A complex environment railway intelligent route selection gravity field dynamics modeling and adaptive guidance optimization method

Abstract

The present application relates to the technical field of railway engineering route selection design, and discloses a complex environment railway intelligent route selection gravitational field dynamics modeling and adaptive guidance optimization method. The method comprises the following steps: constructing a multi-criteria index system, establishing a hierarchical model of the adaptation degree space field, and using the DBSCAN density clustering algorithm to identify high and low adaptation environment patches; constructing an intelligent route selection environment static field model; establishing and solving the dynamics equation of gravitational propagation to realize intelligent route selection environment dynamic field modeling; introducing a Transformer neural network model to complete gravitational field feature situation awareness, proposing a dynamic adaptation route setting method, and outputting the optimal route selection scheme of the railway. The method of the present application converts the route selection environment adaptation degree space field into a gravitational field by constructing the route selection environment adaptation degree space field, combines dynamic evolution mechanism analysis and adaptive guidance intelligent search strategy, realizes efficient optimization design of railway lines in dense constraint environment, improves the strain capacity of route setting decision, and can support the dynamic optimization demand of engineering route selection in complex environment.

Description

Technical Field

[0001] This invention relates to the field of railway engineering route selection design technology, and in particular to a method for intelligent railway route selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization. Background Technology

[0002] In the field of railway construction, the complexity of the construction environment has become a key factor restricting the efficiency and quality of engineering projects. This complexity is mainly reflected in the interweaving of multi-dimensional environmental factors: on the one hand, natural geographical conditions vary significantly, including complex terrain (such as mountains, canyons, and swamps), variable geological structures (such as faults, karst caves, and loose rock layers), and the influence of climate and hydrology (such as heavy rain, permafrost, and groundwater level fluctuations); on the other hand, human and artificial environmental interference is equally prominent, involving sensitive areas such as existing building complexes, transportation networks, areas with dense underground pipelines, and ecological protection zones, resulting in multiple constraints that engineering construction must consider in terms of safety, economy, and environmental harmony.

[0003] Given the aforementioned complex environmental characteristics, railway engineering route selection faces a severe challenge of dynamic adaptation. Traditional route selection schemes often rely on static survey data and experience-based judgments, making it difficult to respond in real time to dynamic changes in environmental parameters. For example, in areas with unstable geological conditions, the stress distribution of soil and rock may dynamically adjust with the construction process. If the route selection scheme fails to adapt to these changes in a timely manner, it can easily lead to safety risks such as landslides and settlement. At the same time, when the project line passes through ecologically sensitive areas, dynamic factors such as the migration of flora and fauna due to seasonal changes and fluctuations in hydrological rhythms also place higher demands on the ecological adaptability of the route selection. In addition, in the context of urban renewal, the need for upgrading and transforming existing infrastructure (such as expanding underground pipelines and widening roads) may lead to sudden changes in the surrounding environmental load, further exacerbating the contradiction between the route selection scheme and the dynamic environment. Traditional static planning models are unable to meet the dynamic adjustment needs throughout the entire life cycle of the project.

[0004] To address the aforementioned issues, existing technical solutions mostly adopt a "line-based environment search" approach, first generating a route plan and then evaluating its suitability for the environment. This approach has limitations in applications to complex environments. In recent years, the "environmental attraction line" strategy for complex environments has been proposed. This strategy uses the environment to attract the route and establishes a gravitational field model. However, the characterization of the environmental gravitational field mostly remains under static or quasi-static conditions, and its core flaw lies in ignoring the dynamic nature of the environment. Static field analysis assumes that environmental parameters remain constant throughout the engineering cycle, using fixed mechanical equilibrium equations for route selection evaluation. However, in actual engineering, dynamic changes in environmental factors (such as the creep effect of geological bodies, instantaneous impacts of external loads, and periodic fluctuations in hydrological conditions) can disrupt the equilibrium state assumed by the static field, leading to significant deviations in route selection schemes based on this model.

