A space-time intelligent prediction and optimization processing method and system for road soft foundation settlement

Abstract

The application discloses a kind of road soft foundation settlement space-time intelligent prediction, optimization processing method and system, belong to geotechnical engineering and artificial intelligence cross technical field.The method obtains high-precision deformation data by time series InSAR technology and monitoring point data fusion, constructs space-time physical information fusion deep learning network, excavates the space correlation and time evolution law of settlement, realizes regional settlement field accurate prediction and influence partition;Foundation treatment scheme parameter is differentiable embedded deep learning network, combined with multi-objective optimization reward function and gradient optimization algorithm, automatically optimized to obtain the optimal treatment scheme that meets settlement control, cost saving and low carbon environmental protection.The application breaks through the limitations of traditional method space-time fragmentation, weak physical mechanism, evaluation and optimization disconnection, realizes closed loop from settlement prediction to scheme optimization, provides intelligent support for soft soil area road engineering safety and design optimization.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering and artificial intelligence, specifically to a method and system for spatiotemporal intelligent prediction and optimization of soft soil settlement in roads. Background Technology

[0002] When constructing long-distance roads in soft soil areas, the settlement of the soft soil foundation directly affects the construction safety, long-term stability, and operational life of the project. Accurately predicting the spatiotemporal distribution of post-construction settlement and developing scientifically sound foundation treatment plans are crucial for ensuring project quality and controlling construction costs.

[0003] In recent years, the development of InSAR technology has provided an effective means to acquire large-scale, high-precision time-series data on surface deformation, making macroscopic monitoring of soft soil settlement over long distances along roads possible. However, existing technologies still face many bottlenecks in practical applications:

[0004] The physical mechanisms of the prediction models are weak: Traditional settlement prediction methods, such as the layered summation method, are based on the assumption of homogeneous soil layers, which makes it difficult to characterize the spatial variability of long-distance soft soil. Furthermore, they do not fully consider the inherent physical mechanisms such as soil consolidation and creep, resulting in unclear physical meaning of the prediction results and insufficient accuracy and reliability.

[0005] The spatiotemporal correlation is fragmented: Existing methods mostly focus on time series prediction of single-point monitoring data, failing to effectively integrate the spatial correlation characteristics between multiple monitoring points, and also failing to fully capture the temporal evolution of settlement development. This makes it difficult to form a comprehensive assessment of the overall settlement distribution in the region and cannot support the analysis and control of regional construction impacts.

[0006] The assessment and optimization are disconnected: Existing technologies often stop at predicting and assessing settlement, failing to form a closed loop with the design and optimization of foundation treatment schemes. The selection of foundation treatment schemes and the adjustment of parameters rely heavily on engineers' experience-based calculations, which is inefficient and makes it difficult to achieve a comprehensive balance between multiple objectives such as settlement control, cost savings, and environmental friendliness.

[0007] Therefore, how to construct an intelligent framework that integrates multi-source data, physical mechanisms and artificial intelligence technologies to achieve high-precision spatiotemporal prediction of road soft soil settlement and automated optimization of foundation treatment schemes has become an urgent technical problem to be solved in the field of road engineering in soft soil areas. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for intelligent prediction and optimization of settlement in the spatiotemporal space of road soft soil, realizing intelligentization of the entire process from settlement spatiotemporal prediction to foundation treatment scheme optimization, and solving the above-mentioned technical bottlenecks of traditional methods.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0010] In a first aspect, the present invention provides a method for spatiotemporal intelligent prediction and optimization of road soft soil settlement, comprising the following steps:

[0011] S1: Obtain multi-source subsidence related data of the regional surface over a period of time, and perform spatiotemporal alignment, denoising and standardization preprocessing on the multi-source subsidence related data;

[0012] S2: Construct and train a spatiotemporal physical information fusion deep learning network. The deep learning network takes preprocessed multi-source settlement-related data as input, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, and outputs settlement prediction values ​​of each monitoring point at multiple times.

[0013] S3: Based on the recursive prediction results of the deep learning network, the regional settlement impact is divided into zones, and a set of initial foundation treatment suggestions is generated by combining historical engineering big data.

