METHOD FOR DETERMINING A TRACK OCCUPANCY AND AXLE COUNTING DEVICE

Description

[0001] The invention relates to a method for computer-aided determination of the occupancy status of a track section, comprising the steps of the physical wheel influence of trains traveling on the track section is converted into a reference voltage, the reference voltage is read into a processor as raw data via an analog-digital converter, an evaluation is carried out with the aid of at least one algorithm to check whether or not there is a wheel influence, and an influence state is derived from this, the influence state is output to determine the occupancy state.

[0002] The invention also relates to an axle counting device for detecting wheel influences on a track section with a sensor with which the physical wheel influence of trains traveling on the track section can be converted into a reference voltage, an analog-digital converter with which raw data can be generated from the reference voltage, a processor with which the raw data can be processed and evaluated with regard to an occupancy status by at least one algorithm implemented in the processor, an interface S3 for outputting an occupancy status

[0003] Furthermore, the invention relates to a computer program product with program instructions for carrying out the above-mentioned method and a provision device for the computer program product, wherein the provision device stores and / or provides the computer program product.

[0004] Methods for computer-aided determination of the occupancy status of track sections are known. Typically, this determination is performed using axle counters, which are also known. The occupancy status of a track section indicates whether a train is currently using that section. In this case, an occupancy report is sent, for example, to a control center. However, a track section may never be used by two trains at the same time. Only when the train in question has left the section in question is the track section reported as clear, allowing another train to enter it.

[0005] Typically, the movement of a train on a track section is monitored using axle counters. These sensors emit a signal indicating a wheel passing the sensor. This allows the axles of an incoming train to be counted, thus identifying the train. If the same sensor pattern is detected when the train exits the track section, this means that the train has completely exited the track section in question.

[0006] The reliability of the method depends on various factors. Among other things, there are specific considerations for each track section, for example, if the sensors are located on curves or switch areas. There are also track sections that are primarily used by certain trains, such as branch lines used exclusively by freight trains, or high-speed lines used primarily by high-speed trains. The desire is for axle counting methods to always be able to be carried out reliably, regardless of the individual requirements of the track section being monitored.

[0007] The document EUGEN BERLIN ET AL: "Sensor Networks for Railway Monitoring: Detecting Trains from their Distributed Vibration Footprints", DISTRIBUTED COMPUTING IN SENSOR SYSTEMS (DCOSS), 2013 IEEE INTERNATIONAL CONFERENCE ON, IEEE, May 20, 2013 (2013-05-20), pages 80-87, XP032444282, DOI: 10.1109 / DCOSS.2013.38 ISBN: 978-1-4799-0206-4 describes a wireless sensor network on railway tracks for monitoring and analyzing vibration patterns caused by passing trains.

[0008] The object of the invention is therefore to provide a method for the computer-aided determination of the occupancy status of a track section, which can be installed cost-effectively and yet functions as reliably as possible, regardless of individual differences on all track sections. Furthermore, the object of the invention is to provide an axle counting device with which such a method can be carried out reliably and cost-effectively. Finally, the object of the invention is to provide a computer program product and a provision for such a computer program product, which allow the implementation of the method.

[0009] Unless otherwise stated in the following description, the terms "create," "calculate," "compute," "determine," "generate," "configure," "modify," and the like preferably refer to actions and / or processes and / or processing steps that change and / or generate data and / or convert the data into other data, wherein the data can be represented or present in particular as physical quantities, for example, as electrical impulses. In particular, the term "computer" is to be broadly interpreted to cover all electronic devices with data processing capabilities. Computers can thus be, for example, personal computers, servers, handheld computer systems, Pocket PC devices, mobile radio devices, and other communication devices capable of computer-aided data processing, processors, and other electronic devices for data processing, which can preferably also be connected to form a network.

[0010] In the context of the invention, "computer-aided" can be understood as meaning, for example, an implementation of the method in which one or more computers carry out or carry out at least one method step of the method.

[0011] In the context of the invention, a "processor" can be understood, for example, as a machine or an electronic circuit. A processor can be, in particular, a central processing unit (CPU), a microprocessor, or a microcontroller, for example, an application-specific integrated circuit or a digital signal processor, possibly in combination with a memory unit for storing program instructions, etc. A processor can also be, for example, an IC (integrated circuit), in particular an FPGA (field programmable gate array), an ASIC (application-specific integrated circuit), or a DSP (digital signal processor). A processor can also be understood as a virtualized processor or a soft CPU.It may, for example, also be a programmable processor which is equipped with a configuration for carrying out the said method according to the invention.

