Highway change design method driven by large language model

Abstract

The invention relates to the technical field of road engineering, in particular to a large language model-driven road change design method, which comprises the following steps of: 1) performing rough selection by using a large language model and a prompt project, and selecting a corridor zone; and 2) performing optimization through an improved artificial potential field MAPE algorithm. According to the method provided by the invention, by applying a natural language processing technology, automatic generation of a preliminary corridor belt design scheme is realized, the manpower and time cost of a route pre-selection stage is saved, a large number of alternative schemes can be produced to assist design, and the optimization effect is also higher than that of a relatively mature genetic algorithm; the method is a successful exploration of a large language model in intelligent road line selection application.

Description

Technical Field

[0001] The present invention relates to the technical field of road engineering, and in particular to a highway relocation design method driven by a large language model. Background Art

[0002] During the construction of railways, it is inevitable that they will pass through the existing road network, cut off the existing roads and form many intersections. In order to restore the traffic capacity of the road network, it is necessary to relocate the roads. Since it is necessary to make a three-dimensional intersection with the railway at this time, it is usually necessary to meet multiple constraints such as azimuth, intersection angle and linear geometry to ensure smooth connection with the existing roads. Otherwise, it is easy to cause traffic safety hazards. This is a difficult situation in road relocation and has a large application scenario. Due to the shortage of urban land and the scarcity of high-quality transportation corridors, road relocation is prone to problems such as chaotic construction sequence and engineering safety contradictions. In addition, the actual relocation project involves a large number of roads. It is difficult to find the optimal solution when the time cost is limited and multiple schemes are involved. Therefore, it is necessary to use intelligent line selection methods for auxiliary design. At present, domestic and foreign scholars have carried out a lot of research in the field of intelligent road line selection. Among them, heuristic algorithms are widely used due to their high efficiency and high robustness: for example, heuristic algorithms such as genetic algorithms, ant colony optimization algorithms, and electromagnetic-like elements are used to optimize the line shape of a plane or longitudinal section; some scholars have proposed the Hybrid A* algorithm based on a heuristic navigation algorithm, which can perform near-global optimization exploration in three-dimensional space, generate routes that retain accurate horizontal, vertical, and cross-sectional geometric information and conform to spatial constraints, and provide a new perspective for intelligent highway line selection; some scholars have proposed a line selection model based on game theory that takes into account the risks and costs of dangerous goods transportation, and uses a heuristic shortest path algorithm to solve it; some scholars have used an improved motion planning algorithm to sample horizontal intersections; some scholars have used discrete algorithms to solve multiple approximate optimal path solutions in different spaces as alternative solutions in the decision-making process to find the optimal corridor.

[0003] Since heuristic algorithms may fall into local optimality and fail to guarantee the optimal solution, some scholars have used BIM and GIS-related technologies to accurately calculate and solve cross-sectional parameters for linear optimization. For example, some scholars have verified and tested the feasibility of BIM technology for cross-sectional calculation in three-dimensional road linear optimization. However, in order to reduce the computational cost, many strategies are also considered to simplify the cross-sectional parameters, such as simplifying the cross-sectional shape or slope, and using predefined decision spaces to reduce the search space.

[0004] It can be seen that although there has been a wealth of research in the field of intelligent highway route selection, there are few studies on highway relocation, especially at railway intersections: some scholars have developed a railway intersection optimization model and used the distance transform (DT) algorithm to solve it, but the speed and accuracy of the solution are greatly limited; some scholars have proposed a custom Dijkstra algorithm to search for auxiliary highways, but due to the limitations of the algorithm, a series of highway constraints are not fully considered during the search. It can be seen that this type of highway relocation design problem currently lacks a more mature solution.

[0005] In recent years, the relevant technologies of generative artificial intelligence (AI Generated Content, AIGC), including generative adversarial networks (GAN) and large pre-trained deep learning models, namely large language models (LLM), have developed rapidly. Especially after the large models have multimodal capabilities, the application prospects and development space of large models have been significantly improved, and they have gradually demonstrated excellent capabilities in natural language processing tasks and various professional fields. Therefore, some professional fields have also begun to conduct research and application on them: for example, LLM can be applied to robot path planning to improve the efficiency of traditional search algorithms and help programming performance drive; in addition, a series of large model frameworks such as TrafficGPT and TrafficSafetyGPT have emerged in the transportation field, which can be applied to the planning and design of transportation infrastructure projects, project management, operation and maintenance, disaster prevention and other aspects, and provide data drive for multi-faceted decision-making tasks in complex multimodal transportation systems; some scholars have proposed a multimodal transportation large model MT-GPT based on the hierarchical "points, lines and surfaces" of the traffic topology structure, and discussed the application prospects of large models in transportation planning and management, infrastructure construction and other aspects. It can be seen that although there are many studies on LLM in the field of urban transportation, there has been no relevant research in the field of road engineering, especially route design. Summary of the invention

[0006] The content of the present invention is to provide a highway relocation design method driven by a large language model, which realizes the automatic generation of preliminary corridor design schemes by using natural language processing technology, saves manpower and time costs in the route pre-selection stage, and can produce a large number of alternative schemes to assist in the design.

