Learning model generation method, estimation device and wireless train control system
By generating machine learning models to improve spectrum perception accuracy, the problem of insufficient spectrum perception accuracy in wireless train control systems is solved, high-reliability wireless communication allocation is achieved, and the communication quality of wireless train control systems is improved.
Patent Information
- Application Number
- CN202080082431.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-11-27
- Filing Date
- 2020-11-02
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2040-11-02
AI Technical Summary
In high-reliability wireless train control systems, the prior art is difficult to achieve high-precision spectrum perception, especially in cognitive radio communications using the ISM frequency band, there are challenges in finding idle wireless communication channels and switching wireless communication standards.
By generating a machine learning model with radio wave environment data as input and signal-to-noise ratio output according to frequency communication status information, using training data to correlate radio wave environment and communication standards of wireless base stations, the first and second learning models are built to improve spectrum perception accuracy .
High-precision spectrum perception in the wireless train control system is realized, wireless channels can be dynamically allocated, and wireless communication reliability between the ground and the vehicle is improved.
Smart Images

Figure CN114762375B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a learning model generation method, an estimation device and a wireless train control system for generating a machine learning model. Background Art
[0002] For a wireless train control system called CBTC (Communication Based Train Control) that utilizes wireless communication in train control, the use of a frequency band called the ISM (Industrial Scientific and Medical) band that does not require a license in wireless communication between the ground and the train has been studied. As wireless communication standards that utilize this ISM band, there are various wireless communication standards such as wireless LAN (IEEE802.11) and Bluetooth (registered trademark) (IEEE802.15.1). The above wireless communication standards are utilized by various other wireless communication systems and therefore cannot avoid interference. Therefore, as a technology for reducing the effects of interference and improving the reliability of wireless communication, for example, Patent Document 1 discloses a technology for changing the channel of wireless communication based on a frequency hopping pattern table.
[0003] Prior art literature
[0004] Patent Literature
[0005] Patent Document 1: Japanese Patent Application Publication No. 2013-239935 Summary of the invention
[0006] Problem that the invention aims to solve
[0007] However, in the above-mentioned patent document 1, it is necessary to actually communicate between the base station and the vehicle-mounted device (mobile station). Therefore, there are problems such as the occupation of wireless communication channels including retransmission and how to find idle wireless communication channels in advance. In wireless communication using the ISM frequency band, there is a technology called cognitive radio communication, which searches for idle wireless communication channels and switches the wireless method to be used. In this cognitive radio communication, spectrum sensing for finding idle wireless communication channels is required. However, when cognitive radio communication is applied to a wireless train control system that requires high reliability, higher accuracy is also required for spectrum sensing.
[0008] The present invention has been made in view of the above-mentioned situation, and an object of the present invention is to provide a new technology of high-precision spectrum sensing applied to a wireless train control system performing cognitive radio communication.
[0009] Solutions for solving problems
[0010] A first approach to solving the above problem is a learning model generation method, which is executed by a computer system, wherein:
[0011] The computer system performs the following steps:
[0012] Acquiring training data, wherein the training data is data obtained by associating data representing the radio wave environment of the wireless base station when the vehicle-mounted device performs cognitive radio communication with the wireless base station installed along the railway with frequency-based communication status information based on the SNR (Signal to Noise Ratio) when the wireless base station performs wireless communication in the radio wave environment, in which communication can be performed by switching the wireless communication standard; and
[0013] A generation step uses the training data to generate a machine learning model that takes data representing the radio wave environment as input and takes the frequency-based communication status information as output.
[0014] According to the first embodiment, a technology for high-precision spectrum sensing applied to a wireless train control system that performs cognitive radio communication can be realized. That is, a machine learning model can be generated that takes as input data representing the radio wave environment of a wireless base station when a wireless base station in a wireless train control system performs cognitive radio communication with an onboard device and outputs frequency-based communication status information based on the SNR when the wireless base station performs wireless communication in the radio wave environment. Furthermore, by using the machine learning model, high-precision spectrum sensing can be realized.
[0015] The learning model generation method can also be constructed as:
[0016] The training data also includes the wireless communication standard used by the wireless base station in the radio wave environment.
