A method and system for adjusting the spacing between soft cross-threads

By deploying monitoring image data acquisition devices and cloud data centers in the railway contact network system, using image recognition and reinforcement learning algorithms, automated detection and adjustment of soft cross-cross clues is realized, and the high cost and low cost caused by manual inspection is solved. Efficiency issues improve real-time and security.

CN118627847BActive Publication Date: 2024-11-05CREC RAILWAY ELECTRIFICATION RAILWAY OPERATIONS MANAGEMENT
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411080495.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-05
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

In the prior art, the adjustment of soft cross-cluster spacing requires manual inspection, resulting in high labor costs, large workload, poor real-time, low safety and low efficiency.

Method used

Deploy monitoring image data acquisition devices in the railway contact network system, connect to the cloud data center, use image recognition algorithm to build a cross-clue detection model and reinforcement learning algorithm to build a spacing adjustment model, and realize automated detection and adjustment.

Benefits of technology

It realizes automation, accurate and efficient detection and adjustment of soft cross-cross clues, reduces labor costs, improves real-time and security, and enhances the reliability and effect of spacing adjustment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118627847B_ABST
    Figure CN118627847B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of soft crossing of electrified railways, and discloses a method and system for adjusting the spacing of soft crossing crossing clues. The method comprises the following steps: deploying a monitoring image data acquisition device and connecting it to a cloud data center; constructing a soft crossing crossing clue detection model and a soft crossing crossing clue spacing adjustment model based on the cloud data center; collecting real-time soft crossing monitoring image data based on the monitoring image data acquisition device, and uploading it to the cloud data center; based on the cloud data center, performing soft crossing crossing clue detection according to the real-time soft crossing monitoring image data, and obtaining real-time soft crossing crossing clue detection results; adjusting the spacing of soft crossing crossing clues according to the real-time soft crossing crossing clue detection results, and obtaining a real-time soft crossing crossing clue spacing adjustment strategy. The present invention solves the problems of large labor cost investment, large workload, poor real-time performance, low safety, low efficiency and poor effect in the prior art.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of soft spanning of electrified railways, and in particular relates to a method and system for adjusting the spacing of soft spanning crossing lines. Background Art

[0002] At present, in the design structure of the electrified railway contact network engineering, the plane layout of the two (or three) cables of the contact network, such as the line switch, anchor section joint, and cross-track anchor, all have the situation of cable crossing (no crossing line switch is not an example). In the soft cross-over (soft cable hard beam) structure, the cables in some sections are affected by the combined effects of curves, cable angles, and cable tension. The suspension pulley is raised by the horizontal force, resulting in insufficient three-dimensional cross height difference of the load-bearing cables at the plane intersection position, and there is a phenomenon of mutual extrusion and friction, leaving hidden dangers in operation. Therefore, timely processing of soft cross-over cross cables and eliminating hidden dangers in operation are important measures to ensure the stable operation of the railway contact network system.

[0003] In the prior art, manual inspections need to be arranged to detect soft crossing clues and adjust the spacing. This method requires high manpower costs and a large workload, and is unable to detect the positions of soft crossing clue points in a timely manner. Furthermore, adjusting the spacing of soft crossing clues requires relying on the experience of staff, which has low safety. Moreover, when the contact network system is more complex, the efficiency of spacing adjustment is low and the effect is poor. Summary of the invention

[0004] In order to solve the problems of high manpower cost, heavy workload, poor real-time performance, low security, low efficiency and poor effect in the prior art, the present invention aims to provide a method and system for adjusting the spacing of soft cross-clues.