[0007] A highway relocation design method driven by a large language model according to the present invention comprises the following steps:

[0008] Step 1) Use a large language model and prompt engineering to perform rough selection and select corridors;

[0009] Step 2) Optimize using the improved artificial potential field MAPE algorithm.

[0010] Preferably, in step 1), the specific steps are:

[0011] 1.1) Pre-training large language models;

[0012] 1.2) Pass prompt words to the big language model, including data, constraints and requirements;

[0013] 1.3) Use prompt words to fine-tune the path and then export the path points;

[0014] 1.4) Process the path points produced by the large language model into a series of control points;

[0015] 1.5) All corridor control points are used as the gravity sources of artificial potential fields.

[0016] Preferably, in step 2), the specific steps are:

[0017] 2.1) Establish a particle model and initialize the velocity, direction of movement and position;

[0018] 2.2) Select the repulsive force source and target range of the artificial potential field according to actual conditions;

[0019] 2.3) Initialize the attraction-repulsion gain coefficient and use the particle model to perform MAPF;

[0020] 2.4) Determine whether it is within the range of attraction and repulsion. If so, update the particle velocity, position, and motion direction information according to the resultant force and the current state, and proceed to the next step; if not, maintain the particle motion state according to the current state, and proceed to the next step;

[0021] 2.5) Determine whether the constraint conditions are met. If yes, proceed to the next step; if not, return to step 2.3);

[0022] 2.6) Determine whether it has passed all control points. If yes, proceed to the next step; if not, return to step 2.4);

[0023] 2.7) Use the least squares method to fit the curve segment to make it conform to the standard.

[0024] As a preferred method, in the improved artificial potential field MAPE algorithm, the control points output by the large language model in the rough selection stage of the corridor are used as the source of the gravitational potential field, and the obstacles, restricted areas and vegetation areas of the actual project are regarded as the source of the repulsive potential field. The path planning is iteratively optimized by dynamically and real-time adjusting the motion state of the particles in the gravitational potential field.

[0025] As a preferred method, in the improved artificial potential field MAPE algorithm, the gravitational potential field formula U att (q) and the gravitational function F att (q) is composed of a series of vectors, as shown below:

[0026]

[0027] U att,n =0, when ρ(q,q contral,n )>ρ n

[0028]

[0029] F att,n =0, when ρ(q,q contral,n )>ρ n

[0030] Where: α is the gravitational gain coefficient, and different control points use the same value; q is the current spatial position of the particle; q contral,n is the spatial position of the nth control point; ρ(q,q contral,n ) is the Euclidean distance from the current point to the nth control point; ρ n is the target range of the nth control point, that is, the minimum distance that will affect the movement of the particle; U att,n (q), F att,n (q) are the gravitational potential field and gravitational force of the nth control point respectively;

[0031] The repulsive potential field formula U rep (q) and the repulsion function F rep (q) is as follows:

[0032]

[0033] Where: β n is the repulsive force gain coefficient of the nth repulsive force source; q is the current spatial position of the particle; q obs is the spatial position of the obstacle; q goal is the spatial position of the target endpoint; ρ n is the target range of the nth repulsive force source; ρ(q,q obs ) is the Euclidean distance from the current point to the obstacle; U rep,n (q), F rep,n (q) are the repulsive potential field and repulsive force of the nth control point respectively.

