine interface manually and set it before starting traveling. As the contents of the set values, characteristic parameters may be extracted from information that can be obtained in advance, such as train specifications and line characteristics, and input.
[0041]
Next, using the characteristic parameters set by the pre-operation characteristic initial value setting means 21, the pre-operation test running train automatic operation means 22 performs test driving of the train by automatic operation. As an automatic train driving method, an optimal travel plan is created when the station stops, and based on this, automatic operation is performed.If the deviation from the optimal travel plan is large, the travel plan is replanned or There is a method of performing correction by error feedback. In addition, since the vehicle is traveling in advance before business, here, for example, in the case of a notch traveling train, traveling for the purpose of characteristic estimation can be performed, such as performing test traveling using a notch whose characteristic is to be estimated.
[0042]
Next, the result of the automatic operation performed by the pre-business test traveling train automatic operation means 22 is accumulated by the traveling result accumulation means 23. In the case of storage, a target travel plan and speed data and position data measured during travel can be stored as an electronic file in a medium such as a hard disk (HD).
[0043]
Next, the characteristic parameters are estimated by the pre-operation characteristic estimating means 24 using the test traveling results accumulated by the traveling result accumulating means 23. The characteristic parameters to be estimated before business hours include weight, acceleration characteristics, deceleration characteristics, and the like.
[0044]
The weight of the entire train set is a test run before the start of business. Therefore, since no passengers are on the train, the weight or the deceleration during coasting and the train running resistance can be estimated. Here, a case where the target train is represented by a simple physical expression such as Expression (1) is considered.
[0045]
Regarding the train running resistance, it is possible to calculate by a formula in consideration of the line characteristics such as the slope and the curvature, the air resistance and the frictional resistance. For the calculation of the train running resistance, refer to the document “Operation Theory (DC AC Electric Locomotive)”, edited by Koyusha. Generally, the train running resistance Fr is
Fr = Frg + Fra + Frc
= S + (A + Bv + Cv ² ) + 800 / r (5)
Can be expressed as Here, Fr is the train resistance [kg weight / ton], Frg is the gradient resistance [kg weight / ton] (positive on the up, negative on the down), Fra is the running resistance [kg weight / ton], and Frc is the curve Resistance force [kg weight / ton], s is gradient [‰], A, B, C are coefficients, v is train speed, and r is radius of curvature.
[0046]
Taking these into consideration, the weight is obtained by transforming equation (1),
M = (F−Fr) / α (6)
Can be estimated by In the equation (6), in the case of coasting, the powering traction force F may be set to 0 (zero). Further, the acceleration (or deceleration) α can be obtained by using the least squares method or the like and calculating from the measurement result (train running speed). The weight M can be estimated by the above processing.
[0047]
When the estimation calculation of the weight M is completed, the power running characteristics and the brake characteristics can be estimated by using the estimated weight values.
[0048]
First, the weight estimate M _est , Acceleration during power running α _acc , And the train running resistance Fr, the power running characteristics (such as the relationship between the power running notch and the power running traction force) are estimated. Acceleration during power running α _acc The train running resistance Fr can be obtained by the same processing as the above-described calculation of the weight. Using these and the weight estimation value, the powering traction force F is calculated as
F = M _est α _acc + Fr (7)
Can be estimated by
[0049]
According to equation (7), in the case of a train that performs a powering operation using a notch, the powering traction force of each notch can be estimated. Based on this, the relationship between the power running notch and the power running traction force can be estimated.
[0050]
Further, the braking force characteristic can be estimated using the estimated weight value, the deceleration during deceleration, and the train running resistance. The deceleration at the time of deceleration and the running resistance of the train can be obtained by the same processing as the above-described calculation at the time of weight. Using these and the estimated weight, the braking force F is calculated as
F = M _est α _dec + Fr (8)
Can be estimated by Where α _dec Is the deceleration (negative acceleration).
[0051]
From the equation (8), the braking force of each notch can be estimated in the case of a train that performs the braking operation by the notch. From this result, the relationship between the brake notch and the braking force can be estimated.