[0017] The machine learning model includes a first learning model and a second learning model for each of the wireless communication standards,
[0018] The generating step comprises the following steps:
[0019] generating the first learning model using the training data, the first learning model having data representing the radio wave environment as input and the wireless communication standard as output; and
[0020] The second learning model is generated for each of the wireless communication standards using the training data, the second learning model having data representing the radio wave environment as input and having the frequency-specific communication status information as output.
[0021] Accordingly, a machine learning model can be generated in a manner including a first learning model and a second learning model for each wireless communication standard, wherein the first learning model takes data representing the radio wave environment as input and takes the wireless communication standard as output, and the second learning model takes data representing the radio wave environment as input and takes frequency-based communication status information as output. When the wireless communication standards are different, the communication status may also be different, so the accuracy of spectrum sensing using the machine learning model can be improved.
[0022] A second aspect is an estimation device comprising:
[0023] a radio wave environment data acquisition unit that acquires data indicating the radio wave environment of a radio base station installed along a railway when the vehicle-mounted device performs cognitive radio communication with the radio base station, in which communication is performed by switching the wireless communication standard; and
[0024] An estimation processing execution unit performs the following estimation processing, i.e., inputs the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into a machine learning model obtained by learning using training data that takes data representing the radio wave environment as input and takes frequency-based communication status information based on the SNR (Signal to Noise Ratio) when the wireless base station performs wireless communications in the radio wave environment as output, and outputs the frequency-based communication status information.
[0025] According to the second embodiment, a technology for high-precision spectrum sensing applied to a wireless train control system that performs cognitive radio communication can be realized. That is, high-precision spectrum sensing using a machine learning model that performs an estimation process can be realized, wherein data representing the radio wave environment of a wireless base station when a wireless base station in the wireless train control system performs cognitive radio communication with an on-board device is input to the machine learning model to output frequency-specific communication status information based on the SNR when the wireless base station performs wireless communication in the radio wave environment.
[0026] The estimation device can also be constructed as:
[0027] The training data also includes the wireless communication standard used by the wireless base station in the radio wave environment.
[0028] The machine learning model includes a first learning model and a second learning model for each of the wireless communication standards,
[0029] The estimation processing execution unit includes:
[0030] a first estimation processing execution unit for executing a first estimation processing of inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the first learning model obtained by learning using the training data and taking the data representing the radio wave environment as input and taking the wireless communication standard used by the wireless base station in the radio wave environment as output, so as to output the wireless communication standard; and
[0031] A second estimation processing execution unit, which executes the following second estimation processing, i.e., selecting the second learning model suitable for the wireless communication standard output by the first estimation processing execution unit from the second learning models of each of the wireless communication standards obtained by learning according to the wireless communication standard using the training data and taking the data representing the radio wave environment as input and taking the frequency-based communication status information as output, and inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the selected second learning model to output the frequency-based communication status information.
[0032] Accordingly, the machine learning model includes a first learning model and a second learning model for each wireless communication standard, wherein the first learning model takes data representing the radio wave environment as input and the wireless communication standard as output, and the second learning model takes data representing the radio wave environment as input and frequency-based communication status information as output. When the wireless communication standards are different, the communication status may also be different, so by using the machine learning model, the accuracy of spectrum perception can be improved.
[0033] The estimation device can also be constructed as:
[0034] The training data also includes, in association with the wireless propagation characteristics related to the wireless communication standard used by the wireless base station in the radio wave environment.
[0035] The second learning model exists for each combination of the wireless communication standard and the wireless propagation characteristic.
[0036] The first learning model is a model obtained by learning using data representing the radio wave environment as input and using the wireless communication standard used by the wireless base station in the radio wave environment and the wireless propagation characteristics related to the wireless communication standard as output.
[0037] The first estimation processing execution unit executes the following first estimation processing: inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the first learning model to output a combination of the wireless communication standard and the wireless propagation characteristics,
[0038] The second estimation processing execution unit performs the following second estimation processing: selecting the second learning model suitable for the combination output by the first estimation processing execution unit, inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the selected second learning model, and outputting the frequency-based communication status information.
