Cable arrangement method and system in power supply and distribution system based on reinforcement learning, and storage medium

Abstract

The invention discloses a reinforcement learning-based cable arrangement method and system in a power supply and distribution system, and a storage medium. The method comprises the following steps of S1, obtaining cable information, laying environment information and constraints required to be met by cable laying; s2, abstracting a bridge system into a weighted undirected topological graph; s3, calling a path planning algorithm based on reinforcement learning, and calculating an optimal cable laying path by taking a minimum comprehensive cost function including path length cost, plot ratio cost and precious penalty terms as an optimization target under the constraint that cable laying needs to be met; s4, calculating the center coordinate of each cable to be arranged in each bridge by adopting a greedy algorithm according to the optimal cable laying path, and arranging the cables according to the center coordinate; according to the method, the optimal laying and arrangement path can be planned on the premise of meeting various physical constraints of cable laying, and the method can automatically and dynamically adapt to different laying scenes.

Description

Technical Field

[0001] This invention relates to cable routing in power supply and distribution systems, and more particularly to a cable routing method, system, and storage medium in power supply and distribution systems based on reinforcement learning. Background Technology

[0002] Currently, in the field of building electrical design and construction, the layout design of power supply and distribution system cables mainly relies on manual operations by designers in Building Information Modeling (BIM) software. This method requires designers to drag cables one by one from the starting equipment to the ending equipment in a complex three-dimensional space and manually specify the cable tray paths they pass through. This process is not only inefficient and labor-intensive, but also, due to the complexity of three-dimensional space, it is difficult for designers to optimize from a global perspective. The final cable paths often only achieve the function of "connectivity" rather than the "optimal" path that balances economy and safety. This directly leads to an unnecessary increase in cable material usage, increases project costs, and may create hidden dangers for subsequent operation and maintenance.

[0003] To address the inefficiency of manual methods, algorithm-based automated cable laying solutions have emerged in existing technologies. For example, Chinese invention patent CN116720294A discloses an automated cable laying method based on BIM technology and a local dynamic domain search algorithm. This method establishes a cable channel routing network by reading power grid identification information and uses a Floyd algorithm based on dynamic programming for automatic path finding to achieve optimal cable layout. However, this existing technology still has significant limitations: First, its optimization objective is relatively singular, mainly pursuing the "shortest path," without dynamically and proactively considering the key engineering constraint of "cable tray volume ratio" during the path optimization process. This can easily lead to excessive concentration of planned cables in certain cable tray sections, causing local congestion, violating electrical design specifications, and creating difficulties for system heat dissipation and future capacity expansion. Second, traditional graph search algorithms such as Floyd are essentially calculating the global optimal solution for static networks, making it difficult to adapt to actual engineering scenarios where cables are laid sequentially and the remaining capacity of cable trays changes dynamically. They lack the ability to adjust online and adapt to complex dynamic environments. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a cable layout method, system, and storage medium based on reinforcement learning in a power supply and distribution system that can plan the optimal laying and layout path under various physical constraints of cable laying and can autonomously and dynamically adapt to different laying scenarios.

[0005] Technical solution: The cable layout method for power supply and distribution systems based on reinforcement learning described in this invention includes the following steps:

[0006] S1. Obtain cable-related information, user-selected start and end devices, and constraints that cable laying must meet;

[0007] S2. Abstract the cable tray system into a weighted undirected topology graph. The nodes in the topology graph include the start and end points of the cable tray, cable tray connectors, and electrical connection points of the start and end point equipment. The edges are the cable trays and carry information on the length, available cross-sectional area, and occupied cross-sectional area of ​​the cable trays.

[0008] S3. Call the reinforcement learning-based path planning algorithm to calculate the optimal cable laying path under the constraints that cable laying needs to meet, with the optimization objective of minimizing the comprehensive cost function that includes path length cost, volume ratio cost and cost penalty term.

[0009] S4. Based on the optimal cable laying path, a greedy algorithm is used to calculate the center coordinates of each cable to be laid in each cable tray, and the cables are laid accordingly.

