Train cooperative control method and system based on brain-like spiking neural network

Abstract

The application provides a kind of train cooperative control method and system based on brain-like pulse neural network, belongs to train operation control technical field.The application realizes the automatic tracking of train for given target speed under the limitation factors such as external environmental interference by establishing train automatic operation control framework based on pulse neural network, and realizes the train safety active protection of multi-train cooperative operation, guarantees the safe distance constraint of train not overspeed operation and cooperative state, meets the safety requirement in the process of operation in various scenes, can control train cooperative operation under the condition of guaranteeing safe operation, and reduces the running interval by more than 40%, the corresponding train efficiency can be improved by more than 2 times;With pulse neuron as basic unit, based on brain-like structure and bionic principle, the overall control effect is improved by dynamically adjusting synaptic weight, the biological interpretability and bionics of control framework are improved, which is beneficial to the migration and calculation efficiency improvement of brain-like computing platform.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of train operation control, in particular to a train cooperative control method and system based on a brain-like spiking neural network. BACKGROUND

[0002] To further improve train operation efficiency, multi-train cooperative operation control technology based on many automatic control algorithms has attracted widespread attention. In the multi-train cooperative operation mode, the coupling relationship between trains is more complex, the tracking distance between trains is closer, and the final cooperative state requires same-speed operation, which puts higher requirements on the performance of control technology. In the process of completing the tracking and cooperative tasks, environmental disturbances, line speed limits and other limiting factors cannot be ignored, and the calculation method of the safety protection distance between trains is also different from the existing operation mode. In view of these new requirements and challenges, the automatic control technology for multi-train cooperative operation needs to develop towards intelligence and be able to integrally process various safety constraints, complex disturbances and the like in the cooperative process, and finally ensure safe and efficient operation of trains.

[0003] In addition, there are certain requirements for the calculation efficiency, power consumption and the like of the automatic control algorithm in the final application of train operation. Intelligent control methods represented by deep learning and reinforcement learning, while meeting the requirements of improving the intelligence of control technology, also have problems such as high requirements for training data and high requirements for computing power. In fact, there are still differences in structure and information processing between the above artificial intelligence methods and real biological intelligence, and the biological interpretability is poor, which is not conducive to fully exerting the advantages of human flexibility and low power consumption in dealing with complex problems. Therefore, a new type of intelligent control technology is needed, which is more in line with the mechanism of human intelligence and is conducive to exerting the advantages of high computing power and low power consumption in future applications.

[0004] In the process of train operation, the control system needs to ensure that it realizes the set target speed tracking and prevents train overspeed operation from affecting the operation safety; considering multi-train cooperation, the train needs to be controlled to run at the same speed and stable interval while avoiding train collision. Therefore, considering the speed and tracking distance constraints, how to establish a new type of control and protection integrated architecture under the premise of fully utilizing the advantages of the spiking neural network control architecture is a problem to be solved. With the development of brain-like computing and brain-like chip technologies, control methods based on brain-like architecture have become a research focus. As the third generation of neural networks, spiking neural networks simulate biological mechanisms by establishing neuron cells, synapse models and the like, and realize information transmission through pulse signals. The brain-like computing structure based on spiking neural networks is conducive to further simulating the understanding and completion ability of biological tasks, and is conducive to the implementation of the overall architecture on the brain-like computing platform, and finally establishes an efficient, low-power and fast control system.

[0005] The existing train cooperative automatic control technology mainly includes traditional control methods represented by adaptive control and model predictive control, and control methods based on artificial intelligence technology such as deep learning and reinforcement learning. The traditional control method mainly realizes the cooperative operation control of multiple trains through the establishment of a train dynamics model, taking the error between the real-time state and the expected value of multiple trains as the input, and finally realizing the cooperative operation control of multiple trains, safety protection or through a train operation protection system independent of the control system, or through state constraints. In order to further improve the intelligence of train control, control methods based on artificial intelligence technology such as deep learning and reinforcement learning have been widely concerned. Taking the deep learning method as an example, a large amount of operation data is used to train the deep learning network, so as to output control signals or protection signals acting on the train, that is, intelligent control based on data.

[0006] The existing train cooperative automatic control technology mainly includes traditional control methods represented by adaptive control and model predictive control, and control methods based on artificial intelligence technology such as deep learning and reinforcement learning. The traditional control method usually does not fully consider safety protection in the process of train cooperative control, and some methods do not have active safety protection function; the train cooperative operation control and protection method based on artificial intelligence method such as deep learning usually requires high data and high computing power in the early stage, and has high power consumption in actual application. In addition, when the neural network is trained and used, the information is transmitted between neurons in the form of continuous signals, which is different from the mechanism of real biological neurons, and the biological interpretability is poor. SUMMARY

[0007] The purpose of the present application is to provide a train cooperative control method and system based on brain-like spiking neural network, to solve at least one of the technical problems in the background art.

[0008] In order to achieve the above purpose, the present application adopts the following technical scheme:

[0009] On the one hand, the present application provides a train cooperative control method based on brain-like spiking neural network, comprising:

[0010] Obtaining the expected target information, real-time state information, and train safety protection information and front car safety protection information in the process of train operation, and completing data preprocessing and coding to obtain coded pulse signals;

[0011] The pre-trained brain-like control model includes: the front lobe simulation calculation module adjusts the hidden layer weight online, and calculates the control signal transmission to the protection and control driving module according to the target information and the speed protection information; in the cooperative operation mode, the cerebellum simulation calculation module calculates the control signal transmission to the cooperative control driving module according to the preset cooperative target, and dynamically adjusts the train speed and the distance between the front and rear trains until the controlled train and the front train cooperatively operate; the protection and control driving module and the cooperative control driving module decode the pulse output, and calculate the final control output, to further control the train operation.