[0034] As a preferred method, the objective function of the improved artificial potential field MAPE algorithm is:

[0035] The total cost of route construction is selected as the objective function, as shown in the following formula:

[0036] C c =C e +C b +C l +C v

[0037] Where: C c is the total construction cost; C e C is the earthwork cost; b is the bridge fee; C l The construction costs related to the length; C v Cost of vegetation destruction;

[0038] Earthwork cost C e The calculation method is as follows:

[0039]

[0040] Where: when a0=1, it represents a full excavation section; when a1=1, it represents a full filling section; when a2=1, it represents a half-filled and half-excavation section; A wi , A ti is the area of ​​cut and fill in the ith cross section; p t 、p w L is the unit price of filling and excavation; Ei is the length of the i-th cross section;

[0041] Bridge fee C b The calculation method is as follows:

[0042]

[0043] Where: n b is the number of bridges along the entire line; p bi is the unit construction cost of the i-th bridge; l bi is the length of the i-th bridge; C ai is the construction cost of the abutment of the i-th bridge;

[0044] Construction costs related to length C l As shown below:

[0045] C l =p li ×L li

[0046] Where: p li The unit land acquisition cost for different areas that the route passes through; L li is the route length;

[0047] The vegetation destruction cost is calculated by multiplying the vegetation area obtained by supervised classification by the unit cost of the vegetation, and the vegetation area is obtained according to the green area in the raster file passed by the route, as shown in the following formula:

[0048] C v =p v×Σn grid ×s grid

[0049] Where: C v is the vegetation destruction cost; p v is the unit cost of regional vegetation; n grid is the number of vegetation grids that the route passes through; s grid is the area size of the grid.

[0050] As a preference, the relevant constraints of the improved artificial potential field MAPE algorithm are:

[0051] a) Geometric constraints

[0052] The plane geometric constraints usually include: combined constraints of curves and straight lines, minimum curve radius constraints, minimum curve length constraints, and maximum slope constraints. The above plane geometric constraints are expressed as φ i (P HPI )≤0,i=0,1,...,N+1, where i represents the i-th plane intersection point and N represents the total number of plane intersection points;

[0053] The longitudinal section geometric constraints include: maximum longitudinal slope constraint, minimum vertical curve radius constraint, and minimum slope length constraint. The above longitudinal section geometric constraints are expressed as ψ j (P VPI )≤0,j=0,1,...,M+1, where j represents the jth longitudinal section intersection point and M represents the total number of longitudinal section intersection points;

[0054] b) Building or structure constraints

[0055] Expressed as U R and U F are the set of route areas and the set of all restricted areas respectively;

[0056] c) Railway crossing restrictions

[0057] The railway crossing constraint of a route is expressed as:

[0058] θ min ≤θ≤90°

[0059] L p ≤L pmin

[0060] Δh≤|h road -h rail

[0061] Where: θ min is the minimum value of the railway crossing angle; L p and L pminare the distance between the bridge piers and the minimum distance in the crossable area; Δh is the vertical clearance of the crossing; h road and h rail Represents the height of highway and railway respectively.

[0062] Preferably, in step 1), the corridor is selected by taking into account building or structure constraints and railway crossing constraints.

[0063] The present invention firstly utilizes the core theoretical framework of large language model and prompt engineering, introduces typical examples in the training stage to guide the model to learn and reproduce these behavior patterns, and refines complex problems into a series of manageable subtasks, so as to realize the efficient generation of a large number of preliminary screening schemes for corridor belt design using only natural language. After obtaining the rough selection results of the first stage, the multi-control points artificial potential field algorithm (MAPF) based on the robot path planning algorithm is used to further optimize the rough selection results of the corridor belt, including constraint checking, circular curve fitting, etc. The method proposed in the present invention realizes the automatic generation of preliminary corridor belt design schemes by using natural language processing technology, saves the manpower and time costs in the route pre-selection stage, can produce a large number of alternative schemes to assist in design, and is also higher than the more mature genetic algorithm in terms of optimization effect. It is a successful exploration of the application of large language model in intelligent highway route selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A flowchart of a highway relocation design method driven by a large language model in an embodiment;

[0065] Figure 2 A schematic diagram of the MAPF algorithm in the embodiment;

[0066] Figure 3 A schematic diagram of preprocessing of terrain and ecological data in the embodiment;

[0067] Figure 4 This is a schematic diagram for visual comparison of the solutions in the embodiments;

[0068] Figure 5 It is a schematic diagram for visualizing and comparing the longitudinal sections in the embodiments. DETAILED DESCRIPTION

[0069] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are only for explaining the present invention and are not intended to limit it.

[0070] Example

[0071] like Figure 1As shown, this embodiment provides a highway relocation design method driven by a large language model, which includes the following steps:

[0072] Step 1) Use a large language model and prompt engineering to perform rough selection and select corridors;

[0073] Step 2) Optimize using the improved artificial potential field MAPE algorithm.

[0074] Large language model corridor selection method based on prompt engineering principle

[0075] Since the algorithm used by the untrained large language model is relatively simple and difficult to fine-tune and optimize for specific actual situations, it is possible to achieve a similar fine-tuning effect by prompting engineering-related principles, thereby improving the ability of the large model to handle professional tasks.

