Road investigation design method and system based on mobile network

Abstract

The invention discloses a road investigation design method and system based on a mobile network, and relates to the technical field of road investigation, and the method comprises the steps: dividing geological anomaly probability distribution data into dynamic grid units, constructing a reinforcement learning state space, and generating an optimal sampling path instruction in combination with a dual-network deep Q learning architecture; analyzing the optimal sampling path instruction into executable parameters, executing the executable parameters, collecting multi-source spatio-temporal data, establishing data association through a spatio-temporal hash algorithm, and generating a reconnaissance data set; and on the basis of the survey data set, through parameterized spline curve modeling, generating a candidate road design scheme, and in combination with a non-dominated sorting genetic algorithm, performing optimization to generate a three-dimensional road design scheme. According to the method, the exploration path is optimized by using the dynamic grid coding and the dual-network architecture, the high-risk area coverage and the moving efficiency are balanced, meanwhile, the method adapts to real-time geological changes in combination with priority experience playback, and the exploration reliability and the resource utilization rate are improved.

Description

Technical Field

[0001] This invention relates to the field of road surveying technology, and in particular to a road surveying and design method and system based on mobile networks. Background Technology

[0002] With the rapid development of intelligent transportation and digital twin technologies, the field of road surveying and design is gradually transforming from traditional manual surveying to an intelligent, data-driven model. Currently, mainstream technologies mainly rely on remote sensing mapping, ground-penetrating radar detection, and BIM tools to achieve road alignment planning and geological adaptation. Furthermore, the widespread adoption of 5G and edge computing has further promoted the real-time transmission and processing capabilities of survey data. In the process of technological evolution, the industry continues to explore how to improve the reliability and completeness of survey data, and how to optimize the efficiency and accuracy of multi-objective decision-making. These research directions provide important references for the intelligent development of road surveying and design.

[0003] Current road survey and design still face significant challenges in data integration and intelligent decision-making. On the one hand, although blockchain technology has been applied to engineering data storage, its deep integration with the entire survey-design process has not yet been achieved, especially in multi-agency collaborative scenarios where data consistency verification and real-time synchronization are delayed. On the other hand, while existing design engines support parametric modeling, their response to dynamic geological conditions is lagging, relying on manual intervention to adjust design parameters, leading to extended solution iteration cycles. Furthermore, although digital twin platforms can visualize and simulate solutions, their mechanical simulation modules and optimization algorithms lack real-time interaction capabilities, making it difficult to synchronously provide structural reliability indicators during the solution generation stage. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a road survey and design method based on mobile networks to solve the problems of lack of reliable data evidence and low efficiency of multi-objective dynamic optimization.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a road survey and design method based on a mobile network, comprising: real-time acquisition of spatiotemporal continuous data streams of the road survey area and preprocessing them to generate a structured spatiotemporal data stream; extraction of a standardized spatiotemporal feature set from the structured spatiotemporal data stream using a sliding window mechanism; construction of a deep spatiotemporal convolutional model based on a multi-scale feature fusion architecture and input of the standardized spatiotemporal feature set to generate geological anomaly probability distribution data; division of the geological anomaly probability distribution data into dynamic grid cells and construction of a reinforcement learning state space; generation of optimal sampling path instructions by combining a dual-network deep Q-learning architecture; parsing the optimal sampling path instructions into executable parameters and executing them; acquiring multi-source spatiotemporal data; establishing data associations through a spatiotemporal hashing algorithm to generate a survey dataset; and generating candidate road design schemes by modeling with parametric spline curves based on the survey dataset, and optimizing them by combining a non-dominated sorting genetic algorithm to generate a three-dimensional road design scheme.

[0007] As a preferred embodiment of the road survey and design method based on mobile networks described in this invention, the preprocessing includes data cleaning, format standardization, noise removal, and timestamp alignment.

[0008] As a preferred embodiment of the road survey and design method based on mobile networks described in this invention, the specific steps of extracting a standardized spatiotemporal feature set from a structured spatiotemporal data stream using a sliding window mechanism are as follows: Based on structured spatiotemporal data streams, dynamic coupling parameters of the geological environment are generated through a sliding window segmentation strategy and short-time Fourier transform. Based on dynamic coupling parameters of the geological environment, spatiotemporal feature fragments are generated through spatial interpolation and time series alignment; The spatiotemporal feature fragments are normalized and then spliced ​​together to form a standardized spatiotemporal feature set.