[0052]
It is desirable to calculate these estimated values after traveling between stations or when stopping, but it is also possible to calculate these values while the train is running, and to check the calculation results while the train is running. By estimating the weight / powering characteristics and the brake characteristics in this manner, it is possible to adjust even for each train organization in a shorter time than before in the course of commercial operation.
[0053]
Next, the estimation result correction unit 25 corrects the characteristic estimated value estimated by the pre-business characteristic estimation unit 24. In performing the correction, it is necessary to set an allowable range of the characteristic parameter that can be theoretically realized, and to correct the characteristic parameter to be within the allowable range. For example, when the characteristic estimated value exceeds the allowable range, a set value calculated in advance may be used, or a limit value in the allowable range may be used. If the allowable range is deviated too much, it is necessary to perform an operation such as performing a test run again.
[0054]
Next, the characteristic estimated value corrected by the estimation result correction unit 25 is stored in the learning characteristic DB 30 by the characteristic estimated value storage unit 26. As a method of accumulation, the same method as that of the traveling result accumulation means 23 described above can be used. The learning characteristic DB 30 stores not only the characteristic estimation results obtained in the test driving before the commercial driving but also the characteristic learning results obtained in the learning after the commercial driving described later.
[0055]
The case where the pre-business running determination means 20 determines that the vehicle is running after business will be described below.
[0056]
In the case of business travel, first, characteristic initial value setting means 31 sets initial values of characteristic parameters. In the case of the first business trip, the characteristic parameters (characteristic estimation results) accumulated by the characteristic estimation value accumulation means 26 are acquired from the learning characteristic DB 30 and used. When the business travel is advanced and the learning is advanced, the characteristic parameter (characteristic learning result) obtained as a result of the learning is used.
[0057]
Next, using the characteristic parameters set by the characteristic initial value setting means 31, the automatic train operation of the train is performed by the automatic train operation means 32. The automatic operation of the train is basically the same as the automatic train operation means 22 for test running before the operation, but after the operation, an unspecified number of passengers are on the train, so that the weight varies. Therefore, it is necessary to estimate the weight when traveling between stations at the time of initial powering from departure of the station. As a method for estimating the weight, when an adaptive load is obtained, the adaptive load may be used. When the adaptive load cannot be used, the weight can be estimated by performing the same operation as the pre-operation characteristic estimating means 24 and the estimation result correcting means 25 at the time of initial powering after departure from the station. If the result of the estimation is different from the value set by the characteristic initial value setting means 31, processing such as creating a travel plan again is necessary. FIG. 3 shows an outline of the case of estimating the weight at the time of initial powering after departure from the station.
[0058]
In FIG. 3, the horizontal axis indicates the distance from the departure station to the next station, that is, the position, and the vertical axis indicates the speed at each position as a speed pattern. After starting traveling based on the optimum traveling pattern 31 (thin broken line) created based on the characteristic estimation value when the departure station stops, weight estimation is performed according to the actual traveling result in the initial powering section 30, that is, the actual traveling pattern 32 (thick solid line). The travel pattern 32 (thick broken line) corrected by the recalculation based on the estimated weight value is created, and the actual travel operation is performed in accordance with this.
[0059]
Next, the result of the automatic operation by the automatic train operation means 32 is accumulated by the post-operation traveling result accumulation means 33. As a method of accumulation, the same method as that of the traveling result accumulation means 23 described above can be used.
[0060]
Next, characteristics are learned by the after-sales characteristic learning unit 34 using the traveling results accumulated by the after-sales traveling result accumulation unit 33. For regular learning of this property,
(1) Learning based on station-to-station driving results
(2) Learning based on all route driving results
(3) Learning based on running results for one day
(4) Learning based on driving results for several days
(5) Learning based on driving results for several months
And conduct for each case.
[0061]
The cases (1) to (5) will be described below.
(1) Learning based on station-to-station driving results
The learning is performed based on the inter-station traveling results obtained after traveling between the stations, and the learning result is reflected when traveling between stations. This learns, for example, how to deal with a decrease in braking force when it starts to rain. An example where it is determined that learning is necessary based on the traveling results between one station includes a response to a decrease in braking force in rainy weather. In rainy weather, when an air brake is used in a train, the braking force (deceleration performance) may be reduced because the friction of the brake shoes decreases due to rain. In such a case, a decrease in the deceleration performance should be observed after the rain starts. The characteristics of the braking force may be learned based on the result. Since the learning result in this case is often temporary, it is better to separately hold and use the temporary characteristic parameter.