[0012] In the context of the invention, a "memory unit" can be understood as meaning, for example, a computer-readable memory in the form of a random-access memory (RAM) or a hard disk.

[0013] The above-mentioned object is achieved according to the invention with the subject matter of the claim (method) specified at the outset in that, for the detection of wheel influences within the scope of the evaluation, the algorithm uses reinforcement learning to optimize the parameters taken into account for the detection.

[0014] "Reinforcement learning" refers to the operation of a processor that aims to optimize the process executed by the processor. The process's performance is measured against a specific success, whereby modifying parameters can lead to greater or lesser success. Parameter modifications that lead to greater success are pursued, while modifications that reduce success are discarded. In this way, the process is gradually optimized, whereby the invention addresses the task of optimizing an axle counting process. This optimization is advantageously carried out without precise knowledge of the specific application situation of the axle counter. Rather, by modifying the parameters, an improvement is achieved over time until an optimum is found. Thus, learning reinforces the success of the process in question.

[0015] The advantage of using reinforcement learning to optimize the axle counting process is that the process can be installed with a single default setting in different environments, with optimization occurring automatically during operation. After a new installation of such a process, train operations are always subject to more stringent monitoring, allowing initial inaccuracies to be compensated for. Optimization during operation leads to a continuous improvement in the reliability of the process and thus also in the reliability of train operations.

[0016] Automatic optimization advantageously results in low costs. Furthermore, changes in operations, such as the introduction of long-distance traffic on a route previously used primarily for local traffic (or vice versa), can be accommodated. If the optimization objective changes during operation, the process can subsequently be automatically adapted to the new task. The process therefore remains reliable even under changing operating conditions, as it constantly adapts to the current requirements.

[0017] Reinforcement learning can be implemented using both software and hardware. In software, reinforcement learning is implemented using a program-based algorithm. Another possibility is to implement the reinforcement learning process using hardware. According to one embodiment of the invention, the reinforcement learning algorithm uses a neural network.

[0018] Neural networks consist of a circuit that generates outputs depending on the input variables and the implemented circuit. The circuit is designed in such a way that a modification of the input variables also leads to a modification of the output variables, with the parameters to be considered being implemented in hardware. However, a modification of these parameters is possible in the circuit.

[0019] According to one embodiment of the invention, it is provided that the at least one algorithm carries out a threshold value detection and / or a temporal filtering and / or the consideration of a hysteresis between the detection of the wheel start and the wheel end.

[0020] The parameters that can be modified by axle counters include, for example, a so-called switch-on threshold and a switch-off threshold for wheel detection. In other words, there is a threshold for the measured value determined by the sensor, above which it is assumed that a wheel has passed the axle counter. If the value then falls below the switch-off threshold again, this is interpreted as the wheel having completely passed the axle counter. Another parameter can be introduced through time filtering, which incorporates empirical values ​​regarding how long (speed-dependent) a wheel takes to pass an axle counter. If a hysteresis between the switch-on and switch-off thresholds must be taken into account, the switch-off threshold may be higher than the switch-on threshold.

[0021] According to the invention, the method steps are carried out in parallel in a first subsystem and in a second subsystem.

[0022] The use of both a first subsystem and a second subsystem is advantageous for achieving greater fault tolerance. For example, a voting process can be used to determine whether the first subsystem and the second subsystem indicate the same result, i.e., the presence or absence of an interference condition. If both subsystems arrive at different results, there is a risk of a fault, and safety measures can be initiated during train operation. This advantageously increases safety. The failure of one of the two subsystems can also be compensated for in an emergency.

[0023] According to the invention, it is provided that the first subsystem and the second subsystem both output the influence state to an evaluation device and the evaluation device compares an actual occupancy state with the influence state.

[0024] The actual occupancy state helps the evaluation system assess the influence state output by the first subsystem and the second subsystem in terms of reliability and certainty of results. This allows a statement to be made as to whether the occupancy state derived from the influence state is sufficiently reliable or not. For example, new axle counters, if an optimum has not yet been approximated, can initially be given lower priority with regard to their reliability in order to minimize errors during operation. This advantageously increases reliability in rail operations. As rail operations progress, the prioritization of the then optimized axle counters can be increased.