[0014] S4: Differentiablely embed the key parameters of the foundation treatment scheme into the deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through the gradient optimization algorithm to obtain the optimal foundation treatment scheme;

[0015] S5: Output the optimal solution and prediction evaluation results. Adjust the parameters dynamically according to the actual needs on site and repeat step S4 to achieve iterative optimization of the solution.

[0016] As a further improvement to the technical solution of the present invention, in step S1, the multi-source settlement-related data includes regional surface deformation field data obtained by persistent scatterer interferometry (PS-InSAR) or small baseline set interferometry (SBAS-InSAR), as well as synchronous settlement time series data of monitoring points within the engineering area.

[0017] As a further improvement to the technical solution of the present invention, in step S2, the physical constraint of soil consolidation is constructed based on the discrete form of Terzaghi's one-dimensional consolidation equation and large deformation consolidation equation. By introducing a physical information loss term, the network prediction results are constrained to conform to the soil consolidation and creep mechanism.

[0018] As a further improvement to the technical solution of the present invention, in step S3, the settlement impact zone is automatically divided into four levels: major impact zone, large impact zone, general impact zone, and slight impact zone based on the predicted settlement contour lines or settlement gradient and the settlement control threshold of the engineering specifications.

[0019] As a further improvement to the technical solution of the present invention, the historical engineering big data includes a library of engineering-verified foundation treatment schemes corresponding to similar settlement impact zones within the region, soil physical and mechanical parameters, construction records, and post-construction settlement monitoring data, providing data support for the generation of the initial scheme.

[0020] As a further improvement to the technical solution of the present invention, in step S4, the key parameter vector of the foundation treatment scheme... ,in This includes the spacing and length of drainage boards, vacuum degree, preloading time, arrangement, spacing, length and replacement rate of mixing piles.

[0021] As a further improvement to the technical solution of the present invention, in step S4, the multi-objective optimization function is:

[0022] ,

[0023] in, , , These are the weighting coefficients. Let be the objective function for settlement control. For cost function, This is a carbon emission function.

[0024] As a further improvement to the technical solution of the present invention, in step S4, the foundation treatment scheme includes one of the following methods: replacement method, vacuum preloading method, surcharge preloading method, cement-soil mixing pile composite foundation method, and pipe pile composite foundation method, using different parameter values, or multiple methods using different parameter values ​​and then combining them.

[0025] As a further improvement to the technical solution of the present invention, during the dynamic adjustment process, the design allowable deformation value of adjacent buildings, underground pipelines or roads is used as the dynamic adjustment threshold to ensure the safety and compatibility of the solution.

[0026] A second aspect of this invention provides a spatiotemporal intelligent prediction and optimization system for road soft soil settlement, comprising:

[0027] The data acquisition and preprocessing unit is used to acquire multi-source settlement-related data throughout the entire construction cycle of the regional surface, and to perform spatiotemporal alignment, noise reduction, and standardization preprocessing.

[0028] The network construction and training unit is used to build a spatiotemporal physical information fusion deep learning network. It inputs preprocessed multi-source settlement-related data into the deep learning network, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, completes the training of the deep learning network and outputs settlement prediction values.

[0029] The settlement zoning and initial scheme generation unit is used to zon the regional settlement impact based on the prediction results of deep learning network, and to generate a set of initial foundation treatment suggestion schemes by combining historical engineering big data.

[0030] The parameter optimization unit is used to embed the key parameters of the foundation treatment scheme into a differentiable deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through a gradient optimization algorithm to obtain the optimal foundation treatment scheme.

[0031] The solution output and iteration unit is used to output the optimal solution and prediction evaluation results, dynamically adjust parameters according to actual on-site needs, and trigger the parameter optimization unit to perform iterative optimization.

[0032] A third aspect of the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor being configured to acquire the computer program and execute the above-described method for intelligent prediction and optimization of road soft soil settlement in time and space.

[0033] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for intelligent spatiotemporal prediction and optimization of road soft soil settlement.