[0009] As a preferred embodiment of the road survey and design method based on mobile networks described in this invention, the specific steps for constructing a deep spatiotemporal convolutional model based on a multi-scale feature fusion architecture and inputting a standardized spatiotemporal feature set to generate geological anomaly probability distribution data are as follows: Based on a multi-scale feature fusion architecture, a feature extraction network is constructed by combining deformable spatiotemporal convolution, and trained using structured spatiotemporal data streams. Based on the trained feature extraction network, a deep spatiotemporal convolution model framework is generated through a dynamic dilatation rate adjustment algorithm. Based on the deep spatiotemporal convolution model framework, a deep spatiotemporal convolution prediction model is generated through a gating attention mechanism and a course learning strategy. Then, a standardized spatiotemporal feature set is input to generate geological anomaly probability distribution data.

[0010] As a preferred embodiment of the mobile network-based road survey and design method of the present invention, the steps of dividing the geological anomaly probability distribution data into dynamic grid cells, constructing a reinforcement learning state space, and generating optimal sampling path instructions using a dual-network deep Q-learning architecture are as follows. Based on the probability distribution data of geological anomalies, the exploration area is divided into dynamic grid units using a dynamic grid coding algorithm, and a reinforcement learning state space is constructed. A dual-network deep Q-learning architecture is constructed using a priority-based empirical replay method driven by the time difference error method and a delay parameter update method. In the dual-network deep Q-learning architecture, the main network transforms the reinforcement learning state space into a path planning action space and combines it with the target network for value evaluation to generate the optimal sampling path instruction.

[0011] As a preferred embodiment of the road survey and design method based on mobile networks described in this invention, the generation of the survey dataset refers to parsing the optimal sampling path instruction into executable parameters and collecting multi-source spatiotemporal data during execution using an inverse kinematics algorithm, dividing the multi-source spatiotemporal data using a latitude and longitude rasterization method, and generating the survey dataset by combining it with gridded hash encoding.

[0012] As a preferred embodiment of the road survey and design method based on mobile networks described in this invention, the generation of a three-dimensional road design scheme refers to generating a three-dimensional road design scheme by performing a three-dimensional trade-off based on candidate road design schemes, screening the Pareto optimal solution set, and verifying it using ANSYS mechanical simulation.

[0013] Secondly, this invention provides a road survey and design system based on a mobile network, comprising a data acquisition module, a risk modeling module, a path planning module, a path parsing module, and a scheme generation module. The data acquisition module is used to collect spatiotemporal continuous data streams of the road survey area in real time and preprocess them to generate a structured spatiotemporal data stream. The risk modeling module is used to extract a standardized spatiotemporal feature set from the structured spatiotemporal data stream using a sliding window mechanism, construct a deep spatiotemporal convolution model based on a multi-scale feature fusion architecture, and input the standardized spatiotemporal feature set to generate geological anomaly probability distribution data. The path planning module is used to divide the geological anomaly probability distribution data into dynamic grid units, construct a reinforcement learning state space, and generate optimal sampling path instructions using a dual-network deep Q-learning architecture. The path parsing module parses the optimal sampling path instructions into executable parameters and executes them, collects multi-source spatiotemporal data, establishes data associations through a spatiotemporal hashing algorithm, and generates a survey dataset. The scheme generation module is used to generate candidate road design schemes based on the survey dataset through parametric spline curve modeling, and optimizes them using a non-dominated sorting genetic algorithm to generate a three-dimensional road design scheme.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein the computer program, when executed by the processor, implements any step of the road survey and design method based on a mobile network as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the road survey and design method based on a mobile network as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: by using deformable convolution and dynamic void ratio adjustment, it adaptively identifies geological anomalies at different scales, and by combining a gating attention mechanism to focus on key spatiotemporal features, it significantly improves the accuracy of geological detection and avoids the risk of misjudgment caused by the single scale of traditional methods; by using dynamic grid coding and dual network architecture to optimize the exploration path, it balances the coverage of high-risk areas and the efficiency of movement, and at the same time, it combines priority experience playback to adapt to real-time geological changes, thereby improving the reliability of exploration and the utilization rate of resources. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a road survey and design method based on mobile networks.

[0019] Figure 2 This is a schematic diagram of a road survey and design system based on a mobile network.

[0020] Figure 3 A flowchart for deep spatiotemporal convolution modeling of road survey and design methods based on mobile networks.