[0025] According to the invention, the evaluation device determines the actual occupancy status by evaluating the data from sensor devices, in particular axle counters, on the track section or from external data, in particular timetable data.

[0026] Sensor devices that have been in operation for some time are particularly advantageous, allowing the data collected by these sensors (preferably also axle counters) to be used for comparison. Timetable data can also be used, if the timetable can be reliably mapped, since the result of the axle counter can be compared with the pattern of the train specified in the timetable.

[0027] According to the invention, the evaluation device reports the actual occupancy state or information regarding the correspondence of the actual occupancy state with the output influence state to the neural networks of the first subsystem and the second subsystem.

[0028] By providing feedback on the occupancy state, which was determined from the influence state, or a value that allows the degree of correspondence between the occupancy state and the influence state, the self-learning system can adjust the parameters as needed. This triggers an iterative process that advantageously contributes to parameter optimization. Parameter modification can thus be carried out in a targeted manner (rather than randomly), which advantageously allows optimization to be completed much more quickly.

[0029] According to the invention, in the event that one of the subsystems of the first subsystem and the second subsystem, i.e. the first subsystem or the second subsystem, has achieved a higher level of security against misinterpretations in determining the influence state than the other through reinforcement learning, the parameters taken into account for the detection are transferred to the other subsystem.

[0030] By adjusting the parameters of the subsystem that delivers the less favorable results, greater reliability is advantageously ensured. In particular, it can prevent the self-learning system from straying from the desired optimum in the event of an unfavorable constellation. A race can thus take place between the first subsystem and the second subsystem to find the optimum, in order to achieve it in the shortest possible time.

[0031] According to one embodiment of the invention, it is provided that in a processing step of the raw data, a temperature influence is eliminated, whereby compensated data is generated for the evaluation.

[0032] Compensating for temperature influences further increases the reliability of the method. Temperature differences that can occur on rail lines (summer, winter) can be considerable. Eliminating the temperature influence makes the output data on the influence state comparable, regardless of the season and weather conditions.

[0033] According to one embodiment of the invention, it is provided that a test cycle is carried out at time intervals with a set of input variables for which the result of the determination of the influence state is predictable, and that the determined influence state is compared with an expected, predictable influence state.

[0034] A test cycle, which can be performed at regular intervals, advantageously increases the reliability of the method. This is achieved by detecting a parameter drift that leads to a deviation from the optimum before it is already producing incorrect results. This is because the result is reliably known for the parameters of the test cycle and will only fail if the self-learning system has moved in the wrong direction over an extended period of time, i.e., away from the desired optimum.

[0035] The above-mentioned object is alternatively achieved according to the invention with the subject matter of the claim specified at the outset (axle counting device) in that the processor has a neural network for detecting wheel influences during the evaluation.

[0036] A neural network advantageously offers a hardware-based and thus highly reliable environment for providing the parameters for the above-mentioned method. Reinforcement learning can be used to find an optimum for the parameters. It is also possible to perform reinforcement learning in preparation before installing the axle counter according to the invention at the site. In this case, the operating conditions for the axle counter must be determined in advance.

[0037] According to one embodiment of the invention, it is provided that a method according to one of claims 1-10 can be carried out with it. The associated advantages have already been explained in connection with the above method.

[0038] According to one embodiment of the invention, a standardized set of parameters for implementing the algorithm is stored in the processor.

[0039] The advantage of using a standardized parameter set is that the 8-counting device is already fully functional prior to optimization. This optimization then uses the standardized parameter set to perform the above procedure. Only during subsequent operation are the parameters then adapted to the specific application environment.

[0040] Furthermore, a computer program product with program instructions for carrying out the said method according to the invention and / or its embodiments is claimed, wherein the method according to the invention and / or its embodiments can be carried out by means of the computer program product.

[0041] Furthermore, a provision device for storing and / or providing the computer program product can be provided. The provision device is, for example, a data storage medium that stores and / or provides the computer program product. Alternatively and / or additionally, the provision device is, for example, a network service, a computer system, a server system, in particular a distributed computer system, a cloud-based computer system, and / or a virtual computer system, which stores and / or provides the computer program product, preferably in the form of a data stream.