[0076] Prompt Engineering is a technique used to guide and optimize pre-trained models, control and guide the output or behavior of large language models, and influence the response of large models by enforcing rules, automating processes, etc., and ensuring that the generated output has a specific quality and quantity. The key lies in understanding how to construct effective prompts to stimulate the best performance of large models, including understanding the capabilities of the model, clear definition of task requirements, and in-depth mastery of language structure and patterns. Even by adjusting the content, format, or tone of the prompt, the quality and relevance of the results generated by the large model can be significantly affected.

[0077] The core of prompt engineering is the divisive algorithm, that is, "divide and conquer" or chain of thought (CoT), which decomposes complex problems into a series of sub-problems and encourages large models to reason about intermediate steps, such as Figure 1 Specific methods include self-consistent demonstration, supervised fine-tuning, and generative knowledge, that is, providing examples of correct behavior during training to guide the large model to imitate these examples, allowing the large model to continuously learn from examples with derivation steps demonstrated by humans, or organizing question-and-answer samples in the form of facts and knowledge before asking targeted questions.

[0078] The corridor selection in this embodiment can start from a simple path planning problem, and use a combination of prompt engineering and fine-tuning to allow the large model to gradually understand geographic information data, professional terms and more complex constraints, and break down the final task into gradually progressive sub-tasks, so as to finally complete the more complex task.

[0079] Since the prompt words are different in different specific tasks and scenarios, the types of base models used may also be different, and the output results of the large model are uncertain, prompt words need to be used in a timely manner to reduce the possibility of hallucinations in the large model. The following uses a relatively simple numerical case as an example to demonstrate the application method of corridor selection of large language models based on prompt engineering. The base model used is qwen2:72b open sourced by Alibaba, and the LORA method is used to fine-tune the relevant professional terms.

[0080] By utilizing the idea of ​​​​division and smelting of prompt engineering, a more complex corridor selection problem is started from a simple path planning problem. In the pre-training stage, the large model is allowed to slowly learn from the steps of human demonstration, gradually understand the meaning of relevant geographic information data and the constraints of the final case, and after the pre-training stage, the result of the complex problem can be obtained by further describing the actual case in natural language. Among them, the content of pre-training can be adjusted according to the actual problem, including adding obstacles, changing different algorithms, adding other constraints, etc. After the pre-training stage, the selection results of the corridor can be quickly generated using only natural language, and multiple solutions can be generated or the output style can be modified later without secondary training.

[0081] In step 1), the specific steps are:

[0082] 1.1) Pre-training large language models;

[0083] 1.2) Pass prompt words to the big language model, including data, constraints and requirements;

[0084] 1.3) Use prompt words to fine-tune the path and then export the path points;

[0085] 1.4) Process the path points produced by the large language model into a series of control points;

[0086] 1.5) All corridor control points are used as the gravity sources of artificial potential fields.

[0087] Optimization algorithm based on multi-control point artificial potential field (MAPF)

[0088] The artificial potential field method is a method of using a virtual force field that is widely used in mobile robot path planning. It can use repulsion to make the robot avoid restricted areas and obstacles in advance before a certain distance. The problem of unreachable local minimum target points is successfully solved by introducing a piecewise function of the relative distance between the current point and the target point in the traditional repulsion function, thereby improving the accuracy of the robot's path planning. This embodiment introduces multiple gravitational targets on the basis of the artificial potential field method, and adds relevant constraints on highway design, thereby improving the multi-control points artificial potential field algorithm (MAPF) applicable to the field of highway intelligent line selection. The resultant force F acting on the particle model is total is the vector sum of the repulsive force of the obstacle or constraint and the gravitational force of all control points in front. The simple schematic diagram is as follows Figure 2 shown.

[0089] The main parameters of the model are shown in Table 1, where the gain coefficient can be changed according to different sources. For example, the repulsion gain coefficient can be enlarged in strict restricted areas, while smaller repulsion gain coefficients are used in areas with crossing costs such as vegetation areas. The target range ρ needs to be flexibly adjusted according to the actual situation of the source of attraction or repulsion.

