Railway line selection method based on artificial intelligence

Abstract

The invention relates to the technical field of path planning, and provides a railway line selection method based on artificial intelligence. According to the method, a dynamic bidirectional A algorithm and a distributed Bellman trajectory optimization algorithm are fused, and efficient generation and intelligent optimization of a railway line path are realized; the method comprises the following steps: firstly, constructing a path comprehensive cost map based on multi-source spatial data, and generating a path set with diversity and globality by utilizing a dynamic bidirectional A algorithm through a bidirectional synchronous propulsion and multi-disturbance strategy; then, in combination with a distributed Bellman trajectory optimization algorithm, the rationality and risk control capability of the path structure are further improved; by means of the technical path, the defects of an existing railway line selection method in the aspects of path diversity control, trajectory optimization modeling and multi-source data fusion and utilization are effectively overcome, and the intelligent level and the global optimal performance of railway path planning are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a railway line route selection method based on artificial intelligence. Background Art

[0002] As a key link in railway engineering design, railway line route selection not only affects construction costs and operation efficiency, but also involves various factors such as complex topography, geological risks, hydrological conditions, land use, ecological protection, and urban planning. Therefore, how to scientifically and efficiently complete railway route selection under the premise of ensuring safety, economy, and environmental coordination has become a major technical challenge in engineering practice. Although current railway line route selection work has gradually introduced intelligent means such as geographic information systems (GIS), remote sensing data processing, spatial analysis, and multi-factor evaluation models, realizing the transformation from manual route selection to digital-assisted route selection, there are still several significant deficiencies in practical engineering applications: First, existing intelligent route selection methods mostly rely on static evaluation models for comprehensive cost analysis of paths, usually through fixed weighted superposition methods to comprehensively process each influencing factor, lacking dynamic adaptability under different geographical regions, construction constraints, and engineering objectives. Second, in the path search stage, traditional methods mostly use one-way heuristic search or graph theory optimization algorithms, and their search strategies have limited balance capabilities among global optimality, path diversity, and local feasibility. Finally, the current path optimization process generally lacks behavior decision-making layer modeling and cannot effectively characterize the dynamic feedback, trajectory distribution characteristics, and strategy evolution mechanism behind path selection. Summary of the Invention

[0003] The present invention provides a railway route selection method based on artificial intelligence. By integrating the dynamic bidirectional A* algorithm and the distributed Bellman trajectory optimization algorithm, a systematic modeling of path generation and optimization in the railway route selection process is realized. First, a path comprehensive cost map is constructed based on multi-source spatial data, and the dynamic bidirectional A* algorithm is used to perform preliminary path search on the path comprehensive cost map. This algorithm adopts a bidirectional synchronous advancement architecture, dynamically adjusts the advancement ratio of the forward and reverse searches, and generates a path set through heuristic function perturbation, cost map perturbation, and start / end point local perturbation strategies, thereby effectively improving path diversity and global optimality. Secondly, in the path optimization stage, the present invention introduces the distributed Bellman trajectory optimization algorithm. By constructing an observable Markov decision process, a trajectory return distribution model is established for the preliminary path set, and the ψ-vector structure is used to characterize the trajectory distribution characteristics. Based on the distributed Bellman expectation operator, the difference between trajectory distributions is measured by the Wasserstein distance, and the path strategy is continuously optimized. Finally, a railway line path plan with global optimality and risk control characteristics is generated. The present invention comprehensively considers diversity control in the path search process and distribution modeling in the trajectory optimization process, realizes the global optimization and dynamic adaptation of the railway route selection plan while improving the path generation efficiency and optimization effect.

[0004] The present invention provides a railway route selection method based on artificial intelligence, and the method includes the following steps:

[0005] Step S1: Collect data on terrain, landform, land use, geological risks, water system distribution, traffic, population density, and urban distribution in the area along the railway, and unify the projection coordinate system, resample the spatial resolution, fill in missing values, and standardize the format to generate a standardized spatial raster layer;

[0006] Step S2: Based on the standardized spatial raster layer, construct a railway route selection impact factor system, build a weight model through the principal component analysis method, and perform weighted superposition processing on each factor of the railway route selection impact factor system based on the weight model to generate a path comprehensive cost map; the path comprehensive cost map is modeled in units of nodes;