[0010] By sequentially executing the steps of "information acquisition - topology abstraction - intelligent planning," this invention transforms the traditional discrete, experience-dependent design model into a systematic, data-driven intelligent decision-making process. Specifically, acquiring cable-related information and constraints clarifies the design objectives and required conditions; abstracting the cable tray system into a weighted undirected topological graph transforms the complex spatial geometry problem into a computer-processable graph theory problem, laying the foundation for algorithm application; most importantly, it calls a path planning algorithm based on reinforcement learning and uses a comprehensive cost function that integrates multiple objectives for optimization, enabling the algorithm to autonomously find an approximate optimal solution within the exploration space while satisfying hard constraints. Furthermore, reinforcement learning, compared to Floyd, allows for online learning, thus enabling it to autonomously adapt to dynamically changing and complex laying scenarios. Using a greedy algorithm as the solver ensures rapid computation while maintaining rationality, making it suitable for efficiently processing cable layouts in a large number of cable tray sections. In summary, this method can plan the optimal, or relatively shortest, laying path while meeting various hard cable laying requirements, and it can also adapt to dynamically changing scenarios.

[0011] Preferably, the constraints that the cable laying needs to meet include cable tray volume ratio constraints, cable bending radius constraints, cable voltage level isolation constraints, and cable strong and weak current isolation constraints.

[0012] Cable tray volume ratio constraints directly affect the space utilization efficiency and heat dissipation safety of the cable tray; cable bending radius constraints ensure the mechanical safety and electrical performance of cables during laying and operation; voltage level isolation constraints and strong / weak current isolation constraints are core regulations for preventing electromagnetic interference and ensuring stable system operation. By explicitly defining and integrating these constraints at the algorithm level, it is ensured that any cable path generated in the end naturally complies with industry specifications and safety standards, avoiding compliance risks that may arise from traditional methods or single-objective algorithms.

[0013] Preferably, the expression for the comprehensive cost function is:

[0014]

[0015] in, Let be the comprehensive cost function of the P-th path; , and These are weighting coefficients, and the sum of the three is 1; , and These are the length cost function, the volume ratio cost function, and the violation penalty function for the Pth path, respectively.

[0016] The expression is

[0017]

[0018] in, For the i-th cable tray, Let be the length of the i-th cable tray;

[0019] The expression is

[0020]

[0021] Where n is the number of cable trays contained in path P. The cross-sectional area already occupied by the i-th cable tray is... Let i be the original usable cross-sectional area of ​​the i-th cable tray. The cross-sectional area of ​​the cable to be laid out;

[0022] The expression is

[0023] .

[0024] The comprehensive cost function directly optimizes economics through the length cost function, aiming to minimize cable usage; it optimizes resource balance through the volume ratio cost function, encouraging uniform cable distribution in the cable tray network and avoiding local overload; and it imposes extremely high penalties on violations of hard constraints through a violation penalty function, ensuring the feasibility of the solution. The comprehensive cost function is composed of a weighted average of these three functions, allowing designers to flexibly adjust the optimization bias according to specific project needs. The design of the comprehensive cost function ensures that the algorithm outputs not only an extreme optimal solution in a single dimension, but also a comprehensive approximate optimal solution under multiple engineering objective constraints, making the decision result more consistent with complex engineering realities.

[0025] Preferably, the and It needs to be normalized before being substituted into the comprehensive cost function for calculation.

[0026] Through the and Normalization is performed to map them to the same quantity, so that the two functions can truly and fairly reflect the relative importance the designers attach to the two objectives of "shortening the path" and "balancing the floor area ratio", thus ensuring that the comprehensive cost function can scientifically and effectively balance different optimization objectives.

[0027] Preferably, the reinforcement learning uses Markov decision-making for modeling, where the state is the real-time capacity and length information of the current node and its connected bridges; the action is moving from the current node to any adjacent node; and the rewards include goal achievement rewards, single-step guidance rewards, and violation negative rewards.

