Transformer-based transmission line tower ranking-to-ground distance rapid verification method, system, computer readable storage medium and computing device

Abstract

This invention belongs to the field of power transmission line engineering design technology, specifically relating to a method, system, computer-readable storage medium, and computing device for rapid verification of the distance between power transmission line tower placement and the ground based on Transformer. S1, construct a labeled multi-feature-dimensional parametric simulation dataset; S2, construct a fused deep neural network model, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network; S3, train and optimize the fused deep neural network model in step S2 based on the labeled multi-feature-dimensional parametric simulation dataset in step S1; S4, input the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model, and output the feasibility prediction results. This achieves efficient and accurate verification of tower parameters. It also improves the model's adaptability to complex terrain and multi-obstacle scenarios.

Description

Technical Field

[0001] This invention belongs to the field of power transmission line engineering design technology, specifically relating to a method, system, computer-readable storage medium, and computing device for rapid verification of the distance between the tower arrangement and the ground of power transmission lines based on Transformer. Background Technology

[0002] The placement of transmission line towers is a crucial aspect of power system planning, directly impacting construction costs and operational safety. Traditional tower placement design typically employs manual calculations or point-by-point verification methods based on physical models to verify parameters such as ground clearance and span distance. For large-span terrain (e.g., terrain span ≥ 500m), the verification process involves numerous complex calculations, including conductor sag calculations and ground clearance detection, which presents the following main drawbacks: First, the computational efficiency is low: the ordinary for loop verification method takes about 30ms to verify the ground distance for large-span terrain, and the verification time is linearly related to the number and length of the segments to be verified, which is difficult to meet the needs of large-scale route planning.

[0003] Second, limited data processing capabilities: Existing methods are unable to effectively process high-dimensional data containing complex terrain data and multiple obstacle information, resulting in insufficient accuracy of verification results.

[0004] Third, poor scalability: When multiple tower placement scenarios need to be verified simultaneously, efficient concurrent calculation cannot be achieved, and the efficiency of comparing multiple solutions is low.

[0005] To address the aforementioned issues, those skilled in the art have attempted to introduce machine learning methods to improve verification efficiency. However, problems remain, such as insufficient model generalization ability and inadequate fusion processing of different types of data (e.g., sequence data and scalar data), leading to low verification accuracy and robustness. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method, system, computer-readable storage medium and computing device for rapid verification of the distance between the tower arrangement and the ground based on Transformer, so as to achieve efficient and accurate verification of the tower arrangement scenario to be verified.

[0007] The technical solution adopted by this invention to solve its technical problem is: a method for rapid verification of the distance between the tower arrangement and the ground of transmission lines based on Transformer, including the following steps: S1, based on the digital elevation model of the target transmission line corridor and the preset engineering constraints, generates multiple tower placement scenarios through parametric random sampling, performs physical simulation calculations on each tower placement scenario, and automatically generates feasible or infeasible labels according to the preset safety distance threshold, thus constructing a labeled multi-feature dimension parametric simulation dataset. S2. Construct a fusion deep neural network model, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network. The Transformer encoder encodes the elevation sequence and the sequence of objects crossing the terrain segment to generate a first feature vector representing the global terrain features. The fully connected neural network encodes the scalar parameters of the tower height and span at both ends to generate a second feature vector representing the scalar semantics of the engineering. The feature fusion module fuses the first and second feature vectors to generate a joint feature vector. The classifier module of another fully connected neural network receives the joint feature vector and processes it through at least one fully connected layer, outputting a probability value representing the feasibility of the tower placement scenario. S3, Based on the labeled multi-feature dimension parameterized simulation dataset in step S1, train and optimize the fused deep neural network model in step S2; S4 inputs the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model and outputs the feasibility prediction results.