[0090] Table 1 Main parameters of the model

[0091] project Numeric <![CDATA[Initial velocity v0]]> 20m / s Sampling frequency / time t 0.025s <![CDATA[Initial gravitational gain coefficient α0]]> 1 <![CDATA[Initial repulsive gain coefficient β0]]> 0.8 Target range distance ρ /

[0092] In step 2), the specific steps are:

[0093] 2.1) Establish a particle model and initialize the velocity, direction of movement and position;

[0094] 2.2) Select the repulsive force source and target range of the artificial potential field according to actual conditions;

[0095] 2.3) Initialize the attraction-repulsion gain coefficient and use the particle model to perform MAPF;

[0096] 2.4) Determine whether it is within the range of attraction and repulsion. If so, update the particle velocity, position, and motion direction information according to the resultant force and the current state, and proceed to the next step; if not, maintain the particle motion state according to the current state, and proceed to the next step;

[0097] 2.5) Determine whether the constraint conditions are met. If yes, proceed to the next step; if not, return to step 2.3);

[0098] 2.6) Determine whether it has passed all control points. If yes, proceed to the next step; if not, return to step 2.4);

[0099] 2.7) Use the least squares method to fit the curve segment to make it conform to the standard.

[0100] The core of the improved artificial potential field MAPE algorithm is mainly the attractive and repulsive potential fields and attractive and repulsive functions. att (q) and the gravitational function F att (q) is composed of a series of vectors, as shown below:

[0101]

[0102] U att,n =0, when ρ(q,q contral,n )>ρ n

[0103]

[0104] F att,n =0, when ρ(q,q contral,n )>ρ n

[0105] Where: α is the gravitational gain coefficient, and different control points use the same value; q is the current spatial position of the particle; q contral,n is the spatial position of the nth control point; ρ(q,q contral,n ) is the Euclidean distance from the current point to the nth control point; ρ n is the target range of the nth control point, that is, the minimum distance that will affect the movement of the particle; U att,n (q), F att,n (q) are the gravitational potential field and gravitational force of the nth control point respectively;

[0106] The repulsive potential field formula U rep (q) and the repulsion function F rep (q) is as follows:

[0107]

[0108] Where: β n is the repulsive force gain coefficient of the nth repulsive force source; q is the current spatial position of the particle; q obs is the spatial position of the obstacle; q goal is the spatial position of the target endpoint; ρ n is the target range of the nth repulsive force source; ρ(q,q obs ) is the Euclidean distance from the current point to the obstacle; U rep,n (q), F rep,n (q) are the repulsive potential field and repulsive force of the nth control point respectively.

[0109] In the improved artificial potential field MAPE algorithm, the control points output by the large language model in the rough selection stage of the corridor are used as the source of the gravitational potential field, and the obstacles, restricted areas and vegetation areas of the actual project are regarded as the source of the repulsive potential field. The path planning is iteratively optimized by dynamically and in real time adjusting the motion state of the particles in the gravitational potential field.

[0110] The objective function of the improved artificial potential field MAPE algorithm is:

[0111] The selection of the objective function is directly related to the algorithm efficiency and route optimization results. Since the construction cost is the most considered factor in engineering practice, the total cost of route construction is selected as the objective function, as shown in the following formula:

[0112] C c =C e +C b +C l +C v

[0113] Where: C c is the total construction cost; C e C is the earthwork cost; b is the bridge fee; C l The construction costs related to the length; C v Cost of vegetation destruction;

[0114] Earthwork cost C e The calculation method is as follows:

[0115]

[0116] Where: when a0=1, it represents a full excavation section; when a1=1, it represents a full filling section; when a2=1, it represents a half-filled and half-excavation section; A wi , A ti is the area of ​​cut and fill in the ith cross section; p t 、p w L is the unit price of filling and excavation; Ei is the length of the i-th cross section;

[0117] Bridge fee C b The calculation method is as follows:

[0118]

[0119] Where: n b is the number of bridges along the entire line; p bi is the unit construction cost of the i-th bridge; l bi is the length of the i-th bridge; C ai is the construction cost of the abutment of the i-th bridge;

[0120] Construction costs related to length C l As shown below:

[0121] C l =p li ×L li

[0122] Where: p li The unit land acquisition cost for different areas that the route passes through; L li is the route length;

[0123] Since the vegetation type in the region is single, the vegetation destruction cost is calculated by multiplying the vegetation area obtained by supervised classification by the unit cost of the vegetation. The vegetation area is obtained based on the green area in the raster file that the route passes through, as shown in the following formula:

[0124] C v =p v ×Σn grid ×s grid

[0125] Where: C v is the vegetation destruction cost; p v is the unit cost of regional vegetation; n grid is the number of vegetation grids that the route passes through; s grid is the area size of the grid.

