Pipeline real-time tuning system and method based on edge calculation and dynamic evolutionary algorithm

Abstract

The invention relates to the technical field of pipeline engineering, in particular to a pipeline real-time adjusting and optimizing system and method based on edge calculation and a dynamic evolution algorithm, five modules of distributed edge perception, position space prediction, real-time finite element analysis, multi-target optimization and layout adjustment are fused, firstly, an edge perception network collects multi-source data, and a unified tensor is generated; then, a position space prediction module predicts multi-time scale position space change based on tensor data by using a Shenchang differential equation; the real-time finite element analysis module receives prediction data, pipeline stress is calculated through a grid self-adaptive refinement technology, and hot spots are identified; the multi-objective optimization module adopts a dynamic variation differential evolution algorithm to generate a plurality of pipeline layout candidate schemes based on the hotspot data; the layout adjustment module performs multi-constraint conflict detection, adaptively adjusts candidate schemes and outputs a final pipeline layout; the system realizes intelligentization, precision and high efficiency of pipeline layout.

Description

Technical Field

[0001] This invention relates to the field of pipeline engineering technology, and in particular to a real-time pipeline optimization system and method based on edge computing and dynamic evolutionary algorithms, which is applicable to real-time monitoring, prediction and optimization of underground pipeline projects such as subways, water supply and drainage, and power. Background Technology

[0002] With the rapid development of urban construction, the complexity of underground pipeline projects is constantly increasing. Existing technologies, such as the Chinese patent application CN117057022A, disclose an automatic modeling method and system based on multi-source data. This method includes converting the acquired CAD files into a unified file; extracting training data from this unified file; inputting this training data into an algorithm model for training and then optimizing the algorithm model; after selecting the algorithm model to distinguish feature points through image recognition, comparing the historical pipeline data of these feature points from multiple sources to obtain the final data to be generated; and converting this data into a CAD 3D model. This technology can identify discrepancies in multi-source data, allowing staff to directly determine which parts have problems within the 3D model.

[0003] However, the above-mentioned technologies have the following shortcomings: 1. It lacks dynamic response capabilities, can only passively identify problems, and cannot provide proactive prediction and optimization functions; 2. The real-time impact of spatial changes on pipeline structural safety was not considered, and real-time stress analysis was lacking; 3. The centralized computing architecture requires data to be transmitted and processed remotely, resulting in high processing latency and difficulty in responding to rapid changes at the construction site; 4. Although it can mark the problem area, it does not provide a solution and requires manual adjustment by staff; 5. The algorithm model is fixed and lacks the ability to evolve on its own, making it difficult to adapt to complex and ever-changing construction environments. Summary of the Invention

[0004] The purpose of this invention is to provide a pipeline real-time optimization system and method based on edge computing and dynamic evolutionary algorithms, which can monitor spatial changes in location in real time, predict potential risks, automatically generate optimization and adjustment schemes, and support multi-pipeline collaborative optimization, thereby significantly improving the safety, efficiency and intelligence level of pipeline construction and maintenance.

[0005] This invention proposes a real-time pipeline optimization system based on edge computing and dynamic evolutionary algorithms, comprising: The distributed edge sensing network module is used to collect multi-source sensor data and generate a unified tensor representation; The location space prediction module is communicatively connected to the distributed edge sensing network module and is used to receive the unified tensor representation and generate multi-time-scale location space change prediction data based on the neural ordinary differential equation model. The real-time finite element analysis module is communicatively connected to the location space prediction module. It is used to receive the location space change prediction data at multiple time scales, calculate the pipeline stress distribution through adaptive mesh refinement technology, and generate stress hotspot data. The multi-objective optimization module is communicatively connected to the real-time finite element analysis module and is used to receive the stress hotspot data and generate multiple candidate pipeline layout schemes based on the differential evolution algorithm of the dynamic mutation strategy. The layout adjustment module is communicatively connected to the multi-objective optimization module. It is used to receive the multiple candidate pipeline layout schemes, perform adaptive adjustment of pipeline layout based on multi-constraint conflict detection, and generate the final pipeline adjustment scheme.

[0006] Preferably, the distributed edge-aware network module includes: A multi-layer sensor deployment unit is used to deploy a multi-layer sensor network including a surface-layer RTK-GNSS base station network, a soil layer strain fiber optic sensor array, and an equipment layer 3D laser scanner. An adaptive sampling control unit is used to dynamically adjust the sensor sampling frequency according to the rate of change of position space; The tensor fusion processing unit is used to convert multi-source heterogeneous data into third-order tensor representations and perform compression processing.

[0007] Preferably, the location spatial prediction module includes: The constant differential equation modeling unit is used to establish a mathematical model of the continuous change of location space over time. An adaptive sliding time window unit is used to dynamically adjust the size of the prediction window based on the rate of change of location space, construction speed, and acceleration. Multi-scale prediction unit is used to generate location and spatial change predictions at multiple time scales of 5 seconds, 15 seconds, 30 seconds, and 60 seconds.

[0008] Preferably, the real-time finite element analysis module includes: Lightweight finite element modeling unit for constructing simplified six-degree-of-freedom shell element models specifically for pipelines; Mesh adaptive refinement elements are used to dynamically adjust the mesh density in key areas based on stress gradients. Multiphysics coupling analysis unit, used to realize thermo-mechanical-fluid multiphysics coupling analysis.