[0005] Therefore, existing static field technology is insufficient to support the dynamic optimization requirements of engineering route selection in complex environments. There is an urgent need to design a method to support engineering route selection in complex environments in order to solve the problems existing in the current technology. Summary of the Invention

[0006] The main objective of this invention is to provide a method for dynamic modeling and adaptive guidance optimization of gravitational fields in intelligent railway alignment selection under complex environments, addressing the challenge that existing static field techniques cannot support the dynamic optimization requirements of engineering alignment selection in complex environments. The specific technical solution is as follows:

[0007] A method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization includes the following steps:

[0008] Step S1: Construct a multi-criteria index system, establish a hierarchical model of the fitness spatial field, and use the DBSCAN density clustering algorithm to identify high-fitting and low-fitting environmental patches; construct an intelligent route selection environmental static field model.

[0009] Step S2: Establish and solve the dynamic equations of gravity propagation to realize the dynamic field modeling of the intelligent route selection environment;

[0010] Step S3: Introduce the Transformer neural network model to complete the situational awareness of gravitational field characteristics, propose a dynamic adaptation and alignment method, and output the optimal railway alignment scheme.

[0011] Preferably, step S1 specifically includes:

[0012] Step S1.1: For the multi-source heterogeneous data involved in the engineering route selection, the FP-tree association rule mining algorithm is used to screen key fitness indicators and identify core indicators that are strongly correlated with the route selection fitness; the analytic hierarchy process and the entropy weight method are combined to construct a multi-criteria comprehensive evaluation model to obtain the subjective and objective weights of the core indicators; the comprehensive constraint weights of each core indicator are obtained through a coupling algorithm to complete the construction of the multi-criteria indicator system.

[0013] Step S1.2: Standardize the study area into grids to obtain grid cells; based on the structural tomography principle and the isotropic assumption, and relying on the grid cells, perform layering of the engineering area according to the vertical depth gradient to construct a layered model of the fitness spatial field; and use the spatial interpolation technique of Kriging interpolation based on the grid cells to obtain the fitness numerical field.

[0014] Step S1.3, Combining The hotspot analysis algorithm performs spatial statistical analysis on the fitness numerical field obtained in step S1.2 to obtain high-fitting environmental patches and low-fitting environmental patches.

[0015] Step S1.4: Traverse the high-fitting and low-fitting environmental patches obtained in step S1.3, define the high-fitting environmental patches as gravity sources and the low-fitting environmental patches as repulsion sources, and construct an intelligent route selection environment static field model.

[0016] Preferably, the layering process in step S1.2 specifically involves classifying engineering structures from top to bottom, including: inaccessible layers, ultra-high bridge layers, high-pier bridge layers, ordinary bridge layers, embankment layers, road cutting layers, shallow buried tunnel layers, deep buried tunnel layers, ultra-deep buried tunnel layers, and inaccessible layers.

[0017] Preferably, in step S1.3, the DBSCAN density clustering algorithm is used for automatic extraction. By setting the minimum number of neighborhood points and the scanning radius, discrete points that are spatially adjacent and whose fitness meets the threshold condition are aggregated into independent patches, and high-fitness spatial hotspot areas, i.e., high-fitness environmental patches, and low-fitness spatial hotspot areas, i.e., low-fitness environmental patches, are identified.

[0018] Preferably, the spatial distribution characteristics and multimodal attributes of the identified high-fit environmental patches and low-fit environmental patches are analyzed respectively; the spatial distribution characteristics include at least one of area, morphology and spatial correlation; the multimodal attributes include at least one of geological stability, engineering feasibility and environmental friendliness.

[0019] Preferably, in step S1.4, when constructing the static field model of the intelligent route selection environment, the following formula is used to calculate the impact of the line element at any spatial point P on the first... The force of each patch :

[0020] ;

[0021] Wherein: force When the value is positive, it represents gravity; force. When the value is negative, it represents a repulsive force. The gravitational field proportionality constant; For the first The source intensity of each patch; The sensitivity coefficient of the line element; For spatial point P to the th The straight-line distance between the centers of the patches; For from the first The unit vector pointing from the center of each patch to a spatial point P.

