Intelligent railway ballast distributing vehicle side plough obstacle avoidance and automatic shaping system and method

Abstract

The invention provides an intelligent railway ballast distribution vehicle side plough obstacle avoidance and automatic shaping system and method, and aims to improve the safety and efficiency of railway maintenance operation. By integrating advanced sensor technology, machine learning algorithm and automatic shaping algorithm, the system realizes intelligent recognition of railway obstacles, obstacle avoidance and automation of track shaping operation. According to the system, the operation safety of the railway ballast distributing vehicle is improved, manual intervention is reduced, the operation efficiency and quality are improved through the intelligent technology, and the system is suitable for railway maintenance work under various complex environments and conditions.

Description

Technical Field

[0001] The invention relates to railway maintenance equipment, in particular to a railway ballast vehicle side plow device integrating intelligent obstacle avoidance and automatic shaping functions. Background Art

[0002] When ballast shaping is performed by a railway ballast shaping vehicle, there is a risk of collision between the side plow of the ballast shaping vehicle and obstacles on the roadside, that is, railway facilities such as overhead wire pillars and signal machines. At present, the safety of the operation is mainly ensured by workers' real-time on-site protection and drivers' observation, which poses a great safety hazard. Accidents such as side plows colliding with overhead wire pillars have occurred many times, causing great economic losses.

[0003] The quality of ballast shaping cannot be guaranteed or monitored, and the side plows can only be adjusted based on experience. Summary of the invention

[0004] The present invention aims to solve the above problems in the prior art and provides an intelligent railway ballast-shaping vehicle side plow obstacle avoidance and automatic shaping system, which can effectively meet the side plow obstacle avoidance needs of the ballast-shaping vehicle under complex obstacle conditions, ensure the effectiveness and safety of the operation, and continuously improve the obstacle avoidance strategy through the learning module and the automatic shaping module, monitor and continuously improve the ballast-shaping effect.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] An intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system, comprising:

[0007] at least one laser radar and at least one ultrasonic radar for detecting obstacles;

[0008] The first camera is located at the front side of the ballast truck and is used to capture obstacle images;

[0009] The second camera is located at the rear side of the ballast truck and is used to capture the image of the ballast after shaping;

[0010] a measurement and control system for controlling the retraction and extension of the side moldboard;

[0011] A machine learning module for identifying and classifying obstacles and optimizing obstacle avoidance strategies; obstacle identification uses the convolutional neural network formula Y = σ(W*X+b), where:

[0012] Y: output feature map of the convolutional layer;

[0013] W: weight of the convolution kernel;

[0014] X: input feature map;

[0015] b: bias term of the convolutional layer;

[0016] σ: activation function;

[0017] Classification using the support vector machine formula in:

[0018] f(x): prediction function given input x;

[0019] α i : Lagrange multiplier, determined by the optimization process;

[0020] y i : The label of the training sample, usually +1 or -1;

[0021] K(x,x i ): kernel function, calculates input x and support vector x i The similarity between

[0022] b: bias term;

[0023] sgn: symbolic function, returns -1 or 1, indicating the classification result;

[0024] The obstacle avoidance strategy is optimized using the Q-learning formula Q(s,a)←Q(s,a)+α[r+γmax a′ Q(s′,a′)-Q(s,a)], where:

[0025] Q(s,a): the expected utility of taking action a in state s;

[0026] α: learning rate, which controls the speed at which new information covers the original Q value;

[0027] r: the immediate reward obtained after taking action a in state s;

[0028] γ: discount factor, measuring the current value of future rewards;

[0029] s′: next state;

[0030] a′: the next possible action;

[0031] max a′ Q(s′,a′): the maximum Q value of all possible actions in the next state s′;

[0032] An automatic shaping module is used to automatically adjust the working angle and depth of the side plowboard according to the location and type of obstacles; the shaping strategy is generated using a genetic algorithm and the real-time adjustment uses a PID controller formula in,