[0042] This provision takes place, for example, as a download in the form of a program data block and / or command data block, preferably as a file, in particular as a download file, or as a data stream, in particular as a download data stream, of the complete computer program product. This provision can, however, also take place, for example, as a partial download consisting of several parts and, in particular, downloaded via a peer-to-peer network or provided as a data stream. Such a computer program product is read into a system, for example, using the provision device in the form of the data carrier and executes the program instructions, so that the method according to the invention is carried out on a computer or the creation device is configured such that it produces the workpiece according to the invention.

[0043] Further details of the invention are described below with reference to the drawings. Identical or corresponding elements of the drawings are provided with the same reference numerals and are explained several times only to the extent that differences arise between the individual figures.

[0044] The exemplary embodiments explained below are preferred embodiments of the invention. In the exemplary embodiments, the described components of the embodiments each represent individual, independently considered features of the invention, which also further develop the invention independently of one another and are thus also to be considered as components of the invention, either individually or in a combination other than that shown. Furthermore, the described embodiments can also be supplemented by further features of the invention already described.

[0045] They show: Figure 1 is a schematic flow diagram for an embodiment of the method according to the invention, wherein the process is carried out in parallel for both subsystems, ie for subsystem 1 and subsystem 2, Figure 2 is a schematic of an embodiment of the axle counter according to the invention, with which the method according to Figure 1 can be carried out, with both subsystems, ie subsystem 1 and subsystem 2, being shown.

[0046] Figure 1 With a track section GA represented as a basis, shows a process step 1 VS1, which consists in the sensor SEN1, SEN2 receiving a wheel pulse IPR (physical wheel influence) from the wheels of a train ZUG traveling on the track section GA. The sensor SEN1, SEN2 converts this wheel pulse IPR into a rectified voltage RSP.

[0047] In Figure 1Furthermore, a method step 2 VS2 can be seen, which consists in reading the reference voltage RSP into a processor PRC as raw data RDT via an analog-digital converter ADU.

[0048] Furthermore, Figure 1 a method step 3 VS3 can be recognized, which consists in the elimination of a temperature influence by a processing step of the raw data RDT, whereby compensated data KDT are generated, which are relevant in the later course of the method for the evaluation of an influence state 1 and 2 BZ1, BZ1.

[0049] Figure 1Furthermore, a process step 4 VS4 is shown, which consists in the processor PRC being able to use an algorithm that, after examining the raw data RDT, evaluates whether or not a wheel influence is present. The reinforcement learning (VSL) used in the algorithm optimizes the determination of the axle counting result using data acquired during train operation (more on this below). The reinforcement learning process can be performed, for example, by a neural network (not shown in detail).

[0050] In Figure 1 It can also be seen that the result of influence states 1 and 2 (BZ1, BZ2) can be transmitted to the AWE evaluation device via an output OUT. Based on the evaluation generated by the evaluation device, the occupancy status of the track section GA can thus be determined.

[0051] The AWE evaluation unit is integrated into an axle counting computer (AZR). The AZR axle counting computer is provided by the interlocking infrastructure. It can operate in a network of computers and take into account additional inputs, as described below.

[0052] Furthermore, Figure 1that the AWE evaluation device can determine the influence state 3, 4, and thus the actual occupancy state of the track section GA, from external data, in particular timetable data FPD, since the result of the occupancy state BZ1, BZ2 can be compared with the pattern of the train planned in the timetable FP. In addition, the further influence states 3 and 4 BZ3, BZ4 can be taken into account by the AWE evaluation device via the OUT outputs, whereby these further influence states were determined by additional axle counters not shown. These axle counters can, for example, be provided at the other end of the track section, since track sections require axle counters at both the beginning and the end in order to generate and check vacancy reports and occupancy reports.

[0053] In Figure 2It can be seen that the electrical pulse generated by the train ZUG traveling on track section GA due to the physical influence of the wheels can be converted by sensors SEN into a rectified voltage RSP. The sensors SEN are designed as part of an axle counting device AZE, which is connected to the axle counting computer AZR via an interface 3.