[0009] Preferably, the multi-objective optimization module includes: Dynamic adaptive variation unit, used to dynamically adjust the variation factor based on stress state; Pipeline layout coding unit, used to represent pipeline paths using B-spline curve compression; Multi-objective evaluation unit, used to comprehensively evaluate the objectives of stress minimization, displacement constraint and cost control.

[0010] Preferably, the layout adjustment module includes: A conflict detection unit is used to detect and classify constraint conflicts based on a spatial octree algorithm. Priority negotiation unit, used to score pipeline adjustment priorities using the analytic hierarchy process; A progressive adjustment unit is used to execute pipeline fine-tuning algorithms based on virtual force fields.

[0011] Preferably, the mathematical model of the neural ordinary differential equation modeling unit is expressed as follows: ,in For location space state, For time, For parameterized neural networks, the neural ordinary differential equation modeling unit also includes a loss function for physical constraints. ,in For data fitting loss, Loss due to physical laws constraints For regularization loss, and Here, represents the weighting coefficient. Preferably, the formula for calculating the mutation factor of the dynamic adaptive mutation unit is: ,in This is the stress weighting coefficient. The attenuation coefficient is... For the first The maximum stress value of each solution For the allowable stress value, As a diversity factor, For the first A measure of the diversity of solutions.

[0012] Preferably, the system further includes: Edge computing nodes, connected to the distributed edge-aware network module, are used to perform local data preprocessing; The central server is connected to the edge computing nodes via an incremental data transmission protocol and is used to perform complex model training and simulation. The task scheduler is used to dynamically allocate computing tasks between edge computing nodes and the central server based on the computing load.

[0013] The pipeline real-time optimization method based on edge computing and dynamic evolutionary algorithms, using the aforementioned system, includes: Multi-source sensor data is collected through a distributed edge sensing network and a unified tensor representation is generated. Receive the unified tensor representation and generate multi-timescale location and spatial change prediction data based on the neuron constant differential equation model; Receive the multi-timescale location and spatial change prediction data, calculate the pipeline stress distribution through grid adaptive refinement technology, and generate stress hotspot data; Receive the stress hotspot data and generate multiple candidate pipeline layout schemes based on the differential evolution algorithm with dynamic mutation strategy; The system receives multiple candidate pipeline layout schemes, performs adaptive adjustment of the pipeline layout based on multi-constraint conflict detection, and generates the final pipeline adjustment scheme.

[0014] The beneficial effects of the present invention include: (1) significantly improved safety, with potential risks warned 30 to 60 seconds in advance; (2) greatly improved efficiency, with the optimized pipeline layout reducing construction conflicts by 30%; (3) effective cost control, with accurate location of risk areas and a 20% reduction in maintenance costs; (4) strong system scalability, supporting the needs of projects of different scales; and (5) high degree of intelligence, with self-learning ability, and better performance the longer it is used. Attached Figure Description

[0015] Figure 1 This is a system architecture diagram according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the distributed edge-aware network module in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the workflow of the location spatial prediction module in an embodiment of the present invention. Figure 4 This is a schematic diagram of the adaptive mesh refinement of the real-time finite element analysis module in an embodiment of the present invention; Figure 5 This is a schematic diagram of the dynamic mutation strategy of the multi-objective optimization module in an embodiment of the present invention; Figure 6 This is a flowchart illustrating the conflict detection and negotiation process of the layout adjustment module in this embodiment of the invention. Figure 7 This is a schematic diagram illustrating the interaction between the edge computing node and the central server in an embodiment of the present invention. Figure 8 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0016] Example 1: System Overall Architecture like Figure 1As shown, the pipeline real-time optimization system based on edge computing and dynamic evolution algorithm provided by the present invention includes: a distributed edge sensing network module 10, a location space prediction module 20, a real-time finite element analysis module 30, a multi-objective optimization module 40, and a layout adjustment module 50.

[0017] The distributed edge sensing network module 10 is used to collect multi-source sensor data and generate a unified tensor representation, realizing the unified expression and compression of multi-source heterogeneous data. The location space prediction module 20 is communicatively connected to the distributed edge sensing network module 10, used to receive the unified tensor representation, and generate multi-time-scale location space change prediction data based on the neural ordinary differential equation model. The real-time finite element analysis module 30 is communicatively connected to the location space prediction module 20, used to receive multi-time-scale location space change prediction data, calculate pipeline stress distribution through mesh adaptive refinement technology, and generate stress hotspot data. The multi-objective optimization module 40 is communicatively connected to the real-time finite element analysis module 30, used to receive stress hotspot data, and generate multiple candidate pipeline layout schemes based on the differential evolution algorithm of dynamic mutation strategy. The layout adjustment module 50 is communicatively connected to the multi-objective optimization module 40, used to receive multiple candidate pipeline layout schemes, perform adaptive adjustment of pipeline layout based on multi-constraint conflict detection, and generate the final pipeline adjustment scheme.

[0018] The system's modules communicate using an asynchronous message queue mechanism to ensure efficient and reliable data flow. The entire system adopts a three-layer edge-fog-cloud computing architecture, where data preprocessing and preliminary analysis are performed on edge devices, medium-complexity calculations are conducted in the fog computing layer, and complex model training and simulation are executed in the cloud, achieving rational allocation and utilization of computing resources.