[0008] Furthermore, step S1 specifically includes the following steps: S101, based on the digital elevation model of the target transmission line corridor, obtains a continuous three-dimensional terrain profile of the target transmission line corridor; presets a set of engineering constraint parameters, including the feasible range of spans. Feasible range of tower height Ground distance threshold and the sag coefficient K; S102, Perform physical simulation calculations and label assignment for a single tower placement scenario, specifically including the following steps: S1021, randomly extract terrain segments within a preset range, obtain the elevation sequence of the current terrain segment, and record the position and height information of all intersecting objects within the current terrain segment; S1022, Randomly select a starting point, and based on a random sampling interval. , Extract the elevation point sequence of the current terrain segment. ,in Determined by the sampling resolution; simultaneously, the positions of all intersecting objects within the current terrain segment are recorded. and height ; S1023, randomly generate the tower heights at both ends of the current terrain segment; randomly generate the tower heights at both ends of the current terrain segment. and , ≤ H ≤ ; S1024, based on the equation of the flat parabola, calculates the spatial shape curve of the conductor according to the horizontal position of any point between the two towers relative to one of the towers, the span, the tower heights of the two towers, and the elevation of the topographic points where the two towers are located. S1025, extract multiple discrete points on the spatial morphology curve of the conductor along the span direction, and calculate the vertical distance from each discrete point to the ground. And the minimum distance from each discrete point to the top of each intersection. ; and compare it with the safe distance threshold for the corresponding voltage level; when And all If feasible, assign a feasible label; otherwise, assign an infeasible label. S103, repeat step S102 to perform physical simulation calculations and label assignments for multiple tower placement scenarios, and then construct a labeled multi-feature dimension parameterized simulation dataset.

[0009] Furthermore, after step S103, step S104 is also included: by dynamically adjusting the parameter range during the sampling generation process, the ratio of the number of samples with feasible labels to the number of samples with infeasible labels in the labeled multi-feature dimension parameterized simulation dataset is kept within a preset range.

[0010] Furthermore, step S2 specifically includes the following steps: S201, using the Transformer encoder to process high-order sequences of terrain segments The sequence of objects to be crossed is standardized and concatenated to form a multi-feature dimension input sequence; the global context features of the multi-feature dimension input sequence are extracted through the multi-head self-attention mechanism in the Transformer encoder; the hidden states of the entire multi-feature dimension input sequence output by the last layer of the Transformer encoder are labeled with the position state through global average pooling to generate the first feature vector representing the global terrain features. S202, regarding the height of the towers at both ends and and gear ratio The scalar parameters are standardized and then input into a fully connected neural network for encoding to generate a second feature vector representing the semantics of the engineering scalar. S203, the first feature vector and the second feature vector are fused through the feature fusion module to generate a joint feature vector; S204 inputs the joint feature vector into another fully connected neural network classifier and obtains the probability value representing the feasibility of the ranking scheme through the Sigmoid function.

[0011] Furthermore, in step S3, based on the labeled multi-feature dimension parameterized simulation dataset in step S1, the fused deep neural network model in step S2 is trained and optimized using the cross-entropy loss function and the Adam optimization algorithm.

[0012] Furthermore, in step S3, overfitting is prevented by using an early stopping strategy and regularization techniques.

[0013] Furthermore, step S4 specifically includes the following steps: S401 converts the trained and optimized fused deep neural network model into a high-performance inference format and encapsulates it as a microservice interface or embedded library. S402, the tower placement scenario to be verified is processed through steps S201, S202 and S203, and then input into the trained and optimized fused deep neural network model, and outputs the feasibility prediction result.

[0014] Furthermore, in step S4, the verification time for a single tower ranking scenario is less than 10ms; for multiple tower ranking scenarios input in batches, concurrent calculation is supported, and the average verification time for a single tower ranking scenario is less than 0.5ms.