[0039] Accordingly, the machine learning model includes a first learning model and a second learning model for each combination of a wireless communication standard and a wireless propagation characteristic, wherein the first learning model takes data representing a radio wave environment as input and takes a combination of a wireless communication standard and a wireless propagation characteristic as output, and the second learning model takes data representing a radio wave environment as input and takes frequency-based communication status information as output. When the combination of a wireless communication standard and a wireless propagation characteristic related to the wireless communication standard is different, the communication status may also be different, so by using a multi-stage machine learning model, such as using a second learning model suitable for the combination estimated by the first learning model, the accuracy of spectrum perception can be improved.
[0040] The estimation device can also be constructed as:
[0041] The machine learning model is a convolutional neural network.
[0042] Based on this, a machine learning model based on convolutional neural network is generated.
[0043] The estimation device can also be constructed as:
[0044] The radio wave environment data acquisition unit acquires data indicating the radio wave environment based on a wireless signal received by a monitoring receiving device installed along the railway line.
[0045] Thus, by providing a monitoring receiver as a device other than the wireless base station, data indicating the radio wave environment of the wireless base station can be acquired without affecting the wireless communication between the vehicle-mounted device and the ground base station. The monitoring receiver can be provided at any location, for example, near the wireless base station.
[0046] The third method is a wireless train control system having:
[0047] The estimation device as described above; and
[0048] An allocating device dynamically allocates a wireless channel related to the cognitive radio communication based on the frequency-specific communication status information output from the estimation processing execution section of the estimation device.
[0049] According to the third aspect, in the wireless train control system, wireless channels related to cognitive radio communication between the wireless base station and the onboard device can be dynamically allocated based on the frequency-specific communication status information of the wireless base station estimated by the estimation device. This can improve the reliability of wireless communication between the ground and the train. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 This is a configuration example of a wireless train control system.
[0051] Figure 2 This is an explanatory diagram of the estimation of the radio wave environment using the communication status estimation model.
[0052] Figure 3 This is an example of classification of the frequency usage status in the frequency-based communication status information.
[0053] Figure 4 This is a diagram for explaining the learning of the communication status estimation model.
[0054] Figure 5 This is a diagram for explaining the learning of the communication status estimation model.
[0055] Figure 6 This is an example of the functional structure of a communication status estimation device. DETAILED DESCRIPTION
[0056] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. In addition, the present invention is not limited to the embodiments described below, and the mode in which the present invention can be applied is not limited to the embodiments described below. In addition, in the description of the accompanying drawings, the same elements are marked with the same reference numerals.
[0057] [System Structure]
[0058] Figure 1 This is a configuration example of the wireless train control system 1 of this embodiment. Figure 1 As shown, the wireless train control system 1 of this embodiment includes a wireless base station 10, a monitor receiving device 20, and a control device 100. The control device 100 is a device that manages the operation of each train 50 in a unified manner. The control device 100 has a distribution device 30 and a communication status estimation device 40 as main devices related to the wireless control of this embodiment. The wireless base station 10 and the monitor receiving device 20 are installed along the railway. The control device 100 is installed in a central command station, etc. In addition, each device is connected to each other through a ground transmission line N for communication.
[0059] The wireless communication between the wireless base station 10 and the onboard device 52 carried by the train 50 running on the line is wireless communication using ISM (Industrial Scientific and Medical) bands such as the 2.4 GHz band, the 5.7 GHz band, and the 920 MHz band. In this wireless communication, cognitive radio communication is performed to search for frequencies not used by other communication systems and switch wireless communication standards and wireless channels to perform wireless communication.
[0060] A plurality of wireless base stations 10 are installed along the railway line so that their communicable ranges partially overlap each other, so that the onboard devices 52 of the trains 50 running on the line can continuously communicate with the control device 100. In addition, the wireless base station 10 can establish communication channels with a plurality of predetermined number of trains 50, and can communicate with the onboard devices 52 of each train 50 through the wireless channels allocated based on the instructions of the allocation device 30.
[0061] The monitoring receiver 20 is a device provided to obtain data indicating the radio wave environment of the corresponding wireless base station 10. The monitoring receiver 20 is provided near the corresponding wireless base station 10, receives wireless signals around the wireless base station 10, and outputs the received wireless signals to the communication status estimation device 40. Figure 1 Although two wireless base stations 10 and two monitoring receiving devices 20 are shown, in reality, a plurality of them are installed along the railway line.