[0019] The system workflow is summarized as follows: First, the distributed edge sensing network collects multi-source sensor data and generates a unified representation through a tensor fusion model; then, the location space prediction module predicts location space changes based on a neural network constant differential equation model; next, the real-time finite element analysis module performs stress analysis on the predicted location space changes; subsequently, the multi-objective optimization module generates candidate adjustment schemes based on the stress analysis results; finally, the layout adjustment module optimizes the schemes according to actual constraints and outputs the final implementation scheme.

[0020] The design of the entire system fully considers the balance between real-time performance, accuracy, and adaptability. It solves the real-time response problem through edge computing, improves prediction accuracy through neural ordinary differential equations, and enhances the adaptability of the algorithm through dynamic mutation strategies, ultimately achieving intelligent real-time optimization of the pipeline system.

[0021] Example 2: Distributed Edge Awareness Network Module like Figure 2As shown, the distributed edge sensing network module 10 includes a multi-layer sensor deployment unit 11, an adaptive sampling control unit 12, and a tensor fusion processing unit 13.

[0022] The multi-layer sensor deployment unit 11 is used to deploy a multi-layer sensor network including a surface-layer RTK-GNSS base station network, a soil layer strain fiber optic sensor array, and an equipment layer 3D laser scanner. The surface-layer RTK-GNSS base station network has an accuracy of ±5mm and is used to accurately monitor surface displacement; the soil layer strain fiber optic sensor array has a sensitivity of 1με (micro-strain) and is used to monitor strain changes within the soil; the equipment layer 3D laser scanner is integrated with millimeter-wave radar on equipment such as excavators and tunnel boring machines for real-time scanning of the surrounding environment.

[0023] The adaptive sampling control unit 12 is used to dynamically adjust the sensor sampling frequency according to the rate of change of position space. This unit implements a sampling frequency adjustment algorithm based on the rate of change: , in, The sampling frequency at the current moment. The base sampling frequency (usually set to 10Hz), This is an adjustment factor (range 0.5-2.0). This represents the magnitude of spatial change in location within the most recent time window. The sampling frequency is set as a threshold. When the location space changes drastically, the sampling frequency can be automatically increased to above 50Hz to ensure the capture of rapid changes; when the location space is stable, the sampling frequency is reduced to save computing and storage resources. The tensor fusion processing unit 13 is used to convert multi-source heterogeneous data into a third-order tensor representation and perform compression processing. This unit innovatively proposes a third-order tensor representation. ,Z,t,p, It integrates location, time, pressure, and temperature information. Tensor fusion employs a weighted strategy: , in, For the fused tensor representation, For the tensor data of the i-th sensor, The weighting coefficients for this sensor are dynamically adjusted based on its reliability. The weighting calculation considers the sensor's historical accuracy, current operating status, and the stability of the measured values ​​to ensure the quality of data fusion.

[0024] Preferably, the tensor fusion processing unit 13 also implements a tensor compression algorithm based on Tucker decomposition, achieving a compression ratio of up to 8:1 while maintaining information integrity. The compressed data is sent to the location spatial prediction module through an optimized communication protocol, significantly reducing the transmission bandwidth requirements.

[0025] Through the above design, the distributed edge sensing network module 10 can obtain the most comprehensive and accurate monitoring data with optimal resource consumption, laying a solid foundation for subsequent location spatial prediction and stress analysis.

[0026] Example 3: Location Spatial Prediction Module like Figure 3 As shown, the location space prediction module 20 includes a neural ordinary differential equation modeling unit 21, an adaptive sliding time window unit 22, and a multi-scale prediction unit 23.

[0027] The Ordinary Differential Equation Modeling Unit 21 is used to establish a mathematical model of the continuous change of location space over time. This unit innovatively introduces ordinary differential equations to represent the continuous change of location space over time: , in, For location space state, For time, For parameterized neural networks, These are the network parameters. Unlike traditional discrete-time series models, the neural network frequent differential equation model can generate predictions at any time point, achieving true continuous-domain prediction. Model training uses a physically constrained loss function: ; in, For data fitting loss, Loss due to physical laws constraints For regularization loss, and These are the weighting coefficients, which are 0.3 and 0.1 respectively.

[0028] The adaptive sliding time window unit 22 is used to dynamically adjust the prediction window size based on the rate of change of location space, construction speed, and acceleration. This unit is designed with a dynamic window algorithm. , in, The current window size. Based on the base window size (usually 60 seconds), These are weighting coefficients, taken as 0.4, 0.3, and 0.3 respectively. The rate of change of location space, To speed up construction, For acceleration, , , These are the maximum reference values ​​for the respective parameters.

[0029] The multi-scale prediction unit 23 is used to generate location and spatial change predictions at multiple time scales of 5 seconds, 15 seconds, 30 seconds, and 60 seconds. This unit adopts a hierarchical prediction architecture, with short-term predictions (5–15 seconds) focusing on accuracy and medium- to long-term predictions (30–60 seconds) focusing on trends. Multi-scale prediction also quantifies prediction uncertainty, providing a confidence interval for each prediction point to guide risk assessment in subsequent optimization decisions.