[0015] The Transformer-based system for rapid verification of the distance between the towers and the ground in transmission line positions adopts a Transformer-based method for rapid verification of the distance between the towers and the ground in transmission line positions. It includes a multi-feature dimension parameterized simulation dataset construction unit, a deep neural network model fusion unit, a training and optimization unit, and a verification unit. The multi-feature-dimensional parametric simulation dataset generation unit is configured to generate multiple tower placement scenarios through parametric random sampling based on the digital elevation model of the target transmission line corridor and preset engineering constraints, and to perform physical simulation calculations on each tower placement scenario. It automatically generates feasible or infeasible labels according to preset safety distance thresholds to construct a labeled multi-feature-dimensional parametric simulation dataset. The fusion deep neural network model unit is configured to construct a fusion deep neural network model, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network. The Transformer encoder encodes the elevation sequence and the sequence of objects crossing the terrain segment to generate a first feature vector representing global terrain features. The fully connected neural network encodes the scalar parameters of the tower height and span at both ends to generate a second feature vector representing engineering scalar semantics. The feature fusion module fuses the first and second feature vectors to generate a joint feature vector. The classifier module of another fully connected neural network receives the joint feature vector and processes it through at least one fully connected layer, outputting a probability value representing the feasibility of the tower placement scenario. The training and optimization unit is configured to train and optimize the fused deep neural network model in step S2 based on the labeled multi-feature dimension parameterized simulation dataset in step S1. The verification unit is configured to input the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model, and output the feasibility prediction result.

[0016] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a Transformer-based method for rapid verification of the distance between the tower arrangement and the ground.

[0017] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a Transformer-based method for rapid verification of the distance between the tower arrangement and the ground of transmission lines.

[0018] Compared with existing technologies, the beneficial effects of this invention are: This invention provides a method, system, computer-readable storage medium, and computing device for rapid verification of the distance between transmission line tower positions and the ground based on Transformer, achieving efficient and accurate verification of the tower position scenarios to be verified. It improves the model's adaptability to complex terrain and multi-obstacle scenarios. Specifically: First, improve data processing capabilities and enhance model robustness. Deep encoding of complex terrain and obstacle data using a Transformer encoder improves the model's adaptability to complex scenarios. By designing a well-balanced training dataset and maintaining the ratio of feasible to infeasible labeled samples within a predetermined range in the labeled multi-feature-dimensional parameterized simulation dataset, the model's robustness in real-world applications can be improved.

[0019] Second, a leap in verification efficiency. By transforming the complex physical computation process into a single forward propagation of a neural network, the time for a single inference can be reduced to milliseconds or even sub-milliseconds. More importantly, the model supports batch input, enabling parallel instantaneous verification of tens of thousands of samples, increasing overall throughput by several orders of magnitude and meeting the stringent requirements of optimization algorithms for real-time feedback.

[0020] Third, it provides a core engine for automated tower placement optimization. This invention can be integrated into dynamic programming and other automated tower placement optimization processes as a rapid evaluation function for scheme feasibility. This enables the optimization algorithm to efficiently explore the huge solution space, quickly eliminating a large number of invalid schemes that do not meet the safety distance requirement, thereby significantly accelerating the search process for the globally optimal or near-optimal placement scheme.

[0021] Fourth, the method is highly versatile and reliable in accuracy. Training on a large-scale dataset generated based on strict physical rules ensures that the model learns universally applicable verification logic that aligns with engineering realities. As long as the training data covers the terrain and parameter range of the target engineering scenario, the model can provide high-precision and rapid verification within that scenario, avoiding the need to develop dedicated verification procedures for each specific project.

[0022] Fifth, it reduces reliance on specialized software. By encapsulating the verification function as an independent model or service, it eliminates dependence on specific large-scale commercial CAD or simulation software, reducing deployment costs and lowering the barrier to entry, making it easier to integrate applications on cloud or edge computing devices. Attached Figure Description

[0023] Figure 1 This is a flowchart of the present invention; Figure 2 This is a schematic diagram of the structure of the deep neural network model fused in this invention; Figure 3 These are samples with feasible labels generated from the dataset of this embodiment of the invention; Figure 4 These are samples with infeasible labels generated from the dataset of this embodiment of the invention; Figure 5 This is a graph showing the accuracy of the validation set in an embodiment of the present invention. Figure 6 This is a graph showing the accuracy of the test set in an embodiment of the present invention. Figure 7 These are sample diagrams illustrating the application and verification of this invention in hilly areas. Figure 8 This is a sample diagram demonstrating the application of this invention in mountainous terrain. Detailed Implementation

[0024] The following is in conjunction with the appendix Figures 1-8The invention is further illustrated by the embodiments.