[0012] Preferably, the real-time state information of the controlled train includes the current position and speed of the train; the safety protection information includes the limited maximum speed or the safety protection distance between the train and the front train;

[0013] The relative braking distance protection rule is used in the cooperative tracking process of the two trains, and the safety protection distance d is calculated according to the following rule w (t):

[0014]

[0015] Wherein, ΔL is the reserved safety boundary, v(t) is the speed of the controlled train (rear train), v l (t) is the speed of the front train, a u is the maximum normal braking rate of the controlled train, a l,max is the maximum emergency braking rate of the front train;

[0016] Under the line speed limit v lim (t) and the safety protection distance d w (t), the safety protection error e(t) is defined as: Wherein, ò v and ò s are the speed and position safety margins set by human beings, and Δp(t) is the real-time distance between the front and rear trains.

[0017] Preferably, for a given reference speed signal v ref (t), the speed tracking error e v (t) = v ref (t) - v(t) is defined; considering the scale gap of each state information, each information is preprocessed, and the processing rule is Wherein, x is the signal before processing, is the signal after processing, P x is a normalization parameter defined according to different signals, and η is a scaling coefficient; each state information is converted into a pulse signal satisfying the Poisson distribution and having different frequencies by using the Poisson encoding mode, and the excitation frequency is each preprocessed signal.

[0018] Preferably, the prefrontal cortex simulation module comprises a perception layer, an intermediate layer and a driving layer; the perception layer contains 3 groups of neurons, receives external stimuli and generates input pulses; the intermediate layer contains 5 neurons, which are fully connected with the perception layer, and the four neurons in the intermediate layer are fully connected, and the connection weight is adjusted according to the R-STDP rule; the perception layer inputs the preprocessed and Poisson encoded reference speed v ref (t), real-time speed v(t), tracking error e v (t) and safety protection information e(t) in pulse form, which has an excitation and inhibition effect on the neurons of the driving layer after the intermediate layer, and the weight adjustment mainly consists of dopamine signal acquisition and weight update by using the R-STDP rule; in a sampling period, DA(t) = ωe v (t) is defined, where DA(t) is the dopamine signal, and ω is a given scaling factor; DA is converted into a pulse signal by using Poisson encoding; the dopamine signal acts on the synapses between the intermediate neurons to further adjust the weight values, and after repeated calculation and update, it enters the next sampling period.

[0019] Preferably, the pre-trained brain-like control model adjusts the weight online based on biological mechanisms, and the pre-training includes: acquiring real-time speed and position information under different control signals, and stimulating weight adjustment by using dopamine signal;

[0020]

[0021]

[0022] wherein m is the mass of the train, R(v) = c0+c v v i +c a v 2 is the air resistance during the train operation; c0, c v and c a are Davis coefficients, which can be inferred from historical data and experience; d(t) is an uncertain disturbance introduced by the external environment; p(t), v(t) are position and speed information, is the calculable traction / braking force, is the control signal corresponding to the unit mass.

[0023] Preferably, when the train is running in a cooperative mode, the cerebellum simulation module is activated, mossy fibers receive the real-time running state of the front and rear trains and the cooperative target, and calculate the corresponding control signal, and the calculation process relies on the trained granule cells, Purkinje cells, deep cerebellar nuclei and olivary nucleus cells; the olivary nucleus cells realize the adjustment of the weight of the connection synapses between the granule cells and the Purkinje cells by using the input correction signal, wherein the correction signal is obtained by spring damping system training, and the target signal is the expected tracking distance d des, the input signal is the real-time distance Δp(t) between the controlled train and the front train and its differential and integral information, and the output signal is the unit mass traction / braking force of the train; the synaptic weight between the granule cell and the Purkinje cell is adjusted according to the STDP rule after being excited by the correction signal, specifically, when the correction signal and the granule cell excitation arrival time are similar, the corresponding synaptic weight is increased, and vice versa; the deep cerebellar nucleus receives the excitatory input from the mossy fiber and the inhibitory input from the Purkinje cell, and generates an output signal input into the cooperative control driving module; and the final actual control signal is the fusion signal of the protection and control module and the cooperative control driving module.

[0024] Preferably, the output pulse signal is decoded into a control signal by using an agonist-antagonist model:

[0025]

[0026]

[0027] wherein is the output of the prefrontal cortex simulation module, and are the driving layer excitation and inhibition output pulse numbers in the prefrontal cortex simulation module respectively; sigmoid is a variable mapping function for mapping the output signal to the range [0, 1]; u max is the maximum range of the control signal given according to the actual situation; is the output of the cerebellum simulation module, and are the deep cerebellar nucleus excitation and inhibition output pulse numbers in the cerebellum simulation module respectively; and α and β are scaling coefficients. is the changed control signal, that is, the total output of the method, corresponding to the control input in the train dynamics model; W is a weight value in the range [0, 1], and H is a selection coefficient, which is 1 when the train is running in a cooperative state, and 0 otherwise.

[0028] In the second aspect, the application provides a train cooperative control system based on a brain-like pulse neural network, comprising:

[0029] An acquisition module is configured to acquire expected target information, real-time state information, and self-vehicle safety protection information and front-vehicle safety protection information during the running of a controlled train, and to complete data preprocessing and coding to obtain coded pulse signals.