[0030] Preferably, the location spatial prediction module 20 also includes an anomaly detection function, which can identify potential geological anomalies or abnormal location spatial changes caused by equipment malfunctions and trigger early warnings. Anomaly detection is based on statistical learning methods, calculating in real time the degree of deviation between the current observation and historical patterns, and issuing an early warning when the deviation exceeds a preset threshold.

[0031] The output of the location space prediction module 20 is location space change prediction data with multiple time scales and uncertainty estimation, which provides input to the real-time finite element analysis module and realizes the key transformation from monitoring to prediction.

[0032] Example 4: Real-time Finite Element Analysis Module like Figure 4 As shown, the real-time finite element analysis module 30 includes a lightweight finite element modeling unit 31, a mesh adaptive refinement unit 32, and a multiphysics coupling analysis unit 33.

[0033] The lightweight finite element modeling element 31 is used to construct a simplified six-DOF shell element model specifically for pipelines. This element innovatively designs a simplified shell element specifically for pipelines, reducing computational load by 80% while maintaining accuracy compared to traditional general-purpose shell elements. This simplified element retains only key degrees of freedom, adapting to the structural characteristics of pipelines and significantly improving computational efficiency. Simultaneously, this element realizes a nonlinear constitutive model of the pipe material, accurately simulating the mechanical behavior of different pipe materials such as PVC, HDPE, and ductile iron under large deformations.

[0034] The adaptive mesh refinement element 32 is used to dynamically adjust the mesh density in key regions based on stress gradients. This element innovatively proposes a stress gradient-driven mesh refinement algorithm, automatically increasing the mesh density in key regions. Mesh quality is evaluated using a comprehensive index: , in, To score the quality of the grid, , , These are the shape, size, and gradient indices, respectively. , , For the weighting coefficients, take respectively When the mesh quality falls below a threshold, the mesh reconstruction algorithm is activated, supporting sub-second model reconstruction.

[0035] Multiphysics coupling analysis element 33 is used to perform thermo-mechanical-fluid multiphysics coupling analysis. This element considers the impact of temperature changes on pipe performance, especially for thermal pipelines, where temperature gradients can lead to significant thermal stress. Simultaneously, this element incorporates soil-pipe interaction interface elements to accurately simulate the interaction between the pipeline and the surrounding soil, including friction, bonding, and separation behaviors.

[0036] The computational process of real-time finite element analysis is as follows: First, receive the predicted spatial changes and generate an initial mesh; then perform mesh quality assessment to determine the areas that need to be refined; next, perform adaptive mesh refinement; then solve the finite element equations in parallel; finally, calculate the stress, displacement, and safety factor at key points and generate a stress hotspot diagram and risk assessment report.

[0037] Preferably, the real-time finite element analysis module 30 employs edge computing acceleration technology, distributing large-scale parallel computing tasks to multiple edge nodes to achieve distributed collaborative computing. For a model with 1 million grid nodes, the computation time is controlled within 3 seconds, meeting the requirements for real-time analysis.

[0038] The output of the real-time finite element analysis module 30 is pipeline stress distribution data and stress hotspot diagram, which provide key decision-making basis for the multi-objective optimization module and realize the key link from prediction to evaluation.

[0039] Example 5: Multi-objective optimization module like Figure 5 As shown, the multi-objective optimization module 40 includes a dynamic adaptive mutation unit 41, a pipeline layout coding unit 42, and a multi-objective evaluation unit 43.

[0040] The dynamic adaptive variation unit 41 is used to dynamically adjust the variation factor based on the stress state. This unit innovatively proposes a formula for calculating the adaptive variation factor based on the stress state: , in, Let be the variation factor of the i-th solution at time t. This is the stress weighting coefficient (value 0.6). The attenuation coefficient (value 2.5) is used. The maximum stress value of the i-th solution. For the allowable stress value, The diversity factor (value 0.2) Let be the diversity measure of the i-th solution. This formula allows solutions with higher stress to obtain a larger variation factor, enhancing the search for high-risk areas.

[0041] Pipeline layout coding unit 42 is used to compress and represent pipeline paths using B-spline curves. This unit designs a pipeline path compression representation based on B-spline curves, reducing the dimensionality of the search space. For complex pipelines, traditional representations may require hundreds of points, while B-spline representations only require 5-20 control points for accurate description, significantly reducing optimization complexity. Simultaneously, this unit innovatively proposes a layout topology protection operator to ensure that pipeline connection relationships are not disrupted, guaranteeing the practicality of the optimization results.

[0042] The multi-objective evaluation unit 43 is used to comprehensively evaluate the objectives of stress minimization, displacement constraints, and cost control. This unit uses a weighted method to construct a comprehensive scoring function: , Wherein, Score is the overall score of the solution. , , The scores are based on stress, displacement, and cost. , , These are weighting coefficients, which are dynamically adjusted according to project requirements. Preferably, the weighting coefficients are set to 0.5, 0.3, and 0.2 by default, emphasizing security considerations.

[0043] The algorithm parameters for multi-objective optimization include: dynamic adjustment of population size (50-200, depending on problem complexity); initial mutation factor F0 of 0.5; crossover probability CR of 0.9; maximum number of iterations of 100; and convergence threshold of 1e. -4 The optimization process employs the classic differential evolution framework, but incorporates dynamic mutation strategies and adaptive crossover, significantly improving algorithm performance.