[0028] Defining the state as the real-time capacity and length information of the current node and its connected bridges enables the agent to perceive dynamic changes in the environment. Defining actions as movement to adjacent nodes defines the agent's decision space. Designing a composite reward signal that includes goal achievement rewards, single-step guidance rewards, and violation negative rewards decomposes the global optimization objective into gradually learnable local signals, while also taking into account hard constraints. This modeling approach transforms the algorithm from an offline search of a static graph into an agent capable of learning online and making sequential decisions through continuous interaction with the dynamic environment.

[0029] Preferably, the function expression corresponding to the target achievement reward is:

[0030]

[0031] in, To achieve the goal, a reward function is provided. and These are the normalized values ​​of the length cost function and the volume ratio cost function of the Pth path, respectively.

[0032] The function expression for single-step guided reward is:

[0033]

[0034] in, For a single-step guided reward function, The selected cable tray length for this step. and These represent the occupied cross-sectional area and the original available cross-sectional area of ​​the cable tray selected in this step, respectively.

[0035] The violation reward is a pre-set negative reward for any action that violates the constraints that cable laying must meet, and the exploration round will be terminated.

[0036] The goal-achievement reward provides feedback when the agent successfully finds a path. The reward value is linked to the overall quality of the path's "length" and "volume ratio," directly reinforcing the pursuit of the global dual objectives. Single-step guidance rewards are given immediately after each move, encouraging the agent to choose edges with "shortest current step length" and "relatively empty current bridge," providing immediate optimization direction for each decision and significantly accelerating the learning and convergence process. Violation-related negative rewards act as a "high-voltage line," imposing severe penalties and terminating the round for any action that violates the constraints, forcing the agent to learn to avoid all violations from the early stages of exploration.

[0037] As a preferred option, after calculating the optimal cable layout path, a greedy algorithm is used to calculate the center coordinates of each cable to be laid in each cable tray in a triangular arrangement.

[0038] The center coordinates of each cable are calculated by arranging them in a "triangular" pattern. This technique strictly follows industry best practices for power cable laying and is beneficial for cable heat dissipation and reducing electromagnetic interference.

[0039] The cable routing system in the power supply and distribution system based on reinforcement learning described in this invention includes:

[0040] Information acquisition module: used to acquire cable-related information, user-selected starting and ending devices, and constraints that cable laying must meet;

[0041] Topology building module: used to abstract the cable tray system into a weighted undirected topology graph. The nodes in the topology graph include the start and end points of the cable tray, cable tray connectors, and electrical connection points of the start and end point equipment. The edges are the cable trays and carry information on the length, available cross-sectional area, and occupied cross-sectional area of ​​the cable trays.

[0042] The cable laying path planning module is used to call a reinforcement learning-based path planning algorithm to calculate the optimal cable laying path under the constraints that cable laying must meet, with the optimization objective of minimizing the comprehensive cost function that includes path length cost, volume ratio cost, and cost penalty term.

[0043] Cable routing module: Used to calculate the center coordinates of each cable to be laid in each cable tray using a greedy algorithm based on the optimal cable laying path, and then lay the cables accordingly.

[0044] The computer-readable storage medium for storing one or more programs according to the present invention includes one or more programs comprising instructions that, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0045] Beneficial effects: This invention solves the challenge of dynamic multi-constraint optimization in automated cable routing. Specifically, it abstracts the power supply and distribution cable tray system into a topology network and introduces a reinforcement learning algorithm based on Markov decision process modeling as the core planning engine. A comprehensive cost function integrating path length, cable tray volume ratio, and violation penalties is used as the optimization solution function for the reinforcement learning algorithm. A hierarchical reward mechanism is designed for the reinforcement learning algorithm, guiding the agent to autonomously explore the optimal cable path that balances economy and safety while strictly adhering to electrical design specifications. Furthermore, it can autonomously adapt to dynamically changing laying scenarios. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0047] As shown in the figure, the cable layout method in a power supply and distribution system based on reinforcement learning according to the present invention includes the following steps:

[0048] S1. Obtain cable-related information, user-selected start and end devices, and constraints that cable laying must meet;

[0049] This method acquires information and then abstracts it into a topology diagram based on Revit. In addition to Revit, other software used for building BIM can also be used.