[0034] The technical solution of the present invention has the following advantages over the prior art:

[0035] This invention integrates regional deformation field data obtained through temporal InSAR technology with settlement time-series data from monitoring points. After preprocessing, the data is input into a spatiotemporal physical information fusion deep learning network. This network leverages convolutional neural networks and long short-term memory neural networks to uncover the spatial correlation features and temporal evolution patterns of settlement. Furthermore, it enhances the model's physical mechanism conformity by embedding physical information loss terms related to soil consolidation theory. This overcomes the limitations of traditional methods, which suffer from fragmented spatiotemporal representations and weak physical mechanisms, enabling high-precision prediction and automated impact zoning of the overall regional settlement field. The invention further embeds differentiable key parameters of foundation treatment schemes into the network. Combined with a multi-objective optimization reward function and gradient optimization algorithm to control post-construction settlement and minimize engineering workload and carbon emissions, it achieves closed-loop intelligent optimization from settlement prediction to scheme optimization. It also supports dynamic adjustment of parameters for iterative optimization based on actual site conditions and allowable deformation values ​​of surrounding facilities, replacing the traditional trial-and-error calculation model that relies on experience. This significantly improves the adaptability, economy, and environmental friendliness of foundation treatment schemes, providing efficient and quantitative technical support for the safety assurance and design optimization of road engineering in soft soil areas. Attached Figure Description

[0036] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0037] Figure 1 This is a schematic diagram of the framework of a method for intelligent spatiotemporal prediction and optimization of soft soil settlement for roads according to an embodiment of the present invention.

[0038] Figure 2 This is a schematic diagram of the module framework of a road soft soil foundation settlement spatiotemporal intelligent prediction and optimization processing system according to an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the composition of a computing device according to an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the regional settlement impact zoning results in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0042] The present invention will be further described in detail below with reference to the accompanying drawings.

[0043] Reference Figure 1 In a first aspect, the present invention provides a method for spatiotemporal intelligent prediction and optimization of road soft soil settlement, comprising the following steps:

[0044] S1: Obtain multi-source subsidence related data of the regional surface over a period of time, and perform spatiotemporal alignment, denoising and standardization preprocessing on the multi-source subsidence related data;

[0045] S2: Construct and train a spatiotemporal physical information fusion deep learning network. The deep learning network takes preprocessed multi-source settlement-related data as input, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, and outputs settlement prediction values ​​of each monitoring point at multiple times.

[0046] S3: Based on the recursive prediction results of the deep learning network, the regional settlement impact is divided into zones, and a set of initial foundation treatment suggestions is generated by combining historical engineering big data.

[0047] S4: Differentiablely embed the key parameters of the foundation treatment scheme into the deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through the gradient optimization algorithm to obtain the optimal foundation treatment scheme;

[0048] S5: Output the optimal solution and prediction evaluation results. Adjust the parameters dynamically according to the actual needs on site and repeat step S4 to achieve iterative optimization of the solution.

[0049] In practice, high-precision deformation field data of the regional surface at multiple time points over a period of time is first obtained using temporal InSAR technology. Simultaneously, contemporaneous settlement time-series data from monitoring points within the engineering area are collected. The two types of data undergo spatiotemporal alignment, denoising, and standardization preprocessing. Next, a spatiotemporal physical information fusion deep learning network is constructed. Using the preprocessed settlement data as input, a convolutional neural network layer extracts spatial correlation features between monitoring points, generating a spatial feature time series. This series is then input into a long short-term memory neural network layer to learn the time dependence of settlement. During training, a physical information loss term based on soil consolidation theory is introduced to constrain the network and improve prediction accuracy. The network outputs subsequent settlement prediction values ​​for each monitoring point at multiple time points. Then, the trained network is used to recursively predict the regional settlement field at different future times. Based on the prediction results, the settlement impact is divided into zones, and preliminary foundation treatment suggestions with different probability levels are set in combination with historical engineering big data of the region. After vectorizing the key design parameters of the foundation treatment scheme, the deep learning network is embedded through differentiable function relations to form a settlement prediction model with differentiable parameters. Based on the multi-objective optimization reward function of controlling post-construction settlement, minimizing engineering volume and carbon emissions, the gradient optimization algorithm is used to optimize the parameter combination and output the optimal construction method combination scheme. Finally, based on the optimal design parameters, the specific foundation treatment construction method combination scheme and prediction evaluation chart are output. If the parameters need to be adjusted, the optimization steps are updated and repeated.