[0033] u(t): control signal at time t;

[0034] Kp: proportional gain;

[0035] Ki: integral gain;

[0036] Kd: differential gain;

[0037] e(t): error signal at time t, i.e. the difference between the expected value and the actual value;

[0038] ∫e(t)dt: integral of the error, used to eliminate steady-state error;

[0039] The rate of change of error, used to reduce overshoot and oscillation of the system;

[0040] Decision making accuracy improved using data fusion formula in,

[0041] Z: result after fusion;

[0042] w i : The weight of the i-th sensor data, reflecting the relative importance of the data in the fusion result; Xi: The output data of the i-th sensor.

[0043] Furthermore, the machine learning module includes:

[0044] at least one convolutional neural network for extracting features from the camera image;

[0045] at least one classifier for classifying obstacles;

[0046] At least one reinforcement learning algorithm for learning an optimal obstacle avoidance path.

[0047] Furthermore, the automatic shaping module comprises:

[0048] at least one genetic algorithm for generating an optimal shaping strategy;

[0049] at least one PID controller for adjusting the position and angle of the side moldboard in real time;

[0050] At least one data fusion technique is used to improve the accuracy of the shaping decision.

[0051] Furthermore, it also includes:

[0052] At least one speed sensor, used to measure the running speed of the vehicle;

[0053] at least one attitude sensor for measuring the attitude of the vehicle;

[0054] At least one display screen, used to display data monitoring results;

[0055] At least one sound and light alarm for alarm prompts.

[0056] Furthermore, the measurement and control system can:

[0057] Automatically control the retraction and deployment of the side moldboard based on the output of the machine learning module;

[0058] The working angle and depth of the side moldboard are automatically adjusted according to the output of the automatic shaping module.

[0059] Furthermore, the system is capable of:

[0060] Remote monitoring and operation are achieved through wireless communication technology;

[0061] Implement online learning mechanisms that allow the system to continuously optimize its performance based on new data;

[0062] The model is updated regularly to adapt to environmental changes and new obstacle types.

[0063] A method for avoiding obstacles and automatically shaping the side plow of a railway ballast vehicle using the above system comprises the following steps:

[0064] Step 1: Detect obstacles using LiDAR and ultrasonic radar;

[0065] Step 2: Use the first camera to capture an image of an obstacle;

[0066] Step 3: Use machine learning module to identify and classify obstacles;

[0067] Step 4: Use the automatic shaping module to automatically adjust the working angle and depth of the side moldboard according to the type and location of the obstacle;

[0068] Step 5: Use the measurement and control system to control the retraction and deployment of the side moldboard;

[0069] Step 6: Adjust the side moldboard position and angle based on real-time feedback for precise shaping.

[0070] It also includes optimizing the plastic surgery module according to the image captured by the second camera to improve the accuracy of plastic surgery decisions.

[0071] The present invention can greatly improve the effectiveness of railway ballast distribution operations under the premise of ensuring safety through the machine learning module, laser radar and ultrasonic radar. The operation is completely performed by machines and can continuously learn and improve itself. Through the setting of dual cameras, not only the automation of obstacle avoidance is solved, but also the shaping effect can be monitored through the second camera, and feedback can be given to optimize the shaping module and improve the accuracy of shaping decisions. It is suitable for railway maintenance work under various complex environments and conditions.

[0072] Specifically, the system has the following advantages:

[0073] 1. Update obstacle avoidance strategy in real time:

[0074] The system uses an online learning algorithm that can instantly process new data and update the model to adapt to environmental changes and new obstacle types. This real-time feedback mechanism enables the system to respond quickly to new challenges and improve the accuracy and efficiency of obstacle avoidance.

[0075] 2. Online Learning Algorithms:

[0076] Online learning algorithms such as online gradient descent and stochastic gradient descent allow the system to train a single sample and update the model in real time, which reduces the amount of computation and speeds up the training process, while enabling the model to quickly adapt to new data and improve the timeliness of the data.