[0062] The allocation device 30 dynamically allocates wireless channels related to cognitive radio communication between the wireless base station 10 and the onboard device 52 of the train 50 based on the communication status estimated by the communication status estimation device 40. For example, when the wireless base station 10 is switched (hand-over), a wireless channel of an idle frequency is selected and allocated based on the communication status around the next wireless base station 10 estimated in real time by the communication status estimation device 40. Then, the onboard device 52 of the train 50 and the wireless base station 10 are notified of the allocated wireless channel.
[0063] The communication status estimation device 40 estimates the communication status of the periphery of each wireless base station 10 based on the wireless signal received by the monitor receiving device 20 corresponding to the wireless base station 10 .
[0064] [Estimate of communication status]
[0065] Figure 2 4 is a diagram for explaining the estimation of the communication status by the communication status estimation device 40. Figure 2As shown, the communication status estimation device 40 estimates the communication status by frequency as the communication status around the corresponding wireless base station 10 using the communication status estimation model 60 based on the wireless signal received by the monitor receiving device 20, that is, performs so-called spectrum sensing. In addition, the communication status estimation model 60 is generated for each wireless base station 10.
[0066] The communication status estimation model 60 is implemented by a convolutional neural network in the present embodiment, but may be a neural network other than the convolutional neural network. In addition, the communication status estimation model 60 is a machine learning model obtained by learning using training data as described later, and the weight coefficients are determined. Moreover, when estimating the communication status using the communication status estimation model 60, N data obtained by normalizing the amplitude and phase of the wireless signal received by the monitor receiving device 20 and performing a high-speed Fourier transform process are input to the communication status estimation model 60 as data representing the radio wave environment, instead of inputting the spectrum diagram of the wireless signal to the communication status estimation model 60 as data representing the radio wave environment. The monitor receiving device 20 is set in the vicinity of the wireless base station 10. Therefore, the wireless signal received by the monitor receiving device 20 can be regarded as a wireless signal received by the wireless base station 10. As a result, the processing time required for estimating the communication status can be greatly shortened. Therefore, it is possible to achieve near real-time or real-time estimation of the communication status.
[0067] Then, the communication status estimation model 60 outputs the frequency-based communication status information of the vicinity of the monitor receiving device 20 when the input wireless signal is received by the monitor receiving device 20. The frequency-based communication status information is the communication status of the vicinity of the corresponding wireless base station 10. The frequency-based communication status information is information showing the communication status by frequency in units of, for example, wireless channels. The communication status is classified into a plurality of stages based on the SNR (Signal to Noise Ratio) when the wireless base station 10 performs wireless communication in the wireless environment.
[0068] Figure 3 An example of the classification of communication conditions based on SNR is shown. Figure 3 An example of classification into four stages in order of lightness and darkness, namely black (B), gray (G), silver (S), and white (W) is shown in FIG. 1 , but the classification may be set to less than three stages, or may be classified into five or more stages. Figure 3 The example shows that in the order of darkness from dark to light, the SNR is from large to small, that is, the possibility of wireless communication at this frequency is from high to low (= used).
[0069] In addition, if Figure 2As shown, the communication status estimation model 60 includes a first model 62 and a plurality of second models 64. The first model 62 and the second model 64 are respectively implemented as convolutional neural networks. In other words, it can be said that the communication status estimation model 60 is configured to include a plurality of convolutional neural networks.
[0070] The first model 62 is a learning model that takes data representing a radio wave environment as input and outputs a combination of a wireless communication standard to be used in cognitive radio and wireless propagation characteristics of a wireless channel, and is a learning model for selecting one from a plurality of second models 64 .
[0071] The second model 64 is a learning model that takes data representing the radio wave environment as input and frequency-based communication status information as output, and is generated by combining each set of wireless communication standards and wireless channel propagation characteristics. The wireless communication standards include wireless methods that can be switched for use in cognitive radio in this embodiment. The wireless communication standards include, for example, communication standards for wireless LAN (Wi-Fi) such as IEEE802.11b / 11a / 11g / 11n / 11ac, and short-range wireless communication standards such as Bluetooth. Wireless propagation characteristics include, for example, SNR, center frequency offset (CFO: Center Frequency Offset), sampling rate offset (SRO: Sample Rate Offset), frequency selective fading (Frequency selective fading), and power delay profile (PDP: Power Delay Profile).