[0021] Figure 4 A flowchart for generating the optimal sampling path instruction for a road survey and design method based on mobile networks. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a road survey and design method based on a mobile network, including the following steps: S1. Real-time acquisition of spatiotemporal continuous data streams of the road survey area, and preprocessing to generate structured spatiotemporal data streams; Spatiotemporal continuous data streams include geological vibration data, temperature and humidity gradients, and electromagnetic field distributions; Furthermore, vibration waveforms are recorded using a seismic detector array, and geological vibration data are extracted from the waveforms through short-time Fourier transform and wavelet packet decomposition. Environmental parameters are collected through a temperature and humidity sensor network, and a three-dimensional temperature field is formed through Kriging interpolation to obtain the temperature and humidity gradient. Magnetic field strength data is measured using a triaxial magnetometer, and electromagnetic field distribution is generated after Gaussian filtering for noise reduction and inverse distance weighted interpolation.

[0026] For spatiotemporal continuous data streams, data cleaning, format standardization, noise removal, and timestamp alignment are performed to generate structured spatiotemporal data streams.

[0027] It should be noted that the data cleaning process involved: extracting effective frequency bands from the geological vibration data using a Butterworth bandpass filter, smoothing the temperature and humidity gradients using a sliding window mean filter, and suppressing high-frequency noise in the electromagnetic field distribution using a Gaussian low-pass filter. Format standardization: Geological vibration data, temperature and humidity gradients, and electromagnetic field distributions are converted into uniform sampling rate, JSON format, and CSV format, respectively; Noise Removal: Kalman filtering is used to remove noise from geological vibration data, moving median filtering is used to remove temperature and humidity gradient noise, and a noise cancellation algorithm is used to remove electromagnetic field distribution noise. Timestamp alignment: The device clock is synchronized via the network time protocol, and precise time calibration is achieved using cross-correlation and linear interpolation methods, and unified to the GPS time reference.

[0028] S2. A sliding window mechanism is used to extract a standardized spatiotemporal feature set from the structured spatiotemporal data stream. Based on a multi-scale feature fusion architecture, a deep spatiotemporal convolution model is constructed and the standardized spatiotemporal feature set is input to generate geological anomaly probability distribution data. Based on structured spatiotemporal data streams, dynamic coupling parameters of the geological environment are generated through a sliding window segmentation strategy and short-time Fourier transform. These dynamic coupling parameters include geological vibration spectrum characteristics, temperature and humidity gradient vectors, and electromagnetic field fluctuation modes. The specific steps are as follows: the geological vibration data is segmented in the time dimension using a fixed time step segmentation method to generate windowed geological vibration data; the temperature and humidity gradient is segmented in the spatial dimension using a spatial grid alignment method to generate windowed temperature and humidity gradients; and the electromagnetic field distribution is segmented in both time and space using a combination of time overlap windows and spatial grids to generate windowed electromagnetic field distributions.

[0029] Frequency domain transformation is applied to windowed geological vibration data to generate geological vibration spectrum characteristics; standardized temperature and humidity gradient vector representations are generated by establishing temperature field distribution for windowed temperature and humidity gradients; and electromagnetic field fluctuation characteristics are generated by wavelet transform time-frequency analysis for windowed electromagnetic field distributions.

[0030] Based on dynamic coupling parameters of the geological environment, spatiotemporal feature fragments are generated through spatial interpolation and time series alignment; Spatial interpolation processing: Kriging spatial interpolation method is used to fill the spatial gaps in the geological vibration spectrum characteristics, inverse distance weighted interpolation method is used to establish the continuous spatial distribution of temperature and humidity gradient vectors, and radial basis function interpolation method is applied to perform spatial smoothing of electromagnetic field fluctuation characteristics.

[0031] Time series alignment: The time series of geological vibration spectrum characteristics is dynamically time-warped with the GPS time reference using the time warping method. The temperature and humidity gradient vector is sampled using the linear interpolation method to achieve time synchronization. The observation timestamps of the electromagnetic field fluctuation mode are aligned with the unified time axis using the timestamp matching method.

[0032] Geographic coordinate labeling: Add standard coordinate information to the geological vibration spectrum characteristics, temperature and humidity gradient vector and electromagnetic field fluctuation characteristics to generate spatiotemporal feature fragments with geographic coordinate labels, including UTM coordinates, WGS84 geographic coordinates and latitude and longitude coordinates.

[0033] The spatiotemporal feature fragments are normalized and then spliced ​​together to form a standardized spatiotemporal feature set.