[0077] 3. Model evaluation and update:

[0078] The system regularly evaluates model performance and updates the model when performance degrades, ensuring continuous optimization and accuracy of the model. This continuous model evaluation and update mechanism ensures the stability and reliability of the system in long-term operation.

[0079] 4. Feedback loop:

[0080] The system's feedback loop provides the ability to optimize system performance, improve decision-making efficiency and promote continuous improvement. By constantly adjusting and optimizing performance, the system can adapt to environmental changes and improve the safety and efficiency of operations.

[0081] 5. Model persistence:

[0082] Model persistence technology allows trained models to be saved for later reuse, including not only the model's parameters and weights, but also the model's configuration and structural information. This persistence capability makes it easier to save, migrate, and deploy models, and is a key step in realizing model service and productization.

[0083] 6. Data fusion technology:

[0084] Data fusion technology improves data accuracy and quality by integrating information from multiple data sources. It also provides more comprehensive analysis results, optimizes resource utilization, and improves data privacy and security. This technology enables the system to understand and respond to complex environmental conditions more accurately.

[0085] In summary, this system provides a more intelligent, automated and adaptable solution for obstacle avoidance and automatic shaping of the side plow of railway ballast vehicles through real-time updating of obstacle avoidance strategies, online learning, model evaluation and updating, feedback loop, model persistence and data fusion. The comprehensive application of these technologies has significantly improved the safety, efficiency and quality of railway maintenance operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 It is a schematic diagram of the overall arrangement structure of the present invention; DETAILED DESCRIPTION

[0087] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0088] Example:

[0089] An intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system, comprising:

[0090] At least one laser radar 1 and at least one ultrasonic radar 2 for detecting obstacles;

[0091] The first camera 3 is located at the front side of the ballast truck and is used to capture obstacle images;

[0092] The second camera 4 is located at the rear side of the ballast truck and is used to capture the image after ballast shaping;

[0093] a measurement and control system 5 for controlling the retraction and deployment of the side moldboard;

[0094] A machine learning module 6 for identifying and classifying obstacles and optimizing obstacle avoidance strategies; wherein obstacle identification uses a convolutional neural network formula Y = σ(W*X+b), where:

[0095] Y: output feature map of the convolutional layer;

[0096] W: weight of the convolution kernel;

[0097] X: input feature map;

[0098] b: bias term of the convolutional layer;

[0099] σ: activation function;

[0100] Classification using the support vector machine formula in:

[0101] f(x): prediction function given input x;

[0102] α i : Lagrange multiplier, determined by the optimization process;

[0103] y i : The label of the training sample, usually +1 or -1;

[0104] K(x,x i ): kernel function, calculates input x and support vector x i The similarity between

[0105] b: bias term;

[0106] sgn: symbolic function, returns -1 or 1, indicating the classification result;

[0107] The obstacle avoidance strategy is optimized using the Q-learning formula Q(s,a)←Q(s,a)+α[r+γmax a′ Q(s′,a′)-Q(s,a)], where:

[0108] Q(s,a): the expected utility of taking action a in state s;

[0109] α: learning rate, which controls the speed at which new information covers the original Q value;

[0110] r: the immediate reward obtained after taking action a in state s;

[0111] γ: discount factor, measuring the current value of future rewards;

[0112] s′: next state;

[0113] a′: the next possible action;

[0114] max a′ Q(s′,a′): the maximum Q value of all possible actions in the next state s′;

[0115] An automatic shaping module 7 is used to automatically adjust the working angle and depth of the side plowboard according to the location and type of obstacles; wherein the shaping strategy is generated using a genetic algorithm and the real-time adjustment is made using a PID controller formula in,

[0116] u(t): control signal at time t;

[0117] Kp: proportional gain;

[0118] Ki: integral gain;

[0119] Kd: differential gain;

[0120] e(t): error signal at time t, i.e. the difference between the expected value and the actual value;

[0121] ∫e(t)dt: integral of the error, used to eliminate steady-state error;

[0122] The rate of change of error, used to reduce overshoot and oscillation of the system;

[0123] Decision making accuracy improved using data fusion formula in,

[0124] Z: result after fusion;

[0125] w i : The weight of the i-th sensor data, reflecting the relative importance of the data in the fusion result;

[0126] Xi: Output data of the i-th sensor.