[0005] The technical solution adopted by the present invention is:

[0006] A method for adjusting the spacing of soft cross-clues comprises the following steps:

[0007] Deploy monitoring image data acquisition devices at each soft span of the railway overhead line system, and connect all monitoring image data acquisition devices to the cloud data center;

[0008] Based on the cloud data center, a soft crossing clue detection model and a soft crossing clue spacing adjustment model are constructed according to a number of pre-collected historical soft crossing monitoring image data;

[0009] Based on the monitoring image data acquisition device, real-time soft-span monitoring image data is collected, and the real-time soft-span monitoring image data is uploaded to the cloud data center;

[0010] Based on the cloud data center, according to the real-time soft crossing monitoring image data, the soft crossing clue detection model is used to perform soft crossing clue detection, and the real-time soft crossing clue detection result is obtained;

[0011] According to the real-time soft crossover clue detection result, the soft crossover clue spacing adjustment model is used to adjust the soft crossover clue spacing, and the real-time soft crossover clue spacing adjustment strategy is obtained.

[0012] Furthermore, after the monitoring image data acquisition device is deployed, it also includes:

[0013] Based on a trusted organization, an asymmetric encryption algorithm is used to generate keys for all monitoring image data acquisition devices connected to the cloud data center to obtain a public-private key pair for each monitoring image data acquisition device; the public-private key pair includes a public key and a private key.

[0014] Furthermore, based on the cloud data center, according to a number of pre-collected historical soft crossing monitoring image data, a soft crossing intersection clue detection model and a soft crossing intersection clue spacing adjustment model are constructed, including the following steps:

[0015] Based on the cloud data center, image preprocessing is performed on a number of pre-collected historical soft span monitoring image data to obtain a number of pre-processed historical soft span monitoring image data;

[0016] Based on a number of pre-processed historical soft crossing monitoring image data, an image recognition algorithm is used to construct a soft crossing clue detection model, and a number of historical soft crossing clue detection results are generated;

[0017] According to several historical soft-spanning-crossing clue detection results and several randomly generated historical soft-spanning-crossing clue spacing adjustment strategies, a soft-spanning-crossing clue spacing adjustment model is constructed using a reinforcement learning algorithm.

[0018] Furthermore, the soft spanning intersection clue detection model is constructed based on the CNN-SVM-GNN-RF algorithm, and the soft spanning intersection clue detection model includes an image data feature extraction module constructed based on the CNN algorithm, a soft spanning component recognition module constructed based on the SVM algorithm, a soft spanning structure connection graph generation module, a structure connection graph feature extraction module constructed based on the GNN algorithm, and a soft spanning intersection clue recognition module constructed based on the RF algorithm.

[0019] Furthermore, a soft cross-cue spacing adjustment model is constructed based on the DQN algorithm.

[0020] Furthermore, based on the monitoring image data acquisition device, real-time soft crossing monitoring image data is collected, and the real-time soft crossing monitoring image data is uploaded to the cloud data center, including the following steps:

[0021] Based on the monitoring image data acquisition device, real-time soft-span monitoring image data is collected;

[0022] According to the private key, the real-time soft cross-over monitoring image data is encrypted using an asymmetric encryption algorithm to obtain encrypted real-time soft cross-over monitoring image data;

[0023] Upload the encrypted real-time soft cross-monitoring image data to the cloud data center.

[0024] Furthermore, before performing the soft crossover clue detection, the method further includes:

[0025] Based on the cloud data center, the encrypted real-time soft cross-over monitoring image data is decrypted according to the public key using an asymmetric decryption algorithm to obtain the decrypted real-time soft cross-over monitoring image data.

[0026] Furthermore, based on the cloud data center, according to the real-time soft crossing monitoring image data, the soft crossing clue detection model is used to perform soft crossing clue detection to obtain the real-time soft crossing clue detection result, including the following steps:

[0027] Based on the cloud data center, the real-time soft crossing monitoring image data is input into the soft crossing clue detection model;

[0028] According to the real-time soft-span monitoring image data, an image data feature extraction module is used to extract image data features to obtain real-time image data features;

[0029] According to the real-time image data features, a soft span component recognition module is used to recognize the soft span component, and a real-time soft span component recognition result including soft span parts and soft span clues is obtained;

[0030] According to the real-time soft span component identification results, a soft span structure connection graph generation module is used to take the soft span components as real-time nodes and the soft span clues as real-time edges to obtain a real-time soft span structure connection graph;

[0031] According to the real-time soft spanning structure connection diagram, a structure connection diagram feature extraction module is used to extract the structure connection diagram features to obtain the real-time soft spanning structure connection diagram features;

[0032] According to the real-time soft-spanning structure connection graph features, a soft-spanning cross clue recognition module is used to perform soft-spanning cross clue recognition to obtain real-time soft-spanning cross clue detection results.