[0022] Preferably, step S2 includes the following steps:

[0023] Step 2.1: Based on the theory of system propagation dynamics, partial differential equations are introduced to characterize the propagation mechanism of gravitational increments, and the dynamic equations of gravitational propagation are constructed as follows:

[0024] ;

[0025] in: It is a time-varying term; For spatial diffusion, It is the diffusion coefficient. It is the Laplace operator; It is a decay term; It is the attenuation coefficient; The term "gravitational source" represents the effect of changes in the gravitational source on the gravitational field, and is the intensity of the gravitational source. ,Location and time The function;

[0026] Step 2.2: In the numerical simulation of gravity propagation, the dynamic equations of gravity propagation are solved step by step using an iterative recursive method to realize the modeling of the dynamic field of the intelligent route selection environment.

[0027] Preferably, step S3 specifically includes the following steps:

[0028] Step S3.1: Based on the dynamic mechanical features of the gravitational field and the line information obtained in step S2, construct a multi-dimensional training dataset;

[0029] Step S3.2: Construct a multimodal attention neural network model; Based on the multidimensional training dataset from step S3.1, train and dynamically adapt the multimodal attention neural network model to obtain an intelligent route selection model; The multimodal attention neural network model is a multimodal attention model built based on the Transformer architecture;

[0030] Step S3.3: Apply the intelligent route selection model obtained in step S3.2 to each step of the route selection action decision from the starting point to the end point, and finally generate the optimal railway route selection scheme.

[0031] Preferably, the multimodal attention model includes an encoder module and a decoder module; the encoder module extracts gravitational field features; the decoder module is used for spatial alignment feature extraction and cross-modal feature interaction decision-making; the cross-modal feature interaction decision-making involves introducing a gravitational field-line heterogeneous attention mapping network into the decoder, calculating the association weights between gravitational field features and spatial alignment features, generating a gravitational field state attention feature map, and outputting the optimal alignment action suggestion adapted to the current gravitational field.

[0032] Preferably, the training and dynamic adaptation optimization are based on the gradient descent optimization strategy and the backpropagation algorithm, and the model parameters are iteratively optimized through the following process:

[0033] Define the loss function and calculate the result of the loss function;

[0034] Parameter tuning specifically involves dynamically adjusting the weights and bias parameters of the Transformer encoder and decoder based on the results of the loss function.

[0035] Parallel acceleration specifically refers to using GPU parallel computing technology to process computations during model training.

[0036] This invention discloses a method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization. The method includes constructing a multi-criteria index system, establishing a hierarchical model of the fit degree spatial field, and using the DBSCAN density clustering algorithm to identify high and low fit environment patches; constructing a static field model of the intelligent alignment selection environment; establishing and solving the dynamic equations of gravity propagation to achieve dynamic field modeling of the intelligent alignment selection environment; introducing a Transformer neural network model to complete gravitational field characteristic situational awareness; proposing a dynamic adaptation alignment method; and outputting the optimal railway alignment scheme. This invention integrates dynamic environmental perception and real-time adaptation decision-making, supporting the dynamic optimization needs of engineering alignment selection in complex environments. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the adaptation space field layering model in an embodiment of the present invention;

[0039] Figure 2 This is a schematic diagram of the gravitational field situational awareness and adaptive alignment action mapping mechanism in an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of the optimal railway route selection scheme output by the intelligent route selection model in this embodiment of the invention.

[0041] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0043] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0044] Furthermore, in this invention, descriptions involving "first," "second," etc., are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0045] In this invention, unless otherwise explicitly specified and limited, the terms "connection," "fixed," etc., should be interpreted broadly. For example, "fixed" can mean a fixed connection, a detachable connection, or an integral part; it can mean a mechanical connection or an electrical connection; it can mean a direct connection or an indirect connection through an intermediate medium; it can mean the internal communication of two components or the interaction between two components, unless otherwise explicitly limited. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0046] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0047] A method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization includes the following steps:

[0048] Step S1: Construct a multi-criteria index system, establish a hierarchical model of the fitness spatial field, and use the DBSCAN density clustering algorithm to identify high-fitting and low-fitting environmental patches; construct an intelligent route selection environmental static field model.