[0127] Furthermore, the machine learning module includes:

[0128] at least one convolutional neural network for extracting features from the camera image;

[0129] at least one classifier for classifying obstacles;

[0130] At least one reinforcement learning algorithm for learning an optimal obstacle avoidance path.

[0131] Furthermore, the automatic shaping module comprises:

[0132] at least one genetic algorithm for generating an optimal shaping strategy;

[0133] at least one PID controller for adjusting the position and angle of the side moldboard in real time;

[0134] At least one data fusion technique is used to improve the accuracy of the shaping decision.

[0135] Furthermore, it also includes:

[0136] At least one speed sensor, used to measure the running speed of the vehicle;

[0137] at least one attitude sensor for measuring the attitude of the vehicle;

[0138] At least one display screen, used to display data monitoring results;

[0139] At least one sound and light alarm for alarm prompts.

[0140] Furthermore, the measurement and control system can:

[0141] Automatically control the retraction and deployment of the side moldboard based on the output of the machine learning module;

[0142] The working angle and depth of the side moldboard are automatically adjusted according to the output of the automatic shaping module.

[0143] Furthermore, the system is capable of:

[0144] Remote monitoring and operation are achieved through wireless communication technology;

[0145] Implement online learning mechanisms that allow the system to continuously optimize its performance based on new data;

[0146] The model is updated regularly to adapt to environmental changes and new obstacle types.

[0147] A method for avoiding obstacles and automatically shaping the side plow of a railway ballast vehicle using the above system comprises the following steps:

[0148] Step 1: Detect obstacles using LiDAR and ultrasonic radar;

[0149] Step 2: Use the first camera to capture an image of an obstacle;

[0150] Step 3: Use machine learning module to identify and classify obstacles;

[0151] Step 4: Use the automatic shaping module to automatically adjust the working angle and depth of the side moldboard according to the type and location of the obstacle;

[0152] Step 5: Use the measurement and control system to control the retraction and deployment of the side moldboard;

[0153] Step 6: Adjust the side moldboard position and angle based on real-time feedback for precise shaping.

[0154] It also includes optimizing the plastic surgery module according to the image captured by the second camera to improve the accuracy of plastic surgery decisions.