[0030] A control module is configured to process the coded pulse signal by using a pre-trained brain-like control model; the pre-trained brain-like control model comprises: an imitation prefrontal cortex calculation module is configured to adjust the hidden layer weight online, and calculate a control signal according to target information and speed protection information and transmit the control signal to a protection and control driving module; in a cooperative operation mode, an imitation cerebellum calculation module is configured to calculate a control signal according to a preset cooperative target and transmit the control signal to a cooperative control driving module, and dynamically adjust the train speed and the distance between the front and rear trains until the controlled train and the front train cooperatively operate; the protection and control driving module and the cooperative control driving module are configured to decode the pulse output and calculate a final control output to further control the train operation.

[0031] In a third aspect, the application provides a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implement the train cooperative control method based on the brain-like pulse neural network.

[0032] In a fourth aspect, the application provides a computer program product comprising a computer program, which, when running on one or more processors, is configured to implement the train cooperative control method based on the brain-like pulse neural network.

[0033] In a fifth aspect, the application provides an electronic device comprising a processor, a memory and a computer program; the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for implementing the train cooperative control method based on the brain-like pulse neural network.

[0034] The application has the advantages that: the train can automatically track a given target speed or a cooperative operation target state under the safety constraint condition that the train meets its own speed and the protection distance between trains, and the operation efficiency and reliability are further improved; meanwhile, the application is inspired by the human brain structure, the pulse neurons in the network transmit information in the form of pulses, and the synaptic weights between neurons are adjusted in a biomimetic form, which has strong biological interpretability and is conducive to the application of high-performance and low-power platforms based on brain-like chips in the future.

[0035] The advantages of the additional aspects of the application will be more apparent from the following description part or be understood through the practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description only show some of the embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort based on these drawings.

[0037] Figure 1 The whole framework schematic diagram of the train cooperative control system based on the brain-like spiking neural network is described in the embodiments of the present application.

[0038] Figure 2 The internal structure schematic diagram of the prefrontal cortex simulation calculation module is described in the embodiments of the present application.

[0039] Figure 3 The internal structure schematic diagram of the cerebellum simulation calculation module is described in the embodiments of the present application. DETAILED DESCRIPTION

[0040] The embodiments of the present application will be described in detail below, and the examples of the embodiments are shown in the drawings, wherein the same or similar notations represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below with the drawings are exemplary and are only used to explain the present application, and cannot be explained as the limitation of the present application.

[0041] Those skilled in the art can understand that, unless otherwise defined, all the terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of the ordinary skilled in the art to which the present application belongs.

[0042] It should also be understood that the terms such as those defined in the general dictionary should be understood as having the same meaning as in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as such.

[0043] Those skilled in the art can understand that, unless otherwise stated, the singular forms "a", "an" and "the" used herein also include the plural forms. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or groups exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements and / or groups thereof.

[0044] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0045] In order to facilitate the understanding of the present application, the present application will be further explained and described in specific embodiments in conjunction with the accompanying drawings, and the specific embodiments do not constitute a limitation on the embodiments of the present application.

[0046] The person skilled in the art should understand that the drawings are only schematic diagrams of the embodiments, and the components in the drawings are not necessarily necessary for the implementation of the present application.

[0047] Embodiment 1

[0048] In this embodiment 1, a train cooperative control system based on a brain-like pulse neural network is first provided, comprising:

[0049] The acquisition module is used to acquire the expected target information, real-time state information, and the safety protection information of the train and the front vehicle during the running process of the controlled train, and to complete data preprocessing and coding to obtain the coded pulse signal;

[0050] The control module is used to process the coded pulse signal by using the pre-trained brain-like control model; the pre-trained brain-like control model comprises: the prefrontal cortex simulation module adjusts the hidden layer weight online, and calculates the control signal transmission to the protection and control driving module according to the target information and the speed protection information; in the cooperative running mode, the cerebellum simulation module calculates the control signal transmission to the cooperative control driving module according to the preset cooperative target, and dynamically adjusts the train speed and the distance between the front and rear vehicles until the controlled train and the front vehicle run cooperatively; the protection and control driving module and the cooperative control driving module decode the pulse output, and calculate the final control output to further control the train running.

[0051] In this embodiment 1, the train cooperative control method based on the brain-like pulse neural network is realized by using the above system, comprising:

[0052] The expected target information, real-time state information, and safety protection information of the train and the front vehicle during the running process of the controlled train are acquired, and data preprocessing and coding are completed to obtain the coded pulse signal;

[0053] The pre-trained brain-like control model includes: the front lobe simulation calculation module adjusts the hidden layer weight online, and calculates the control signal transmission to the protection and control driving module according to the target information and the speed protection information; in the cooperative operation mode, the cerebellum simulation calculation module calculates the control signal transmission to the cooperative control driving module according to the preset cooperative target, and dynamically adjusts the train speed and the distance between the front and rear trains until the controlled train and the front train cooperatively operate; the protection and control driving module and the cooperative control driving module decode the pulse output, and calculate the final control output, to further control the train operation.