[0126] The relevant constraints of the improved artificial potential field MAPE algorithm are:

[0127] a) Geometric constraints

[0128] The plane geometric constraints usually include: combined constraints of curves and straight lines, minimum curve radius constraints, minimum curve length constraints, and maximum slope constraints. The above plane geometric constraints are expressed as φ i (P HPI )≤0,i=0,1,...,N+1, where i represents the i-th plane intersection point and N represents the total number of plane intersection points;

[0129] The longitudinal section geometric constraints include: maximum longitudinal slope constraint, minimum vertical curve radius constraint, and minimum slope length constraint. The above longitudinal section geometric constraints are expressed as ψ j (P VPI )≤0,j=0,1,...,M+1, where j represents the jth longitudinal section intersection point and M represents the total number of longitudinal section intersection points;

[0130] b) Building or structure constraints

[0131] Expressed as U R and UF are the set of route areas and the set of all restricted areas respectively;

[0132] c) Railway crossing restrictions

[0133] The railway crossing constraint of a route is expressed as:

[0134] θ min ≤θ≤90°

[0135] L p ≤L pmin

[0136] Δh≤|h road -h rail |

[0137] Where: θ min is the minimum value of the railway crossing angle; L p and L pmin are the distance between the bridge piers and the minimum distance in the crossable area; Δh is the vertical clearance of the crossing; h road and h rail Represents the height of highway and railway respectively.

[0138] The selection of corridors takes into account the constraints of buildings or structures and railway crossings.

[0139] Case Study

[0140] Taking a railway intersection in an actual high-speed railway project in East China as an example, the method proposed in this embodiment is verified again. Since the highway in this case crosses the railway by passing under it, the problem can be transformed into a multi-scheme decision problem for the route passing through different must-pass areas, that is, the schemes passing through different bridge piers can be compared to obtain the best scheme, and compared with the manual route and the current more mature genetic algorithm results.

[0141] like Figure 3 As shown in the figure, the contour data and DOM data are first preprocessed, mainly by processing the contour data through ArcGIS and other software to obtain a digital elevation model (DEM), and then the ecological data is processed by supervised classification of the DOM data to obtain a vegetation distribution map.

[0142] Then, in the preliminary selection stage, multiple routes were quickly and batch-generated using a large model as references for the corridors in the first phase, with only a few must-pass areas, namely different railway intersections, added as constraints. The optimal route in each must-pass area was optimized using the MAPF algorithm, with the attraction source being the control points of the corridor in the first phase, and the repulsion source being the vegetation area and the railway roadbed and bridge pier area. Based on the actual situation of the project, the main parameters for calculation are: the width of a single-lane fourth-class highway is 3.5m, and the approximate equivalent earthwork cost is assumed to be 20￥ / m 3 , road surface cost 660￥ / m, vegetation cost 1500￥ / m 2 .

[0143] Since the length of the route will lead to a significant increase in engineering costs, only the best schemes among the first three intersections are selected for comparison. The route length, excavation and filling, ecological data and other indicators are calculated and compared with the manual route. The results are shown in Table 2. The optimal intelligent route selection scheme is Scheme 1, which selects a closer intersection than the manual scheme and causes less vegetation damage costs. It saves 4.8% of the engineering cost compared with the manual route scheme and is better than the final result of the genetic algorithm iteration. Among them, Scheme 2 selects the same intersection as the manual scheme, and the route bypasses more vegetation areas. Therefore, although the total length of the route is longer than the manual scheme, the significant optimization of the vegetation area makes the engineering cost optimized by 2.87% compared with the manual scheme, and the optimization result is also better than the genetic algorithm. The main parameters of the genetic algorithm are: crossover probability is 0.8, mutation probability is 0.2, initial population size is 100, and number of iterations is 500.

[0144] Table 2 Comparison of schemes

[0145]

[0146] like Figure 4 As shown in the figure, the artificial route and the optimized route are exported and embedded into the topographic map (part a) and the vegetation distribution base map obtained by supervised classification (part b), so that the overall plan can be visually compared. Green, yellow and white represent trees, cultivated land and original roads respectively. It can be clearly seen that compared with the artificial route results, Scheme 1 and Scheme 2 pass through fewer green areas (forests), significantly reducing the vegetation cost caused by ecological damage.

[0147] like Figure 5 Shown is a visual comparison of longitudinal sections of all schemes and the manual scheme.

