CN-121998210-A - Cable path planning method based on deep learning and multi-objective optimization

CN121998210ACN 121998210 ACN121998210 ACN 121998210ACN-121998210-A

Abstract

The invention relates to the technical field of path planning, in particular to a cable path planning method based on deep learning and multi-objective optimization, which comprises the following steps of adopting a multi-scale semantic segmentation deep learning model based on remote sensing image data, combining auxiliary accessibility labeling samples, identifying the types of all ground objects, and constructing a ground object accessibility confidence field with continuous spatial distribution; constructing a confidence coupling path cost function, giving punishment weight to a region with low ground feature confidence and giving passing rewards weight to a region with high confidence, generating the confidence coupling path cost function for path searching, constructing an improved heuristic function, and executing path searching by adopting a multi-objective optimization algorithm to obtain a cable path scheme. According to the method and the device, not only is the history total cost of the current node considered, but also the average confidence risk in the direction from the current node to the target is predicted, so that prospective avoidance is realized.

Inventors

ZHAO QIANG
ZHOU WEIHANG
XU YUN
WANG ZHIKAI
ZHOU QIYUE
REN YANJIN

Assignees

无锡市广盈电力设计有限公司

Dates

Publication Date: 20260508
Application Date: 20251217

Claims (10)

1. The cable path planning method based on deep learning and multi-objective optimization is characterized by comprising the following steps of: S1, based on remote sensing image data, a multi-scale semantic segmentation deep learning model is adopted, and an auxiliary accessibility labeling sample is combined, so that each ground object type is identified, accessibility confidence level of each pixel point for cable construction is calculated, and a ground object accessibility confidence field with continuous spatial distribution is constructed; s2, defining a confidence coefficient adjustment factor based on the feature reachable confidence field, and introducing the confidence coefficient adjustment factor into a multi-objective cost modeling process of path optimization to construct a confidence coupling path cost function, wherein the cost function fuses a path length, construction cost, terrain gradient and confidence coefficient inverse factor, and gives punishment weight to a region with low feature confidence coefficient and a region with high confidence coefficient and gives passing rewards weight to a region with high feature confidence coefficient to generate the confidence coupling path cost function for path search; and S3, constructing an improved heuristic function by taking the confidence coupling path cost function as a path search input, and executing path search by adopting a multi-objective optimization algorithm to obtain a cable path scheme meeting the requirements of construction continuity, risk avoidance and path smoothness.
2. The cable path planning method based on deep learning and multi-objective optimization according to claim 1, wherein the S1 comprises performing geometric correction, radiometric calibration and image enhancement preprocessing on acquired remote sensing image data, wherein the multi-scale semantic segmentation deep learning model is constructed based on a convolutional neural network of an encoder-decoder structure, the encoder part extracts multi-scale ground object features through a pyramid pooling module, the decoder part performs up-sampling and fusion on the features, and outputs the ground object type initial classification and corresponding class probability of each pixel point; The auxiliary accessibility marking sample is correspondingly acquired with the remote sensing image area, the auxiliary accessibility marking sample comprises an in-situ survey accessibility score of a known surface coverage type, the accessibility score is comprehensively determined according to the passing difficulty, geological stability and ecological sensitivity of the cable construction machinery, the auxiliary accessibility score is used as an additional supervision signal in a multi-scale semantic segmentation deep learning model training stage, a multi-task learning target is formed together with a semantic segmentation type label, and the model is subjected to end-to-end training, so that the model is enabled to recognize the type of a ground feature and simultaneously the deep feature of the model is associated with accessibility priori knowledge.
3. The cable path planning method based on deep learning and multi-objective optimization according to claim 1, wherein the calculation of the accessibility confidence level comprises the steps of obtaining a prediction probability distribution of each feature type of any pixel point output by a trained model, weighting and fusing the probability distribution into a comprehensive accessibility expected score of the pixel point according to a feature type-standard accessibility score mapping table established in the auxiliary accessibility labeling sample, and converting the comprehensive accessibility expected score into an accessibility confidence level between 0 and 1 through a preset monotone mapping function, wherein a high score is mapped into the high confidence level, represents accessibility determination and risk is low, a low score is mapped into the low confidence level, and represents accessibility difference or risk uncertainty is high.
4. The cable path planning method based on deep learning and multi-objective optimization according to claim 3, wherein the construction of the accessibility confidence field comprises the steps of performing an accessibility confidence calculation process on all pixel points in a remote sensing image coverage area, generating an initial discrete confidence map which is consistent with the spatial resolution of an original image and corresponds to one accessibility confidence of each pixel point, processing the initial discrete confidence map by adopting a spatial interpolation algorithm, eliminating discontinuity caused by local outliers or small ranges, and generating a ground feature accessibility confidence field which is continuously and smoothly changed in space.