[0054] The real-time state information of the controlled train includes the current position and speed of the train; the safety protection information includes the limited maximum speed or the safety protection distance between the train and the front train;

[0055] The relative braking distance protection rule is used in the cooperative tracking process of the two trains, and the safety protection distance d is calculated according to the following rule w (t):

[0056]

[0057] Wherein ΔL is the reserved safety boundary, v(t) is the speed of the controlled train (rear train), v l (t) is the speed of the front train, a u is the maximum normal braking rate of the controlled train, a l,max is the maximum emergency braking rate of the front train;

[0058] Under the line speed limit v lim (t) and the safety protection distance d w (t), the safety protection error e(t) is defined as: Wherein, ò v And ò s is the speed and position safety margin set by human, and Δp(t) is the real-time distance between the front and rear trains.

[0059] For a given reference speed signal v ref (t), the speed tracking error e v (t) = v ref (t) - v(t) is defined; considering the scale gap of each state information, each information is preprocessed, and the processing rule is Wherein x is the signal before processing, is the signal after processing, P x is a normalization parameter defined according to different signals, and η is a scaling coefficient; using the Poisson coding mode, each state information is converted into a pulse signal satisfying the Poisson distribution and having different frequencies, and the excitation frequency is each preprocessed signal.

[0060] The simulation prefrontal calculation module comprises a perception layer, an intermediate layer and a driving layer; the perception layer contains 3 groups of neurons, receives external stimulation and generates input pulses; the intermediate layer contains 5 neurons, which are fully connected with the perception layer, and the four neurons in the intermediate layer are fully connected, and the connection weight is adjusted according to the R-STDP rule; the perception layer inputs the preprocessed and Poisson encoded reference speed v ref (t), real-time speed v(t), tracking error e v (t) and safety protection information e(t) in pulse form, after the intermediate layer, the driving layer neurons are excited and inhibited, and the weight adjustment mainly consists of dopamine signal collection and weight update; in a sampling period, DA(t) = ωe v (t) is defined, wherein DA(t) is the dopamine signal, and ω is a given scaling coefficient; DA is converted into a pulse signal by Poisson coding; the dopamine signal acts on the synapses between the intermediate neurons to further adjust the weight values, and after repeated calculation and update, it enters the next sampling period.

[0061] The pre-trained brain-like control model adjusts the weight online based on biological mechanisms, and the pre-training includes: acquiring the speed and position information under different control signals in real time, and stimulating the weight adjustment by dopamine signal;

[0062]

[0063]

[0064] Wherein, m is the mass of the train, R(v) = c0+c v v i +c a v 2 is the air resistance during train operation; c0, c v and c a are Davis coefficients, which can be inferred from historical data and experience; d(t) is the uncertain disturbance introduced by the external environment; p(t), v(t) are position and speed information, is the calculable traction / braking force, is the control signal corresponding to the unit mass.

[0065] When the train runs in the cooperative mode, the simulation cerebellum calculation module is activated, the mossy fiber receives the real-time running state of the front and rear trains and the cooperative target, calculates the corresponding control signal, and the calculation process mainly depends on the trained granule cells, Purkinje cells, deep cerebellar nuclei and olivary nucleus cells. The olivary nucleus cell realizes the adjustment of the weight of the connection synapse between the granule cell and the Purkinje cell by using the input correction signal, wherein the correction signal is obtained by spring damping system training, and the target signal is the expected tracking distance d des, the input signal is the real-time distance Δp(t) between the controlled train and the front train and its differential and integral information, and the output signal is the unit mass traction / braking force of the train; the synaptic weight between the granule cell and the Purkinje cell is adjusted according to the STDP rule after being excited by the correction signal, specifically, when the correction signal and the granule cell excitation arrival time are similar, the corresponding synaptic weight increases, and vice versa; the deep cerebellar nucleus receives excitatory input from the mossy fiber and inhibitory input from the Purkinje cell, and generates an output signal input into the cooperative control driving module; and the final actual control signal is the fusion signal of the protection and control module and the cooperative control driving module.

[0066] The output pulse signal is decoded into a control signal by using an agonist-antagonist muscle model:

[0067]

[0068]

[0069] wherein is the output of the prefrontal cortex simulation module, and are the driving layer excitation and inhibition output pulse numbers in the prefrontal cortex simulation module; sigmoid is a variable mapping function for mapping the output signal to the range [0, 1]; u max is the maximum range of the control signal given according to the actual situation; is the output of the cerebellum simulation module, and are the deep cerebellar nucleus excitation and inhibition output pulse numbers in the cerebellum simulation module; α and β are scaling coefficients; is the changed control signal, that is, the total output of the method, corresponding to the control input in the train dynamics model; W is a weight value with a value range of [0, 1], and H is a selection coefficient, which is 1 when the train is running in a cooperative state, and 0 otherwise.

[0070] Embodiment 2

[0071] As Figures 1 to 3 shown, a train cooperative control method based on a brain-like pulse neural network is provided in this embodiment 2. The train running control and safety protection of the brain-like pulse neural network mainly includes the following steps: Figure 1The main function module implementation is shown. The state perception module is directly connected with the train in operation, mainly realizing real-time state information collection; the protection and control driving calculation module completes safety risk assessment and control signal calculation, and is converted by the driving module to ensure that the train realizes target state tracking operation; the cerebellum calculation module completes multi-train cooperative control signal calculation, and the driving module further generates suitable control force until the train reaches the cooperative state, i.e. the controlled train and the preceding train run at the same speed and stable spacing. The main implementation steps of the application are as follows:

[0072] (1) The state perception module collects the expected target information, real-time state information and safety protection information of the train during operation, and completes data preprocessing and coding.