[0025] A rapid verification method for the distance to ground of transmission line towers based on Transformer, such as Figure 1 As shown, the steps include: S1, based on the digital elevation model of the target transmission line corridor and the preset engineering constraints, generates multiple tower placement scenarios through parametric random sampling, performs physical simulation calculations on each tower placement scenario, and automatically generates feasible or infeasible labels according to the preset safety distance threshold, thus constructing a labeled multi-feature dimension parametric simulation dataset.

[0026] Specifically, step S1 includes the following steps: S101, based on the digital elevation model of the target transmission line corridor, obtains a continuous three-dimensional terrain profile of the target transmission line corridor; presets a set of engineering constraint parameters, including the feasible range of spans. Feasible range of tower height Ground distance threshold And the sag coefficient K.

[0027] Among them, the ground distance threshold The distance is determined based on voltage level and land cover type, including distance to the ground, distance to trees, and distance to buildings.

[0028] From the digital elevation model (DEM) obtained by lidar scanning of a 500kV transmission line project in a mountainous area, longitudinal profile data of the line center was extracted, with points taken every 5m. The sag coefficient K was set to... .

[0029] Define the set of engineering constraint parameters, including: feasible range of spans. Feasible range of tower height According to the "Design Code for 110kV~750kV Overhead Transmission Lines" (GB 50545-2010), the following threshold values ​​for distance to the ground are set: 14m for residential areas, 11m for non-residential areas, 8.5m for areas with difficult transportation, 9.0m for buildings, 7m for trees, 14m for railways, 6m for overhead contact lines, 14m for highways, 6m for crossing power lines, 6.5m for cableways, and 11.5m for rivers.

[0030] S102, Perform physical simulation calculations and label assignment for a single tower placement scenario, specifically including the following steps: S1021, within the preset gear range Randomly extract terrain segments, obtain the elevation sequence of the current terrain segment, and record the position and elevation information of all intersecting objects within the current terrain segment; S1022, Randomly select a starting point, and based on a random sampling interval. , Extract the elevation point sequence of the current terrain segment. ,in Determined by the sampling resolution. Simultaneously, the positions of all intersecting features within the current terrain segment are recorded. and height .

[0031] The objects that cross include highways, railways, rivers, and buildings.

[0032] S1023 is used to randomly generate the heights of the two towers at both ends of the current terrain segment; Randomly generate the tower heights at both ends of the current terrain segment. and , ≤ H ≤ ; when and When they are equal, simulate a high-speed rail tower scenario; when... and The simulation of unequal high-speed rail tower scenarios is conducted.

[0033] S1024, based on the equation of the flat parabola, calculates the spatial shape curve of the conductor according to the horizontal position of any point between the two towers relative to one of the towers, the span, the tower heights of the two towers, and the elevation of the topographic points where the two towers are located. Taking towers A and B at both ends as examples: The equation of the horizontal parabola is: ; in, ; In the formula, Let A be the horizontal position of any point between towers A and B at both ends relative to tower A. K is the sag coefficient; For gear distance; Let B be the height of the tower. The elevation of the topographic point where tower B is located; Let A be the height of tower A; The elevation of the topographic point where tower A is located; S1025, extract multiple discrete points on the spatial morphology curve of the conductor along the span direction. The distance between any two adjacent discrete points is 5m. Calculate the vertical distance from each discrete point to the ground. And the minimum distance from each discrete point to the top of each intersection. ; and compare it with the safe distance threshold for the corresponding voltage level; when And all At that time, assign feasible labels, using Indicate; otherwise, assign an infeasible label, using This indicates that a total of 50,000 samples were generated. For example... Figure 3 As shown, the samples with feasible labels generated from the example dataset meet the requirements for conductor-to-ground distance. Figure 4 As shown, the sample generated from the example dataset has an infeasibility label, and the guide wire intersects with the terrain line.