[0007] Step S3: Construct a dynamic bidirectional A* algorithm through a bidirectional search architecture, a front synchronization advancement mechanism, and a multi-perturbation strategy optimization mechanism, and use the dynamic bidirectional A* algorithm to search for preliminary feasible paths on the path comprehensive cost map to generate a preliminary path set;

[0008] Step S4: Construct a distributed Bellman trajectory optimization algorithm through observable Markov decision modeling, trajectory distribution modeling, and strategy optimization driven by the Wasserstein distance; combine the preliminary path set, and obtain the optimal path strategy through the distributed Bellman trajectory optimization algorithm;

[0009] Step S5: According to the optimal path strategy, perform path reconstruction and trajectory generation on the path comprehensive cost map to obtain the optimal railway line path plan, including path trajectory, cost distribution, and risk characteristics.

[0010] Further, step S3 specifically includes the following steps:

[0011] Step S31: Based on the path comprehensive cost map, synchronously promote the forward search and backward search in the bidirectional A* algorithm to generate a bidirectional promotion rhythm;

[0012] Step S32: Adopt a dynamic search ratio adjustment strategy to control the bidirectional promotion rhythm and continuously synchronously promote the forward search and backward search;

[0013] During the continuous synchronous promotion process, identify the intersection node, judge the path closure, and respectively backtrack the forward path segment and the backward path segment centered on the intersection node to generate a preliminary path;

[0014] Step S34: Based on the preliminary path, introduce perturbation strategies, including heuristic function perturbation, cost map perturbation, and start / end point local perturbation. Under different perturbation conditions, use the Bidirectional A* algorithm to repeatedly execute steps S31 - S33 to obtain a set of preliminary paths;

[0015] Further, step S4 specifically includes the following steps:

[0016] Step S41: Define the state space, action space, and observation space, construct the state transition probability, observation probability, and reward function, and construct an observable Markov decision process;

[0017] Step S42: Based on the observable Markov decision process, use the set of preliminary paths to construct an initial path strategy, and perform sample trajectory replay. For the cumulative return value of each path of the initial path strategy, obtain the trajectory return distribution. Introduce the ψ - vector structure to perform distribution modeling on the trajectory return distribution of each path to obtain an initial set of ψ - vectors, which is used to characterize the structural characteristics of the path distribution;

[0018] Step S43: Use the distributed Bellman optimality operator to perform distributed backup on the trajectory return distribution, and maintain the consistency of the distribution mean through mean - preserving projection; Use the Wasserstein distance to measure the difference between distributions, and perform policy improvement to update the initial path strategy; And update the initial set of ψ - vectors to generate an updated set of ψ - vectors, and record the optimized path distribution structure;

[0019] Step S44: Set the convergence threshold. Combine the updated ψ-vector set, calculate the Wasserstein distance between the trajectory return distributions of two adjacent rounds, and judge the convergence status. If the Wasserstein distance is less than the convergence threshold, it means convergence, stop the policy update. If the Wasserstein distance is greater than the convergence threshold, continue to execute Step S43 for policy improvement to obtain the optimal path policy.

[0020] Adopting the above solution, the beneficial effects achieved by the present invention are as follows:

[0021] By introducing the dynamic bidirectional A* algorithm, the present invention realizes the effective control of path diversity and global optimality in the process of railway line path generation. Compared with the traditional path search algorithm, through the bidirectional synchronous advancement mechanism, heuristic function perturbation, and cost map perturbation strategy, the present invention dynamically adjusts the advancement rhythm of forward and backward searches, thus breaking through the limitation that the traditional algorithm is prone to falling into local optimal solutions and generating diverse and globally optimal path schemes. This mechanism not only improves the efficiency of path search but also enhances the robustness of path generation, providing a more comprehensive and reasonable path alternative set for railway route selection in complex terrain environments.

[0022] In the path optimization stage, the present invention realizes the in-depth optimization of the preliminary path set through the distributed Bellman trajectory optimization algorithm. Based on the observable Markov decision-making modeling, the present invention constructs a trajectory return distribution model and introduces the ψ-vector structure to perform distributed modeling and optimization of the trajectory return. By measuring the trajectory distribution difference with the Wasserstein distance and combining the distributed Bellman expectation operator to iteratively optimize the path policy, the present invention realizes the accurate characterization and structural adjustment of the trajectory distribution characteristics in the path optimization process, thus effectively solving problems such as multi-stage policy incoordination and local cost bias in the path optimization process, and further improving the global consistency and optimality of the path optimization result.