[0033] Further, according to the real-time soft crossing clue detection result, the soft crossing clue spacing adjustment model is used to adjust the soft crossing clue spacing, and the real-time soft crossing clue spacing adjustment strategy is obtained, which includes the following steps:

[0034] Inputting the randomly generated real-time soft cross-over cue spacing adjustment strategy and the real-time soft cross-over cue detection result into the soft cross-over cue spacing adjustment model;

[0035] According to the randomly generated real-time soft cross-clue spacing adjustment strategy, the action space of the soft cross-clue spacing adjustment model is updated to obtain an updated action space;

[0036] According to the real-time soft cross-over clue detection result, the state space of the soft cross-over clue spacing adjustment model is updated to obtain an updated state space;

[0037] Use the updated state space as input to the soft cross-cue spacing adjustment model and obtain the predicted Q-values ​​of possible actions in the updated action space;

[0038] Use the reward function to obtain the reward value of the possible action, and update the predicted Q value of the possible action according to the reward value to obtain the updated Q value of the possible action;

[0039] Repeat the above steps until the iteration threshold is reached, and use the greedy strategy to take the possible action corresponding to the highest updated Q value as the execution action;

[0040] The execution actions are output as real-time soft cross-cue spacing adjustment strategies.

[0041] A soft cross-over clue spacing adjustment system is used to implement a soft cross-over clue spacing adjustment method. The system includes a cloud data center and a plurality of monitoring image data acquisition devices. The cloud data center is respectively connected to the plurality of monitoring image data acquisition devices for communication;

[0042] The cloud data center is provided with a model building unit, a soft-spanning cross clue detection unit and a soft-spanning cross clue spacing adjustment unit which are connected in sequence.

[0043] The beneficial effects of the present invention are:

[0044] The invention discloses a method and system for adjusting the spacing of soft spanning intersection clues. A soft spanning intersection clue detection model constructed by an image recognition algorithm can automatically, accurately and efficiently perform soft spanning intersection clue detection, and find soft spanning intersection clue points at the first time, thereby reducing human cost investment, alleviating workload and improving real-time performance. A soft spanning intersection clue spacing adjustment model constructed by a reinforcement learning algorithm can intelligently and accurately generate a soft spanning intersection clue spacing adjustment strategy according to actual conditions such as the position, quantity and soft spanning structure connection information of the soft spanning intersection clue points, thereby avoiding the defect of relying on manual judgment by staff, improving the reliability and accuracy of the soft spanning intersection clue spacing adjustment strategy, and improving the safety of staff and the effect after spacing adjustment.

[0045] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of the method for adjusting the spacing of soft cross-clues in the present invention.

[0047] Figure 2 It is a structural block diagram of the soft cross-clue spacing adjustment system in the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments.

[0049] Embodiment 1:

[0050] like Figure 1 As shown, this embodiment provides a method for adjusting the spacing of soft cross-clues, including the following steps:

[0051] S1: Deploy monitoring image data acquisition devices at each soft span of the railway overhead line system, and connect all monitoring image data acquisition devices to the cloud data center;

[0052] S1.1: Based on a trusted institution, using an asymmetric encryption algorithm, keys are generated for all monitoring image data acquisition devices connected to the cloud data center to obtain a public-private key pair for each monitoring image data acquisition device, including the following steps:

[0053] S1.1-1: Based on a trusted institution, collect attribute information of all monitoring image data collection devices connected to the cloud data center ;

[0054] S1.1-2: Based on attribute information , generate a key for the monitoring image data acquisition device to obtain a public-private key pair for each monitoring image data acquisition device; the public-private key pair includes a public key and a private key;

[0055] The formula is:

[0056]