[0050] This invention overcomes the limitations of traditional methods in spatiotemporal representation and single-point prediction. By combining multi-source data fusion with deep learning networks, it achieves regional, high-precision intelligent assessment of soft soil foundation settlement. By embedding the parameters of foundation treatment schemes into the model in a differentiable manner and combining them with multi-objective optimization, a closed loop from settlement prediction to scheme optimization is formed, replacing the traditional trial-and-error model that relies on experience. This greatly improves the adaptability, economy, and environmental friendliness of foundation treatment schemes, and provides efficient and quantitative technical support for engineering safety assurance and design optimization in soft soil areas.

[0051] In some embodiments, in step S1, the multi-source settlement-related data includes regional surface deformation field data obtained by persistent scatterer interferometry (PS-InSAR) or small baseline set interferometry (SBAS-InSAR), as well as contemporaneous settlement time series data of monitoring points within the engineering area.

[0052] It should be noted that the temporal InSAR technology used in step S1 is specifically Persistent Scatterer Interferometry (PS-InSAR) or Small Baseline Set Interferometry (SBAS-InSAR). Utilizing the technical characteristics of these two mature InSAR technologies, multi-time-point observations of the regional surface are conducted to capture subtle changes in surface deformation, thereby obtaining high-precision time-series data of surface deformation. This provides reliable basic data support for subsequent subsidence prediction. This invention ensures the accuracy and stability of the regional surface deformation field data acquisition. The two technologies can be flexibly selected according to the actual conditions of the engineering area, such as topography and observation conditions, improving the applicability and data reliability of the method and laying a solid foundation for achieving high-precision subsidence prediction.

[0053] In some embodiments, in step S2, the physical constraints of soil consolidation are constructed based on the discrete forms of Terzaghi's one-dimensional consolidation equation and large deformation consolidation equation. By introducing physical information loss terms, the network prediction results are constrained to conform to the soil consolidation and creep mechanism.

[0054] It should be noted that the physical information loss term in step S2 is constructed based on the discrete forms of Terzaghi's one-dimensional consolidation equation and large deformation consolidation equation. This transforms the fundamental physical laws of soil consolidation into constraints for network training. This loss term constrains the settlement change process predicted by the network, ensuring it conforms to the mechanical properties of soil during consolidation and preventing predictions from deviating from physical mechanisms. This strengthens the physical mechanism of the prediction model, ensuring that the prediction results not only conform to statistical data patterns but also follow the fundamental physical principles of soil consolidation. This effectively solves the problem of unclear physical meaning in traditional prediction models, significantly improving the model's generalization ability and extrapolation reliability, and ensuring accurate prediction of subsequent settlement.

[0055] Reference Figure 4 In some embodiments, in step S3, the settlement impact zone is automatically divided into four levels: major impact zone, large impact zone, general impact zone, and minor impact zone, based on the predicted settlement contour lines or settlement gradients and the settlement control thresholds in the engineering specifications.

[0056] It should be noted that in step S3, based on the overall settlement distribution map obtained from the network recursive prediction, settlement contour lines are extracted or settlement gradients are calculated. Combined with the settlement control thresholds set in the engineering specifications, the engineering area is automatically classified and zoned into major impact areas, large impact areas, general impact areas, and minor impact areas. The zoning result map is then output, visually presenting the settlement risk level of different areas. This invention achieves accurate division and visualization of settlement impact areas, clearly defining the settlement risk level of different areas. This provides a clear basis for subsequently developing targeted foundation treatment schemes, avoiding the subjectivity and ambiguity of traditional zoning methods, and improving the scientific rigor and targeted nature of foundation treatment scheme design.

[0057] In some embodiments, the historical engineering big data includes a library of engineering-verified foundation treatment schemes corresponding to similar settlement impact zones within the region, soil physical and mechanical parameters, construction records, and post-construction settlement monitoring data, providing data support for the generation of initial schemes.

[0058] It should be noted that a large database of historical engineering projects in this region is being constructed. This database must contain at least a collection of verified and actually adopted foundation treatment schemes from historical engineering cases within the region that are in similar influence zones to the current target area. When generating preliminary foundation treatment recommendations, relevant data from this database is used to provide data support and practical experience for the initial generation and optimization scope of the current design, and to initially select suitable foundation treatment schemes. By fully utilizing the practical experience of historical engineering projects, reliable data references are provided for the generation of preliminary foundation treatment schemes, reducing the blind spots in scheme design, shortening the scheme design cycle, and improving the feasibility and rationality of the preliminary schemes, thus laying a good foundation for subsequent refined optimization.