[0034] S103, repeat step S102 to perform physical simulation calculations and label assignments for multiple tower placement scenarios, and then construct a labeled multi-feature dimension parameterized simulation dataset. The step S103 is followed by step S104, which involves dynamically adjusting the parameter range during the sampling generation process to keep the ratio of feasible labeled samples to infeasible labeled samples in the labeled multi-feature dimension parameterized simulation dataset within a preset range. Preset range meets the condition: Number of samples with feasible labels : Number of samples with infeasible labels = Where P is a real number greater than 1, such as 2.

[0035] S2, as Figure 2 As shown, a fusion deep neural network model is constructed, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network. The Transformer encoder encodes the elevation sequence and the sequence of objects crossing the terrain segment to generate a first feature vector representing the global terrain features. The fully connected neural network encodes the scalar parameters of the tower height and span at both ends to generate a second feature vector representing the scalar semantics of the engineering. The feature fusion module fuses the first and second feature vectors to generate a joint feature vector. The classifier module of another fully connected neural network receives the joint feature vector and processes it through at least one fully connected layer to output a probability value representing the feasibility of the tower placement scenario.

[0036] Both the first eigenvector and the second eigenvector are one-dimensional vectors.

[0037] Specifically, step S2 includes the following steps: S201, using the Transformer encoder to process high-order sequences of terrain segments The sequence of objects to be crossed is standardized and concatenated to form a multi-feature-dimensional input sequence. The global context features of the multi-feature-dimensional input sequence are extracted through the multi-head self-attention mechanism in the Transformer encoder. The hidden states of the entire multi-feature-dimensional input sequence output by the last layer of the Transformer encoder are labeled with the position state through global average pooling (GAP) to generate the first feature vector representing the global terrain features.

[0038] The terrain and overpass sequence encoding channels employ a 6-layer Transformer encoder, with each layer containing 8 attention heads and a hidden layer dimension of 512. The input consists of a terrain array and overpass data, and the output is a 512-dimensional feature vector.

[0039] In this process, the multi-feature dimension input sequence is first embedded into a high-dimensional space through a linear projection layer, and then a learnable or fixed positional encoding is added. S202, regarding the height of the towers at both ends and and gear ratio The scalar parameters are standardized and then input into a fully connected neural network for encoding to generate a second feature vector representing the semantics of the engineering scalar. The engineering scalar parameter encoding channel adopts a 3-layer fully connected neural network (256-128-64), inputting scalar parameters such as tower height and spacing, and outputting a 64-dimensional feature vector.

[0040] S203, the first feature vector and the second feature vector are fused through the feature fusion module to generate a joint feature vector; Feature fusion module: It concatenates 512-dimensional sequence features and 64-dimensional scalar features to form 576-dimensional fused features.

[0041] S204 inputs the joint feature vector into another fully connected neural network classifier and obtains the probability value representing the feasibility of the ranking scheme through the Sigmoid function.

[0042] A 3-layer fully connected neural network (256-128-2) is used to output binary classification results (feasible and infeasible) and regression results (minimum distance to the ground).

[0043] S3. Based on the labeled multi-feature dimension parameterized simulation dataset from step S1, train and optimize the fused deep neural network model from step S2.

[0044] In step S3, based on the labeled multi-feature dimension parameterized simulation dataset in step S1, the fused deep neural network model in step S2 is trained and optimized using the cross-entropy loss function and the Adam optimization algorithm.

[0045] Preferably, the labeled multi-feature dimension parameterized simulation dataset includes a training set, a validation set, and a test set, wherein the training set accounts for 80%, the validation set accounts for 10%, the test set accounts for 10%, and the validation set accuracy is set to 95%.