[0044] Preferably, the multi-objective optimization module 40 also implements an interactive decision support function, which provides an intuitive visualization interface after generating multiple Pareto optimal solutions, integrates expert experience and algorithm recommendations, and supports decision-makers in selecting the most suitable solution.

[0045] The output of the multi-objective optimization module 40 is multiple candidate pipeline layout schemes, which provide optimization suggestions for the layout adjustment module and realize the key link from evaluation to optimization.

[0046] Example 6: Layout Adjustment Module like Figure 6 As shown, the layout adjustment module 50 includes a conflict detection unit 51, a priority negotiation unit 52, and a progressive adjustment unit 53.

[0047] The conflict detection unit 51 is used to detect and classify constraint conflicts based on a spatial octree algorithm. This unit innovatively designs a highly efficient conflict detection algorithm based on a spatial octree, dividing the three-dimensional space into units with increasing precision to quickly locate potential conflict areas and improve detection efficiency. Conflict detection considers not only geometric interference but also various constraints such as spacing required by specifications, maintenance space, thermal isolation, and signal interference. The conflict classification system categorizes conflicts into critical conflicts (must be resolved), important conflicts (should be resolved), and minor conflicts (acceptable), providing a decision-making basis for subsequent negotiations.

[0048] Priority negotiation unit 52 is used to score pipeline adjustment priorities using the Analytic Hierarchy Process (AHP). This unit designs a pipeline priority scoring system based on AHP, comprehensively considering factors such as pipeline importance, flexibility, and cost sensitivity. The priority scoring formula is: , in, The priority of the i-th pipeline is scored. Let j be the weight of the j-th evaluation factor. This represents the score of the i-th pipeline on the j-th factor. Evaluation factors include functional importance (weight 0.4), adjustment flexibility (weight 0.3), and adjustment cost (weight 0.4). .

[0049] The incremental adjustment unit 53 is used to execute a pipeline fine-tuning algorithm based on a virtual force field. This unit designs a pipeline fine-tuning algorithm based on a virtual force field to achieve flexible obstacle avoidance. The virtual force field model transforms conflict constraints into repulsive forces and optimization objectives into attractive forces, finding the optimal adjustment scheme through force field balance. Unlike direct optimization, the incremental adjustment strategy gradually achieves the objective through multiple small adjustments, avoiding the implementation difficulties that may arise from large adjustments.

[0050] Preferably, the layout adjustment module 50 innovatively proposes a hybrid adjustment strategy of elastic deformation and rigid displacement. For flexible materials such as plastic pipes, elastic deformation adjustment is given priority; for rigid materials such as steel pipes, rigid displacement adjustment is given priority, fully taking into account the characteristics of different pipe materials.

[0051] The data flow and processing logic for layout adjustment are as follows: First, candidate solutions provided by the optimization module are received; then, conflict detection is performed to build a conflict map; next, conflicts are sorted according to priority; then, the minimum adjustment solution is calculated; finally, adjustment instructions are generated and the final pipeline layout solution and implementation suggestions are output.

[0052] The output of the layout adjustment module 50 is the final pipeline adjustment plan, which provides direct guidance for on-site construction and realizes the key link from optimization to implementation.

[0053] Example 7: Mathematical Model of a Modeling Unit for Neuronormal Differential Equations This embodiment details the mathematical model of the modeling unit for the ordinary differential equations. As mentioned earlier, this mathematical model is expressed as: , in, The location space state is represented as a three-dimensional tensor, which includes the position, elevation, and physical parameters of each point in space; It is a time variable; For parameterized neural networks, These are network parameters.

[0054] Neural Networks A deep residual network structure is employed, comprising eight residual blocks, each containing two convolutional layers and one skip connection. The network input is the current state. and time The output is the rate of change of state. State prediction is achieved through numerical integration: , The integration employs the Runge-Kutta method with adaptive step sizes to ensure numerical stability and computational efficiency. Model training uses a loss function with physical constraints. , in, The data fitting loss is represented as the mean square error between the predicted and observed values. Losses are constrained by physical laws, including the laws of conservation of mass and energy. This is the regularization loss, used to prevent overfitting; and These are the weighting coefficients, which are 0.3 and 0.1 respectively.

[0055] This embodiment details the mathematical model of the neural ordinary differential equation modeling unit involved in claim 7. As mentioned above, the mathematical model is expressed as: , in, The priority of the i-th pipeline is scored. Let j be the weight of the j-th evaluation factor. This represents the score of the i-th pipeline on the j-th factor. Evaluation factors include functional importance (weight 0.4), adjustment flexibility (weight 0.3), and adjustment cost (weight 0.4). .

[0056] Physical constraint loss One of the innovative aspects of this model is its design, which encodes the physical laws governing the evolution of location space into the loss function, ensuring that the model's predictions conform to physical laws. For example, for soil motion, continuity and momentum equations can be introduced as constraints: , in, and These represent the residuals of the continuity equation and the momentum equation, respectively. and The corresponding weights are assigned. By combining data-driven approaches and physical constraints, the neural frequent differential equation model can maintain physical plausibility in sparse data regions, significantly improving prediction accuracy and generalization ability. Compared with traditional time series models, the prediction accuracy is improved by more than 30%, with a particularly significant advantage in long-term predictions.