[0072] In the estimation of the communication status using the communication status estimation model 60, first, data representing the radio wave environment based on the wireless signal received by the monitor receiving device 20 is input to the first model 62. Through the first estimation process performed using the first model 62, a combination of a wireless communication standard and a wireless propagation characteristic suitable for the input data representing the radio wave environment is output. Next, a second model 64 corresponding to the combination of the wireless communication standard and the wireless propagation characteristic output from the first model 62 is selected. Then, the data representing the radio wave environment based on the wireless signal received by the monitor receiving device 20 is input to the selected second model 64. Through the second estimation process performed using the second model 64, frequency-specific communication status information suitable for the input data representing the radio wave environment is output.
[0073] [Learning of Communication Status Estimation Model Data]
[0074] Figure 4 4 is a diagram for explaining the learning (machine learning) of the communication status estimation model 60 by the communication status estimation device 40. Figure 4As shown, the communication status estimation model 60 is learned using a plurality of training data 318 prepared in advance. One training data 318 is data obtained by associating the following information: data indicating the radio wave environment of the wireless base station 10; the wireless communication standard used by the wireless base station 10 in the radio wave environment; the wireless propagation characteristics related to the wireless communication standard used by the wireless base station 10 in the radio wave environment; and the frequency-based communication status information based on the SNR when the wireless base station 10 performs wireless communication in the radio wave environment. The communication status estimation model 60 is generated by completing learning with the first three data (data indicating the radio wave environment, the wireless communication standard, and the wireless propagation characteristics) as input and the last one data (frequency-based communication status information) as output. The type of learning is set to deep learning in this embodiment, but it can also be set to any other type of machine learning.
[0075] The training data 318 can be generated by, for example, simulation calculations performed using a computer in addition to using data of actual wireless communication. Specifically, the communication status estimation device 40 sets the parameters of the wireless communication standard and the wireless propagation characteristics, and generates a transmission signal when arbitrary transmission data is transmitted in accordance with the set wireless communication standard. Next, the communication status estimation device 40 obtains a wireless signal assumed to be received by the wireless base station 10 when the transmission signal is transmitted in a wireless channel of the set wireless propagation characteristics. Then, the communication status estimation device 40 calculates the spectrum diagram of the obtained wireless signal or the N data obtained by normalizing the amplitude and phase of the wireless signal and performing a fast Fourier transform process as data representing the radio wave environment. In addition, the communication status estimation device 40 obtains the SNR of the transmission signal by frequency based on the spectrum of the wireless signal and the transmission signal, thereby calculating the communication status information by frequency. Then, the set wireless communication standard and the parameters of the wireless propagation characteristics, the data representing the radio wave environment, and the communication status information by frequency are regarded as one training data 318. By setting various wireless communication standards and parameters of the wireless propagation characteristics, a large amount of training data 318 can be generated.
[0076] As described above, the communication status estimation model 60 has a first model 62 and a plurality of second models 64. Figure 5As shown, machine learning using the training data 318 is completed for each learning model of the first model 62 and the plurality of second models 64. That is, for the first model 62, learning is completed with the data representing the radio wave environment as input and the combination of the wireless communication standard and the wireless propagation characteristics as output. In addition, the second model 64 is generated for each combination of the wireless communication standard and the wireless propagation characteristics, so for each second model 64, learning is completed with the data representing the radio wave environment included in the training data 318 including the corresponding combination of the wireless communication standard and the wireless propagation characteristics as input and the communication status information by frequency as output.
[0077] More specifically, all the prepared training data 318 are grouped according to the combination of wireless communication standards and wireless propagation characteristics, and machine learning is performed using the training data 318 of each group, thereby generating the second model 64 corresponding to the combination of the wireless communication standard and wireless propagation characteristics of the group. Then, learning is performed using the combination of the wireless communication standard and wireless propagation characteristics of each group as an output, thereby generating the first model 62.