[0073] (2) The coded pulse signal is input into the pre-trained brain-like control architecture, the hidden layer weight is adjusted online by the prefrontal cortex calculation module, and the control signal is calculated according to the target information and speed protection information and transmitted to the protection and control driving module.

[0074] (3) In the cooperative operation mode, the cerebellum calculation module calculates the control signal according to the preset cooperative target and transmits it to the cooperative control driving module, dynamically adjusts the train speed and the distance between the front and rear trains until the controlled train (rear train) and the front train run cooperatively.

[0075] (4) The protection and control driving module and the cooperative control driving module decode the pulse output and calculate the final control output to further control the train operation.

[0076] In step (1), the real-time state information of the controlled train includes: the current position and speed of the train; the safety protection information includes: the maximum speed limit or the safety protection distance between the train and the front train. The relative braking distance protection rule is used in the cooperative tracking process of the two trains, and the safety protection distance d w (t) is calculated according to the following rule:

[0077]

[0078] Where ΔL is the reserved safety boundary, v(t) is the speed of the controlled train (rear train), v l (t) is the speed of the front train, a u is the maximum normal braking rate of the controlled train, a l,max is the maximum emergency braking rate of the front train. Under the line speed limit v lim (t) and the safety protection distance d w (t), the safety protection error e(t) is defined as:

[0079]

[0080] Where, ò v ​s The speed and position safety margin set by human, and Δp(t) is the real-time distance between the front and rear vehicles. Considering the tracking operation of the two vehicles, it is assumed that the front vehicle runs at a speed of 350 km / h, the rear vehicle runs at a speed of 348 km / h, the front vehicle emergency braking deceleration is -1.0 m / s 2 , and the rear vehicle braking deceleration is -0.8 m / s 2 . Without considering the sudden stop of the front vehicle, compared with the protection rule under the moving block, the train cooperative operation is controlled under the safety operation condition, and the running interval is reduced by more than 40%, and the train operation efficiency can be improved by more than 2 times.

[0081] For a given reference speed signal v ref (t) (the reference speed signal in the cooperative state is v ref (t) = v l (t)), a speed tracking error e v (t) = v ref (t) - v(t) can be defined. Considering the scale gap of each state information, first, each information is preprocessed, and the processing rule is where x is the signal before processing, is the signal after processing, P x is a normalization parameter defined according to different signals, and η is a scaling factor. In addition, since the signal of the spiking neural network architecture is propagated in the form of pulses, the normalized information needs to be encoded. Using the Poisson encoding method, each state information is converted into a pulse signal that satisfies the Poisson distribution and has different frequencies. The excitation frequency is the preprocessed signal, and the excited pulse signal is input into the subsequent brain-like control system.

[0082] In step (2), the spiking neural network adopts the LIF model, and the driving calculation model mainly includes a perception layer, an intermediate layer and a driving layer. The perception layer includes 3 groups of neurons, receives external stimulation and generates input pulses, the intermediate layer includes 5 neurons, is fully connected with the perception layer, and the four neurons in the intermediate layer are fully connected. The connection weight is adjusted according to the R-STDP rule. The driving layer is two neurons, which are stimulated by the perception layer and the intermediate layer, connected with the external driving unit, and further act on the train. The perception layer inputs the reference speed v ref (t), the real-time speed v(t), the tracking error e v (t) and the safety protection information e(t) after preprocessing and Poisson encoding in the form of pulses, and the intermediate layer has an excitation and inhibition effect on the neurons of the driving layer. The weight adjustment using the R-STDP rule mainly consists of two parts of collecting dopamine signals and weight updating. First, in a sampling period, DA(t) = ωe v(t), where DA(t) is the dopamine signal, and ω is a given scaling factor. Similarly, DA is converted into a spike signal using Poisson encoding. The dopamine signal then acts on the interneuron synapses to further adjust the weights, which are updated after repeated calculations and enter the next sampling period.

[0083] In addition, the network in step (2) adjusts the weights online based on biological mechanisms. To improve the initial performance, the network is pre-trained using the following train model simulation data. The specific method is to connect the driving calculation part with the train simulation model to real-time feedback the speed and position information under different control signals, and use the dopamine signal to stimulate the weight adjustment.

[0084]

[0085]

[0086] where m is the mass of the train, R(v) = c0+ c v v i + c a v 2 is the air resistance during train operation; c0, c v and c a are Davis coefficients, which can be inferred from historical data and experience; d(t) is the uncertain disturbance introduced by the external environment; p(t), v(t) are position and speed information, is the calculable traction / braking force, is the control signal corresponding to the unit mass.

[0087] In step (3), the cerebellum-like computing model completes the cooperative control function, mainly composed of mossy fibers, granular cells and deep cerebellar nuclei. When the train is running in the cooperative mode, the cerebellum-like computing module is activated, the mossy fiber receives the real-time running state of the front and rear trains and the cooperative target, and calculates the corresponding control signal. The calculation process mainly depends on the trained granular cells, Purkinje cells, deep cerebellar nuclei and inferior olive cells. The inferior olive cells in the model realize the adjustment of the weights between the granular cells and the Purkinje cells by using the input correction signal, where the correction signal is obtained through a spring damping system training, and the target signal is the expected tracking distance d des, the input signal is the real-time distance Δp(t) between the controlled train and the front train and its differential and integral information, and the output signal is the unit mass traction / braking force of the train (corrected signal). The synaptic weight between the granule cell and the Purkinje cell is adjusted according to the STDP rule after being excited by the corrected signal, and specifically, when the corrected signal and the granule cell excitation arrival time are similar, the corresponding synaptic weight is increased, and vice versa. The deep cerebellar nucleus receives excitatory input from the mossy fiber and inhibitory input from the Purkinje cell, and generates an output signal input into the cooperative control driving module. The actual control signal received by the train is the fusion signal of the protection and control module and the cooperative control driving module.