[0049] Step S2: Establish and solve the dynamic equations of gravity propagation to realize the dynamic field modeling of the intelligent route selection environment;

[0050] Step S3: Introduce the Transformer neural network model to complete the situational awareness of gravitational field characteristics, propose a dynamic adaptation and alignment method, and output the optimal railway alignment scheme.

[0051] In this preferred embodiment, step S1 specifically includes:

[0052] Step S1.1: For the multi-source heterogeneous data (including geological exploration data, engineering design parameters, and environmental monitoring data) involved in the engineering route selection, the FP-tree association rule mining algorithm is used to screen key fitness indicators (specifically, the algorithm performs deep association analysis on multi-dimensional data by constructing a frequent pattern tree) to efficiently identify core indicators that are strongly correlated with the route selection fitness.

[0053] A multi-criteria comprehensive evaluation model is constructed by combining the Analytic Hierarchy Process (AHP) and the entropy weight method. The AHP is used to assign subjective weights to core indicators based on expert experience, while the entropy weight method is used to calculate the objective weights of core indicators based on data dispersion. Finally, the comprehensive constraint weights of each indicator are obtained through a coupling algorithm, thus achieving the organic integration of subjective judgment and objective data and completing the construction of a multi-criteria indicator system.

[0054] Step S1.2: Standardize the study area into a grid to obtain grid cells. Determine a unified spatial analysis cell (the planar grid size is set based on the accuracy of multi-source data and the scale of the engineering structure, and the vertical direction is matched with the subsequent layer gradient). All subsequent structural tomography-related processing is based on this grid.

[0055] Based on the principle of structural tomography and the isotropic assumption (i.e., the type of structure layout corresponding to a design elevation traversed by the route in any direction is determined only by the traversing elevation and the difference in ground elevation), a layered model of the fit-degree spatial field can be constructed by relying on grid cells and according to the vertical depth gradient to divide the engineering area into several levels, such as the above-ground elevated layer, the surface cover layer, the shallow soil and rock layer, and the deep geological structure layer, even when the route scheme is unknown. In this embodiment, the engineering structure types are divided from top to bottom, including: inaccessible layer, ultra-high bridge layer, high-pier bridge layer, ordinary bridge layer, embankment layer, road cutting layer, shallow buried tunnel layer, deep buried tunnel layer, ultra-deep buried tunnel layer, and inaccessible layer. See details. Figure 1 .

[0056] To transform discrete monitoring data into a continuous spatial distribution, a spatial interpolation technique using Kriging interpolation is employed based on grid cells to obtain a fitness numerical field. Specifically, in this embodiment, the spatial interpolation technique using Kriging interpolation quantifies the spatial autocorrelation of data within the grid using a semi-variogram function, generating a fitness numerical field with minimal estimation error. This visually presents the spatial gradient characteristics of fitness indices at different depth levels, providing continuous data support for subsequent hotspot identification.

[0057] Step S1.3, Combining The hotspot analysis algorithm performs spatial statistical analysis on the fitness numerical field obtained in step S1.2 by calculating the local Getis-Ord statistic for each spatial grid and determining the statistical significance of hotspot areas through Z-score testing. The DBSCAN density clustering algorithm is used for automatic extraction. By setting the minimum number of neighborhood points and the scanning radius, spatially adjacent discrete points that meet the fitness threshold are aggregated into independent patches, effectively eliminating interference from isolated noise points and identifying high-fitness spatial hotspot areas (i.e., high-fitness environmental patches) and low-fitness spatial hotspot areas (i.e., low-fitness environmental patches). For the extracted high-fitness and low-fitness environmental patches, their spatial distribution characteristics (including at least one of area, morphology, and spatial correlation) and multimodal attributes (covering at least one of geological stability, engineering feasibility, and environmental friendliness) are further analyzed, providing multi-dimensional decision-making basis for the dynamic adaptation of subsequent route selection schemes.