[0054] The rectified voltage RSP is transmitted from the sensors SEN to the analog-to-digital converter ADU via interface S1. The raw data RDT from the analog-to-digital converter ADU is sent to the processor PRC via interface S2. Interface S3 is used to transmit this data to the axle counting computer AZR for detecting wheel influences / influence states, i.e., the influence states BZ1 and BZ2. The data for the occupancy status of the track section GA is transmitted to the control center via interface S4, where it is used by the axle counting computer to generate the vacant or occupied report for the track section GA. List of reference symbols:

[0055] GAG Track section TS1 Subsystem 1 TS2 Subsystem 2 AZR Axle counting computer IPR Wheel pulse SEN Sensor VS1 Process step 1 RSP Directed voltage ADU Analog-to-digital converter VS2 Process step 2 RDT Raw data PRC Processor VS3 Process step 3 KDT Compensated data VS4 Process step 4 VSL Reinforcement learning OUT Output BZ1 Influence state 1 BZ2 Influence state 2 AWE Evaluation device BZ3, BZ4 Influence states (of another axle counting device) FP Timetable FPD Timetable data ZUG Train AZE Axle counting device LZ Control center S1 ... S4 Interfaces

Claims

1. Method for computer-assisted determining of the occupancy state of a track section (GA) with the steps of • the physical wheel influence of trains (ZUG) travelling on the track section (GA) is converted into a directional voltage (RSP), • the directional voltage (RSP) is read into a processor (PRC) as raw data (RDT) via an analogue-digital converter (ADU), • an assessment to check whether or not a wheel influence is present takes place with the aid of at least one algorithm, and an influence state is derived therefrom, • the influence state is output to determine the occupancy state, characterised in that to identify wheel influences as part of the assessment of the algorithm, a reinforcement learning is used to optimise the parameters taken into consideration for the identification, wherein • the algorithm uses neural networks for the reinforcement learning (VSL), • the method steps are performed in parallel in a first subsystem (TS1) as well as in a second subsystem (TS2) and the first subsystem (TS1) as well as the second subsystem (TS2) both output the influence state to an evaluation facility (AWE) and the evaluation facility (AWE) compares an actual occupancy state with the influence state, • the evaluation facility (AWE) ascertains the actual occupancy state by evaluation of the data of sensor facilities, in particular axle counters, on the track section (GA) or of external data, in particular timetable data (FPD) • the evaluation facility (AWE) reports the actual occupancy state or an item of information with regard to the actual occupancy state corresponding to the output influence state back to the neural networks of the first subsystem (TS1) as well as of the second subsystem (TS2), • in the event that one of the subsystems of the first subsystem (TS1) as well as of the second subsystem (TS2), by way of the reinforcement learning (VSL), has achieved a higher level of certainty in relation to misinterpretations during the ascertaining of the influence state than the other, then the parameters taken into consideration for the identifying are transferred to the other subsystem (TS1, TS2).

2. Method according to claim 1, characterised in that the at least one algorithm performs a threshold value identification and / or a temporal filtering and / or the consideration of a hysteresis between identification of the wheel start and the wheel end.

3. Method according to one of the preceding claims, characterised in that in a processing step of the raw data (RDT) a temperature influence is eliminated, wherein compensated data (KDT) is generated for the assessment.

4. Method according to one of the preceding claims, characterised in that a check cycle with a set of input variables is performed in time intervals, for which the result of the ascertaining of the influence state can be predicted, and the ascertained influence state is compared with an expected, predictable influence state.

5. Axle counting facility (AZE) for identifying wheel influences on a track section (GA) with • a sensor (SEN), with which the physical wheel influence of trains (ZUG) travelling on the track section (GA) can be converted into a directional voltage (RSP), • an analogue-digital converter (ADU), with which raw data (RDT) can be generated from the directional voltage (RSP), • a processor (PRC), with which the raw data (RDT) can be processed and can be assessed with regard to an occupancy state by at least one algorithm implemented in the processor (PRC), • an interface (S3) for outputting an occupancy state characterised in that the processor has a neural network for the identification of wheel influences as part of the assessment, wherein the axle counting facility with the neural networks is configured to carry out a method for computer-assisted determining of the occupancy state of a track section (GA) according to one of the preceding claims.

6. Axle counting facility (AZE) according to claim 4 or 5, characterised in that a standardised parameter set for performing the algorithm is stored in the processor (PRC).

7. Computer program product with program commands for performing the method according to one of claims 1-4 when it is performed by a processor (PRC) of an axle counting facility (AZE) according to claim 5.