Method and system for setting decision of ordinary-speed railway annunciator
By integrating multi-source data and using deep learning decision models, the subjectivity and complexity of signal setting on conventional railways have been addressed. This has enabled scientific, accurate, and efficient decision-making regarding signal location and type, reducing safety risks and workload, and improving transportation efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-04-07
AI Technical Summary
In the existing technology, the setting of signal lights for conventional railways relies on human experience, which is highly subjective and makes it difficult to achieve an optimal balance of multiple factors. This results in safety risks and a large workload, and it is also difficult to optimize the position and type of signal lights in complex environments.
By employing multi-source data fusion and a deep learning decision model, and combining geographic information, signal history records, and on-site survey data, a deep reinforcement learning model and decision fusion layer are constructed to generate a scientific and accurate signal setting scheme. Through data preprocessing, model training and optimization, and combined with engineering constraints, the final decision is generated.
It enables scientific, accurate, and efficient signal setting decisions, reduces safety risks, improves transportation efficiency, reduces manual workload, and supports continuous learning and optimization.
Smart Images
Figure CN121808692A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of railway signaling technology, and more specifically to a method and system for setting up signaling devices on conventional railways. Background Technology
[0002] In conventional railway systems, the placement of signal lights is crucial for ensuring train operation safety and improving transportation efficiency. Signal lights are mainly classified into high-mast signals and low-mast signals based on their installation method. High-mast signals have a long line of sight but are more expensive and have a relatively greater impact on the surrounding environment; low-mast signals are cheaper and more flexible to install, but have a relatively shorter line of sight. Traditional signal placement methods mainly rely on design specifications, engineers' experience, and qualitative analysis of specific line conditions (such as curve radius, gradient, and surrounding obstacles). While this method is mature, it has a certain degree of subjectivity, and in actual operation, signal placement is easily affected by adverse weather and environmental conditions, leading to safety risks such as signal tilting and falling, affecting the normal operation of railway transportation. Existing methods struggle to comprehensively and quantitatively handle numerous complex and interrelated influencing factors, potentially resulting in suboptimal placement schemes. Furthermore, for long lines or complex station designs, manually determining the location and type of each signal is labor-intensive and time-consuming. Therefore, there is an urgent need for a more scientific, precise, and data-driven decision-making method to optimize the placement of signal lights on conventional railways.
[0003] Therefore, there is an urgent need to study a decision-making method and system for setting up signal lights on conventional railways. Summary of the Invention
[0004] In view of this, the present invention provides a method and system for decision-making on the setting of conventional railway signals, in order to solve the problems of reliance on human experience, subjective decision-making, low efficiency, and difficulty in achieving optimal balance of multiple factors in the prior art.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for setting up signal lights on conventional railways, comprising: Step 1: Collect and fuse multi-source data of railway lines; Step 2: Preprocess the fused data; Step 3: Build a deep learning decision model and train and optimize the model using the preprocessed data; Step 4: Construct a decision fusion layer to fuse the optimized deep learning decision model with the decision fusion layer; Step 5: After preprocessing the multi-source data to be designed or modified, input it into the trained deep learning decision model and fuse it with the decision fusion layer to generate a decision scheme.
[0006] Preferably, the multi-source data in step 1 includes: geographic information data, signal record data, field survey data, and data on the location, geological conditions, environmental conditions, and geometric parameters of the railway line.
[0007] Preferably, step 2 includes: verifying the collected multi-source data, identifying and processing outliers, default values, and duplicate values; filling and completing default values; processing numerical data using a normalization method; and processing categorical data using a one-hot encoding method. The coordinates of the multi-source data are aligned, and the preprocessed data is associated and aligned according to the route mileage information to ensure data dimension consistency and complete the dataset construction.
[0008] Preferably, the deep learning decision model is based on deep reinforcement learning theory and neural network architecture, and designs an input layer, an intermediate layer and an output layer as the basic framework. The intermediate layers use multilayer perceptrons, convolutional neural networks, or graph neural networks. The output layer uses the Softmax activation function and outputs a two-dimensional probability vector.
[0009] Preferably, the training and optimization process of the deep learning decision model includes: The preprocessed dataset is divided into a training set, a validation set, and a test set; The cross-entropy loss function is used as the optimization objective, and an adaptive learning rate optimizer is used for parameter updates. Early stopping is used during training to prevent overfitting. The model performance was evaluated using accuracy, precision, recall, F1 score, and AUC-ROC curves on the test set.