[0062]
(2) Learning based on all route driving results
The learning is performed based on the result of traveling one route from the beginning to the end, and the learning result is reflected when the next route traveling is started. For example, at the end of traveling on one line, when each station stops by causing an excess or deficiency (deviation amount) with respect to the target stop position, the braking is performed according to the deviation amount in order to eliminate such deviation amount. What is necessary is just to learn a force characteristic. For example, if the vehicle goes beyond the target stop position, the set value of the braking force characteristic may be slightly larger than the actual value. That is, it is conceivable that the expected deceleration is not obtained because a braking force larger than the actual one is considered. In this case, learning may be performed so as to slightly reduce the set value of the braking force characteristic.
[0063]
(3) Learning based on running results for one day
Learning is performed based on the running results for one day, and the learning results are reflected in running on the next day. For example, when looking at the running results for one day (for example, the running results for several runs on one route), almost the same amount of overshoot as to the target stop position occurs between certain stations. In such a case, there is a possibility that a deviation has occurred in the setting of the route characteristic parameters such as the gradient and the curve between the stations. In this case, learning may be performed so as to gradually adjust the route characteristic parameters such as the gradient and the curve according to the traveling result.
[0064]
(4) Learning based on driving results for several days
It accumulates running results for several days and performs learning based on the accumulated results. For example, when looking at the driving results for several days, if there is a deviation from the driving plan only in the same time zone, due to some influence, the powering traction force characteristic or the braking force characteristic only consequently differs from the actual It is possible that the situation has shifted. If there is no deviation in other time zones, it is considered that the characteristic parameter itself is not actually shifted, so the characteristic is corrected only in the target time zone, and the correction value is then learned. It should be corrected by.
[0065]
(5) Learning based on driving results for several months
When running results for several months are accumulated, learning is performed based on the accumulated results. For example, it is conceivable that learning is performed based on traveling results accumulated during maintenance and inspection. For example, when viewing the running results for three months, there may be a situation where the braking force gradually decreases over time, such as three months ago, two months ago, and one month ago. This assumes a situation where it is difficult to judge from the learning situation based on the traveling results for several days. If an air brake is used, the friction shoe may be worn due to friction. Therefore, it is necessary to change (learn) the characteristic parameters based on the result or take measures such as replacement of the brake shoes depending on the degree. In addition, measures against aging such as fluctuations in wheel diameter are also taken.
[0066]
The above learning can be selectively realized by, for example, performing the learning as shown in the flow of FIG. In FIG. 4, if the result of the judgment (step 51) between the pre-business test run and the post-business test run by the pre-business run determination means 20 is the former (pre-business test run), the pre-business test run ( Step 52) is performed, the initial parameters are estimated (step 53), and the process ends. If the result of determination in step 51 is that the vehicle is traveling, any one of the five types of learning is performed according to the content of traveling. That is, the mode of the end of travel in the business travel is determined (step 54). If the travel between stations is completed, “(1) learning based on the result of travel between stations” (step 55) is performed. "(2) Learning based on all route driving results" (step 56) is performed. If it is determined in step 54 that the driving for one day has been completed, it is determined how many more days of data have been stored (step 57). 3) Learning based on running results for one day "(step 58). If data storage for several days is completed," (4) learning based on running results for several days "(step 59) is performed, and data for several months is performed. If the accumulation is completed, “(5) Learning based on traveling results for several months” (step 60) is performed.
[0067]
However, in each of the learning steps 55, 56, 58, 59, and 60 indicated by a thick line in FIG. 4, learning is performed only when the running result shows the following tendency to be learned. That is,
a) When the same tendency of deviation continues (for example, in the case where the result of traveling on all routes shows that the target stop position of the same degree is excessively exceeded between all stations), and
b) When the deviation is remarkable
It is.