5. The cable path planning method based on deep learning and multi-objective optimization according to claim 1, wherein S2 includes definition of a basic cost factor, specifically including: The cost factor of the path length is that for any path segment on the path, the cost is proportional to the actual geometric length of the path segment; The construction cost factor is that based on the ground object type and the history construction data of the path crossing area, standard construction cost coefficients of unit length are given to different ground object types, and the total construction cost of the path line segment is the product of the length and the corresponding area cost coefficient; And (3) calculating the average gradient of the area where the path line segment passes based on the digital elevation model data, and mapping the gradient value into gradient cost through a nonlinear function, wherein the larger the gradient is, the higher the cost is.
6. The method of claim 5, wherein the defining of the confidence adjustment factor includes extracting an accessibility confidence value for each pixel covered by the path segment from the feature accessibility confidence field generated in S1, and defining an inverse confidence factor based on the accessibility confidence value, wherein the inverse confidence factor is inversely related to the accessibility confidence.
7. The cable path planning method based on deep learning and multi-objective optimization according to claim 6, wherein S2 further comprises a penalty and rewarding weight mechanism, and the cost values of different confidence regions in the path are differentiated to realize avoidance of high-risk regions and excitation of low-risk regions, and specifically comprises setting a accessibility confidence threshold, and dividing the path segments into two categories: a low confidence region is defined when the accessibility confidence value < accessibility confidence threshold; The accessibility confidence value is more than or equal to the accessibility confidence threshold, and is defined as a high confidence region; According to the comparison result, different adjusting weights are distributed: Multiplying a punishment weight to amplify the local path cost of the low confidence region, so that the local cost is raised in the path searching process, and the algorithm is promoted to automatically avoid the current region; And multiplying the high confidence region by a reward weight to reduce the local path cost, thereby attracting the path to preferentially select the current region.
8. The method for cable path planning based on deep learning and multi-objective optimization of claim 7, wherein the constructing of the confidence coupled path cost function specifically comprises, for any one candidate path from a start point to an end point, discretizing it into a series of continuous path segments, calculating a local cost for each path segment separately, and accumulating the local costs to form a total cost of the whole path, wherein: The local cost of each path segment includes a sum of base costs and a reachability adjustment factor; The sum of the basic costs comprises path length cost, construction cost and gradient cost, and the three costs are weighted and combined through set basic cost factor weight coefficients; the accessibility adjustment factors include a confidence inverse factor and a weighting factor; and (5) all the local costs of the candidate paths are summed up to obtain the total cost of the current candidate path.
9. The cable path planning method based on deep learning and multi-objective optimization according to claim 8, wherein S3 specifically comprises modeling a planning area as a three-dimensional grid network including geographic coordinates and the location confidence attribute as a basic search map based on a digital elevation model and the feature accessibility confidence field; The improved heuristic function comprises a first part, a second part and an improvement heuristic function, wherein the first part is a geometric cost estimated based on Euclidean distance from a current node to a target point, the second part is a confidence cost estimated based on a confidence field average value in the direction from the current node to the target point, particularly a desired value of a confidence penalty possibly encountered by the rest part of a predicted path, and the improvement heuristic function is used for guiding a search direction to avoid a wide range of low-confidence high-risk areas in front in advance while tending towards a target.
10. The cable path planning method based on deep learning and multi-objective optimization of claim 9, wherein S3 further comprises employing multi-objective The searching algorithm executes path exploration, takes the constructed confidence coupling path cost function as a core evaluation function, and has multiple targets When the searching algorithm expands the searching node, not only accumulating the actual cost from the starting point to the current node, but also estimating the cost from the current node to the target point by using the constructed improved heuristic function, and calculating an evaluation value, wherein the evaluation value is the sum of the actual cost and the estimated cost of the improved heuristic function; the searching process synchronously considers constraint and optimization targets, and specifically comprises the following steps: a) Setting minimum turning angle limitation in the definition of the search space, and ensuring the mechanical constructable operation of the path by restricting the direction change of the adjacent path sections; b) The algorithm preferentially explores the node with smaller evaluation value when the node is expanded, and naturally guides the path to deviate from a low confidence unit to realize risk avoidance because the confidence coupling path cost function comprises confidence punishment weight; c) The multi-objective balance is realized by adjusting the weight coefficient of the basic cost factor, cooperatively optimizing multiple objectives such as path length, construction cost, terrain fitness and the like in one search, and finally outputting a cable path with the minimum total cost under the set weight.