[0034] Furthermore, Z-score standardization is applied to the spatiotemporal feature segments, and a sliding window mean-variance update method is used to generate scale-uniform normalized spatiotemporal feature segments. Time series matching is used to align the data points of the scale-uniform normalized spatiotemporal feature segments along the time dimension. Spatial grid coordinate mapping is used to map the spatial locations of the scale-uniform normalized spatiotemporal feature segments. Finally, a multidimensional array stacking method is used to integrate the scale-uniform normalized spatiotemporal feature segments into a continuous and structurally regular standardized spatiotemporal feature set.

[0035] The structured spatiotemporal data stream is divided into a training set and a validation set; It should be noted that the structured spatiotemporal data stream is divided into a set of continuous data blocks according to timestamps using a time series segmentation method. Based on the data block set, a stratified random sampling method is used to randomly allocate data blocks, generating training set data blocks and validation set data blocks with the same spatial coverage. The training set data blocks and validation set data blocks are then verified for time series integrity and spatial coverage integrity using a spatiotemporal continuity check method to generate the training set and validation set.

[0036] Based on a multi-scale feature fusion architecture, a feature extraction network is constructed through deformable spatiotemporal convolution and trained using a training set. The specific steps are as follows: based on the multi-scale feature fusion architecture, a parallel processing path containing convolutional kernels of different sizes is established through a multi-branch parallel construction method, forming a multi-scale feature extraction path and original path features.

[0037] Based on multi-scale feature extraction paths, a deformable spatiotemporal convolution kernel deformation offset prediction method is used to predict and generate convolution kernel parameters. Geometric deformation features are extracted from the original path features through deformable convolution operations to generate processed path features. The processed path features are then fused into a unified feature map through cross-scale feature concatenation. The network structure of the unified feature map is initialized using the He initialization method and the zero-value initialization method to generate an initialized feature extraction network.

[0038] The training set is read using a parallel data pipeline method to generate a structured spatiotemporal data stream that can be directly input into the initial feature extraction network. The initialized feature extraction network is then trained using a mixed-precision training method: multi-scale spatiotemporal features are extracted from the structured spatiotemporal data stream through low-precision forward propagation to generate the network's output features; based on the network's output features, the network weight parameters are updated through high-precision backpropagation to generate the optimized feature extraction network. Based on the validation set, an early stopping strategy is used to verify the loss change; the training process is terminated when the loss no longer decreases, and the trained feature extraction network is output.

[0039] Based on the trained feature extraction network, a multi-scale spatiotemporal feature map set is extracted from the structured spatiotemporal data stream, and combined with the dynamic dilatation rate adjustment algorithm, a deep spatiotemporal convolution model framework is generated. The specific steps are as follows: the trained feature extraction network extracts a multi-scale spatiotemporal feature map set from the structured spatiotemporal data stream. The multi-scale spatiotemporal feature map set contains both local details and global semantic information. The dynamic dilation rate adjustment algorithm generates the convolutional kernel interval parameters for each layer based on the spatial resolution of the multi-scale spatiotemporal feature map set. Combined with the hierarchical dilation rate scheduling strategy, it assigns a dilation rate to each layer. Finally, by integrating the convolutional kernel interval parameters and dilation rates of each layer, it generates a complete dilated convolutional parameter configuration table.

[0040] Based on a complete table of dilated convolution parameters, convolutions with corresponding dilation rates are inserted into the corresponding layers of the feature extraction network. Simultaneously, skip connections are used to achieve channel-level fusion of shallow high-resolution features and deep high-semantic features. Batch normalization layers are then inserted to stabilize the feature distribution, thus constructing a deep spatiotemporal convolution model framework. Based on the deep spatiotemporal convolution model framework, a deep spatiotemporal convolution prediction model is generated through a gating attention mechanism and a course learning strategy. The specific steps are as follows: Based on the deep spatiotemporal convolutional model framework, a multi-head attention weight allocation method is used to analyze the long-range dependencies between spatiotemporal features. Combined with a gating signal generation method, the weight distribution of different feature channels is adjusted to generate feature representations with enhanced spatiotemporal correlation. These enhanced spatiotemporal feature representations are then used to classify training difficulty levels using a feature complexity evaluation method, generating phased optimized network parameters. Furthermore, a hybrid precision acceleration method is employed to improve training speed, an early stopping mechanism ensures the generalization ability of the deep spatiotemporal convolutional prediction model, and a multi-fold cross-validation method is used to comprehensively test the performance of the deep spatiotemporal convolutional prediction model, resulting in a deep spatiotemporal convolutional prediction model that can be directly used for spatiotemporal prediction tasks.