[0068]
Regarding learning, a method of increasing or decreasing a related characteristic parameter at a certain fixed ratio is conceivable. For example, as described above, in the results of running on all routes, if overshoot of the same target stop position is observed between all stations, the set value of the braking force may be slightly larger than the actual braking force. Learning may be performed so that the set value of the braking force characteristic is reduced at a fixed rate.
[0069]
In particular, with regard to learning based on the results of running between stations, it is rare that the same tendency is seen in several cases. Therefore, in this case, the following learning can be considered. That is,
・ Eligible automatic train operation methods:
This is an automatic train operation system in which, when a considerable deviation occurs between a travel plan and an actual measurement value, a control command (powering notch command, brake notch command, etc.) is corrected according to the deviation.
[0070]
・ Learning method:
Learning is performed according to the control command correction situation when a deviation occurs between the travel plan and the actual measurement value. As an example, in the case of the braking force characteristic, for example, if a control command correction that causes a stronger brake notch than planned occurs during braking, it is possible that the assumed deceleration may not be obtained. In this case, since it is considered that the set value of the braking force characteristic is too large, learning may be performed so as to reduce the set value of the braking force characteristic at a fixed rate. If a control command correction that results in a weaker brake notch than planned occurs, learning may be performed so that the set value of the braking force characteristic is increased at a constant rate.
[0071]
The judgment that the estimated characteristics are different from the actual values is based on the acceleration / deceleration obtained as the measured data, the characteristics related to the train running, the characteristics related to the line shape (gradient, curve, etc.), the weight, the power running, etc. The traction or braking force may be used to determine whether equation (1) is satisfied.
[0072]
As described above, the result learned by the after-market characteristic learning means 34 is corrected by the learning result correcting means 35. The correction can be performed by the same processing as that of the estimation result correction unit 25 described above. The correction result here is stored in the learning characteristic DB 30 as a characteristic learning result.
[0073]
As described above, the learning is performed even after entering the commercial operation, and the commercial operation is performed while adjusting the characteristic parameters.
[0074]
Most of the learning described above assumes online automatic learning while the train is stopped, such as when arriving at a station. However, the estimation of the weight at the time of power running assumes online automatic estimation during running.
[0075]
As described above, by performing the automatic operation of the train while performing the learning and the estimation, it is possible to perform the automatic operation satisfactorily corresponding to the variation of the train organization, the secular change, and the like.
[0076]
As described above, according to the automatic train driving device of the first embodiment, by estimating the weight, the powering traction force, and the braking force before traveling, the variation in each train organization is shorter than before. Adjustment can be made depending on time, and by learning characteristic parameters even after business hours, it is possible to realize automatic driving that satisfies ride comfort and stopping accuracy even when characteristic parameters change. Further, in the post-business learning, the learning based on the actual situation can be performed by executing the learning by dividing the distance between the stations, the distance along the route, etc. according to the period of the data using the learning. In addition, in the pre-business estimation and post-business learning, by correcting the estimation / learning result, even in the unlikely event that an impossible result is obtained, it is possible to correct the impossible characteristic parameter by correcting it. Estimation and learning can be performed without using.
[0077]
As described above, it is possible to create an effective optimal travel plan as the learning of the characteristics progresses. In addition, when a large learning occurs during the running of the train, this is used as a trigger to replan the running plan, thereby enabling automatic train operation that satisfies riding comfort, target stop position stop accuracy, and running time.
[0078]
<Embodiment 2>
In the first embodiment, most of the learning assumes online automatic learning while the train is stopped when the train arrives at the station, and the weight estimation during power running assumes online automatic estimation during traveling. However, if there is a man-machine interface that allows the user to check the learning progress while the train is running, it is possible to implement a system that proceeds with online automatic learning while driving and uses the learning results at the discretion of the driver. It is. In this case, only the learning means may be a separate device, and may be an automatic train driving support device.
[0079]
<Embodiment 3>
FIG. 5 shows a main configuration of an automatic train driving device according to the third embodiment. In this embodiment, as the post-sales characteristic learning means, the automatic characteristic learning means 341 according to claim 3, the automatic characteristic learning means 342 according to claim 4, the automatic characteristic learning means 343 according to claim 5, An automatic characteristic learning unit 344 according to claim 6 and an automatic characteristic learning unit 345 according to claim 7, further comprising a learning result comparison unit 36 that receives a learning result obtained by the automatic characteristic learning unit as an input. And a learning result correcting means 37 for correcting the learning result according to the comparison result of the learning result comparing means 36.