[0023] In addition, by comprehensively constructing the path comprehensive cost map, path search using the dynamic bidirectional A* algorithm, and distributed Bellman trajectory optimization, the present invention realizes the systematic modeling of path generation and optimization. This method not only solves the problem of insufficient comprehensive utilization of multi-source data in the existing railway route selection process but also effectively improves the dynamic adaptation ability to complex geographical environments and changing constraint conditions in the path generation process. Through the above multi-algorithm fusion mechanism, the railway route selection scheme generated by the present invention not only has the advantage of cost optimization but also has the characteristics of reasonable path risk distribution and strong path structure rationality, thus significantly enhancing the feasibility and reliability of the railway route selection scheme in practical engineering applications. Brief Description of the Drawings

[0024] Figure 1 It is a module schematic diagram of a railway line route selection method based on artificial intelligence proposed by the present invention;

[0025] Figure 2 It is the cost profile in the cost distribution proposed in Embodiment 5. Specific implementation manner

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present invention.

[0027] Embodiment 1, according to Figure 1 , the present invention provides a railway line route selection method based on artificial intelligence, and the method includes the following steps:

[0028] Step S1: Data collection: Collect data on the terrain, landform, land use, geological risks, water system distribution, traffic, population density, and urban distribution in the area along the railway, and unify the projection coordinate system, resample the spatial resolution, fill in missing values, and standardize the format to generate a standardized spatial raster layer;

[0029] Step S2: Cost map construction: Based on the standardized spatial raster layer, construct a railway route selection impact factor system, build a weight model through the principal component analysis method, and perform weighted superposition processing on each factor of the railway route selection impact factor system based on the weight model to generate a path comprehensive cost map; the path comprehensive cost map is modeled in units of nodes;

[0030] Step S3: Generation of a preliminary path set: Construct a dynamic bidirectional A* algorithm through a bidirectional search architecture, a frontier synchronous advancement mechanism, and a multi-disturbance strategy optimization mechanism, and use the dynamic bidirectional A* algorithm to search for preliminary feasible paths on the path comprehensive cost map to generate a preliminary path set;

[0031] Step S4: Generation of an optimal strategy: Construct a distributed Bellman trajectory optimization algorithm through observable Markov decision-making modeling, trajectory distribution modeling, and Wasserstein distance-driven strategy optimization; combine the preliminary path set, and obtain an optimal path strategy through the distributed Bellman trajectory optimization algorithm;

[0032] Step S5: Generation of a plan: According to the optimal path strategy, perform path reconstruction and trajectory generation on the path comprehensive cost map to obtain an optimal railway line path plan, including path trajectory, cost distribution, and risk characteristics.

[0033] Embodiment 2, this embodiment is based on Embodiment 1. In this embodiment, step S3 specifically includes the following steps:

[0034] Step S31: Based on the path comprehensive cost map, set the starting position and the ending position, initialize the forward search state set and the backward search state set, construct the open list and the closed list to store the nodes to be expanded and the expanded nodes; define the total cost evaluation function to evaluate the node expansion priority; construct the priority queue scheduling strategy to dynamically select the node with the minimum evaluation value for expansion operation, synchronously promote the forward search and the backward search in the bidirectional A* algorithm, generate the bidirectional promotion rhythm, and provide structural support for path intersection and splicing. The used formula is as follows:

[0035] Total cost evaluation function:

[0036] ;

[0037] Among them, represents the node, represents the total evaluation value, represents the actual cost, represents the heuristic cost;

[0038] Step S32: Adopt the dynamic search ratio adjustment strategy to control the bidirectional promotion rhythm, and continuously and synchronously promote the forward search and the backward search. The used formula is as follows:

[0039] ;

[0040] Among them, represents the promotion ratio of the forward search and the backward search, represents the number of nodes expanded in the forward search, represents the number of nodes expanded in the backward search, represents the balance adjustment factor;