Description

Cable path planning method based on deep learning and multi-objective optimization Technical Field The invention relates to the technical field of path planning, in particular to a cable path planning method based on deep learning and multi-objective optimization. Background With the continuous promotion of urban and rural infrastructure intelligent construction, the requirements of electric power communication, power grid capacity expansion and industrial park intelligent upgrading on cable path planning are increasing. Particularly, under the complex environments such as urban edges, mountain areas, farmlands, wetlands and the like, how to scientifically plan a cable path on the premise of ensuring the construction feasibility, reduces the construction cost and avoids the environmental risk becomes one of the core problems in engineering practice. The existing cable path planning method mainly relies on engineering personnel to conduct manual scheme design based on topographic map, satellite map or field survey data, and is assisted by a simple shortest path algorithm to conduct auxiliary calculation. The traditional path planning method mostly aims at the shortest geographic distance, ignores factors such as accessibility, geological stability, environmental sensitivity and the like involved in the construction process, and causes the problems that the path is shortest but the crossing area cannot be constructed or the construction cost is extremely high, and the geometry is optimal but the engineering is not feasible. The existing partial improvement method introduces gradient factors or ground object type cost, but most of the existing partial improvement methods are static weight superposition models, cannot dynamically reflect the uncertainty of areas or the difference of risk levels, lack self-adaptive adjustment capability, and are difficult to meet multi-objective balance requirements. The conventional heuristic search algorithm only considers geometric distance estimation, and cannot predict risk distribution trend in front of the path, so that the path may be mistakenly input into a large-area construction unreachable area in the search process, calculation redundancy is increased, and result quality is affected. Disclosure of Invention The invention provides a cable path planning method based on deep learning and multi-objective optimization, which integrates remote sensing analysis and multi-objective optimization mechanisms and has risk self-adaptive evading capacity. A cable path planning method based on deep learning and multi-objective optimization comprises the following steps: The method comprises the steps of S1, based on remote sensing image data, adopting a multi-scale semantic segmentation deep learning model, combining auxiliary accessibility labeling samples, identifying each feature type, calculating accessibility confidence level of each pixel point to cable construction, and constructing a feature accessibility confidence field with continuous spatial distribution, wherein the accessibility confidence field is used for expressing the reliability level of a surface area on the construction accessibility and the uncertainty level of a potential risk area in remote sensing identification; S2, constructing a confidence coupling path cost function, namely defining a confidence coefficient adjustment factor based on the feature reachable confidence field, introducing the confidence coefficient adjustment factor into a multi-objective cost modeling process of path optimization, and constructing the confidence coupling path cost function, wherein the cost function fuses the path length, the construction cost, the terrain gradient and the confidence coefficient inverse factor, and endows a punishment weight to a region with low feature confidence coefficient and endows a region with high opposite confidence coefficient with a passing rewarding weight to generate the confidence coupling path cost function for path searching; and S3, optimizing path searching based on the confidence coupling path cost function, namely constructing an improved heuristic function by taking the confidence coupling path cost function as path searching input, executing path searching by adopting a multi-objective optimization algorithm to obtain a cable path scheme meeting the requirements of construction continuity, risk evasion and path smoothness, and outputting a cable path vector result for standard engineering deployment after carrying out smoothing treatment on the path. Optionally, the S1 comprises performing geometric correction, radiometric calibration and image enhancement preprocessing on the acquired remote sensing image data, wherein the multi-scale semantic segmentation deep learning model is constructed based on a convolutional neural network of an encoder-decoder structure, the encoder part extracts multi-scale ground object features through a pyramid pooling module, and the decoder part perfor