[0080]
In the automatic characteristic learning means 341 to 345, characteristic learning is performed as described in the first embodiment. The learning result comparing means 36 receives the results learned by the automatic characteristic learning means 341 to 345, compares the learning results, and checks whether there is a great inconsistency between them. In the automatic characteristic learning means 341 to 345, since the learning period, that is, the learning interval is greatly different, basically, the result of the longer learning period may be checked with the result of the shorter learning period. For example, when the learning result of the automatic characteristic learning unit 345 has a value that is clearly n times, for example, 10 times as large as the learning result of the automatic characteristic learning unit 344 in the same time period, it is determined that the abnormality is obviously abnormal, and What is necessary is just to make the result that the learning result of the characteristic learning means 345 largely contradicts. In addition, by performing a check using a plurality of results among the automatic characteristic learning means 341 to 345, the accuracy of the check can be further improved.
[0081]
Next, the learning result correcting means 37 corrects the result of the comparison that the learning result comparing means 36 greatly contradicts. The simplest method of correction is to use the learning result of the automatic characteristic learning means having a short learning period (learning interval) as it is. However, when a plurality of learning results of the automatic characteristic learning means 341 to 345 are used, an average value of the learning results may be used. Further, when the results of the learning of the automatic characteristic learning means 341 to 345 are almost inconsistent, or when there is a large variation between the learning results of the automatic characteristic learning means 341 to 345, those are discarded. An average value may be used.
[0082]
<Embodiment 4>
The automatic characteristic learning means 34 can perform characteristic learning using an adaptive observer. The adaptive observer identifies a parameter from a value that can be observed (measured) when a target plant is mathematically modeled as in Expression (1). Depending on the viewpoint, it is considered that the system identification is being performed, and a kind of adaptive control system is configured by using the identification result of the adaptive observer with the automatic train operation means 81 every moment. In the case of equation (1), by using the adaptive observer, the acceleration / deceleration (which can be calculated from the speed detected by the speed detector 6) and the powering traction force or the braking force which are the control command values are used to calculate the weight and the train. The running resistance can be identified every moment. As an algorithm of the adaptive observer, an extended least squares method, an extended Kalman filter, an adaptive filter, and the like can be used (for details, see Chapter 2 of "Introduction to Robust Adaptive Control" (edited by Mitsuru Terao, Kimio Kanai, published by Ohmsha). See "Estimation of Unknown Plant and Adaptive Observation Device", pages 47 to 87, or "System Control Series 6 Optimal Filtering" (by Kiyoshi Nishiyama, Baifukan), section 3.3, "Adaptive Filter", pages 50 to 57).
[0083]
As described above, by comparing several automatic characteristic learning means having different learning periods (learning intervals) and eliminating inconsistent learning results, it is possible to obtain more accurate characteristic learning results.
[0084]
<Embodiment 5>
The automatic characteristic learning means 34 can also learn characteristics using a disturbance observer. The disturbance observer is often used in motion control and the like, and identifies a disturbance (for details, see "Control System Design Using MATLAB" (edited by Kenzo Nonami, co-authored by Hidekazu Nishimura and Mitsuo Hirata, Tokyo Denki University Press) (See Section 4.4, “Disturbance Observer in Motion Control,” pp. 99-102). By considering the train running resistance in equation (1) as a force disturbance in motion control, the train running resistance can be estimated every moment by a disturbance observer. By performing learning using this estimation result, more accurate learning can be performed.
[0085]
【The invention's effect】
According to the present invention, train characteristics, route characteristics, and control parameters can be automatically learned online while the vehicle is running, and efficient train automatic operation using the learning results can be performed.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an entire train including an automatic train driving device according to the present invention.
FIG. 2 is a block diagram for explaining an internal configuration of the automatic train driving device of FIG.
FIG. 3 is a conceptual diagram of traveling pattern correction based on weight estimation during initial powering.
FIG. 4 is a flowchart showing a learning procedure considering before and after business hours.