[0058] Step S1.4: Traverse the high-fitting and low-fitting environmental patches obtained in step S1.3, define the high-fitting environmental patches as gravity sources and the low-fitting environmental patches as repulsion sources, and construct an intelligent route selection environment static field model.

[0059] In this preferred embodiment, when constructing the static field model of the intelligent route selection environment, the following formula is used to calculate the effect of the first... on the line micro-element at any spatial point P. The force of each patch :

[0060] ;

[0061] Wherein: force When the value is positive, it represents gravity; force. When the value is negative, it represents a repulsive force. This is the gravitational field scaling constant (calibrated according to the engineering scenario). For the first The source intensity of each patch (positive values ​​for high-fit patches and negative values ​​for low-fit patches; the absolute value is positively correlated with the patch fit). This is the sensitivity coefficient of the line element (characterizing the strength of the line's response to patch constraints). For spatial point P to the th The straight-line distance between the centers of the patches; For from the first The center of each patch points to a unit vector at a spatial point P (with the gravitational force in the opposite direction).

[0062] Therefore, a mathematical expression model of attraction and repulsion can be constructed to obtain a static field model of intelligent route selection environment.

[0063] Since environmental changes can easily cause changes in the size, location, and intensity of gravitational sources, further disturbing the static field, step S2 extends the static field into a dynamic field to characterize the propagation process of gravitational increment.

[0064] In this embodiment, step S2 includes the following steps:

[0065] Step 2.1: Characterize the spatiotemporal propagation process of gravitational increments and simulate the impact of environmental changes on field distribution. Specifically:

[0066] Based on the theory of system propagation dynamics, partial differential equations are introduced to characterize the propagation mechanism of gravitational increments, and the dynamic equations for gravitational propagation are constructed as follows:

[0067] ;

[0068] in: It is a time-varying term; For spatial diffusion, D It is the diffusion coefficient (reflecting the rate of gravitational propagation). It is the Laplace operator (representing the gravitational diffusion process in space); It is a decay term; It is the attenuation coefficient, which describes the energy loss of the gravitational field during propagation; The term "gravitational source" represents the effect of changes in the gravitational source on the gravitational field, and is the intensity of the gravitational source. ,Location and time The function;

[0069] Step 2.2: In the numerical simulation of gravity propagation, an iterative recursive method is used to solve the dynamic equations of gravity propagation step by step, realizing the modeling of the dynamic field of the intelligent route selection environment. Specifically:

[0070] Based on the initial gravitational source strength and spatial location information Construct the initial gravitational field state With tiny time steps At intervals, within each time step, based on the gravitational field state of the previous moment and combined with the gravitational source strength... ,Location Dynamically changing, calculating the increment of the gravitational field at the current moment. Substituting these increments into the dynamic equations of gravitational propagation, we update the gravitational field solution at the current moment. Starting from the initial state, through continuous iterative recursion, the propagation process of gravity is accurately simulated in both time and space, and the dynamic evolution of the gravitational field is tracked in real time, thereby revealing the dynamic propagation range, path, and spatiotemporal evolution laws of gravity.

[0071] Thus, the process of establishing and solving the dynamic equations of gravitational propagation was completed, which can track the dynamic evolution of the gravitational field caused by environmental changes in real time and realize intelligent route selection environmental dynamic field modeling.

[0072] After characterizing and deducing the dynamic process of gravitational propagation, step S3 proposes a dynamically adapted alignment method. This method breaks down the entire alignment determination process into several sequential alignment actions, and introduces a Transformer neural network model into each action, which is then trained and iteratively output. Details are as follows:

[0073] Step S3 specifically includes the following steps:

[0074] Step S3.1: Based on the gravitational field dynamic mechanical features and route information obtained in step S2, construct a multi-dimensional training dataset. In this embodiment, the gravitational field dynamic mechanical features include gravitational source strength, spatial location, and gravitational intensity; route information (including historical route selection three-dimensional coordinates and geometric parameters). Here, it is further preferred to integrate the aforementioned gravitational field dynamic mechanical features from the local gravitational field, the global gravitational field, and the temporal gravitational field.