[0088] In step (4), the output of the driving calculation module and the safety protection module is a pulse signal, which needs to be further decoded.

[0089] In the present application, the output pulse signal is decoded into a control signal by using an agonist-antagonist model, and the specific change formula is:

[0090]

[0091]

[0092] Wherein is the output of the prefrontal cortex simulation module, and respectively, the output pulse number of the driving layer excitation (agonist) and inhibition (antagonist) in the prefrontal cortex simulation module; sigmoid is a common variable mapping function, which is used to map the output signal to the range of [0, 1]; u max is the maximum range of the control signal given according to the actual situation; is the output of the cerebellum simulation module, and respectively, the output pulse number of the deep cerebellar nucleus excitation (agonist) and inhibition (antagonist) in the cerebellum simulation module; alpha and beta are scaling coefficients; is the changed control signal, that is, the total output of the method, corresponding to the control input in the train dynamics model; W is a weight value with a value range of [0, 1], and H is a selection coefficient, which takes a value of 1 when the train is running in a cooperative state, and vice versa.

[0093] In summary, in the embodiment, the main control signal calculation part is composed of the prefrontal cortex simulation calculation module and the cerebellum simulation calculation module, and through the cooperation of multiple modules designed, the train expected state tracking and collaborative operation control are finally realized, which can ensure the active safety protection under the speed limit and interval distance constraints in the collaborative operation process. Under the assumed simulation conditions, compared with the protection rules under the mobile block, the train collaborative operation can be controlled under the condition of ensuring safe operation, and the running interval can be reduced by more than 40%, and the train efficiency can be improved by more than 2 times. The signals are transmitted between neurons in the form of pulses, and the measurement state, control signal and information conversion and interaction between the neural network are completed through the designed encoding and decoding method. At the same time, the synaptic weights between neurons are adjusted in a biomimetic way to ensure the overall performance. The method has strong biological interpretation and is beneficial to the further application of the biomimetic brain chip, and the computing power consumption is low.

[0094] Embodiment 3

[0095] The embodiment 3 provides a non-transitory computer readable storage medium for storing computer instructions, which, when executed by a processor, implements the train collaborative control method based on the brain-like pulse neural network as described above, and the method comprises the following steps:

[0096] obtaining expected target information, real-time state information, and safety protection information of the controlled train and safety protection information of the preceding train in the running process of the controlled train, and completing data preprocessing and encoding to obtain encoded pulse signals;

[0097] processing the encoded pulse signals by using a pre-trained brain-like control model; the pre-trained brain-like control model comprises: a prefrontal cortex simulation calculation module for online adjusting hidden layer weights and calculating control signals according to target information and speed protection information and transmitting the control signals to a protection and control driving module; in the collaborative operation mode, a cerebellum simulation calculation module calculates control signals according to a preset collaborative target and transmits the control signals to a collaborative control driving module to dynamically adjust the train speed and the distance between the trains until the controlled train and the preceding train collaboratively operate; the protection and control driving module and the collaborative control driving module decode the pulse output and calculate the final control output to further control the train operation.

[0098] Embodiment 4

[0099] The embodiment 4 provides a computer program product comprising a computer program which, when run on one or more processors, is configured to implement the train collaborative control method based on the brain-like pulse neural network as described above, and the method comprises the following steps:

[0100] obtaining expected target information, real-time state information, and safety protection information of the controlled train and safety protection information of the preceding train in the running process of the controlled train, and completing data preprocessing and encoding to obtain encoded pulse signals;

[0101] The pre-trained brain-like control model includes: an imitation prefrontal calculation module for online adjustment of hidden layer weights and calculation of control signal transmission to a protection and control driving module according to target information and speed protection information; in a cooperative operation mode, an imitation cerebellum calculation module calculates control signal transmission to a cooperative control driving module according to a preset cooperative target, and dynamically adjusts train speed and inter-train distance until the controlled train and the front train cooperatively operate; and the protection and control driving module and the cooperative control driving module decode the pulse output and calculate the final control output to further control train operation.

[0102] Embodiment 5

[0103] The embodiment 5 provides an electronic device, including: a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions for realizing a train cooperative control method based on a brain-like pulse neural network, the method including:

[0104] obtaining expected target information, real-time state information, and train safety protection information and front train safety protection information during train operation, and completing data preprocessing and encoding to obtain encoded pulse signals;

[0105] The pre-trained brain-like control model includes: an imitation prefrontal calculation module for online adjustment of hidden layer weights and calculation of control signal transmission to a protection and control driving module according to target information and speed protection information; in a cooperative operation mode, an imitation cerebellum calculation module calculates control signal transmission to a cooperative control driving module according to a preset cooperative target, and dynamically adjusts train speed and inter-train distance until the controlled train and the front train cooperatively operate; and the protection and control driving module and the cooperative control driving module decode the pulse output and calculate the final control output to further control train operation.