[0057] In the formula, is a public parameter; is the master key; is the initial key; Integer domain A random number; Both are target hash functions; Both are cyclic groups The random number of the generator; is a random number Bilinear mapping of ;

[0058]

[0059] In the formula, For monitoring image data acquisition device The private key of Integer domain A random number; For monitoring image data acquisition device The private key parameters; For public parameters The target hash hash function; To monitor the indication quantity of the image data acquisition device; is the master key; is the initial key; For monitoring image data acquisition device 's public key; Cyclic group The random number of the generator; For monitoring image data acquisition device Attribute information of

[0060] S1.1-3: Send the private key in the public-private key pair to the corresponding monitoring image data acquisition device, and publish the public key in the public-private key pair to the cloud data center;

[0061] Since the overhead line system or railway system is an important department and system, involving a large amount of important data, in order to avoid the defect that the monitoring image data is transmitted in plain text during the data transmission process, an asymmetric encryption algorithm is used to encrypt the monitoring image data, thereby improving data security and avoiding data leakage;

[0062] S2: Based on the cloud data center, a soft crossing clue detection model and a soft crossing clue spacing adjustment model are constructed according to a number of pre-collected historical soft crossing monitoring image data, including the following steps:

[0063] S2-1: Based on the cloud data center, image preprocessing is performed on a number of pre-collected historical soft span monitoring image data to obtain a number of pre-processed historical soft span monitoring image data;

[0064] Image preprocessing includes Gaussian denoising, grayscale processing, and size normalization to enhance image data quality and improve the accuracy of model prediction;

[0065] S2-2: Based on a number of pre-processed historical soft crossing monitoring image data, use an image recognition algorithm to build a soft crossing clue detection model and generate a number of historical soft crossing clue detection results;

[0066] The soft crossover clue detection model is constructed based on Convolutional Neural Networks (CNN)-Support Vector Machine (SVM)-Graph Neural Networks (GNN)-Random Forest (RF) algorithm, and the soft crossover clue detection model includes an image data feature extraction module constructed based on the CNN algorithm, a soft crossover component recognition module constructed based on the SVM algorithm, a soft crossover structure connection graph generation module, a structure connection graph feature extraction module constructed based on the GNN algorithm, and a soft crossover clue recognition module constructed based on the RF algorithm;

[0067] The CNN network extracts deep features of image data through the convolution layer structure. The SVM performs classification and recognition based on the features of the image data to obtain the recognition results of soft spanning components and soft spanning clues. Among them, the recognition results of soft spanning components include overhead wire posts, insulator strings, grounding devices and other components. The recognition results of soft spanning clues include horizontal bearing clues, upper fixing clues, lower fixing clues, vertical hanging clues and other clues. The soft spanning structure connection diagram generation module uses soft spanning components as nodes and soft spanning clues as connection edges to construct a structural connection diagram of the soft spanning structure, which can characterize the connection relationship and connection status of the soft spanning and more accurately characterize information such as cross clues and cross positions. The GNN network can extract node features, edge features and graph features of the structural connection diagram, and more accurately extract feature information of cross clues from the structural connection relationship. The RF structure uses the internal Classification And Regression Tree (CART) to screen the key features of the input graph features, extract key features related to cross clue prediction, and perform soft spanning cross clue recognition, thereby improving the recognition accuracy and efficiency.

[0068] S2-3: Based on several historical soft cross-over clue detection results and several randomly generated historical soft cross-over clue spacing adjustment strategies, a deep Q network (DQN) algorithm is used to construct a soft cross-over clue spacing adjustment model, including the following steps:

[0069] S2-3-1: Based on the problem of adjusting the soft cross-cue spacing, define the simulation environment of the Deep Q Network (DQN) algorithm, and build the Deep Q Network, agent, and experience storage pool of the DQN algorithm;

[0070] S2-3-2: Define the action space of the DQN algorithm based on the randomly generated historical soft cross-cue spacing adjustment strategy, and define the state space of the DQN algorithm based on the historical soft cross-cue detection results;