[0075] Step S3.2: Construct a multimodal attention neural network model; based on the multidimensional training dataset from step S3.1, train and dynamically adapt the multimodal attention neural network model to obtain an intelligent route selection model; the multimodal attention neural network model is a multimodal attention model built based on the Transformer architecture. Specifically:

[0076] For details on the construction and feature extraction of multimodal attention neural network models, please refer to [link / reference]. Figure 2 A multimodal attention model is built based on the Transformer architecture. It achieves deep interaction and long-range dependency mining of gravitational field and line features through a three-level feature fusion mechanism. The details are as follows: The multimodal attention model includes an encoder module and a decoder module.

[0077] The encoder module extracts gravitational field features by employing a Transformer encoder and using a multi-head self-attention mechanism to analyze the local and global spatiotemporal distribution patterns of gravitational field mechanical parameters. This captures the long-range dependence features of the gravitational field in the time dimension (history and future) and the spatial dimension (local and global), generating a structured gravitational field feature vector.

[0078] The decoder module is used for spatial line feature extraction and cross-modal feature interaction decision-making, wherein:

[0079] The specific process of spatial line feature extraction is as follows: a Transformer decoder is constructed, an autoregressive attention mechanism is introduced to build a temporal spatial line model, and the high-dimensional latent features of spatial lines are learned autonomously by analyzing the spatiotemporal continuity of historical line sequences.

[0080] Cross-modal feature interaction decision-making involves introducing a gravitational field-line heterogeneous attention mapping network (cross attention layer) into the decoder to calculate the association weights between gravitational field features and spatial alignment features, generate a gravitational field state attention feature map, and output optimal alignment action suggestions adapted to the current gravitational field. Specifically, these include: latent space features, K-dimensional gravitational field sequences, M-dimensional alignment action sequences, and N-dimensional cross feature action sequences.

[0081] In this embodiment, the training and dynamic adaptation optimization are further preferably based on the gradient descent optimization strategy and the backpropagation algorithm, and the model parameters are iteratively optimized through the following process:

[0082] Define loss function Loss The result of calculating the loss function is used to quantify the difference between the model's output of the fixed-line action and the manually optimized scheme, serving as the basis for parameter updates.

[0083] Parameter adjustment specifically involves dynamically adjusting the weights and bias parameters of the Transformer encoder and decoder based on the results of the loss function, thereby optimizing the mapping relationship between gravitational field features and alignment motion.

[0084] Parallel acceleration specifically involves using GPU parallel computing technology to handle computations during model training. In particular, GPU parallel computing technology is used to handle high-complexity computations during model training (complexity O(n²d), where n is the sequence length and d is the embedding dimension), thereby improving training efficiency.

[0085] Through multiple rounds of iterative training, the model acquires the ability to accurately map gravitational field characteristics to adaptive route-setting actions, enabling intelligent optimization of route schemes under dynamic changes in the gravitational field and improving the real-time performance and responsiveness of route selection decisions.

[0086] Step S3.3, Model Application and Alignment Decision, specifically: The intelligent alignment model obtained in step S3.2 is applied to each alignment action decision from the starting point to the end point; after several repeated alignment actions, the optimal alignment scheme from the starting point to the end point is determined, and the optimal railway alignment scheme is output. A dynamic adaptation alignment method can be proposed to realize "alignment by field".

[0087] Application examples:

[0088] The FP-tree association rule mining algorithm was used to identify five key indicators affecting the environmental-line adaptability evaluation: cost, geological disaster risk, stability, carbon emissions, and deviation from the short-straight direction of the line. The weights of these indicators were determined using the analytic hierarchy process (AHP) and entropy weight method.

[0089] The threshold for inaccessible high bridges is defined as 1000m, the threshold for super-high bridges versus high-pier bridges as 150m, the threshold for high-pier bridges versus ordinary bridges as 50m, the threshold for ordinary bridges versus embankments as 20m, the threshold for road cuts versus shallow tunnels as -20m, the threshold for shallow tunnels versus deep tunnels as -400m, and the threshold for deep tunnels versus ultra-deep tunnels as -1000m. Based on these, spatial structural layers are divided from top to bottom. The environmental-line adaptability value of each spatial unit is calculated using the calculation models and weights of each key adaptability index in different structural layers. A continuous spatial distribution field of adaptability is formed using Kriging spatial interpolation technology.