[0050] Cable-related information includes a set of key attributes that define the physical and electrical characteristics of a cable, including at least:

[0051] Cable model: such as YJV-0.6 / 1kV-4X150+1X70, used to identify the cable type, voltage level and conductor material.

[0052] Cable outer diameter: The outer diameter of the cable, used to calculate the physical space it occupies in the cable tray, and is directly related to the bending radius verification.

[0053] Cable cross-sectional area: The sum of the cross-sectional areas of all conductors, which is a direct input parameter for calculating the cable tray volume ratio.

[0054] Minimum bending radius: The minimum allowable bending radius during cable laying, usually expressed as a multiple of the cable's outer diameter. This parameter is a hard constraint that must be followed in route planning; the algorithm needs to verify whether the actual bending radius at all turns in the route meets this requirement.

[0055] The user-selected start and end devices are the start and end points of cable laying.

[0056] The constraints that cable laying must meet include cable tray volume ratio constraints, cable bending radius constraints, cable voltage level isolation constraints, and cable strong and weak current isolation constraints.

[0057] The expression for the volume ratio constraint of the cable tray is as follows:

[0058]

[0059] in, The occupied cross-sectional area of ​​the i-th cable tray (the occupied cross-sectional area referred to in this article refers to the cross-sectional area occupied by the cable). Let i be the original usable cross-sectional area of ​​the i-th cable tray. The cross-sectional area of ​​the cable to be laid out; The maximum allowed proportion of the cable cross-sectional area of ​​the i-th cable tray to the original available cross-sectional area is defined as the maximum volume ratio of the i-th cable tray. This constraint is one of the core optimization objectives and hard constraints of the reinforcement learning path planning algorithm in this invention. During execution, the algorithm must ensure that the predicted volume ratio η of any cable tray segment after arranging new cables does not exceed the preset maximum allowable value η. max This value is determined by national standards (such as the "GB50303 Code for Acceptance of Construction Quality of Building Electrical Engineering") or specific project regulations.

[0060] In this invention, the cable tray volume ratio is defined as the ratio of the total cross-sectional area of ​​all cables accommodated in any straight cable tray or its connector (such as a tee or elbow) to the theoretical (original) usable internal cross-sectional area of ​​that cable tray.

[0061] Its mathematical expression is:

[0062]

[0063] in, For cable tray volume ratio, This is the cross-sectional area of ​​the cable. This represents the sum of the cross-sectional areas of all cables within the cable tray. This is the theoretical (original) usable cross-sectional area of ​​the cable tray. The calculation needs to take into account the reduction factor caused by the cable tray structure.

[0064] The cable bending radius constraint stipulates that for any turning node on the path, the turning radius shall not be less than the minimum bending radius; the minimum bending radius is defined as the minimum allowable radius of curvature of the centerline of the cable tray when it is installed and turns. The minimum radius of curvature is determined by the cable laying requirements and industry standards, and must comply with the provisions of standards such as "GB50168 Standard for Construction and Acceptance of Cable Lines in Electrical Installations".

[0065] In path planning algorithms, the minimum bending radius serves as a hard geometric constraint. When searching a path, the algorithm continuously verifies the actual turning radius R of each turning point (such as a bend or tee) in the path. actual Does R satisfy? actual≥R bridge_min (Minimum bending radius). The algorithm will exclude path nodes that do not meet this constraint.

[0066] Cable voltage level isolation constraints require that cables of different voltage levels (such as 10kV high voltage and 0.4kV low voltage) be installed in different cable trays or separated by metal partitions in the same cable tray.

[0067] The isolation constraint for strong and weak current cables requires that power cables and information or control cables be installed in cable trays of different voltage levels to prevent electromagnetic interference.

[0068] In addition to the constraints mentioned above, other classification rules (constraints) based on safety specifications and functional zoning can be set to restrict the laying location of different types of cables in the cable tray system, such as:

[0069] Fire protection and emergency power supply requirements: Power supply lines for fire protection equipment or emergency lighting lines, etc., must be laid in cable trays of a specific fire protection rating or isolated in accordance with their specifications.