[0106] In summary, the train cooperative control method and system based on the brain-like pulse neural network according to the embodiments of the present application establish a train automatic operation control architecture based on the pulse neural network, realize automatic tracking of a given target speed of the train under the limitation of external environmental interference and other factors, realize cooperative operation of multiple trains, realize safe active protection of the train, ensure non-exceeding speed operation of the train and safe distance constraint in the cooperative state, and meet the safety requirements in the running process in multiple scenarios. Under the assumed simulation conditions, compared with the protection rules under the moving block, the train cooperative operation can be controlled under the condition of safe operation, and the running interval is reduced by more than 40%, and the train efficiency is improved by more than 2 times. The pulse neuron is used as a basic unit, the brain-like structure and the bionic principle are used, the overall control effect is improved through dynamic adjustment of synaptic weights, the method has high intelligence level. In addition, the biological interpretability and bionics of the control architecture are improved, which is beneficial to migration on the brain-like computing platform and improvement of the computing efficiency.

[0107] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0108] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that carries out the function specified in one or more blocks.

[0109] These computer program instructions can also be stored in a computer-readable memory that can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction devices that implement the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that carries out the function specified in one or more blocks.

[0110] These computer program instructions can also be loaded into a computer or other programmable data processing devices, to cause a series of operational steps to be performed on the computer or other programmable data processing devices, so as to generate a computer implemented process, thus the instructions executed on the computer or other programmable data processing devices provide a process for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 one flow or multiple flows and / or the functions specified in the block

[0111] The above describes the specific embodiments of the present application in conjunction with the drawings, but is not a limitation on the scope of protection of the present application, and those skilled in the art should understand that on the basis of the technical solutions disclosed in the present application, various modifications or changes made by those skilled in the art without the need for creative labor should be covered within the scope of protection of the present application.

Claims

1. A train cooperative control method based on a brain-like spiking neural network, characterized in that, include: The system acquires the desired target information, real-time status information, and safety protection information of the train itself and the train ahead during the operation of the controlled train, and completes data preprocessing and encoding to obtain the encoded pulse signal. The encoded pulse signals are processed using a pre-trained neuromorphic control model; The pre-trained brain-like control model includes: a prefrontal cortex-like computation module that adjusts the weights of the hidden layer online and calculates control signals based on target information and speed protection information, transmitting them to the protection and control drive module; in cooperative operation mode, a cerebellum-like computation module calculates control signals based on preset cooperative targets and transmits them to the cooperative control drive module, dynamically adjusting the train speed and the distance between trains until the controlled train and the preceding train operate in coordination; the protection and control drive module and the cooperative control drive module decode the pulse output and calculate the final control output to further control the train operation; wherein, the prefrontal cortex-like computation module includes a perception layer, an intermediate layer, and a drive layer; the perception layer contains 3 groups of neurons that receive external stimuli and generate input pulses; the intermediate layer contains 5 neurons that are fully connected to the perception layer, and the four neurons in the intermediate layer are fully connected to each other, with the connection weights adjusted according to the R-STDP rule; the perception layer uses the preprocessed and Poisson-encoded reference speed... Real-time speed Tracking error and security protection information Input in pulse form, after passing through the intermediate layer, excites and inhibits neurons in the driver layer. Weight adjustment using the R-STDP rule mainly consists of two parts: acquiring dopamine signals and updating weights. Within one sampling period, the following is defined: ,in This is a dopamine signal, in which Given scaling factors; DA is converted into pulse signals using Poisson coding; dopamine signals act on the synapses between interneurons to further adjust the weight values, and after repeated calculations and updates, the next sampling cycle begins; a pre-trained brain-like control model adjusts the weights online based on biological mechanisms. Pre-training includes: real-time acquisition of velocity and position information under different control signals, and using dopamine signals to stimulate weight adjustment; ； Where m is the mass of the train. Air resistance during train operation; and The Davis coefficient can be inferred from historical data and experience. Uncertain interference introduced by the external environment; , For position and velocity information, For calculable traction / braking force, The control signal corresponding to a unit mass; When the train is running in cooperative mode, the cerebellar computing module is activated. Moss fibers receive the real-time operating status of the trains ahead and behind, as well as the cooperative target, and calculate the corresponding control signals. The calculation process mainly relies on trained granulocytes, Parkinson cells, deep cerebellar nuclei, and inferior olivary nuclei. Olivary nuclei cells adjust the weights of the synapses connecting granulocytes and Parkinson cells using input correction signals. The correction signals are obtained through training with a spring-damped system, and the target signal is the desired tracking distance. The input signal is the real-time distance between the controlled train and the preceding train. The differential and integral information of the signal is used to output the traction / braking force per unit mass of the train. The synaptic weights between granulocytes and Parkinson's cells are adjusted according to the STDP rule after being excited by the correction signal. Specifically, when the arrival time of the correction signal and the granulocyte excitation is similar, the corresponding synaptic weight increases, and vice versa. The deep cerebellar nucleus receives excitatory input from moss fibers and inhibitory input from Parkinson's cells, and generates an output signal that is input to the collaborative control drive module. The final actual control signal is the fusion signal of the protection and control module and the collaborative control drive module.

2. The train cooperative control method based on a neuromorphic spiking neural network according to claim 1, characterized in that, The real-time status information of the controlled train includes the train's current position and speed; safety protection information includes the maximum speed limit or the safe distance between the train and the preceding train. During the coordinated tracking process between the two vehicles, a relative braking distance protection rule is adopted, and the safe protection distance is calculated according to the following rules. : ； in For the reserved safety boundary, The speed of the controlled vehicle (the vehicle behind), The speed of the vehicle in front. Maximum service braking rate of the controlled vehicle The maximum emergency braking rate of the vehicle in front; Speed ​​limits on the line and safe protection distance Below, define the safety protection error. : ;in, and Safety margins for speed and position set by humans. This shows the real-time distance between the vehicles in front and behind.