[0148] This embodiment has the following advantages:

[0149] (1) An intelligent line selection model based on LLM and MAPF algorithm optimization was constructed. In the first stage, a large number of corridor rough selection results can be efficiently generated using only natural language, saving a certain amount of manpower and time costs. Since different typical actual cases can be learned to guide the large model to learn its behavior patterns, and complex problems can be decomposed and the results can be continuously adjusted in the process, this method has strong adaptability and flexibility and can be applied to a wide range of practical scenarios. In addition, as the capabilities of the large model continue to increase, the application prospects and potential of this method will continue to increase.

[0150] (2) An artificial potential field algorithm with multiple control points based on a robot path planning algorithm was proposed. By changing the definition of repulsive and attractive objects, it was applied to the field of intelligent line selection and optimized the corridor results generated by the large model. Multiple constraints can be considered by using different attraction and repulsion gain coefficients and target ranges. The algorithm was compared with a more mature genetic algorithm and had certain advantages in speed and optimization effect.

[0151] (3) This embodiment was applied in a railway intersection of a high-speed rail project in East China. The results showed that the optimal solution for intelligent route selection selected the optimal intersection, reduced more vegetation damage, saved 4.8% of the engineering cost compared with the manual route solution, and was better than the genetic algorithm; the second solution of this method selected the same intersection as the manual solution, and also saved 2.78% of the engineering cost. Therefore, it has shown certain results in actual projects and is an effective exploration of the large language model in the field of highway intelligent route selection.

[0152] (4) The method proposed in this embodiment may be difficult to deal with more complex actual cases, and the relevant potential of the large language model needs to be further explored. Subsequent research can also continue to explore its huge potential: using RAG (retrieval enhanced generation) to build a professional database to enhance the professional knowledge capabilities of the large model; designing and building a professional AI agent (intelligent body) to enable it to have the ability to think independently and call tools, so as to effectively deal with given complex tasks. In addition, the vegetation destruction in this study is calculated in the engineering cost, and subsequent research can also consider the optimization of multiple objectives such as engineering cost and green ecology.

[0153] The present invention and its embodiments are described schematically above, and the description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by it and designs a structural method and an embodiment similar to the technical solution without creativity without departing from the purpose of the invention, they shall all fall within the protection scope of the present invention.

Claims

1. A highway relocation design method driven by a large language model, characterized by: The following steps are involved: Step 1) Use a large language model and prompt engineering to perform rough selection and select corridors; Step 2) Optimize using the improved artificial potential field MAPE algorithm.

2. The highway relocation design method driven by a large language model according to claim 1 is characterized by: In step 1), the specific steps are: 1.1) Pre-training large language models; 1.2) Pass prompt words to the big language model, including data, constraints and requirements; 1.3) Use prompt words to fine-tune the path and then export the path points; 1.4) Process the path points produced by the large language model into a series of control points; 1.5) All corridor control points are used as the gravity sources of artificial potential fields.

3. The highway relocation design method driven by a large language model according to claim 2 is characterized by: In step 2), the specific steps are: 2.1) Establish a particle model and initialize the velocity, direction of movement and position; 2.2) Select the repulsive force source and target range of the artificial potential field according to actual conditions; 2.3) Initialize the attraction-repulsion gain coefficient and use the particle model to perform MAPF; 2.4) Determine whether it is within the range of attraction and repulsion. If so, update the particle velocity, position, and motion direction information according to the resultant force and the current state, and proceed to the next step; if not, maintain the particle motion state according to the current state, and proceed to the next step; 2.5) Determine whether the constraint conditions are met. If yes, proceed to the next step; if not, return to step 2.3); 2.6) Determine whether it has passed all control points. If yes, proceed to the next step; if not, return to step 2.4); 2.7) Use the least squares method to fit the curve segment to make it conform to the standard.

4. The highway relocation design method driven by a large language model according to claim 3 is characterized by: In the improved artificial potential field MAPE algorithm, the control points output by the large language model in the rough selection stage of the corridor are used as the source of the gravitational potential field, and the obstacles, restricted areas and vegetation areas of the actual project are regarded as the source of the repulsive potential field. The path planning is iteratively optimized by dynamically and in real time adjusting the motion state of the particles in the gravitational potential field.