[0070] Special area restrictions: For areas with explosion hazards, corrosive environments, or specific process requirements, the installation of cable trays must comply with the corresponding safety level regulations.

[0071] S2. Abstract the cable tray system into a weighted undirected topology graph. The nodes in the topology graph include the start and end points of the cable tray, cable tray connectors, and electrical connection points of the start and end point equipment. The edges are the cable trays and carry information on the length, available cross-sectional area, and occupied cross-sectional area of ​​the cable trays.

[0072] Based on the information obtained in step S1, the spatial coordinates and electrical connection point information of the cable tray system and equipment (power generation equipment such as distribution boxes and cabinets) are obtained by importing the cable list and using the Revit API.

[0073] Constructing a topology network: The system (Revit) automatically scans and analyzes cable trays (including ladder trays and trough trays, etc., with voltage levels corresponding to the voltage levels of the cables that can be arranged inside) throughout the BIM project, abstracting them into a weighted undirected graph G=(V, E). Where:

[0074] Node (V): Includes all cable tray start and end points, tees, crosses, elbows and reducers, as well as electrical connection points of starting and ending equipment.

[0075] Edge (E): Represents each straight segment of the cable tray connecting these nodes. Each edge e i Includes key engineering attributes:

[0076] : The length of this section of the cable tray.

[0077] The theoretical usable cross-sectional area of ​​this section of the cable tray.

[0078] : The total cross-sectional area of ​​all cables currently accommodated in this section of the cable tray (dynamically updated).

[0079] S3. Call the reinforcement learning-based path planning algorithm to calculate the optimal cable laying path under the constraints that cable laying must meet, with the optimization objective being to minimize the comprehensive cost function that includes path length cost, volume ratio cost, and cost penalty term.

[0080] The path planning algorithm based on reinforcement learning is invoked to find the optimal path P from the starting node S to the target node T in graph G, under the premise of ensuring the volume ratio and minimum radius.

[0081] The expression for the comprehensive cost function is:

[0082]

[0083] in, Let be the comprehensive cost function of the P-th path; , and These are weighting coefficients, and the sum of the three is 1; , and These are the length cost function, the volume ratio cost function, and the violation penalty function for the Pth path, respectively.

[0084] The expression is

[0085]

[0086] in, For the i-th cable tray, Let be the length of the i-th cable tray; to eliminate the influence of dimensions, After normalizing according to the following formula, the result is substituted into the comprehensive cost function for calculation.

[0087]

[0088] in, for The normalized value, The maximum length of all feasible paths. It is the minimum straight line length from the starting end to the ending end.

[0089] The expression is

[0090]

[0091] Where n is the number of cable trays contained in path P. The cross-sectional area already occupied by the i-th cable tray is... Let i be the original usable cross-sectional area of ​​the i-th cable tray. Let this be the cross-sectional area of ​​the cable to be laid out; similarly, to eliminate the influence of dimensions, let... After normalizing according to the following formula, the result is substituted into the comprehensive cost function for calculation.

[0092]

[0093] in, for The normalized value, and These are the average maximum and minimum floor area ratios for all feasible paths, respectively.

[0094] The expression is

[0095] .

[0096] Reinforcement learning uses Markov decision-making for modeling, specifically:

[0097] State: Defined as the graph node where the agent (cable) is currently located, as well as the capacity and length information of the cable tray connected to the current node.

[0098] Action: Defined as moving from the current node to any adjacent node.

[0099] Rewards: We designed sophisticated rewards to guide the agent's learning, specifically including:

[0100] Target achievement reward (sparse positive reward): When the agent reaches the target node T, a graded reward is given based on the overall performance of the path. The better the path (the shorter the length and the lower the volume ratio), the higher the reward, which strengthens the motivation for dual-objective optimization.

[0101] The function expression for the reward for achieving the goal is:

[0102]

[0103] in, To achieve the goal, a reward function is provided. and are the normalized values ​​of the length cost function and the volume ratio cost function of the Pth path, respectively.