[0041] The standardized spatiotemporal feature set is input into the deep spatiotemporal convolutional prediction model to generate geological anomaly probability distribution data, expressed as follows: ; in, It is geological anomaly probability distribution data. It is the Sigmoid activation function. In time step and spatial location Local spatiotemporal feature blocks at the location, It is a depth-space-time convolutional encoder. It is the weight matrix of the gating attention mechanism. In time step and spatial location Multi-head attention weights at the location, It is the total number of time steps. It is the index of the time step. It is the total number of spatial locations. It is an index of spatial location. It is a bias term; The specific steps are as follows: a standardized spatiotemporal feature set establishes cross-spatiotemporal regional dependencies using a multi-head attention weight allocation method, and noise interference is suppressed using a gating feature screening method to generate attention-weighted spatiotemporal feature representations. These spatiotemporal feature representations are then converted into probability space dimensions using a fully connected layer mapping method, and the probability space dimensions are normalized to probability intervals using a Softmax activation function to generate geological anomaly probability distribution data.

[0042] S3. Divide the geological anomaly probability distribution data into dynamic grid units and construct a reinforcement learning state space. Combine this with a dual-network deep Q-learning architecture to generate the optimal sampling path instruction. Based on geological anomaly probability distribution data, the exploration area is divided into dynamic grid units through a dynamic grid coding algorithm, and a reinforcement learning state space is constructed by fusing multi-dimensional state parameters. The specific steps are as follows: high-risk areas are located using the geological anomaly probability density analysis method based on the geological anomaly probability distribution data; grid cell size is adjusted using the resolution adjustment method based on the geological anomaly probability distribution data; fragmented grids are eliminated using the spatial connectivity test method; and a spatial distribution map of dynamic grid cells with complete boundaries is generated.

[0043] The dynamic grid cell spatial distribution map generates the mean anomaly probability within each grid cell using a sliding window statistical method. Principal component analysis is then used to extract core geological features from these mean anomaly probabilities, generating standardized multidimensional grid state parameters. These standardized multidimensional grid state parameters embed location information using a spatial coordinate encoding method. An attention-weighted approach is used to amplify the influence of parameters in high-risk areas, and feature dimensionality reduction methods are combined to simplify the data dimensions, generating a reinforcement learning state space. The dynamic grid cell division achieves refined segmentation of the exploration area, the multidimensional parameter extraction preserves key geological information, and the state space fusion encoding provides a structured environmental representation for reinforcement learning.

[0044] Based on historical exploration records, a dual-network deep Q-learning architecture is constructed using a priority experience replay method driven by the time difference error method and a delay parameter update method. Furthermore, historical exploration records are a complete collection of operational records from previous exploration missions, including equipment movement trajectories, environmental condition observations, action selections, and reward feedback. These records are stored in a spatiotemporal sequence, with each data point labeled with its corresponding geographic coordinates and timestamp, forming a complete historical record of the interaction between the exploration equipment and the environment. Historical exploration records include both successful path samples and suboptimal decision-making cases.

[0045] The specific steps are as follows: quantify each path sample in the historical exploration record using the time difference error method to generate error values; construct a priority experience pool based on the priority sampling method; correct the sampling bias using the importance sampling weight adjustment method; and generate a priority-ranked experience sample set.

[0046] An empirical sample set is used for training the dual-network deep Q-learning architecture. A network cloning method is used to create a master network and a target network with consistent parameters. A parameter freezing method is used to maintain the stability of the target network, and an asynchronous update method separates the parameter update rhythm, thus constructing a dual-network interactive learning framework. After the master network is trained using a priority empirical sample set, a periodic parameter synchronization method is used to transfer the updated parameters to the target network. A soft update method achieves a smooth parameter transition, and a gradient truncation method limits the magnitude of parameter changes, generating a parameter-synchronized and stable dual-network deep Q-learning architecture.