[0071] S2-3-3: Define the reward function of the DQN algorithm according to the action space and state space, and build the initial soft cross-cue spacing adjustment model based on the simulation environment, state space, action space, reward function, deep Q network, intelligent agent and experience storage pool;

[0072] S2-3-4: Optimizing and training the initial soft cross-cross clue spacing adjustment model according to several historical soft cross-cross clue detection results and several historical soft cross-cross clue spacing adjustment strategies, and storing the generated soft cross-cross clue spacing adjustment experience into the experience storage pool to obtain the final soft cross-cross clue spacing adjustment model;

[0073] S3: Based on the monitoring image data acquisition device, real-time soft cross-over monitoring image data is collected, and the real-time soft cross-over monitoring image data is uploaded to the cloud data center, including the following steps:

[0074] S3-1: Based on the monitoring image data acquisition device, collect real-time soft-span monitoring image data;

[0075] S3-2: Encrypt the real-time soft cross-traffic monitoring image data using an asymmetric encryption algorithm according to the private key to obtain encrypted real-time soft cross-traffic monitoring image data;

[0076] The formula is:

[0077]

[0078] In the formula, For monitoring image data acquisition device Encrypted real-time soft-span monitoring image data; is an asymmetric encryption function; For monitoring image data acquisition device Real-time soft-span monitoring image data; For monitoring image data acquisition device The private key of To monitor the indication quantity of the image data acquisition device;

[0079] S3-3: Upload the encrypted real-time soft cross-monitoring image data to the cloud data center;

[0080] S3.0: Based on the cloud data center, the encrypted real-time soft cross-over monitoring image data is decrypted using an asymmetric decryption algorithm according to the public key to obtain the decrypted real-time soft cross-over monitoring image data;

[0081] The formula is:

[0082]

[0083] In the formula, For monitoring image data acquisition device Decrypted real-time soft-span monitoring image data; is an asymmetric decryption function; For monitoring image data acquisition device Encrypted real-time soft-span monitoring image data; For monitoring image data acquisition device 's public key;

[0084] S4: Based on the cloud data center, according to the decrypted real-time soft crossing monitoring image data, using the soft crossing clue detection model, soft crossing clue detection is performed to obtain the real-time soft crossing clue detection result, including the following steps:

[0085] S4-1: Based on the cloud data center, the real-time soft crossing monitoring image data is input into the soft crossing clue detection model;

[0086] S4-2: according to the real-time soft-span monitoring image data, use the image data feature extraction module to extract image data features to obtain real-time image data features;

[0087] S4-3: using a soft span component identification module to identify the soft span component according to the real-time image data features, and obtaining a real-time soft span component identification result including soft span parts and soft span clues;

[0088] S4-4: Based on the real-time soft span component identification result, a soft span structure connection graph generation module is used to take the soft span components as real-time nodes and the soft span clues as real-time edges to obtain a real-time soft span structure connection graph;

[0089] S4-5: according to the real-time soft spanning structure connection diagram, using the structure connection diagram feature extraction module, extracting the structure connection diagram features to obtain the real-time soft spanning structure connection diagram features;

[0090] S4-6: According to the real-time soft-spanning structure connection graph feature, a soft-spanning cross clue identification module is used to perform soft-spanning cross clue identification to obtain a real-time soft-spanning cross clue detection result, including the following steps:

[0091] S4-6-1: input the real-time soft-spanning structure connection graph features into the soft-spanning cross clue recognition module;

[0092] S4-6-2: Use the RF structure trained in the soft cross-clue recognition module to extract the feature contribution of several real-time candidate features in the real-time soft cross-structure connection graph features;

[0093] The formula is:

[0094]

[0095] In the formula, For the Feature contribution of real-time candidate features; For the Real-time candidate features in random forest The feature contribution of each tree; is the CART tree indicator; It is a real-time alternative characteristic indicator; is the total number of CARTs;

[0096]

[0097] In the formula, CART tree node for random forest m ,node and nodes r The Gini index of CART tree node m Medium Category The proportion of is the total number of categories; m , , r is the node indicator; is the category indicator;