[0010] Preferably, the decision fusion layer in step 4 includes a rule base and a cost function. The rule base pre-sets mandatory rules, and the cost function comprehensively considers economic costs, construction and maintenance difficulty, and safety margin. The decision result is determined by the following formula: ; In the formula, α, β, γ, λ These are the weighting coefficients; P m Output for the model; F c This is an economic evaluation coefficient; F m For signal construction and maintenance coefficients; F s This is a safety performance evaluation coefficient; The output signal generator setting recommendations are calculated based on the preset rule base and cost function.
[0011] Preferably, step 5 includes the following steps: After preprocessing, the data of the railway line to be designed or modified is input into a trained deep learning decision model. The deep learning decision model generates preliminary probabilistic suggestions for each signal setting point. The preliminary suggestions are then integrated with engineering constraints to generate the final decision scheme. The signal setting scheme for the entire line is then output in a graphical manner through a human-computer interaction interface, and detailed decision-making basis for each decision point is provided.
[0012] Preferably, a conventional railway signal setting decision system includes: The data acquisition and integration module is used to collect and integrate multi-source data from railway lines. The data preprocessing module preprocesses the merged data; The model building module is used to build deep learning decision models and train and optimize them using preprocessed data. An engineering constraint knowledge base is used to store and manage business rules and constraints. The decision fusion center is used to fuse the output of deep learning decision models with engineering constraints to generate the final decision. A human-computer interaction and visualization platform is used to display decision results and receive user interactions; The system management and maintenance module provides system backend management functions.
[0013] Preferably, it also includes a continuous learning feedback mechanism, which uses the results of manual intervention and on-site acceptance data as new training samples to periodically trigger incremental training and optimization updates of the model.
[0014] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method and system for decision-making regarding the setting of signal lights on conventional railways. By collecting and fusing multi-source data related to signal setting, and after data preprocessing, data cleaning, and feature extraction, a deep learning-based decision model is constructed. Through model training and optimization, the decision result is output by fusing with multi-constraint information. This invention can provide scientific, accurate, and efficient decision support for the setting of signal lights on conventional railways, while ensuring the interpretability of the decision-making process and the flexibility of manual intervention. Attached Figure Description
[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0016] Figure 1The method flowchart provided by the present invention.
[0017] Figure 2 A flowchart illustrating the construction of the deep learning decision model provided by this invention.
[0018] Figure 3 The flowchart illustrates the training and optimization of a deep learning-based signal generator setting decision model provided by this invention.
[0019] Figure 4 The flowchart shows the multi-constraint information fusion decision-making method provided by the present invention.
[0020] Figure 5 The configuration decision system architecture diagram provided by this invention. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] like Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for setting up signal lights on conventional railways, including: Step 1: Collect and fuse multi-source data of railway lines; Step 2: Preprocess the fused data; Step 3: Build a deep learning decision model and train and optimize the model using the preprocessed data; Step 4: Construct a decision fusion layer to fuse the optimized deep learning decision model with the decision fusion layer; Step 5: After preprocessing the multi-source data to be designed or modified, input it into the trained deep learning decision model and fuse it with the decision fusion layer to generate a decision scheme.
[0023] Specifically, the multi-source data in step 1 includes: geographic information data, signal record data, field survey data, as well as the location, geological conditions, environmental conditions, and geometric parameters of the railway line.
[0024] In a specific embodiment of the present invention, the location, geological conditions, environmental conditions and geometric parameters of the railway line include whether the railway is located in a bridge, tunnel or other similar area, the distance from the railway construction clearance and catenary clearance, the geological conditions of the line, the environmental conditions of the line (whether it is located in a windy area, typhoon area, snowy environment, etc.), the geometric parameters of the line (radius of curvature, gradient, etc.), and the type labels of the signals that have been set.
[0025] Specifically, step 2 includes: verifying the collected multi-source data, identifying and processing outliers, default values, and duplicate values; filling and completing default values; processing numerical data using normalization methods; and processing categorical data using one-hot encoding methods. The coordinates of the multi-source data are aligned, and the preprocessed data is associated and aligned according to the route mileage information to ensure data dimension consistency and complete the dataset construction.
[0026] In a specific embodiment of the present invention, the numerical data includes: Line sight distance assessment characteristics: Based on the line curvature radius, surrounding obstruction relationships, and catenary support obstruction relationships, calculate the difference between the effective sight distance range and the specifications; Environmental load impact characteristics: Combining environmental characteristics such as strong winds and snow, a comprehensive risk index is constructed to characterize the impact of the environment on the structural stability of the high-column signal machine.
[0027] Clearance safety distance characteristics: calculate the minimum dynamic distance between the preset installation position of the signal and the building clearance and the overhead contact line clearance, and derive a Boolean characteristic to determine whether the safety margin is met.