[0047] In a dual-network deep Q-learning architecture, the main network transforms the reinforcement learning state space into a path planning action space and combines it with the target network for value evaluation to generate the optimal sampling path instruction, expressed as: ; in, In the reinforcement learning state space The optimal sampling path instruction in the text. In the reinforcement learning state space Candidate actions in the action space of the next path planning Main network The expected long-term returns In the reinforcement learning state space Candidate actions in the action space of the next path planning Target network The expected long-term returns It is a reinforcement learning state space. These are candidate actions in the path planning action space. It is the main network. It is the target network. It is the temperature coefficient. It is the regularization coefficient. It is a divergence constraint. It is a reinforcement learning state space Characteristic distribution, It is the historical reinforcement learning state space Characteristic distribution, It is an uncertainty weighting coefficient. It is the standard deviation of the dual network outputs. It is the mean of the outputs of the two networks; Furthermore, the main network transforms the reinforcement learning state space into a probability distribution of the path planning action space through a policy generation method, generating a multi-dimensional action candidate set containing movement direction, step size, and probe frequency. This multi-dimensional action candidate set is then evaluated using the target network's state-action pair value estimation method to assess the long-term exploration benefits of each action, generating a quantified action value assessment matrix. A greedy policy selection method filters this matrix, generating the highest-value candidate actions. A path continuity constraint method ensures spatial coherence between adjacent actions, forming a preliminary sampled path instruction set. This set is then optimized for trajectory curvature using a path smoothing method, and high-risk areas are eliminated using collision detection, ultimately generating the sampled path instructions that meet safety constraints and achieve optimal exploration efficiency. The main network handles the end-to-end mapping from state to action, while the target network provides a stable value assessment benchmark, collaboratively enabling intelligent decision-making for exploration actions. The action probability distribution captures the potential possibilities of exploration strategies, the action value assessment quantifies decision benefits, and the path continuity constraint ensures instruction feasibility. The final output sampled path instructions achieve an optimal balance between exploration efficiency and safety. The feature representation of the state space of reinforcement learning drives the quality of action generation, the value assessment of the target network suppresses policy bias, and the post-processing of the path ensures the engineering feasibility of the instructions, forming a complete exploration path planning link from environmental perception to decision execution.

[0048] S4. Parse the optimal sampling path instruction into executable parameters and execute it. Collect multi-source spatiotemporal data during the execution process, establish data association through spatiotemporal hashing algorithm, and generate exploration dataset; The optimal sampling path instruction is parsed into executable parameters and executed using inverse kinematics algorithm and real-time path interpolation. Furthermore, the optimal sampling path command is parsed into an angle parameter sequence for each joint of the exploration equipment using an inverse kinematics transformation method, generating joint spatial control quantities containing position, velocity, and acceleration information. The joint spatial control quantities are then used to divide the execution time axis using a time-uniform sampling method, and cubic spline interpolation is employed to fill in the motion trajectory between discrete path points, generating smooth, continuous, and high-precision trajectory parameters. These high-precision continuous trajectory parameters ensure mechanical structural safety through joint limit detection and eliminate potential conflict path segments in interaction with the environment using collision detection methods, outputting safe and executable trajectory parameters that conform to physical constraints. These safe and executable trajectory parameters are then converted into control commands recognizable by the exploration equipment and executed using an industrial communication protocol encapsulation method.

[0049] Multi-source spatiotemporal data are collected during the process and generated into a structured exploration dataset through gridded hash encoding; Furthermore, during the execution process, the multi-source spatiotemporal data is acquired through a sensor synchronous acquisition method to obtain the movement trajectory of the exploration equipment and collect environmental parameters. Combined with a timestamp alignment method, a unified time reference is established, and a time-aligned multi-source data stream is output.

[0050] Time-aligned multi-source data streams are divided into regular spatial grid cells using latitude and longitude rasterization, generating a spatial grid index structure with geographic indexes. This spatial grid index structure guides feature vector encoding to convert multi-source data (vibration, electromagnetic, chemical, etc.) within each grid into fixed-length hash codes. A consistent hashing algorithm generates a globally unique identifier for each grid, forming gridded hash-coded data blocks. These gridded hash-coded data blocks are then integrated into a continuous data volume in the spatiotemporal dimension using tensor concatenation. Metadata association methods bind the hash identifiers to the original spatiotemporal coordinates, generating a structured survey dataset with unified dimensions and a clear index.

[0051] S5. Based on the survey dataset, candidate road design schemes are generated through parametric spline curve modeling, and optimized by non-dominated sorting genetic algorithm to generate three-dimensional road design schemes.

[0052] Based on the survey dataset, candidate road design schemes are obtained through parametric spline curve modeling. The specific steps are as follows: The survey dataset is processed using principal component analysis to extract topographic elevation, slope, and geological stability, generating a set of topographic feature vectors. The topographic parameters in the feature vector set are converted into spline curve control point coordinates using a feature weight mapping method, forming a parameterized control point sequence. The parameterized control point sequence is then used to establish continuity through B-spline basis function construction, and combined with curvature smoothing constraints to eliminate local abrupt curvature changes. A geometrically reasonable candidate road curve model is constructed and validated using a slope-curvature joint detection method, generating a validated candidate road curve model. The survey dataset is then input into the validated candidate road curve model to generate candidate road design schemes.