[0098] S4-6-3: normalizing the feature contributions of several real-time candidate features to obtain corresponding normalized feature contributions;

[0099] The formula is:

[0100]

[0101] In the formula, It is the contribution of the feature after normalization; J is the total number of real-time candidate features;

[0102] S4-6-4: Generate feature selection standard values ​​for several real-time candidate features based on the normalized feature contribution;

[0103] The formula is:

[0104]

[0105] In the formula, For the Feature selection criteria values ​​for real-time candidate features; For the The normalized feature contribution of real-time candidate features; is an alternative characteristic indicator;

[0106] S4-6-5: sort the real-time candidate features in descending order according to the feature selection standard value, and select the first M real-time candidate features as real-time key features to obtain M real-time key features, where M is the total number of real-time key features;

[0107] S4-6-6: According to the M real-time key features, soft crossover clues are identified to obtain real-time soft crossover clues detection results;

[0108] The real-time soft crossover clue detection results include the location of the soft crossover clue point, the soft crossover components on both sides of the soft crossover clue, etc., which can provide the staff with the accurate positioning and accurate connection of the soft crossover clue point;

[0109] S5: According to the real-time soft crossover clue detection result, the soft crossover clue spacing adjustment model is used to adjust the soft crossover clue spacing to obtain the real-time soft crossover clue spacing adjustment strategy, including the following steps:

[0110] S5-1: inputting the randomly generated real-time soft cross-over clue spacing adjustment strategy and the real-time soft cross-over clue detection result into the soft cross-over clue spacing adjustment model;

[0111] S5-2: Update the action space of the soft cross-cue spacing adjustment model according to the randomly generated real-time soft cross-cue spacing adjustment strategy to obtain an updated action space ,in, For the updated Action value, is the action indicator;

[0112] S5-3: Update the state space of the soft cross-clue spacing adjustment model according to the real-time soft cross-clue detection result to obtain an updated state space ,in, For the updated Status value, is the state indicator;

[0113] S5-4: Use the updated state space as input to the soft cross-cue spacing adjustment model and obtain the predicted Q-values ​​of possible actions in the updated action space;

[0114] S5-5: Use the reward function to obtain the reward value of the possible action, and update the predicted Q value of the possible action according to the reward value to obtain the updated Q value of the possible action;

[0115] The formula is:

[0116]

[0117] In the formula, The updated status value and the updated action value The corresponding updated Q value; Status value and action value The corresponding predicted Q value; is the learning rate; is the highest predicted Q value; is the reward value of the possible action;

[0118] S5-6: Repeat the above steps until the iteration number threshold is reached, and use the greedy strategy to take the possible action corresponding to the highest updated Q value as the execution action;

[0119] S5-7: outputting the execution action as a real-time soft cross-cue spacing adjustment strategy;

[0120] The real-time soft cross-clue spacing adjustment strategy includes raising the position height of the horizontal load-bearing clues, lowering the position height of the vertical hanging clues, or lowering the position height of the lower fixed clues, etc., ultimately achieving the staggered position of the load-bearing clues, ensuring the three-dimensional spatial distance at the intersection of the clue planes, and eliminating the hidden dangers of clues rubbing against each other.

[0121] Embodiment 2:

[0122] like Figure 2 As shown, this embodiment provides a soft cross-cross clue spacing adjustment system for implementing a soft cross-cross clue spacing adjustment method, the system includes a cloud data center and a plurality of monitoring image data acquisition devices, the cloud data center is respectively connected to the plurality of monitoring image data acquisition devices for communication;

[0123] A monitoring image data acquisition device is used to collect real-time soft cross-over monitoring image data and upload the real-time soft cross-over monitoring image data to a cloud data center;

[0124] The cloud data center is provided with a model building unit, a soft cross-over clue detection unit and a soft cross-over clue spacing adjustment unit connected in sequence;

[0125] A model building unit, used for building a soft-spanning crossing clue detection model and a soft-spanning crossing clue spacing adjustment model according to a plurality of pre-collected historical soft-spanning monitoring image data;

[0126] A soft crossing clue detection unit is used to perform soft crossing clue detection according to real-time soft crossing monitoring image data using a soft crossing clue detection model to obtain a real-time soft crossing clue detection result;

[0127] The soft crossing clue spacing adjustment unit is used to adjust the soft crossing clue spacing according to the real-time soft crossing clue detection result and the soft crossing clue spacing adjustment model to obtain the real-time soft crossing clue spacing adjustment strategy.