[0028] Specifically, such as Figure 2 As shown, the deep learning decision model is based on deep reinforcement learning theory and neural network architecture, and designs an input layer, intermediate layer and output layer as the basic framework. The intermediate layers use multilayer perceptrons, convolutional neural networks, or graph neural networks. The output layer uses the Softmax activation function and outputs a two-dimensional probability vector.
[0029] In a specific embodiment of the present invention, the deep learning decision model construction method includes: Construct an end-to-end deep learning network whose input layer node count is consistent with the dimension of the feature vector after data preprocessing; The intermediate layers use multilayer perceptrons, convolutional neural networks, or graph neural networks. Multilayer perceptron: The network contains at least two hidden layers, uses activation functions such as ReLU to introduce non-linearity, and introduces a Dropout layer and a batch normalization layer after the hidden layers; Furthermore, convolutional neural networks treat the feature data of continuous line segments as a one-dimensional sequence and extract local spatial features and dependencies through one-dimensional convolutional layers. Graph Neural Networks: Construct a graph of the railway network topology (such as the connections between stations, signals, and sections), and use graph convolutional layers to aggregate neighborhood information in order to capture the mutual influence of signal setting decisions at the network level.
[0030] The intermediate layer uses a multilayer perceptron, and the network contains at least two hidden layers. Non-linearity is introduced using activation functions such as ReLU, and a Dropout layer and a batch normalization layer are introduced after the hidden layers.
[0031] The output layer uses the Softmax activation function and outputs a two-dimensional probability vector [P(tall signal), P(short signal)], which represents the recommended probability of setting tall signal and short signal, respectively.
[0032] Specifically, such as Figure 3 As shown, the training and optimization process of the deep learning decision model includes: The preprocessed dataset is divided into a training set, a validation set, and a test set; The cross-entropy loss function is used as the optimization objective, and an adaptive learning rate optimizer is used for parameter updates. Early stopping is used during training to prevent overfitting. The model performance was evaluated using accuracy, precision, recall, F1 score, and AUC-ROC curves on the test set.
[0033] In a specific embodiment of the present invention, the loss function is selected as the cross-entropy function, which is used to measure the difference between the model's predicted probability distribution and the true label distribution.
[0034] Furthermore, an adaptive learning rate optimizer is selected for updating model parameters.
[0035] The model is trained on the training set and validated and evaluated on the validation set.
[0036] Furthermore, the model is trained on the training set, and its performance is evaluated on the validation set after each training cycle.
[0037] Furthermore, when the validation set loss function no longer decreases over several consecutive periods, an early stopping mechanism is triggered to terminate training and restore the model parameters to the optimal performance on the validation set, in order to prevent overfitting.
[0038] The final trained model is fully evaluated on an independent test set.
[0039] Furthermore, the evaluation metrics include accuracy, precision, recall, F1 score, and AUC-ROC curve.
[0040] Specifically, the decision fusion layer in step 4 includes a rule base and a cost function. The rule base pre-sets mandatory rules, and the cost function comprehensively considers economic costs, construction and maintenance difficulty, and safety margin. The decision result is determined by the following formula: ; In the formula, α, β, γ, and λ are weighting coefficients, respectively; Pm is the model output; Fc is the economic evaluation coefficient; Fm is the signal construction and maintenance coefficient; and Fs is the safety performance evaluation coefficient. The output signal generator setting recommendations are calculated based on the preset rule base and cost function.
[0041] In a specific embodiment of the present invention, the rule base is pre-set with mandatory rules, including: if the installation location does not meet the building clearance or contact wire safety distance requirements, then a high-pole signal shall not be installed; if it is in a specific environment including a windy area with wind speeds exceeding the set wind speed, a typhoon area, a geological area that does not meet the requirements, or an area that is frequently affected by ice and snow, then a high-pole signal shall not be installed. Specifically, such as Figure 4 As shown, step 5 includes the following process: After preprocessing, the data of the railway line to be designed or modified is input into a trained deep learning decision model. The deep learning decision model generates preliminary probabilistic suggestions for each signal setting point. The preliminary suggestions are then integrated with engineering constraints to generate the final decision scheme. The signal setting scheme for the entire line is then output in a graphical manner through a human-computer interaction interface, and detailed decision-making basis for each decision point is provided.
[0042] like Figure 5 As shown, specifically, a conventional railway signal setting decision system includes: The data acquisition and integration module is used to collect and integrate multi-source data from railway lines. The data preprocessing module preprocesses the merged data; The model building module is used to build deep learning decision models and train and optimize them using preprocessed data. An engineering constraint knowledge base is used to store and manage business rules and constraints. The decision fusion center is used to fuse the output of deep learning decision models with engineering constraints to generate the final decision. A human-computer interaction and visualization platform is used to display decision results and receive user interactions; The system management and maintenance module provides system backend management functions.