[0041] Step S33: During the continuous synchronous promotion process, when the forward search and the backward search intersect in the adjacent area, identify the intersection node, judge that the path is closed, take the intersection node as the center, respectively backtrack the forward path segment and the backward path segment, splice and generate a complete path, generate the preliminary path, and record the path structure, the cumulative cost and the topological features;

[0042] Step S34: Based on the preliminary path, introduce the perturbation strategy, including heuristic function perturbation, cost map perturbation and start-end point local perturbation. Under different perturbation conditions, use the Bidirectional A* algorithm, and repeat Steps S31 - S33 to obtain the preliminary path set;

[0043] The specific contents of the heuristic function perturbation, the cost map perturbation and the start-end point local perturbation include:

[0044] .

[0045] Embodiment 3. This embodiment is based on Embodiment 1. In this embodiment, step S3 specifically includes the following steps:

[0046] Step S31: Based on the path comprehensive cost map, set the starting position and the ending position, initialize the forward search state set and the reverse search state set, construct an open list and a closed list for storing nodes to be expanded and expanded nodes; define a total cost evaluation function for evaluating the node expansion priority; construct a priority queue scheduling strategy to dynamically select the node with the minimum evaluation value for expansion operations, synchronously promote the forward search and the reverse search in the bidirectional A* algorithm, generate a bidirectional advancement rhythm, and provide a structural support for path intersection and splicing;

[0047] Step S32: Adopt a dynamic search ratio adjustment strategy to control the bidirectional advancement rhythm, and continuously and synchronously promote the forward search and the reverse search;

[0048] Step S33: During the continuous and synchronous advancement process, when the forward search and the reverse search intersect in the adjacent area, identify the intersection node, judge that the path is closed, take the intersection node as the center, respectively backtrack the forward path segment and the reverse path segment, splice them to generate a complete path, generate a preliminary path, and record the path structure, the cumulative cost, and the topological features;

[0049] Step S34: Based on the preliminary path, adopt the Bidirectional A* algorithm, and repeat steps S31 - S33 to obtain a preliminary path set.

[0050] Embodiment 4. This embodiment is based on Embodiment 2. In this embodiment, step S4 specifically includes the following steps:

[0051] Step S41: Define the state space, the action space, and the observation space, construct the state transition probability, the observation probability, and the reward function, and construct an observable Markov decision process; the state space includes the current node coordinates, the terrain type, the land use type, the geological risk level, the geometric distance from the target point, the adjacent population density, and the current cumulative path cost;

[0052] Step S42: Based on the observable Markov decision process, use the preliminary path set to construct an initial path strategy, and perform sample trajectory playback. For each path cumulative return value of the initial path strategy, obtain the trajectory return distribution, introduce the ψ-vector structure to perform distribution modeling on the trajectory return distribution of each path, and obtain the initial ψ-vector set for characterizing the structural features of the path distribution;

[0053] Step S43: Use the distributed Bellman optimality operator to perform distributed backup on the trajectory return distribution, and maintain the consistency of the distribution mean through mean-preserving projection; use the Wasserstein distance to measure the difference between distributions, and perform policy improvement to update the initial path policy; and update the initial ψ-vector set to generate an updated ψ-vector set, record the optimized path distribution structure. The formulas used are as follows:

[0054] Distributed Bellman expectation formula:

[0055] ;

[0056] where represents the iteration index, represents the belief state, represents at iteration, the trajectory return distribution corresponding to the belief state , represents at iteration, the trajectory return distribution corresponding to the belief state ; represents the immediate reward, represents the discount factor, represents the current policy under which the distribution operator for Bellman-style propagation of the trajectory return distribution; represents the distribution modeling operator;

[0057] Step S44: Set the convergence threshold, combine the updated ψ-vector set, calculate the Wasserstein distance between the trajectory return distributions of two adjacent rounds, and judge the convergence status. If the Wasserstein distance is less than the convergence threshold, it means convergence and stop policy update. If the Wasserstein distance is greater than the convergence threshold, continue to execute Step S43 for policy improvement to obtain the optimal path policy.