[0128] The invention discloses a method and system for adjusting the spacing of soft spanning intersection clues. A soft spanning intersection clue detection model constructed by an image recognition algorithm can automatically, accurately and efficiently perform soft spanning intersection clue detection, and find soft spanning intersection clue points at the first time, thereby reducing human cost investment, alleviating workload and improving real-time performance. A soft spanning intersection clue spacing adjustment model constructed by a reinforcement learning algorithm can intelligently and accurately generate a soft spanning intersection clue spacing adjustment strategy according to actual conditions such as the position, quantity and soft spanning structure connection information of the soft spanning intersection clue points, thereby avoiding the defect of relying on manual judgment by staff, improving the reliability and accuracy of the soft spanning intersection clue spacing adjustment strategy, and improving the safety of staff and the effect after spacing adjustment.

[0129] The present invention is not limited to the above optional implementations, and anyone can derive other various forms of products under the enlightenment of the present invention. The above specific implementations should not be understood as limiting the scope of protection of the present invention. The scope of protection of the present invention should be based on the definition in the claims, and the description can be used to interpret the claims.

Claims

1. A method for adjusting the spacing of soft cross-clues, characterized in that: The steps include: Deploy monitoring image data acquisition devices at each soft span of the railway overhead line system, and connect all monitoring image data acquisition devices to the cloud data center; Based on the cloud data center, a soft crossing clue detection model and a soft crossing clue spacing adjustment model are constructed according to a number of pre-collected historical soft crossing monitoring image data, including the following steps: Based on the cloud data center, image preprocessing is performed on a number of pre-collected historical soft span monitoring image data to obtain a number of pre-processed historical soft span monitoring image data; Based on a number of pre-processed historical soft crossing monitoring image data, an image recognition algorithm is used to construct a soft crossing clue detection model, and a number of historical soft crossing clue detection results are generated; According to several historical soft crossover clue detection results and several randomly generated historical soft crossover clue spacing adjustment strategies, a soft crossover clue spacing adjustment model is constructed using a reinforcement learning algorithm; Based on the monitoring image data acquisition device, real-time soft-span monitoring image data is collected, and the real-time soft-span monitoring image data is uploaded to the cloud data center; Based on the cloud data center, according to the real-time soft crossing monitoring image data, the soft crossing clue detection model is used to perform soft crossing clue detection, and the real-time soft crossing clue detection result is obtained; According to the real-time soft crossover clue detection result, the soft crossover clue spacing adjustment model is used to adjust the soft crossover clue spacing, and the real-time soft crossover clue spacing adjustment strategy is obtained.

2. The method for adjusting the spacing of soft cross-clues according to claim 1, characterized in that: After the monitoring image data acquisition device is deployed, it also includes: Based on a trusted organization, an asymmetric encryption algorithm is used to generate keys for all monitoring image data acquisition devices connected to the cloud data center to obtain a public-private key pair for each monitoring image data acquisition device; the public-private key pair includes a public key and a private key.

3. The method for adjusting the spacing of soft cross-clues according to claim 1, characterized in that: The soft-spanning intersection clue detection model is constructed based on the CNN-SVM-GNN-RF algorithm, and the soft-spanning intersection clue detection model includes an image data feature extraction module constructed based on the CNN algorithm, a soft-spanning component recognition module constructed based on the SVM algorithm, a soft-spanning structure connection graph generation module, a structure connection graph feature extraction module constructed based on the GNN algorithm, and a soft-spanning intersection clue recognition module constructed based on the RF algorithm.