[0104] Single-step guided reward (dense positive reward): After each step, an immediate positive reward is given based on the "length + floor area ratio" of the bridge section traversed, where the shorter the length, the higher the reward; and the lower the floor area ratio, the higher the reward.

[0105] The function expression for single-step guided reward is:

[0106]

[0107] in, For a single-step guided reward function, The selected cable tray length for this step. and These represent the occupied cross-sectional area and the original available cross-sectional area of ​​the cable tray selected in this step, respectively.

[0108] Violation Reward (Sparse Negative Reward): If any action results in a violation of any constraint that must be met for cable laying, an extremely high negative reward of -3000 will be given and the current exploration round will be terminated immediately, forcing the agent to avoid all regulatory violations. The specific negative reward value can be adjusted according to the actual situation.

[0109] Through extensive training in simulated environments, the agent learns a policy by interacting with the environment. Starting from any initial state, it can choose a series of actions (i.e., paths) to maximize cumulative rewards, which is equivalent to finding the compliant path with the lowest overall cost.

[0110] After the algorithm calculates the optimal path P, the system automatically generates a 3D cable model using the Revit API, precisely lays the cable along path P, and correctly connects it to the starting and ending devices. Simultaneously, the system updates all cable tray segments e along the path. i of The value (including the cross-sectional area of ​​the newly laid cable) enables real-time data linkage, providing accurate initial conditions for the intelligent layout of the next cable.

[0111] S4. Based on the optimal cable laying path, a greedy algorithm is used to calculate the center coordinates of each cable to be laid in each cable tray, and the cables are laid accordingly.

[0112] After generating cable laying paths using a reinforcement learning-based path planning algorithm, this invention further executes the "automatic cable arrangement of cable tray sections" function to achieve intelligent management of the entire process from macro-routing to micro-layout. This function treats each planned cable tray section as an independent two-dimensional optimization problem. The system (Revit) reads the outer diameter attributes of all cables passing through that section of the cable tray and runs a sorting module based on a greedy algorithm, using the inner wall boundary of the cable tray as a constraint.

[0113] The module supports various preset layout rules, the most typical of which is the "triangular" arrangement. The system automatically calculates the number of cables in each layer of the "triangular" arrangement and the precise center coordinates (X, Y) of each cable, ensuring no interference between cables or between cables and the inner wall of the cable tray. Ultimately, the path of each cable in three-dimensional space and its precise position in each cable tray cross-section are uniquely determined, generating a fully constructible and highly detailed BIM model.

[0114] The cable routing system in the power supply and distribution system based on reinforcement learning described in this invention includes:

[0115] Information acquisition module: used to acquire cable-related information, user-selected starting and ending devices, and constraints that cable laying must meet;

[0116] Topology building module: used to abstract the cable tray system into a weighted undirected topology graph. The nodes in the topology graph include the start and end points of the cable tray, cable tray connectors, and electrical connection points of the start and end point equipment. The edges are the cable trays and carry information on the length, available cross-sectional area, and occupied cross-sectional area of ​​the cable trays.

[0117] The cable routing planning module is used to call a reinforcement learning-based routing algorithm to calculate the optimal cable route while satisfying the constraints that cable laying must meet. The optimization objective is to minimize the comprehensive cost function, which includes path length cost, volume ratio cost, and cost penalty term.

[0118] The computer-readable storage medium for storing one or more programs according to the present invention includes one or more programs comprising instructions that, when executed by a computing device, cause the computing device to perform the method described above.

Claims

1. A cable layout method in a power supply and distribution system based on reinforcement learning, characterized in that, Includes the following steps: S1. Obtain cable-related information, user-selected start and end devices, and constraints that cable laying must meet; S2. Abstract the cable tray system into a weighted undirected topology graph. The nodes in the topology graph include the start and end points of the cable tray, cable tray connectors, and electrical connection points of the start and end point equipment. The edges are the cable trays and carry information on the length, available cross-sectional area, and occupied cross-sectional area of ​​the cable trays. S3. Call the reinforcement learning-based path planning algorithm to calculate the optimal cable laying path under the constraints that cable laying needs to meet, with the optimization objective of minimizing the comprehensive cost function that includes path length cost, volume ratio cost and cost penalty term. S4. Based on the optimal cable laying path, a greedy algorithm is used to calculate the center coordinates of each cable to be laid in each cable tray, and the cables are laid accordingly.