5. The highway relocation design method driven by a large language model according to claim 4 is characterized by: In the improved artificial potential field MAPE algorithm, the gravitational potential field formula U att (q) and the gravitational function F att (q) is composed of a series of vectors, as shown below: U att,n =0, then ρ(q,q contral,n )>r n F att,n =0, if ρ(q,q contral,n )>ρ n Where: α is the gravity gain coefficient, and different control points use the same value; q is the current spatial position of the particle; q contral,n is the spatial position of the nth control point; ρ(q,q contral,n ) is the Euclidean distance from the current point to the nth control point; ρ n is the target range of the nth control point, that is, the minimum distance that will affect the movement of the particle; U att,n (q), F att,n (q) are the gravitational potential field and gravitational force of the nth control point respectively; The repulsive potential field formula U rep (q) and the repulsion function F rep (q) is as follows: Where: β n is the repulsive force gain coefficient of the nth repulsive force source; q is the current spatial position of the particle; q obs is the spatial position of the obstacle; q goal is the spatial position of the target endpoint; ρ n is the target range of the nth repulsive force source; ρ(q,q obs ) is the Euclidean distance from the current point to the obstacle; U rep,n (q), F rep,n (q) are the repulsive potential field and repulsive force of the nth control point respectively.

6. The highway relocation design method driven by a large language model according to claim 5 is characterized by: The objective function of the improved artificial potential field MAPE algorithm is: The total cost of route construction is selected as the objective function, as shown in the following formula: C c =C e +C b +C l +C v Where: C c is the total construction cost; C e C is the earthwork cost; b is the bridge fee; C l The construction costs related to the length; C v Cost of vegetation destruction; Earthwork cost C e The calculation method is as follows: <h2 style=";text-align:left;direction:ltr"> <h2 style=";text-align:left;direction:ltr"> +a2p<h2 style=";text-align:left;direction:ltr"> t <h2 style=";text-align:left;direction:ltr"> A<h2 style=";text-align:left;direction:ltr"> wi <h2 style=";text-align:left;direction:ltr"> L<h2 style=";text-align:left;direction:ltr"> Ei <h2 style=";text-align:left;direction:ltr"> +a2p<h2 style=";text-align:left;direction:ltr"> w <h2 style=";text-align:left;direction:ltr"> A<h2 style=";text-align:left;direction:ltr"> ti <h2 style=";text-align:left;direction:ltr"> L<h2 style=";text-align:left;direction:ltr"> Ei <h2 style=";text-align:left;direction:ltr"> , Where: when a0=1, it represents a full excavation section; when a1=1, it represents a full filling section; when a2=1, it represents a half-filled and half-excavation section; A wi , A ti is the area of ​​cut and fill in the ith cross section; p t 、p w L is the unit price of filling and excavation; Ei is the length of the i-th cross section; Bridge fee C b The calculation method is as follows: Where: n b is the number of bridges along the entire line; p bi is the unit construction cost of the i-th bridge; l bi is the length of the i-th bridge; C ai is the construction cost of the abutment of the i-th bridge; Construction costs related to length C l As shown below: C l =p li ×L li Where: p li The unit land acquisition cost for different areas that the route passes through; L li is the route length; The vegetation destruction cost is calculated by multiplying the vegetation area obtained by supervised classification by the unit cost of the vegetation, and the vegetation area is obtained according to the green area in the raster file passed by the route, as shown in the following formula: C v =p v ×∑n grid ×s grid Where: C v is the vegetation destruction cost; p v is the unit cost of regional vegetation; n grid is the number of vegetation grids that the route passes through; s grid is the area size of the grid.

7. The highway relocation design method driven by a large language model according to claim 6 is characterized by: The relevant constraints of the improved artificial potential field MAPE algorithm are: a) Geometric constraints The plane geometric constraints usually include: combined constraints of curves and straight lines, minimum curve radius constraints, minimum curve length constraints, and maximum slope constraints. The above plane geometric constraints are expressed as φ i (P HPI )≤0,i=0,1,...,N+1, where i represents the i-th plane intersection point and N represents the total number of plane intersection points; The longitudinal section geometric constraints include: maximum longitudinal slope constraint, minimum vertical curve radius constraint, and minimum slope length constraint. The above longitudinal section geometric constraints are expressed as ψ j (P VPI )≤0,j=0,1,...,M+1, where j represents the jth longitudinal section intersection point and M represents the total number of longitudinal section intersection points; b) Building or structure constraints Expressed as U R and U F are the set of route areas and the set of all restricted areas respectively; c) Railway crossing restrictions The railway crossing constraint of a route is expressed as: i min ≤θ≤90° L p ≤L pmin Δh≤|h road -h rail Where: θ min is the minimum value of the railway crossing angle; L p and L pmin are the distance between the bridge piers and the minimum distance in the crossable area; Δh is the vertical clearance of the crossing; h road and h rail Represents the height of highway and railway respectively.

8. The highway relocation design method driven by a large language model according to claim 7 is characterized by: In step 1), the corridor is selected taking into account building or structure constraints and railway crossing constraints.

Images