[0046] Batch size: 1024; Learning rate: 0.001 (using cosine annealing strategy); Optimizer: AdamW; Loss function: Cross-entropy loss + L1 regression loss; Training rounds: 500 rounds, early stop patience value: 10 rounds; Training process: The training lasted approximately 72 hours; like Figure 5 , 6 As shown in the figure, the final validation accuracy reaches 96.2%, as can be seen from the validation set accuracy curve and the test set accuracy curve.

[0047] Preferably, in step S3, overfitting is prevented by using an early stopping strategy and regularization techniques.

[0048] S4 inputs the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model and outputs the feasibility prediction results.

[0049] Specifically, step S4 includes the following steps: S401 converts the trained and optimized fused deep neural network model into a high-performance inference format (such as ONNX, TensorRT format) and encapsulates it as a microservice interface or embedded library. S402, the tower placement scenario to be verified, after being processed through steps S201, S202, and S203, is then input into the trained and optimized fused deep neural network model, which outputs the feasibility prediction result. The trained and optimized fused deep neural network model returns the feasibility probability in milliseconds. .

[0050] The preset threshold is 0.5. If the condition is met, it is deemed feasible; otherwise, it is deemed infeasible.

[0051] In step S4, the verification time for a single tower ranking scenario is less than 10ms; for multiple tower ranking scenarios input in batches, concurrent calculation is supported, and the average verification time for a single tower ranking scenario is less than 0.5ms.

[0052] Figure 7 The test results for hilly areas were presented, with an accuracy rate of 98.3%. Figure 8The test results in mountainous terrain were presented, with an accuracy rate of 97.7%. This demonstrates that the invention has strong robustness and can adapt to different terrain conditions.

[0053] To quantify the efficiency advantages of this invention, the following comparative experiments were conducted.

[0054] Control group (traditional dynamic programming method): A precise geometric calculation script based on Python was used to simulate the traditional verification process. For a terrain cross-section with a span of 1000 meters, a for loop was used to calculate the distance to the ground point by point and make a judgment, with an average time of about 1 second. The verification time increased approximately linearly with the number of cross-section points (i.e., span).

[0055] Experimental group (method of this invention): Inference was performed using the model trained in Example 1. On the same hardware, the average time for validating a single sample was approximately 5 milliseconds. The advantage was even more pronounced when performing batch validation: for batch inference with 50,000 samples input at once, the total time was approximately 4,500 milliseconds, with an average validation time of only 0.09 milliseconds per sample.

[0056] The Transformer-based system for rapid verification of the distance between the towers and the ground in transmission line positions adopts a Transformer-based method for rapid verification of the distance between the towers and the ground in transmission line positions. It includes a multi-feature dimension parameterized simulation dataset construction unit, a deep neural network model fusion unit, a training and optimization unit, and a verification unit. The multi-feature-dimensional parametric simulation dataset generation unit is configured to generate multiple tower placement scenarios through parametric random sampling based on the digital elevation model of the target transmission line corridor and preset engineering constraints, and to perform physical simulation calculations on each tower placement scenario. It automatically generates feasible or infeasible labels according to preset safety distance thresholds to construct a labeled multi-feature-dimensional parametric simulation dataset. The fusion deep neural network model unit is configured to construct a fusion deep neural network model, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network. The Transformer encoder encodes the elevation sequence and the sequence of objects crossing the terrain segment to generate a first feature vector representing global terrain features. The fully connected neural network encodes the scalar parameters of the tower height and span at both ends to generate a second feature vector representing engineering scalar semantics. The feature fusion module fuses the first and second feature vectors to generate a joint feature vector. The classifier module of another fully connected neural network receives the joint feature vector and processes it through at least one fully connected layer, outputting a probability value representing the feasibility of the tower placement scenario. The training and optimization unit is configured to train and optimize the fused deep neural network model in step S2 based on the labeled multi-feature dimension parameterized simulation dataset in step S1. The verification unit is configured to input the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model, and output the feasibility prediction result.