[0043] Specifically, it also includes a continuous learning feedback mechanism, which uses the results of manual intervention and on-site acceptance data as new training samples to regularly trigger incremental training and optimization updates of the model.
[0044] Those skilled in the art should understand that, Figure 5 The computer software system shown is merely a block diagram of a portion of the structure related to the embodiments of the present invention, and does not constitute a limitation on the computer software system applied thereto in the embodiments of the present invention. A specific computer software system may contain more or fewer components or units than shown in the figure, or combine certain components or units, or have certain components or units with different deployment orientations.
[0045] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0046] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for setting up signal lights on conventional railways, characterized in that, include: Step 1: Collect and fuse multi-source data of railway lines; Step 2: Preprocess the fused data; Step 3: Build a deep learning decision model and train and optimize the model using the preprocessed data; Step 4: Construct a decision fusion layer to fuse the optimized deep learning decision model with the decision fusion layer; Step 5: After preprocessing the multi-source data to be designed or modified, input it into the trained deep learning decision model and fuse it with the decision fusion layer to generate a decision scheme.
2. The method for setting up signal lights on conventional railways according to claim 1, characterized in that, The multi-source data in step 1 includes: geographic information data, signal record data, field survey data, as well as data on the location, geological conditions, environmental conditions, and geometric parameters of the railway line.
3. The method for setting up signal lights on conventional railways according to claim 2, characterized in that, Step 2 includes: verifying the collected multi-source data, identifying and processing outliers, default values, and duplicate values; filling and completing default values; processing numerical data using normalization methods; and processing categorical data using one-hot encoding methods. The coordinates of the multi-source data are aligned, and the preprocessed data is associated and aligned according to the route mileage information to ensure data dimension consistency and complete the dataset construction.
4. The method for setting up signal lights on conventional railways according to claim 3, characterized in that, The deep learning decision model is based on deep reinforcement learning theory and neural networks, and designs an input layer, an intermediate layer and an output layer as the basic framework. The intermediate layers use multilayer perceptrons, convolutional neural networks, or graph neural networks. The output layer uses the Softmax activation function and outputs a two-dimensional probability vector.
5. The method for setting up signal lights on conventional railways according to claim 4, characterized in that, The training and optimization process of the deep learning decision model includes: The preprocessed dataset is divided into a training set, a validation set, and a test set; The cross-entropy loss function is used as the optimization objective, and an adaptive learning rate optimizer is used for parameter updates. Early stopping is used during training to prevent overfitting. The model performance was evaluated using accuracy, precision, recall, F1 score, and AUC-ROC curves on the test set.
6. The method for setting up signal lights on conventional railways according to claim 5, characterized in that, The decision fusion layer in step 4 includes a rule base and a cost function. The rule base pre-sets mandatory rules, and the cost function comprehensively considers economic costs, construction and maintenance difficulty, and safety margin. The decision result is determined by the following formula: ； In the formula, α, β, γ, λ These are the weighting coefficients; P m Output for the model; F c This is an economic evaluation coefficient; F m For signal construction and maintenance coefficients; F s This is a safety performance evaluation coefficient; The output signal generator setting recommendations are calculated based on the preset rule base and cost function.
7. The method for setting up signal lights on conventional railways according to claim 6, characterized in that, Step 5 includes the following process: After preprocessing, the data of the railway line to be designed or modified is input into a trained deep learning decision model. The deep learning decision model generates preliminary probabilistic suggestions for each signal setting point. The preliminary suggestions are then integrated with engineering constraints to generate the final decision scheme. The signal setting scheme for the entire line is then output in a graphical manner through a human-computer interaction interface, and detailed decision-making basis for each decision point is provided.
8. A decision-making system for setting up signal lights on conventional railways, used to implement the method described in any one of claims 1-7, characterized in that, include: The data acquisition and integration module is used to collect and integrate multi-source data from railway lines. The data preprocessing module preprocesses the merged data; The model building module is used to build deep learning decision models and train and optimize them using preprocessed data. An engineering constraint knowledge base is used to store and manage business rules and constraints. The decision fusion center is used to fuse the output of deep learning decision models with engineering constraints to generate the final decision. A human-computer interaction and visualization platform is used to display decision results and receive user interactions; The system management and maintenance module provides system backend management functions.
9. A decision-making system for setting up a conventional railway signal according to claim 8, characterized in that, It also includes a continuous learning feedback mechanism, which uses the results of manual intervention and on-site acceptance data as new training samples to trigger incremental training and optimization updates of the model on a regular basis.