3. The train cooperative control method based on a neuromorphic spiking neural network according to claim 2, characterized in that, For a given reference velocity signal Define speed tracking error Considering the scale differences between various state information, each piece of information undergoes preprocessing, with the following processing rules: ,in To process the pre-processed signal, For the processed signal, These are normalization parameters defined for different signals. The scaling factor is used; using Poisson coding, each state information is converted into pulse signals of different frequencies that satisfy the Poisson distribution, and the excitation frequency is the preprocessed signal.

4. The train cooperative control method based on a neuromorphic spiking neural network according to claim 1, characterized in that, The output pulse signal is decoded into a control signal using an agonist-antagonist muscle model: ； ； in This is the output of the simulated prefrontal cortex calculation module. and These represent the number of excitation and inhibition output pulses in the driving layer of the simulated prefrontal cortex calculation module, respectively; sigmoid is a variable mapping function used to map the output signal to the range [0,1]. The maximum range of control signals given based on actual conditions; For the output of the cerebellum-like computing module, and These represent the number of output pulses for the deep cerebellar nuclei of the cerebellar computing module, respectively, and the number of output pulses for the stimulation and inhibition of the cerebellar nuclei. and This is the scaling factor; The changed control signal is the total output of the proposed method, which corresponds to the control input in the train dynamics model. The weight values ​​are in the range [0,1]. The selection coefficient is set to 1 when the train is in a cooperative state, and 0 otherwise.

5. A train cooperative control system based on a brain-like spiking neural network, characterized in that, include: The acquisition module is used to acquire the desired target information, real-time status information, and safety protection information of the train itself and the train ahead during the operation of the controlled train, and to complete the data preprocessing and encoding to obtain the encoded pulse signal; The control module is used to process the encoded pulse signals using a pre-trained neuromorphic control model; The pre-trained brain-like control model includes: a prefrontal cortex-like computation module that adjusts the weights of the hidden layer online and calculates control signals based on target information and speed protection information, transmitting them to the protection and control drive module; in cooperative operation mode, a cerebellum-like computation module calculates control signals based on preset cooperative targets and transmits them to the cooperative control drive module, dynamically adjusting the train speed and the distance between trains until the controlled train and the preceding train operate in coordination; the protection and control drive module and the cooperative control drive module decode the pulse output and calculate the final control output to further control the train operation; wherein, the prefrontal cortex-like computation module includes a perception layer, an intermediate layer, and a drive layer; the perception layer contains 3 groups of neurons that receive external stimuli and generate input pulses; the intermediate layer contains 5 neurons that are fully connected to the perception layer, and the four neurons in the intermediate layer are fully connected to each other, with the connection weights adjusted according to the R-STDP rule; the perception layer uses the preprocessed and Poisson-encoded reference speed... Real-time speed Tracking error and security protection information Input in pulse form, after passing through the intermediate layer, excites and inhibits neurons in the driver layer. Weight adjustment using the R-STDP rule mainly consists of two parts: acquiring dopamine signals and updating weights. Within one sampling period, the following is defined: ,in This is a dopamine signal, in which Given scaling factors; DA is converted into pulse signals using Poisson coding; dopamine signals act on the synapses between interneurons to further adjust the weight values, and after repeated calculations and updates, the next sampling cycle begins; a pre-trained brain-like control model adjusts the weights online based on biological mechanisms. Pre-training includes: real-time acquisition of velocity and position information under different control signals, and using dopamine signals to stimulate weight adjustment; ； Where m is the mass of the train. Air resistance during train operation; and The Davis coefficient can be inferred from historical data and experience. Uncertain interference introduced by the external environment; , For position and velocity information, For calculable traction / braking force, The control signal corresponding to a unit mass; When the train is running in cooperative mode, the cerebellar computing module is activated. Moss fibers receive the real-time operating status of the trains ahead and behind, as well as the cooperative target, and calculate the corresponding control signals. The calculation process mainly relies on trained granulocytes, Parkinson cells, deep cerebellar nuclei, and inferior olivary nuclei. Olivary nuclei cells adjust the weights of the synapses connecting granulocytes and Parkinson cells using input correction signals. The correction signals are obtained through training with a spring-damped system, and the target signal is the desired tracking distance. The input signal is the real-time distance between the controlled train and the preceding train. The differential and integral information of the signal is used to output the traction / braking force per unit mass of the train. The synaptic weights between granulocytes and Parkinson's cells are adjusted according to the STDP rule after being excited by the correction signal. Specifically, when the arrival time of the correction signal and the granulocyte excitation is similar, the corresponding synaptic weight increases, and vice versa. The deep cerebellar nucleus receives excitatory input from moss fibers and inhibitory input from Parkinson's cells, and generates an output signal that is input to the collaborative control drive module. The final actual control signal is the fusion signal of the protection and control module and the collaborative control drive module.

6. A computer program product, characterized in that, Includes a computer program, which, when run on one or more processors, is used to implement the train cooperative control method based on a neuromorphic spiking neural network as described in any one of claims 1-4.

7. An electronic device, characterized in that, include: The electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions that implement the train cooperative control method based on a neuromorphic spiking neural network as described in any one of claims 1-4.

Images