[0155] When using this system, existing data can be used to set strategies, and laser radar and ultrasonic radar can be used to detect obstacles. The measurement and control system controls the left and right main side plowboards and the front and rear wing plowboards to avoid obstacles based on the detection data and strategy settings. Through the image of the first camera, the machine learning module is used to identify and classify obstacles, and the automatic shaping module is used to automatically adjust the working angle and depth of the side plowboards. In combination with the graphics of the second camera, the shaping module is optimized, and the position and angle of the side plowboards are adjusted through feedback from each sensor.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system, characterized in that include: at least one laser radar and at least one ultrasonic radar for detecting obstacles; The first camera is located at the front side of the ballast truck and is used to capture obstacle images; The second camera is located at the rear side of the ballast truck and is used to capture the image of the ballast after shaping; a measurement and control system for controlling the retraction and extension of the side moldboard; A machine learning module to identify and classify obstacles and optimize obstacle avoidance strategies; The obstacle recognition uses the convolutional neural network formula Y = σ (W * X + b), where: Y: output feature map of the convolutional layer; W: weight of the convolution kernel; X: input feature map; b: bias term of the convolutional layer; σ: activation function; Classification using the support vector machine formula in: f(x): prediction function given input x; α i : Lagrange multiplier, determined by the optimization process; y i : The label of the training sample, usually +1 or -1; K(x,x i ): kernel function, calculates input x and support vector x i The similarity between b: bias term; sgn: symbol function, returns -1 or 1, indicating the classification result; The obstacle avoidance strategy is optimized using the Q-learning formula Q(s,a)←Q(s,a)+α[r+γmax a′ Q(s′,a′)-Q(s,a)], where: Q(s,a): the expected utility of taking action a in state s; α: learning rate, which controls the speed at which new information covers the original Q value; r: the immediate reward obtained after taking action a in state s; γ: discount factor, measuring the current value of future rewards; s′: next state; a′: the next possible action; max a′ Q(s′,a′): the maximum Q value of all possible actions in the next state s′; An automatic shaping module is used to automatically adjust the working angle and depth of the side plowboard according to the location and type of obstacles; the shaping strategy is generated using a genetic algorithm and the real-time adjustment uses a PID controller formula in, u(t): control signal at time t; Kp: proportional gain; Ki: integral gain; Kd: differential gain; e(t): error signal at time t, i.e. the difference between the expected value and the actual value; ∫e(t)dt: integral of the error, used to eliminate steady-state error; The rate of change of error, used to reduce overshoot and oscillation of the system; Decision making accuracy improved using data fusion formula in, Z: result after fusion; w i : The weight of the i-th sensor data, reflecting the relative importance of the data in the fusion result; Xi: Output data of the i-th sensor.

2. The intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system as claimed in claim 1, characterized in that The machine learning module includes: at least one convolutional neural network for extracting features from the camera image; at least one classifier for classifying obstacles; At least one reinforcement learning algorithm for learning an optimal obstacle avoidance path.

3. The intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system as claimed in claim 1, characterized in that The automatic shaping module comprises: at least one genetic algorithm for generating an optimal shaping strategy; at least one PID controller for adjusting the position and angle of the side moldboard in real time; At least one data fusion technique is used to improve the accuracy of the shaping decision.

4. The intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system as claimed in claim 1, characterized in that Also includes: At least one speed sensor, used to measure the running speed of the vehicle; at least one attitude sensor for measuring the attitude of the vehicle; At least one display screen, used to display data monitoring results; At least one sound and light alarm for alarm prompts.

5. The intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system as claimed in claim 1, characterized in that The measurement and control system is capable of: Automatically control the retraction and deployment of the side moldboard based on the output of the machine learning module; The working angle and depth of the side moldboard are automatically adjusted according to the output of the automatic shaping module.

6. The intelligent railway ballast vehicle side plow obstacle avoidance and automatic shaping system as claimed in claim 1, characterized in that The system is capable of: Remote monitoring and operation are achieved through wireless communication technology; Implement online learning mechanisms that allow the system to continuously optimize its performance based on new data; The model is updated regularly to adapt to environmental changes and new obstacle types.

7. A method for avoiding obstacles and automatically shaping the side plow of a railway ballast vehicle using the system as claimed in any one of claims 1 to 6, characterized in that The following steps are involved: Step 1: Detect obstacles using LiDAR and ultrasonic radar; Step 2: Use the first camera to capture an image of an obstacle; Step 3: Use machine learning module to identify and classify obstacles; Step 4: Use the automatic shaping module to automatically adjust the working angle and depth of the side moldboard according to the type and location of the obstacle; Step 5: Use the measurement and control system to control the retraction and deployment of the side moldboard; Step 6: Adjust the side moldboard position and angle based on real-time feedback for precise shaping.

8. The method for avoiding obstacles and automatically shaping the side plow of a railway ballast vehicle according to claim 7, characterized in that: It also includes optimizing the plastic surgery module according to the image captured by the second camera to improve the accuracy of plastic surgery decisions.

Images