[0057] Example 8: Variation Factor of Dynamically Adaptive Mutation Unit This embodiment details the formula for calculating the mutation factor of the dynamic adaptive mutation unit: , in, The mutation factor of the i-th solution at time t determines the intensity of individual mutation in the differential evolution algorithm; This is the stress weighting coefficient, with a value of 0.6, which controls the degree of influence of stress factors. The attenuation coefficient is set to 2.5, which adjusts the steepness of the stress response curve. The maximum stress value of the pipeline layout corresponding to the i-th solution is obtained from finite element analysis. The allowable stress value is determined based on the pipe material and the safety factor; This is a diversity factor with a value of 0.2, used to balance development and exploration; This is a measure of the diversity of the i-th solution, reflecting the degree of difference between this solution and other solutions in the population.

[0058] Diversity measurement The calculation formula is: , in, For population size, To solve and The distance between them is measured using Euclidean distance normalization.

[0059] The design philosophy of the dynamic variation factor is as follows: when the maximum stress of the solution approaches or exceeds the allowable stress, the variation factor is increased to enhance the local search capability and accelerate convergence to the safe region; when the stress level is low, the variation factor is decreased to maintain the stability of the scheme. At the same time, through the diversity factor, sufficient diversity of the population is ensured to prevent premature convergence to a local optimum.

[0060] Preferably, the variation factor The value of is limited to the range [0.1, 2.0] to prevent excessively small or large mutations from causing algorithm instability. During algorithm iteration, the mutation factor is dynamically adjusted according to the individual stress state and population diversity to achieve adaptive optimization, significantly improving the algorithm's ability to solve problems under complex constraints.

[0061] Experimental results show that, compared with a fixed mutation factor, the dynamic adaptive mutation strategy can improve the convergence speed by 40% and improve the quality of the solution, especially when dealing with highly nonlinear and multi-constraint problems.

[0062] Example 9: Edge Computing Architecture like Figure 7 As shown, this embodiment details the edge computing architecture, including edge computing nodes, a central server, and a task scheduler.

[0063] Edge computing nodes connect to a distributed edge-aware network module to perform local data preprocessing. Each edge computing node is equipped with an industrial-grade ARM processor (such as a Jetson AGX Xavier), 8GB of RAM, 128GB of storage, and a dedicated AI acceleration chip. The edge nodes employ a lightweight real-time operating system with optimized memory management and task scheduling, supporting millisecond-level response times. The main functions of the edge nodes include data acquisition and filtering, feature extraction, preliminary analysis, and emergency decision-making. The edge nodes are fault-tolerant, maintaining basic monitoring and early warning functions even in the event of communication interruptions.

[0064] The central server connects to the edge computing nodes via an incremental data transmission protocol for performing complex model training and simulation. The central server is equipped with a high-performance GPU cluster, supporting massively parallel computing. Its main functions include model training and updates, large-scale simulation analysis, historical data mining, and global optimization decision-making. The incremental data transmission protocol is designed to transmit only changed data, significantly reducing bandwidth requirements; for a system with 100 sensor nodes, the bandwidth requirement drops from the traditional 50 Mbps to below 5 Mbps.

[0065] The task scheduler, acting as the system's coordination center, dynamically allocates computing tasks between edge computing nodes and the central server based on computational load. Task scheduling employs a heuristic algorithm, comprehensively considering task urgency, computational complexity, node load, and network conditions to optimize task allocation. Scheduling strategies include: prioritizing urgent tasks to the nearest edge nodes; prioritizing computationally intensive tasks to high-performance servers; and increasing the task weighting of edge nodes when communication is limited.

[0066] Preferably, the edge computing architecture implements a model sinking mechanism, deploying a trained lightweight model to edge nodes and updating it periodically through a central server. For example, a simplified version of the location spatial prediction model can be deployed to edge nodes to support short-term predictions, while the complete model can be kept on the central server for long-term predictions and complex scene analysis.

[0067] Through the edge-fog-cloud three-layer computing architecture, the system achieves optimized allocation of computing resources, ensuring real-time data processing and accurate decision-making, while reducing network bandwidth requirements and system operating costs.

[0068] Example 10: Real-time pipeline optimization method like Figure 8 As shown, this invention also provides a real-time pipeline optimization method based on edge computing and dynamic evolutionary algorithms, including the following steps: Step S1: Collect multi-source sensor data through a distributed edge sensing network and generate a unified tensor representation.

[0069] Specifically, a multi-layer sensor network is first deployed, including a surface-layer RTK-GNSS base station network, a soil-layer strain fiber optic sensor array, and an equipment-layer 3D laser scanner. Then, the sensor sampling frequency is dynamically adjusted according to the rate of change of location space. Next, recursive median filtering is used to denoise the raw data. Finally, a tensor fusion model is used to convert the multi-source heterogeneous data into a unified third-order tensor representation and compress it to reduce the transmission bandwidth requirements.

[0070] Step S2: Receive the unified tensor representation and generate multi-timescale location and spatial change prediction data based on the neural constant differential equation model.