FIG. 5 is a block diagram showing a correction unit for correcting an automatic characteristic learning result according to an embodiment of the present invention.
[Explanation of symbols]
1 train
2 Automatic train control device (ATC)
3 database (DB)
4 cab
5 Adaptive load device
6 Speed detector
7 Ground detector
8 Automatic train operation equipment (ATO)
9 Drive
10 Reduction gear
20 Driving judgment method before business
21 Pre-operation characteristic initial value setting means
22 Pre-sales test train automatic driving means
23 Running result accumulation means
24 Pre-operation characteristic estimation means
25 Estimation result correction means
26 Characteristic estimated value accumulation means
30 Learning characteristic database (Learning characteristic DB)
31 Characteristic initial value setting means
32 Automatic train operation means
33 After-sales driving result accumulation means
34 Post-business characteristic learning means
35 Learning result correction means
36 Learning result comparison means
37 Learning result correction means
80 Data processing means
81 Automatic train operation means
341 to 345 Automatic characteristic learning means

Claims (13)

A data processing means for processing data acquired during train operation online,
Automatic characteristic learning means for automatically learning control parameters during train running, and train characteristics and route characteristics during train running, based on data obtained during train running and data obtained in advance by this data processing means,
Using the train characteristics and the line characteristics learned by the automatic characteristics learning means, a train automatic driving means for automatically driving the train,
An automatic train driving device comprising: A preliminary test run before the operation includes a pre-operation characteristic estimating means for preliminarily estimating a train characteristic and a line characteristic necessary for automatic train operation and an initial value of a control parameter, and the automatic characteristic learning means includes the pre-operation characteristic estimating means. The automatic train operation device according to claim 1, wherein learning is advanced by running after business based on the initial value estimated by (1). The said automatic characteristic learning means performs learning at the time when it is judged that an assumed characteristic value is clearly different from an actual value at the time of train driving, and reflects the learning content on subsequent train driving. 3. The automatic train driving device according to 1 or 2. The automatic train operation device according to claim 1, wherein the automatic characteristic learning unit performs learning based on a result of traveling between one station and reflects the learned content on a train traveling to a next station. . 3. The automatic train operation device according to claim 1, wherein the automatic characteristic learning unit performs learning based on a result of traveling on one route, and reflects the learning content when the vehicle travels on the next route. 4. 3. The automatic train operation device according to claim 1, wherein the automatic characteristic learning unit performs learning based on a running result for one day, and reflects the learning content when the train runs the next day. 4. 3. The automatic train driving apparatus according to claim 1, wherein the automatic characteristic learning unit performs learning based on at least several days of running results, and reflects the learning content when the train runs on the next day or later. . A learning result comparing means comprising at least two of the automatic characteristic learning means according to claims 3 to 7, and comparing the learning results of the provided automatic characteristic learning means, and a comparison result of the learning result comparing means. 3. The automatic train operation device according to claim 1, further comprising a learning result correcting unit that corrects each learning result according to the following. If the result of the estimation by the before-sales characteristic estimating means is a characteristic value that cannot actually occur or is outside the limit characteristic value that can actually occur, the estimation result is within the characteristic value of the limit. 3. The automatic train operation device according to claim 2, further comprising an estimation result correction unit that corrects the estimated result. If the result of the learning by the automatic characteristic learning means is a characteristic value that cannot actually occur or is outside a limit characteristic value that can actually occur, the learning result is corrected to within the limit characteristic value. The automatic train operation device according to any one of claims 1 to 7, further comprising a second learning result correction unit. In an automatic train driving device that performs automatic train operation by correcting a control command based on an error from a target travel plan value, the automatic learning characteristic means includes a function for learning a characteristic during a commercial run, and a target travel plan value. The automatic train operation device according to any one of claims 1 to 10, wherein characteristic learning is performed according to a control command correction amount based on an error between the trains. The automatic train operating device according to any one of claims 1 to 11, wherein the automatic characteristic learning means performs characteristic learning using an adaptive observer. The automatic train operating device according to any one of claims 1 to 11, wherein the automatic characteristic learning means performs characteristic learning using a disturbance observer.

Images