[0078] [Functional structure]
[0079] Figure 6 FIG. 4 is a functional block diagram of the communication status estimation device 40. Figure 6 The communication status estimation device 40 includes an operation unit 102, a display unit 104, a sound output unit 106, a communication unit 108, a processing unit 200, and a storage unit 300, and can be configured as a computer system.
[0080] The operation unit 102 is implemented by input devices such as a keyboard, a mouse, a touch panel, and various switches, and outputs an operation signal corresponding to the completed operation to the processing unit 200. The display unit 104 is implemented by a display device such as a liquid crystal display and a touch panel, and performs various displays corresponding to the display signal from the processing unit 200. The sound output unit 106 is implemented by a sound output device such as a speaker, and performs various sound outputs corresponding to the sound signal from the processing unit 200. The communication unit 108 is implemented by a wireless or wired communication device, connected to the ground transmission line N, and communicates with various external devices such as the wireless base station 10 and the distribution device 30.
[0081] The processing unit 200 is a processor implemented by a computing device such as a CPU (Central Processing Unit), an FPGA (Field Programmable Gate Array), or a computing circuit. The processing unit 200 performs overall control of the communication status estimation device 40 based on the program or data stored in the storage unit 300, the input data from the operation unit 102, and the like. In addition, the processing unit 200 has a radio wave environment data acquisition unit 202, an estimation processing execution unit 204, and an estimation model generation unit 210 as functional processing blocks. These functional units of the processing unit 200 can be implemented in software by executing a program by the processing unit 200, or can be implemented by a dedicated computing circuit. In this embodiment, the former is described as being implemented in software.
[0082] The radio wave environment data acquisition unit 202 acquires data indicating the radio wave environment of the wireless base station 10 when the vehicle-mounted device 52 performs cognitive radio communication with the wireless base station 10 installed along the railway, in which communication can be performed by switching the wireless communication standard. Specifically, N data obtained by normalizing the amplitude and phase of the wireless signal received by the monitoring receiving device 20 installed near the wireless base station 10 and performing a fast Fourier transform process are acquired as data indicating the radio wave environment of the wireless base station 10.
[0083] The estimation processing execution unit 204 performs the following estimation processing: the data representing the radio wave environment acquired by the radio wave environment data acquisition unit 202 is input to the communication status estimation model 60, which is a machine learning model generated by the estimation model generation unit 210, to output the communication status information by frequency. The communication status estimation model 60 includes a first model 62 and a second model 64, wherein the first model 62 is a first learning model that takes the data representing the radio wave environment as input and takes the combination of the wireless communication standard and the wireless propagation characteristics related to the wireless communication standard as output, and the second model 64 is a second learning model that takes the data representing the radio wave environment as input and takes the communication status information by frequency for each combination of the wireless communication standard and the wireless propagation characteristics as output.
[0084] Furthermore, the estimation processing execution unit 204 includes a first estimation processing execution unit 206 and a second estimation processing execution unit 208 .
[0085] The first estimation processing execution unit 206 executes a first estimation processing of inputting data indicating the radio wave environment acquired by the radio wave environment data acquisition unit 202 to the first model 62 which is the first learning model, and outputting a combination of the wireless communication standard and the radio propagation characteristics.
[0086] The second estimation processing execution unit 208 performs the following second estimation processing: a second model 64 suitable for the combination of the wireless communication standard and the wireless propagation characteristics output by the first estimation processing execution unit 206 is selected from the second learning model, i.e., the second model 64, and data representing the radio wave environment acquired by the radio wave environment data acquisition unit 202 is input into the selected second model 64 to output frequency-based communication status information.
[0087] The estimation model generation unit 210 obtains training data 318, which is data obtained by associating data representing the radio wave environment of the wireless base station 10 when the vehicle-mounted device 52 and the wireless base station 10 installed along the railway line perform cognitive radio communication capable of switching wireless communication standards, frequency-based communication status information based on the SNR (Signal to Noise Ratio) when the wireless base station 10 performs wireless communication in the radio wave environment, the wireless communication standard used by the wireless base station 10 in the radio wave environment, and wireless propagation characteristics related to the wireless communication standard used by the wireless base station 10 in the radio wave environment. The estimation model generation unit 210 uses the training data 318 to generate a machine learning model, namely, a communication status estimation model 60, which uses the data representing the radio wave environment as input and uses the frequency-based communication status information as output.