[0053] Based on candidate road design schemes, a three-dimensional trade-off is performed using a non-dominated sorting genetic algorithm to select the Pareto optimal solution set, which is then verified using ANSYS mechanical simulation to obtain the three-dimensional road design scheme.

[0054] The specific steps are as follows: the candidate road design schemes are weighed in three dimensions by a non-dominated sorting genetic algorithm on three types of indicators: construction cost, traffic efficiency and environmental impact. The Pareto front screening method is used to retain non-dominated solutions and generate the Pareto optimal solution set with the best overall performance.

[0055] Based on the Pareto optimal solution set, it is converted into a set of finite element models containing material properties, geometry, and load conditions using the ANSYS parametric scripting method. Then, a set of parametric mechanical simulation models that can be processed in batches is generated using the ANSYS Mechanical APDL batch processing scripting method.

[0056] Based on a set of parametric mechanical simulation models, the stress concentration of the cross section is generated using the ANSYS statics solver method. Combined with fatigue life prediction methods, the structural durability under long-term loads is simulated, and a mechanical performance verification report including stress cloud diagrams and life assessment is generated.

[0057] Based on the structural stress data, deformation displacement data, geological compatibility data, and material durability data in the mechanical performance verification report, the candidate road design schemes are quantitatively scored using a weighted scoring method, with scores for structural safety factor, displacement control compliance rate, geological stability index, and material durability. Candidate road design schemes with collapse risk or material failure are eliminated using a decision tree classification method. Finally, a three-dimensional road design scheme that meets the requirements of multi-objective optimization and mechanical safety is output.

[0058] This embodiment also provides a road survey and design system based on a mobile network, including: a data acquisition module, a risk modeling module, a path planning module, a path parsing module, and a scheme generation module; the data acquisition module is used to collect spatiotemporal continuous data streams of the road survey area in real time, and perform preprocessing to generate structured spatiotemporal data streams; the risk modeling module is used to extract standardized spatiotemporal feature sets from the structured spatiotemporal data streams using a sliding window mechanism, construct a deep spatiotemporal convolution model based on a multi-scale feature fusion architecture, and input the standardized spatiotemporal feature sets to generate geological anomaly probability distribution data; the path planning module is used to divide the geological anomaly probability distribution data into dynamic grid units, construct a reinforcement learning state space, and generate optimal sampling path instructions by combining a dual-network deep Q-learning architecture; the path parsing module parses the optimal sampling path instructions into executable parameters and executes them, collects multi-source spatiotemporal data, establishes data associations through a spatiotemporal hashing algorithm, and generates a survey dataset; the scheme generation module is used to generate candidate road design schemes based on the survey dataset by modeling with parametric spline curves, and optimizes them by combining a non-dominated sorting genetic algorithm to generate three-dimensional road design schemes.

[0059] This embodiment also provides a computer device applicable to the road survey and design method based on a mobile network, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the road survey and design method based on a mobile network as proposed in the above embodiment.

[0060] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0061] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the road survey and design method based on a mobile network as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0062] In summary, this invention significantly improves geological detection accuracy by using deformable convolution and dynamic void ratio adjustment to adaptively identify geological anomalies at different scales, and by combining a gated attention mechanism to focus on key spatiotemporal features, thus avoiding the risk of misjudgment caused by the single scale of traditional methods. Furthermore, by utilizing dynamic grid coding and a dual-network architecture to optimize the exploration path, it balances coverage of high-risk areas with movement efficiency, and by combining priority experience playback to adapt to real-time geological changes, it improves exploration reliability and resource utilization.

[0063] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A road survey and design method based on mobile networks, characterized in that: include, Real-time acquisition of spatiotemporal continuous data streams of the road survey area, followed by preprocessing to generate structured spatiotemporal data streams; A sliding window mechanism is used to extract a standardized spatiotemporal feature set from a structured spatiotemporal data stream. Based on a multi-scale feature fusion architecture, a deep spatiotemporal convolution model is constructed and input with the standardized spatiotemporal feature set to generate geological anomaly probability distribution data. Geological anomaly probability distribution data is divided into dynamic grid cells, and a reinforcement learning state space is constructed. Combined with a dual-network deep Q-learning architecture, the optimal sampling path instruction is generated. The optimal sampling path instruction is parsed into executable parameters and executed to collect multi-source spatiotemporal data. Data associations are established through a spatiotemporal hash algorithm to generate an exploration dataset. Based on the survey dataset, candidate road design schemes are generated through parametric spline curve modeling, and then optimized using a non-dominated sorting genetic algorithm to generate three-dimensional road design schemes.