2. The method according to claim 1, characterized in that: The constraints that cable laying must meet include cable tray volume ratio constraints, cable bending radius constraints, cable voltage level isolation constraints, and cable strong and weak current isolation constraints.

3. The method according to claim 1, characterized in that: The expression for the comprehensive cost function is as follows: , in, Let be the comprehensive cost function of the P-th path; , and These are weighting coefficients, and the sum of the three is 1; , and These are the length cost function, the volume ratio cost function, and the violation penalty function for the Pth path, respectively. The expression is , in, For the i-th cable tray, Let be the length of the i-th cable tray; The expression is , Where n is the number of cable trays contained in path P. The cross-sectional area already occupied by the i-th cable tray is... Let i be the original usable cross-sectional area of ​​the i-th cable tray. The cross-sectional area of ​​the cable to be laid out; The expression is 。 4. The method according to claim 3, characterized in that: The and It needs to be normalized before being substituted into the comprehensive cost function for calculation.

5. The method according to claim 1, characterized in that: The reinforcement learning uses Markov decision-making for modeling, where the state is the real-time capacity and length information of the current node and its connected bridges; the action is to move from the current node to any adjacent node; and the rewards include goal achievement rewards, single-step guidance rewards, and violation negative rewards.

6. The method according to claim 5, characterized in that: The function expression corresponding to the reward for achieving the target is: , in, To achieve the goal, a reward function is provided. and These are the normalized values ​​of the length cost function and the volume ratio cost function of the Pth path, respectively. The function expression for single-step guided reward is: , in, For a single-step guided reward function, The selected cable tray length for this step. and These represent the occupied cross-sectional area and the original available cross-sectional area of ​​the cable tray selected in this step, respectively. The violation reward is a pre-set negative reward for any action that violates the constraints that cable laying must meet, and the exploration round will be terminated.

7. The method according to claim 2, characterized in that: The expression for the cable tray volume ratio constraint is as follows: , in, The maximum proportion of the cable cross-sectional area allowed in the i-th cable tray to the original available cross-sectional area; The cable bending radius constraint is that for any turning node on the path, the turning radius is not less than the minimum bending radius. Cable voltage level isolation constraints require cables of different voltage levels to be installed in different cable trays or separated by metal partitions in the same cable tray; The strong and weak current isolation constraint of cables requires that power cables and information or control cables be installed in cable trays of different voltage levels.

8. The method according to claim 1, characterized in that: The greedy algorithm calculates the center coordinates of each cable to be laid in each cable tray according to the triangular arrangement.

9. A cable routing system in a power supply and distribution system based on reinforcement learning, characterized in that, include: Information acquisition module: used to acquire cable-related information, user-selected starting and ending devices, and constraints that cable laying must meet; Topology building module: used to abstract the cable tray system into a weighted undirected topology graph. The nodes in the topology graph include the start and end points of the cable tray, cable tray connectors, and electrical connection points of the start and end point equipment. The edges are the cable trays and carry information on the length, available cross-sectional area, and occupied cross-sectional area of ​​the cable trays. The cable laying path planning module is used to call a reinforcement learning-based path planning algorithm to calculate the optimal cable laying path under the constraints that cable laying must meet, with the optimization objective of minimizing the comprehensive cost function that includes path length cost, volume ratio cost, and cost penalty term. Cable routing module: Used to calculate the center coordinates of each cable to be laid in each cable tray using a greedy algorithm based on the optimal cable laying path, and then lay the cables accordingly.

10. A computer-readable storage medium for storing one or more programs, characterized in that: The program includes one or more instructions that, when executed by a computing device, cause the computing device to perform any of the methods according to claims 1 to 8.

Images