[0088] The machine learning model, i.e., the communication status estimation model 60, includes a first model 62 and a second model 64, wherein the first model 62 is a first learning model, and the second model 64 is a second learning model of a combination of each set of wireless communication standards and wireless propagation characteristics related to the wireless communication standards. That is, the estimation model generation unit 210 uses the training data 318 to learn with the data representing the radio wave environment as input and the combination of the wireless communication standard and the wireless propagation characteristics related to the wireless communication standard as output, thereby generating the first learning model, i.e., the first model 62. In addition, using the training data 318, learning is performed according to the combination of the wireless communication standard and the wireless propagation characteristics related to the wireless communication standard with the data representing the radio wave environment as input and the communication status information by frequency as output, thereby generating the second learning model, i.e., the second model 64, of a combination of each set of wireless communication standards and the wireless propagation characteristics related to the wireless communication standard.
[0089] The storage unit 300 is implemented by a storage device such as an IC (Integrated Circuit) memory such as a ROM (Read Only Memory), a RAM (Random Access Memory), or a hard disk. The storage unit 300 stores programs and data used by the processing unit 200 to uniformly control the communication status estimation device 40, and is used as a work area of the processing unit 200 to temporarily store calculation results obtained by the processing unit 200, input data from the operation unit 102, etc. In this embodiment, the storage unit 300 stores a communication status estimation program 302 and wireless base station information 310.
[0090] The wireless base station information 310 is generated for each wireless base station 10, and stores, in correspondence with the identification information (base station ID) of the corresponding wireless base station 10: communication status estimation model data 312; radio wave environment data 314, which is data indicating the radio wave environment of the corresponding wireless base station 10 acquired by the radio wave environment data acquisition unit 202; frequency-specific communication status information 316 of the corresponding wireless base station 10 estimated by the estimation processing execution unit 204; and training data 318 used in learning the communication status estimation model 60 performed by the estimation model generation unit 210. The communication status estimation model data 312 is data defining the communication status estimation model 60 of the corresponding wireless base station 10, specifically, data of weight coefficients.
[0091] [Effects]
[0092] According to the present embodiment, a technology for high-precision spectrum sensing applied to a wireless train control system that performs cognitive radio communication can be realized. That is, a machine learning model, i.e., a communication status estimation model 60, can be generated that takes as input data representing the radio wave environment of the wireless base station 10 when the wireless base station 10 in the wireless train control system 1 performs cognitive radio communication with the onboard device 52 and outputs frequency-specific communication status information based on the SNR when the wireless base station 10 performs wireless communication in the radio wave environment. Furthermore, by executing an estimation process, high-precision spectrum sensing using the communication status estimation model 60 can be realized. The estimation process is as follows: data representing the radio wave environment of the wireless base station 10 is input to the communication status estimation model 60, and frequency-specific communication status information based on the SNR when the wireless base station 10 performs wireless communication in the radio wave environment is output.
[0093] In addition, the embodiment to which the present invention can be applied is not limited to the above-mentioned embodiment, and it is of course possible to appropriately change it within the scope not departing from the gist of the present invention.
[0094] (A) Monitor receiving device 20
[0095] In the above-mentioned embodiment, the monitoring receiving device 20 is set in a one-to-one correspondence with the wireless base station 10 near the corresponding wireless base station 10, but it can also be set at any position. In this case, for example, based on the wireless signal received by the monitoring receiving device 20 installed at the position closest to the installation position of the wireless base station 10, the communication status around the wireless base station 10 is estimated.
[0096] Alternatively, the wireless base station 10 may also be used as the monitoring receiver 20 . In this case, the communication status estimation device 40 estimates the communication status around the wireless base station 10 based on the wireless signal received by the wireless base station 10 .
[0097] Description of Reference Numerals
[0098] 1: wireless train control system; 10: wireless base station; 20: receiving device for monitor; 30: distribution device; 40: communication status estimation device; 200: processing unit; 202: radio wave environment data acquisition unit; 204: estimation processing execution unit; 206: first estimation processing execution unit; 208: second estimation processing execution unit; 210: estimation model generation unit; 300: storage unit; 302: communication status estimation program; 310: wireless base station information; 312: communication status estimation model data; 314: radio wave environment data; 316: communication status information by frequency; 318: training data; 50: train; 52: on-board device; 60: communication status estimation model; 62: first model; 64: second model.