[0071] Specifically, first, a pre-trained ordinary differential equation model is loaded; then, the prediction window size is dynamically adjusted based on the rate of change of location space, construction speed, and acceleration; next, the ordinary differential equation model is applied to the current location space state, and future state predictions are generated through numerical integration; finally, location space change predictions at four time scales of 5 seconds, 15 seconds, 30 seconds, and 60 seconds are output, and prediction uncertainty estimates are provided.

[0072] Step S3: Receive multi-timescale location and spatial change prediction data, calculate pipeline stress distribution through adaptive grid refinement technology, and generate stress hotspot data.

[0073] Specifically, a lightweight finite element model is first constructed based on the pipeline material and geometric properties; then, the predicted spatial changes are applied to the model as boundary conditions; next, a mesh quality assessment is performed to determine the areas that need refinement; then, adaptive mesh refinement is performed to reconstruct the computational model; then, the finite element equations are solved in parallel; finally, the stress, displacement, and safety factor at key points are calculated to generate a stress hotspot diagram and a risk assessment report.

[0074] Step S4: Receive stress hotspot data and generate multiple candidate pipeline layout schemes based on the differential evolution algorithm with dynamic mutation strategy.

[0075] Specifically, the pipeline path is first compressed using B-spline curves; then the differential evolution algorithm population is initialized; next, the mutation factor is dynamically adjusted according to the stress state; then selection, mutation, and crossover operations are performed; subsequently, the fitness of the newly generated solution is evaluated, taking into account stress minimization, displacement constraints, and cost control objectives; finally, the Pareto optimal solution is selected as the candidate solution.

[0076] Step S5: Receive multiple candidate pipeline layout schemes, perform adaptive adjustment of pipeline layout based on multi-constraint conflict detection, and generate the final pipeline adjustment scheme.

[0077] Specifically, firstly, conflict detection is performed on candidate schemes to construct a conflict map; then, the hierarchical analysis method is used to score the priority of pipeline adjustments; next, the conflicts are classified and sorted; then, a pipeline fine-tuning algorithm based on virtual force fields is used to calculate the minimum adjustment scheme; finally, adjustment instructions are generated and the final pipeline layout scheme and implementation suggestions are output.

[0078] Preferably, the method further includes a closed-loop verification step: after the scheme is implemented, the system continues to monitor the location space and pipeline status, verify the adjustment effect, and feed the implementation results back into the model training to form a closed-loop optimization and continuously improve the system performance.

[0079] Using the above methods, the system can predict the trend of changes in the location and space of the construction site in real time, assess the impact on pipelines, automatically generate optimization and adjustment plans, and guide on-site implementation, thereby comprehensively improving the safety, efficiency, and intelligence level of pipeline projects.

[0080] During the construction of the subway tunnel, the tunnel boring machine caused deformation of the surrounding strata, threatening the safety of nearby water supply pipelines.

[0081] Before implementation, construction teams primarily relied on manual inspections and periodic measurements to monitor pipeline status, making it difficult to promptly detect and address risks arising from changes in location and space. Once a problem was discovered, work typically had to be halted, causing delays and additional costs.

[0082] After applying the system of this invention, the edge sensing network monitors spatial changes in real time. The spatial prediction module predicts 45 seconds in advance that the tunnel boring machine will cause the ground subsidence in the area where the water supply pipeline is located to be 15mm, exceeding the safety threshold. The real-time finite element analysis module calculates that the subsidence will cause the pipeline stress to exceed 70% of the material limit, posing a safety hazard. The multi-objective optimization module generates three candidate adjustment schemes, and the layout adjustment module determines the optimal implementation scheme after considering the site constraints: installing temporary supports in the critical section and fine-tuning the pipeline position.

[0083] The entire process, from monitoring to solution generation, took only 28 seconds, allowing ample time for on-site implementation. After implementation, pipeline stress was reduced to below 40% of the material's limit, ensuring construction safety. Compared to traditional methods, this avoided a work stoppage, saved two days of construction time, and reduced direct costs by approximately 200,000 yuan.

[0084] Long-term tracking data shows that after one year of system application, subway pipeline-related safety accidents decreased by 85%, emergency repairs decreased by 75%, and pipeline maintenance costs decreased by 23%, fully verifying the practical value and economic benefits of the invention.

[0085] In summary, the present invention has the following beneficial effects: 1. Significantly improve safety: Through predictive analytics, potential risks can be warned 30 to 60 seconds in advance, providing ample time for preventative measures and reducing safety incidents by more than 85%.

[0086] 2. Significantly improve efficiency: The optimized pipeline layout can reduce construction conflicts by 30%, speed up the project progress, reduce downtime, and save on construction time and labor costs.

[0087] 3. Effective cost control: Accurately locate risk areas, reduce unnecessary full-line inspections, reduce maintenance costs by more than 20%, and extend pipeline service life by 15% to 20%.

[0088] 4. Enhance system adaptability: The dynamic adaptive algorithm design enables the system to cope with complex and ever-changing construction environments, continuously optimize performance, and form a self-learning ecosystem.

[0089] 5. Enhance the level of intelligence: The system is upgraded from post-event detection to pre-event prevention, and from passive response to proactive optimization, realizing intelligent management of pipeline projects.