[0058] Example 5, according to Figure 2 , this example is based on Example 4. In this example, Step S5: According to the optimal path policy, perform path reconstruction and trajectory generation in the path comprehensive cost graph to obtain the optimal railway line path plan, including path trajectory, cost distribution, and risk characteristics;

[0059] In the example, the following is selected: Route optimization and line selection for the high-speed railway line from A to B;

[0060] Path trajectory:

[0061] Starting point: A;

[0062] End point: B;

[0063] Total track length: approximately 318 km;

[0064] Spatial step size: 30 m (grid scale);

[0065] Data format: GeoJSON;

[0066] Path form: continuous node trajectories optimized based on the distributed Bellman strategy, approximately containing 10,600 path points;

[0067] Cost distribution (including cost profile):

[0068] ;

[0069] Risk characteristics:

[0070] .

[0071] The present invention and its implementation manners have been described above. Such description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design, without creative efforts, structural manners and embodiments similar to the technical solution without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A railway line route selection method based on artificial intelligence, characterized in that: The method includes the following steps: Step S1: Collect data of the areas along the railway and generate a standardized spatial grid layer; Step S2: Generate a path comprehensive cost map based on the standardized spatial grid layer; Step S3: Construct a dynamic bidirectional A* algorithm, and use the dynamic bidirectional A* algorithm to search for a preliminary feasible path on the path comprehensive cost map to generate a preliminary path set; Step S4: Construct a distributed Bellman trajectory optimization algorithm; combine the preliminary path set, and obtain an optimal path strategy through the distributed Bellman trajectory optimization algorithm; Step S5: Reconstruct the path and generate a trajectory on the path comprehensive cost map according to the optimal path strategy to obtain an optimal railway line path plan.

2. The method for selecting a railway line based on artificial intelligence according to claim 1, wherein: The path comprehensive cost map is modeled in units of nodes.

3. The method for selecting a railway line based on artificial intelligence according to claim 1, wherein: The construction method of the dynamic bidirectional A* algorithm is: construct it through a bidirectional search architecture, a frontier synchronous advancement mechanism, and a multi-perturbation strategy optimization mechanism.

4. A railway line route selection method based on artificial intelligence according to claim 1, characterized in that: The construction method of the distributed Bellman trajectory optimization algorithm is: construct it through an observable Markov decision-making model, a trajectory distribution model, and a Wasserstein distance-driven strategy optimization.

5. The method for selecting a railway line based on artificial intelligence according to claim 3, characterized in that: Step S3 specifically includes the following steps: Step S31: Based on the path comprehensive cost map, synchronously advance the forward search and the backward search in the bidirectional A* algorithm to generate a bidirectional advancement rhythm; Step S32: Use a dynamic search ratio adjustment strategy to control the bidirectional advancement rhythm, and continuously synchronously advance the forward search and the backward search; Step S33: During the continuous synchronous advancement process, identify the intersection node, judge the path closure, and respectively backtrack the forward path segment and the backward path segment centered on the intersection node to generate a preliminary path; Step S34: Based on the preliminary path, introduce a perturbation strategy, including heuristic function perturbation, cost map perturbation, and start and end point local perturbation, use the Bidirectional A* algorithm, and repeat steps S31 - S33 to obtain a preliminary path set.

6. The method for selecting a railway line based on artificial intelligence according to claim 4, wherein: Step S4 specifically includes the following steps: Step S41: Construct an observable Markov decision-making process; Step S42: Based on the observable Markov decision-making process, use the preliminary path set to construct an initial path strategy, and perform sample trajectory replay to obtain a trajectory return distribution. Introduce a ψ-vector structure to perform distribution modeling on the trajectory return distribution to obtain an initial ψ-vector set; Step S43: Use a distributed Bellman optimality operator to perform distributed backup on the trajectory return distribution, and maintain the consistency of the distribution mean through a mean-preserving projection; use the Wasserstein distance to measure the difference between distributions, and perform strategy improvement to update the initial path strategy; and update the initial ψ-vector set to generate an updated ψ-vector set; Step S44: Combine the updated ψ-vector set, calculate the Wasserstein distance between the trajectory return distributions of two adjacent rounds, judge the convergence status, and obtain an optimal path strategy.

7. The method for selecting a railway line based on artificial intelligence according to claim 6, wherein: Step S41 specifically includes: define a state space, an action space, an observation space, construct a state transition probability, an observation probability, and a reward function, and construct an observable Markov decision-making process.

Images