[0057] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a Transformer-based method for rapid verification of the distance between the tower arrangement and the ground.

[0058] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a Transformer-based method for rapid verification of the distance between the tower arrangement and the ground of transmission lines.

[0059] The specific embodiments described are preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent changes made to the structure, shape, and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for rapid verification of the distance to ground of transmission line tower arrangement based on Transformer, characterized in that, Including the following steps: S1, based on the digital elevation model of the target transmission line corridor and the preset engineering constraints, generates multiple tower placement scenarios through parametric random sampling, performs physical simulation calculations on each tower placement scenario, and automatically generates feasible or infeasible labels according to the preset safety distance threshold, thus constructing a labeled multi-feature dimension parametric simulation dataset. S2. Construct a fusion deep neural network model, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network. The Transformer encoder encodes the elevation sequence and the sequence of objects crossing the terrain segment to generate a first feature vector representing the global terrain features. The fully connected neural network encodes the scalar parameters of the tower height and span at both ends to generate a second feature vector representing the scalar semantics of the engineering. The feature fusion module fuses the first and second feature vectors to generate a joint feature vector. The classifier module of another fully connected neural network receives the joint feature vector and processes it through at least one fully connected layer, outputting a probability value representing the feasibility of the tower placement scenario. S3, Based on the labeled multi-feature dimension parameterized simulation dataset in step S1, train and optimize the fused deep neural network model in step S2; S4 inputs the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model and outputs the feasibility prediction results.

2. The method for rapid verification of the distance to the ground of transmission line tower arrangement based on Transformer as described in claim 1, characterized in that, Step S1 specifically includes the following steps: S101, based on the digital elevation model of the target transmission line corridor, obtains a continuous three-dimensional terrain profile of the target transmission line corridor; presets a set of engineering constraint parameters, including the feasible range of spans. Feasible range of tower height Ground distance threshold and the sag coefficient K; S102, Perform physical simulation calculations and label assignment for a single tower placement scenario, specifically including the following steps: S1021, randomly extract terrain segments within a preset range, obtain the elevation sequence of the current terrain segment, and record the position and height information of all intersecting objects within the current terrain segment; S1022, Randomly select a starting point, and based on a random sampling interval. , Extract the elevation point sequence of the current terrain segment. ,in Determined by the sampling resolution; simultaneously, the positions of all intersecting objects within the current terrain segment are recorded. and height ; S1023, randomly generate the tower heights at both ends of the current terrain segment; randomly generate the tower heights at both ends of the current terrain segment. and , ≤ H ≤ ; S1024, based on the equation of a flat parabola, calculates the spatial shape curve of the conductor according to the horizontal position of any point between the two towers relative to one of the towers, the span, the tower heights of the two towers, and the elevation of the topographic points where the two towers are located. S1025, extract multiple discrete points on the spatial morphology curve of the conductor along the span direction, and calculate the vertical distance from each discrete point to the ground. And the minimum distance from each discrete point to the top of each intersection. ; and compare it with the safe distance threshold for the corresponding voltage level; when And all If feasible, assign a feasible label; otherwise, assign an infeasible label. S103, repeat step S102 to perform physical simulation calculations and label assignments for multiple tower placement scenarios, and then construct a labeled multi-feature dimension parameterized simulation dataset.

3. The method for rapid verification of the distance to ground of transmission line tower arrangement based on Transformer as described in claim 2, characterized in that, The step S103 is followed by step S104, which involves dynamically adjusting the parameter range during the sampling generation process to keep the ratio of feasible labeled samples to infeasible labeled samples in the labeled multi-feature dimension parameterized simulation dataset within a preset range.