Claims
1. A learning model generation method, performed by a computer system, wherein: The computer system performs the following steps: Get training data, where Acquiring data indicating the radio wave environment of a wireless base station installed along the railway when the vehicle-mounted device performs cognitive radio communication with the wireless base station installed along the railway based on a wireless signal received by a receiving device for monitoring installed along the railway, wherein communication can be performed by switching the wireless communication standard in the cognitive radio communication; and associating the data indicating the radio wave environment with frequency-specific communication status information based on a signal-to-noise ratio when the wireless base station performs wireless communication in the radio wave environment; and A generation step uses the training data to generate a machine learning model that takes data representing the radio wave environment as input and takes the frequency-based communication status information as output.
2. The learning model generation method according to claim 1, wherein: The training data also includes the wireless communication standard used by the wireless base station in the radio wave environment. The machine learning model includes a first learning model and a second learning model for each of the wireless communication standards, The generating step comprises the following steps: generating the first learning model using the training data, the first learning model having data representing the radio wave environment as input and the wireless communication standard as output; and The second learning model is generated for each of the wireless communication standards using the training data, the second learning model having data representing the radio wave environment as input and having the frequency-specific communication status information as output.
3. An estimation device, comprising: a radio wave environment data acquisition unit configured to acquire data representing the radio wave environment of a radio base station installed along a railway when the vehicle-mounted device performs cognitive radio communication with the radio base station installed along the railway, wherein the communication is performed by switching the wireless communication standard, wherein the radio wave environment data acquisition unit is configured to acquire data representing the radio wave environment based on a wireless signal received by a receiving device for a monitor installed along the railway; and An estimation processing execution unit is configured to perform the following estimation processing, namely, inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into a machine learning model obtained by learning using training data that takes data representing the radio wave environment as input and takes frequency communication status information based on the signal-to-noise ratio when the wireless base station performs wireless communications in the radio wave environment as output, to output the frequency communication status information.
4. The estimation device according to claim 3, wherein: The training data also includes the wireless communication standard used by the wireless base station in the radio wave environment. The machine learning model includes a first learning model and a second learning model for each of the wireless communication standards, The estimation processing execution unit includes: a first estimation processing execution unit for executing the following first estimation processing, namely, inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the first learning model obtained by learning using the training data and taking the data representing the radio wave environment as input and taking the wireless communication standard used by the wireless base station in the radio wave environment as output, so as to output the wireless communication standard; as well as A second estimation processing execution unit, which executes the following second estimation processing, i.e., selecting the second learning model suitable for the wireless communication standard output by the first estimation processing execution unit from the second learning models of each of the wireless communication standards obtained by learning according to the wireless communication standard using the training data and taking the data representing the radio wave environment as input and taking the frequency-based communication status information as output, and inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the selected second learning model to output the frequency-based communication status information.
5. The estimation device according to claim 4, wherein: The training data also includes, in association with the wireless propagation characteristics related to the wireless communication standard used by the wireless base station in the radio wave environment. The second learning model exists for each combination of the wireless communication standard and the wireless propagation characteristic. The first learning model is a model obtained by learning using data representing the radio wave environment as input and using the wireless communication standard used by the wireless base station in the radio wave environment and the wireless propagation characteristics related to the wireless communication standard as output. The first estimation processing execution unit executes the following first estimation processing: inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the first learning model to output a combination of the wireless communication standard and the wireless propagation characteristics, The second estimation processing execution unit performs the following second estimation processing: selecting the second learning model suitable for the combination output by the first estimation processing execution unit, inputting the data representing the radio wave environment acquired by the radio wave environment data acquisition unit into the selected second learning model, and outputting the frequency-based communication status information.
6. The estimation device according to any one of claims 3 to 5, wherein: The machine learning model is a convolutional neural network.
7. A wireless train control system comprising: The estimation device according to any one of claims 3 to 6; and An allocating device dynamically allocates a wireless channel related to the cognitive radio communication based on the frequency-specific communication status information output from the estimation processing execution section of the estimation device.