4. The method for adjusting the spacing of soft cross-clues according to claim 1, characterized in that: The soft cross-cue spacing adjustment model is constructed based on the DQN algorithm.

5. The method for adjusting the spacing of soft cross-clues according to claim 2, characterized in that: Based on the monitoring image data acquisition device, real-time soft cross-over monitoring image data is collected, and the real-time soft cross-over monitoring image data is uploaded to the cloud data center, including the following steps: Based on the monitoring image data acquisition device, real-time soft-span monitoring image data is collected; According to the private key, the real-time soft cross-over monitoring image data is encrypted using an asymmetric encryption algorithm to obtain encrypted real-time soft cross-over monitoring image data; Upload the encrypted real-time soft cross-monitoring image data to the cloud data center.

6. The method for adjusting the spacing of soft cross-clues according to claim 5, characterized in that: Before soft cross-cue detection, it also includes: Based on the cloud data center, the encrypted real-time soft cross-over monitoring image data is decrypted according to the public key using an asymmetric decryption algorithm to obtain the decrypted real-time soft cross-over monitoring image data.

7. The method for adjusting the spacing of soft cross-clues according to claim 3, characterized in that: Based on the cloud data center, according to the real-time soft crossing monitoring image data, the soft crossing clue detection model is used to perform soft crossing clue detection, and the real-time soft crossing clue detection result is obtained, including the following steps: Based on the cloud data center, the real-time soft crossing monitoring image data is input into the soft crossing clue detection model; According to the real-time soft-span monitoring image data, an image data feature extraction module is used to extract image data features to obtain real-time image data features; According to the real-time image data features, a soft span component recognition module is used to recognize the soft span component, and a real-time soft span component recognition result including soft span parts and soft span clues is obtained; According to the real-time soft span component identification results, a soft span structure connection graph generation module is used to take the soft span components as real-time nodes and the soft span clues as real-time edges to obtain a real-time soft span structure connection graph; According to the real-time soft spanning structure connection diagram, a structure connection diagram feature extraction module is used to extract the structure connection diagram features to obtain the real-time soft spanning structure connection diagram features; According to the real-time soft-spanning structure connection graph features, a soft-spanning cross clue recognition module is used to perform soft-spanning cross clue recognition to obtain real-time soft-spanning cross clue detection results.

8. The method for adjusting the spacing of soft cross-clues according to claim 4, characterized in that: According to the real-time soft crossover clue detection result, the soft crossover clue spacing adjustment model is used to adjust the soft crossover clue spacing, and the real-time soft crossover clue spacing adjustment strategy is obtained, including the following steps: Inputting the randomly generated real-time soft cross-over cue spacing adjustment strategy and the real-time soft cross-over cue detection result into the soft cross-over cue spacing adjustment model; According to the randomly generated real-time soft cross-clue spacing adjustment strategy, the action space of the soft cross-clue spacing adjustment model is updated to obtain an updated action space; According to the real-time soft cross-over clue detection result, the state space of the soft cross-over clue spacing adjustment model is updated to obtain an updated state space; Use the updated state space as input to the soft cross-cue spacing adjustment model and obtain the predicted Q-values ​​of possible actions in the updated action space; Use the reward function to obtain the reward value of the possible action, and update the predicted Q value of the possible action according to the reward value to obtain the updated Q value of the possible action; Repeat the above steps until the iteration threshold is reached, and use the greedy strategy to take the possible action corresponding to the highest updated Q value as the execution action; The execution actions are output as real-time soft cross-cue spacing adjustment strategies.

9. A soft cross-thread spacing adjustment system, used to implement the soft cross-thread spacing adjustment method according to any one of claims 1 to 8, characterized in that: The system includes a cloud data center and a plurality of monitoring image data acquisition devices, wherein the cloud data center is respectively connected to the plurality of monitoring image data acquisition devices for communication; The cloud data center is provided with a model building unit, a soft-spanning cross clue detection unit and a soft-spanning cross clue spacing adjustment unit which are connected in sequence.