[0090] This invention addresses the real-time response problem through edge computing, improves prediction accuracy through neural ordinary differential equations, balances computational accuracy and efficiency through adaptive mesh refinement technology, enhances algorithm adaptability through dynamic mutation strategies, and achieves intelligent adjustment of pipeline layout through a multi-constraint negotiation mechanism. It constructs a complete real-time pipeline optimization technology system, providing an innovative solution for the safe construction and intelligent maintenance of underground pipeline projects.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A pipeline real-time optimization system based on edge computing and dynamic evolutionary algorithms, characterized in that, include: The distributed edge sensing network module is used to collect multi-source sensor data and generate a unified tensor representation; The location space prediction module is communicatively connected to the distributed edge sensing network module and is used to receive the unified tensor representation and generate multi-time-scale location space change prediction data based on the neural ordinary differential equation model. The real-time finite element analysis module is communicatively connected to the location space prediction module. It is used to receive the location space change prediction data at multiple time scales, calculate the pipeline stress distribution through adaptive mesh refinement technology, and generate stress hotspot data. The multi-objective optimization module is communicatively connected to the real-time finite element analysis module and is used to receive the stress hotspot data and generate multiple candidate pipeline layout schemes based on the differential evolution algorithm of the dynamic mutation strategy. The layout adjustment module is communicatively connected to the multi-objective optimization module. It is used to receive the multiple candidate pipeline layout schemes, perform adaptive adjustment of pipeline layout based on multi-constraint conflict detection, and generate the final pipeline adjustment scheme.

2. The system according to claim 1, characterized in that, The distributed edge-aware network module includes: A multi-layer sensor deployment unit is used to deploy a multi-layer sensor network including a surface-layer RTK-GNSS base station network, a soil layer strain fiber optic sensor array, and an equipment layer 3D laser scanner. An adaptive sampling control unit is used to dynamically adjust the sensor sampling frequency according to the rate of change of position space; The tensor fusion processing unit is used to convert multi-source heterogeneous data into third-order tensor representations and perform compression processing.

3. The system according to claim 1, characterized in that, The location spatial prediction module includes: The constant differential equation modeling unit is used to establish a mathematical model of the continuous change of location space over time. An adaptive sliding time window unit is used to dynamically adjust the size of the prediction window based on the rate of change of location space, construction speed, and acceleration. Multi-scale prediction unit is used to generate location and spatial change predictions at multiple time scales of 5 seconds, 15 seconds, 30 seconds, and 60 seconds.

4. The system according to claim 1, characterized in that, The real-time finite element analysis module includes: Lightweight finite element modeling unit for constructing simplified six-degree-of-freedom shell element models specifically for pipelines; Mesh adaptive refinement elements are used to dynamically adjust the mesh density in key areas based on stress gradients. Multiphysics coupling analysis unit, used to realize thermo-mechanical-fluid multiphysics coupling analysis.

5. The system according to claim 1, characterized in that, The multi-objective optimization module includes: Dynamic adaptive variation unit, used to dynamically adjust the variation factor based on stress state; Pipeline layout coding unit, used to represent pipeline paths using B-spline curve compression; Multi-objective evaluation unit, used to comprehensively evaluate the objectives of stress minimization, displacement constraint and cost control.

6. The system according to claim 1, characterized in that, The layout adjustment module includes: A conflict detection unit is used to detect and classify constraint conflicts based on a spatial octree algorithm. Priority negotiation unit, used to score pipeline adjustment priorities using the analytic hierarchy process; A progressive adjustment unit is used to execute pipeline fine-tuning algorithms based on virtual force fields.

7. The system according to claim 3, characterized in that, The mathematical model of the aforementioned ordinary differential equation modeling unit is expressed as follows: ,in For location space state, For time, For parameterized neural networks, the neural ordinary differential equation modeling unit also includes a loss function for physical constraints. ,in For data fitting loss, Loss due to physical laws constraints For regularization loss, and These are the weighting coefficients; The formula for calculating the mutation factor of the dynamic adaptive mutation unit is as follows: ,in This is the stress weighting coefficient. The attenuation coefficient is... For the first The maximum stress value of each solution For the allowable stress value, As a diversity factor, For the first A measure of the diversity of solutions.

8. The system according to claim 1, characterized in that, The system also includes: Edge computing nodes, connected to the distributed edge-aware network module, are used to perform local data preprocessing; The central server is connected to the edge computing nodes via an incremental data transmission protocol and is used to perform complex model training and simulation. The task scheduler is used to dynamically allocate computing tasks between edge computing nodes and the central server based on the computing load.

9. A pipeline real-time optimization method based on edge computing and dynamic evolutionary algorithms, employing the system described in any one of claims 1-9, characterized in that, include: Multi-source sensor data is collected through a distributed edge sensing network and a unified tensor representation is generated. Receive the unified tensor representation and generate multi-timescale location and spatial change prediction data based on the neuron constant differential equation model; Receive the multi-timescale location and spatial change prediction data, calculate the pipeline stress distribution through grid adaptive refinement technology, and generate stress hotspot data; Receive the stress hotspot data and generate multiple candidate pipeline layout schemes based on the differential evolution algorithm with dynamic mutation strategy; The system receives multiple candidate pipeline layout schemes, performs adaptive adjustment of the pipeline layout based on multi-constraint conflict detection, and generates the final pipeline adjustment scheme.

Images