2. The road survey and design method based on mobile networks as described in claim 1, characterized in that: The preprocessing includes data cleaning, format standardization, noise removal, and timestamp alignment.

3. The road survey and design method based on mobile networks as described in claim 2, characterized in that: The process of extracting a standardized spatiotemporal feature set from a structured spatiotemporal data stream using a sliding window mechanism involves the following steps: Based on structured spatiotemporal data streams, dynamic coupling parameters of the geological environment are generated through a sliding window segmentation strategy and short-time Fourier transform. Based on dynamic coupling parameters of the geological environment, spatiotemporal feature fragments are generated through spatial interpolation and time series alignment; The spatiotemporal feature fragments are normalized and then spliced ​​together to form a standardized spatiotemporal feature set.

4. The road survey and design method based on mobile networks as described in claim 3, characterized in that, The method involves constructing a deep spatiotemporal convolutional model based on a multi-scale feature fusion architecture, inputting a standardized spatiotemporal feature set, and generating geological anomaly probability distribution data. The specific steps are as follows: Based on a multi-scale feature fusion architecture, a feature extraction network is constructed by combining deformable spatiotemporal convolution, and trained using structured spatiotemporal data streams. Based on the trained feature extraction network, a deep spatiotemporal convolution model framework is generated through a dynamic dilatation rate adjustment algorithm. Based on the deep spatiotemporal convolution model framework, a deep spatiotemporal convolution prediction model is generated through a gating attention mechanism and a course learning strategy. Then, a standardized spatiotemporal feature set is input to generate geological anomaly probability distribution data.

5. The road survey and design method based on mobile networks as described in claim 4, characterized in that: The process involves dividing the geological anomaly probability distribution data into dynamic grid cells, constructing a reinforcement learning state space, and combining this with a dual-network deep Q-learning architecture to generate optimal sampling path instructions. The specific steps are as follows: Based on the probability distribution data of geological anomalies, the exploration area is divided into dynamic grid units using a dynamic grid coding algorithm, and a reinforcement learning state space is constructed. A dual-network deep Q-learning architecture is constructed using a priority-based empirical replay method driven by the time difference error method and a delay parameter update method. In the dual-network deep Q-learning architecture, the main network transforms the reinforcement learning state space into a path planning action space and combines it with the target network for value evaluation to generate the optimal sampling path instruction.

6. The road survey and design method based on mobile networks as described in claim 1, characterized in that: The generation of the exploration dataset refers to the process of parsing the optimal sampling path instruction into executable parameters and collecting multi-source spatiotemporal data during execution using the inverse kinematics algorithm, dividing the multi-source spatiotemporal data using latitude and longitude rasterization, and generating the exploration dataset by combining it with gridded hash encoding.

7. The road survey and design method based on mobile networks as described in claim 1, characterized in that: The generation of three-dimensional road design schemes refers to the process of using a non-dominated sorting genetic algorithm to perform three-dimensional trade-offs based on candidate road design schemes, selecting the Pareto optimal solution set, and verifying it using ANSYS mechanical simulation method to generate three-dimensional road design schemes.

8. A road survey and design system based on a mobile network, based on the road survey and design method based on a mobile network as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a risk modeling module, a path planning module, a path parsing module, and a solution generation module; The data acquisition module is used to collect spatiotemporal continuous data streams of the road survey area in real time, and to preprocess them to generate structured spatiotemporal data streams; The risk modeling module is used to extract a standardized spatiotemporal feature set from the structured spatiotemporal data stream using a sliding window mechanism. Based on a multi-scale feature fusion architecture, it constructs a deep spatiotemporal convolution model and inputs the standardized spatiotemporal feature set to generate geological anomaly probability distribution data. The path planning module is used to divide the probability distribution data of geological anomalies into dynamic grid cells and construct a reinforcement learning state space. Combined with a dual-network deep Q-learning architecture, it generates the optimal sampling path instruction. The path parsing module is used to parse the optimal sampling path instruction into executable parameters and execute it, collect multi-source spatiotemporal data, establish data associations through spatiotemporal hashing algorithms, and generate exploration datasets. The scheme generation module is used to generate candidate road design schemes based on the survey dataset by modeling with parametric spline curves, and then optimize them by combining non-dominated sorting genetic algorithm to generate three-dimensional road design schemes.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the road survey and design method based on mobile networks as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the road survey and design method based on mobile networks as described in any one of claims 1 to 7.

Images