[0090] Introduction Hotspot analysis calculates the Getis-Ord statistic for each spatial cell. Spatial cells with Getis-Ord > 0 and Z > 1.96 (corresponding to 95% confidence level) are defined as high-fit spatial hotspot regions, and the rest are defined as low-fit spatial hotspot regions. The minimum number of neighborhood points for the DBSCAN clustering method is defined as 8, and the scan radius is 180m. High-fit environmental patches and low-fit environmental patches are clustered separately.

[0091] The spatial grid cells are traversed sequentially, and the magnitude and direction of the gravitational force (range 360°) of each spatial grid cell are calculated. The number of gravitational and repulsive sources are 23 and 15, respectively.

[0092] Seventy-nine manual alignment optimization schemes were selected, and the static field model of the intelligent alignment environment was established in step S1, and the dynamic field model of the intelligent alignment environment was established in step S2. The input vector for the Transformer model was generated by combining sliding window and data augmentation methods.

[0093] A Transformer three-layer attention mechanism is used to extract features from the gravitational field state to the optimal trajectory (including the forward direction and step size of the current intersection point).

[0094] The difference between the predicted alignment action and the optimal alignment action of the Transformer is defined as the loss function. The input set is continuously trained under a GPU parallel computing framework to obtain a Transformer network structure that stores the gravitational field state and the optimal alignment action. The gravitational field state is input into the intelligent alignment selection model, which predicts the forward step length and direction at each intersection point, and outputs the optimal railway alignment scheme. See details... Figure 3 .

[0095] By applying the technical solution of this embodiment, by constructing a spatial field of alignment environment adaptability and transforming it into a gravitational field, and combining dynamic evolution mechanism analysis and an adaptive intelligent search strategy, efficient optimization design of railway lines under dense constraints is achieved, improving the adaptability of alignment decision-making and providing a new technical approach for railway alignment selection in complex environments.

[0096] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization, characterized in that, Includes the following steps: Step S1: Construct a multi-criteria index system, establish a hierarchical model of the fitness spatial field, and use the DBSCAN density clustering algorithm to identify high-fitting and low-fitting environmental patches; construct an intelligent route selection environmental static field model. Step S2: Establish and solve the dynamic equations of gravity propagation to realize the dynamic field modeling of the intelligent route selection environment; Step S3: Introduce the Transformer neural network model to complete the situational awareness of gravitational field characteristics, propose a dynamic adaptation and alignment method, and output the optimal railway alignment scheme. Step S1 specifically includes: Step S1.1: For the multi-source heterogeneous data involved in the engineering route selection, the FP-tree association rule mining algorithm is used to screen key fitness indicators and identify core indicators that are strongly correlated with the route selection fitness; the analytic hierarchy process and the entropy weight method are combined to construct a multi-criteria comprehensive evaluation model to obtain the subjective and objective weights of the core indicators; the comprehensive constraint weights of each core indicator are obtained through a coupling algorithm to complete the construction of the multi-criteria indicator system. Step S1.2: Standardize the study area into grids to obtain grid cells; based on the structural tomography principle and the isotropic assumption, and relying on the grid cells, perform layering of the engineering area according to the vertical depth gradient to construct a layered model of the fitness spatial field; and use the spatial interpolation technique of Kriging interpolation based on the grid cells to obtain the fitness numerical field. Step S1.3, Combining The hotspot analysis algorithm performs spatial statistical analysis on the fitness numerical field obtained in step S1.2 to obtain high-fitting environmental patches and low-fitting environmental patches. Step S1.4: Traverse the high-fitting and low-fitting environmental patches obtained in step S1.3, define the high-fitting environmental patches as gravity sources and the low-fitting environmental patches as repulsion sources, and construct a static field model of intelligent route selection environment. In step S1.4, when constructing the static field model of the intelligent route selection environment, the following formula is used to calculate the impact of the line element at any spatial point P on the first... The force of each patch : ； Wherein: force When the value is positive, it represents gravity; force. When the value is negative, it represents a repulsive force. The gravitational field proportionality constant; For the first The source intensity of each patch; The sensitivity coefficient of the line element; For spatial point P to the th The straight-line distance between the centers of the patches; For from the first The unit vector pointing from the center of each patch to a spatial point P.

2. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 1, characterized in that, The specific layering process in step S1.2 is as follows: the engineering structure types are divided from top to bottom, including: inaccessible layer, ultra-high bridge layer, high pier bridge layer, ordinary bridge layer, embankment layer, road cutting layer, shallow buried tunnel layer, deep buried tunnel layer, ultra-deep buried tunnel layer and inaccessible layer.

3. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 1, characterized in that, In step S1.3, the DBSCAN density clustering algorithm is used for automatic extraction. By setting the minimum number of neighborhood points and the scanning radius, discrete points that are spatially adjacent and whose fitness meets the threshold condition are aggregated into independent patches. High fitness spatial hotspot areas, i.e., high fitness environment patches, and low fitness spatial hotspot areas, i.e., low fitness environment patches, are identified.

4. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 3, characterized in that, The spatial distribution characteristics and multimodal attributes of the identified high-fitting and low-fitting environmental patches were analyzed respectively. The spatial distribution characteristics included at least one of area, morphology and spatial correlation. The multimodal attributes included at least one of geological stability, engineering feasibility and environmental friendliness.

5. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 1, characterized in that, Step S2 includes the following steps: Step 2.1: Based on the theory of system propagation dynamics, partial differential equations are introduced to characterize the propagation mechanism of gravitational increments, and the dynamic equations of gravitational propagation are constructed as follows: ； in: It is a time-varying term; For spatial diffusion, It is the diffusion coefficient. It is the Laplace operator; It is a decay term; It is the attenuation coefficient; The term "gravitational source" represents the effect of changes in the gravitational source on the gravitational field, and is the intensity of the gravitational source. ,Location and time The function; Step 2.2: In the numerical simulation of gravity propagation, the dynamic equations of gravity propagation are solved step by step using an iterative recursive method to realize the modeling of the dynamic field of the intelligent route selection environment.

6. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 5, characterized in that, Step S3 specifically includes the following steps: Step S3.1: Based on the dynamic mechanical features of the gravitational field and the line information obtained in step S2, construct a multi-dimensional training dataset; Step S3.2: Construct a multimodal attention neural network model; Based on the multidimensional training dataset from step S3.1, train and dynamically adapt the multimodal attention neural network model to obtain an intelligent route selection model; The multimodal attention neural network model is a multimodal attention model built based on the Transformer architecture; Step S3.3: Apply the intelligent route selection model obtained in step S3.2 to each step of the route selection decision from the starting point to the end point, and finally generate the optimal railway route selection scheme.

7. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 6, characterized in that, The multimodal attention model includes an encoder module and a decoder module; the encoder module extracts gravitational field features. The decoder module is used for spatial line feature extraction and cross-modal feature interaction decision-making; Cross-modal feature interaction decision-making involves introducing a gravitational field-line heterogeneous attention mapping network into the decoder, calculating the association weights between gravitational field features and spatial alignment features, generating a gravitational field state attention feature map, and outputting optimal alignment action suggestions adapted to the current gravitational field.

8. The method for intelligent railway alignment selection in complex environments using gravitational field dynamics modeling and adaptive guidance optimization as described in claim 6, characterized in that, Training and dynamic adaptation optimization are based on gradient descent optimization strategy and backpropagation algorithm, and the model parameters are iteratively optimized through the following process: Define the loss function and calculate the result of the loss function; Parameter tuning specifically involves dynamically adjusting the weights and bias parameters of the Transformer encoder and decoder based on the results of the loss function. Parallel acceleration specifically refers to using GPU parallel computing technology to process computations during model training.

Images