4. The method for rapid verification of the distance to the ground of transmission line tower arrangement based on Transformer as described in claim 1, characterized in that, Step S2 specifically includes the following steps: S201, using the Transformer encoder to process high-order sequences of terrain segments The sequence of objects to be crossed is standardized and concatenated to form a multi-feature dimension input sequence; the global context features of the multi-feature dimension input sequence are extracted through the multi-head self-attention mechanism in the Transformer encoder; the hidden states of the entire multi-feature dimension input sequence output by the last layer of the Transformer encoder are labeled with the position state through global average pooling to generate the first feature vector representing the global terrain features. S202, regarding the height of the towers at both ends and and gear ratio The scalar parameters are standardized and then input into a fully connected neural network for encoding to generate a second feature vector representing the semantics of the engineering scalar. S203, the first feature vector and the second feature vector are fused through the feature fusion module to generate a joint feature vector; S204 inputs the joint feature vector into another fully connected neural network classifier and obtains the probability value representing the feasibility of the ranking scheme through the Sigmoid function.

5. The method for rapid verification of the distance to the ground of transmission line tower arrangement based on Transformer as described in claim 1, characterized in that, In step S3, based on the labeled multi-feature dimension parameterized simulation dataset in step S1, the fused deep neural network model in step S2 is trained and optimized using the cross-entropy loss function and the Adam optimization algorithm.

6. The method for rapid verification of the distance to the ground of transmission line tower arrangement based on Transformer as described in claim 1, characterized in that, In step S3, overfitting is prevented by using an early stopping strategy and regularization techniques.

7. The method for rapid verification of the distance to ground of transmission line tower arrangement based on Transformer as described in claim 1, characterized in that, Step S4 specifically includes the following steps: S401 converts the trained and optimized fused deep neural network model into a high-performance inference format and encapsulates it as a microservice interface or embedded library. S402, the tower placement scenario to be verified is processed through steps S201, S202 and S203, and then input into the trained and optimized fused deep neural network model, and outputs the feasibility prediction result.

8. The method for rapid verification of the distance to ground of transmission line tower arrangement based on Transformer as described in claim 1, characterized in that, In step S4, the verification time for a single tower ranking scenario is less than 10ms; for multiple tower ranking scenarios input in batches, concurrent calculation is supported, and the average verification time for a single tower ranking scenario is less than 0.5ms.

9. A rapid verification system for the distance to ground of transmission line towers based on Transformer, characterized in that, The method for rapid verification of the distance between the tower arrangement and the ground based on Transformer as described in any one of claims 1-8 includes a multi-feature dimension parameterized simulation dataset construction unit, a deep neural network model fusion unit, a training and optimization unit, and a verification unit. The multi-feature-dimensional parametric simulation dataset generation unit is configured to generate multiple tower placement scenarios through parametric random sampling based on the digital elevation model of the target transmission line corridor and preset engineering constraints, and to perform physical simulation calculations on each tower placement scenario. It automatically generates feasible or infeasible labels according to preset safety distance thresholds to construct a labeled multi-feature-dimensional parametric simulation dataset. The fusion deep neural network model unit is configured to construct a fusion deep neural network model, which includes a Transformer encoder, a fully connected neural network, a feature fusion module, and a classifier module of another fully connected neural network. The Transformer encoder encodes the elevation sequence and the sequence of objects crossing the terrain segment to generate a first feature vector representing global terrain features. The fully connected neural network encodes the scalar parameters of the tower height and span at both ends to generate a second feature vector representing engineering scalar semantics. The feature fusion module fuses the first and second feature vectors to generate a joint feature vector. The classifier module of another fully connected neural network receives the joint feature vector and processes it through at least one fully connected layer, outputting a probability value representing the feasibility of the tower placement scenario. The training and optimization unit is configured to train and optimize the fused deep neural network model in step S2 based on the labeled multi-feature dimension parameterized simulation dataset in step S1. The verification unit is configured to input the parameters of the tower placement scenario to be verified into the trained and optimized fused deep neural network model, and output the feasibility prediction result.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method for rapid verification of the distance between the transmission line towers and the ground based on Transformer as described in any one of claims 1-8.

11. A computing device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for rapid verification of the distance between the transmission line towers and the ground based on Transformer as described in any one of claims 1-8.

Images