[0059] In some embodiments, in step S4, the key parameter vector of the foundation treatment scheme ,in This includes the spacing and length of drainage boards, vacuum degree, preloading time, arrangement, spacing, length and replacement rate of mixing piles.

[0060] It should be noted that, in step S3, when generating the preliminary design parameter vector, based on the spatial distribution characteristics of settlement gradients in each affected zone, the probability of recommending each treatment level under different gradient levels is calculated using a specific probability formula. This probability formula is constructed based on parameters such as the standard settlement gradient threshold and its standard deviation, and the adjustable sensitivity coefficient obtained from historical engineering database statistics. A set of preliminary foundation treatment suggestions is generated based on the calculated probability levels, serving as the initial iteration starting point and constraint reference for subsequent refined gradient optimization. This invention achieves precise matching between preliminary foundation treatment schemes and settlement gradient characteristics, providing an objective basis for scheme selection through probability quantification. This makes the initial iteration starting point more aligned with actual engineering needs, effectively narrowing the search range for subsequent optimization and improving optimization efficiency and the reliability of optimization results.

[0061] In some embodiments, in step S4, the multi-objective optimization function is:

[0062] ,

[0063] in, , , These are the weighting coefficients. Let be the objective function for settlement control. For cost function, This is a carbon emission function.

[0064] It should be noted that this invention uses this function to comprehensively measure the performance of various aspects of foundation treatment schemes, providing quantitative indicators for parameter optimization. This invention clarifies the quantitative standards for multi-objective optimization, achieving a comprehensive balance between the three objectives of post-construction settlement control, minimization of engineering workload, and carbon emission control. It avoids the one-sidedness of schemes caused by single-objective optimization, making the optimized foundation treatment scheme more in line with the diverse needs of actual engineering, and taking into account safety, economy, and environmental protection.

[0065] In some embodiments, in step S4, the foundation treatment scheme includes one of the following methods: replacement method, vacuum preloading method, surcharge preloading method, cement-soil mixing pile composite foundation method, and pipe pile composite foundation method, using different parameter values, or multiple methods using different parameter values ​​and then combining them.

[0066] It should be noted that, based on the actual factors such as settlement and geological conditions of the engineering area, a suitable treatment method should be selected from the above schemes, or multiple schemes can be combined to form a targeted foundation treatment plan. This invention lists specific types of foundation treatment schemes, enriching the range of scheme selection. These schemes can be flexibly combined according to the actual engineering situation, improving the adaptability of the method to different soft soil geological conditions and settlement risk levels, and ensuring the practicality and effectiveness of the foundation treatment scheme.

[0067] In some embodiments, during the dynamic adjustment process, the design allowable deformation value of adjacent buildings, underground pipelines, or roads is used as the dynamic adjustment threshold to ensure the safety and compatibility of the scheme.

[0068] It should be noted that when adjusting the foundation treatment parameters in step S5, the design allowable deformation values ​​of adjacent buildings, underground pipelines, or roads are used as dynamic adjustment thresholds. During the parameter adjustment process, it is ensured that after the implementation of the adjusted foundation treatment scheme, the settlement of the soft soil foundation will not exceed the threshold range, thus avoiding adverse effects on surrounding buildings, underground pipelines, or roads. This invention considers the settlement tolerance of surrounding facilities, ensuring that the adjusted foundation treatment scheme not only meets the settlement control requirements of the project itself but also takes into account the safety of surrounding facilities. This improves the compatibility and safety of the scheme, avoids damage to surrounding facilities due to soft soil foundation settlement, and reduces environmental risks and safety hazards during project construction.

[0069] Reference Figure 2 The second aspect of the present invention provides a spatiotemporal intelligent prediction and optimization system for road soft soil settlement, comprising:

[0070] The data acquisition and preprocessing unit is used to acquire multi-source settlement-related data throughout the entire construction cycle of the regional surface, and to perform spatiotemporal alignment, noise reduction, and standardization preprocessing.

[0071] The network construction and training unit is used to build a spatiotemporal physical information fusion deep learning network. It inputs preprocessed multi-source settlement-related data into the deep learning network, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, completes the training of the deep learning network and outputs settlement prediction values.

[0072] The settlement zoning and initial scheme generation unit is used to zon the regional settlement impact based on the prediction results of deep learning network, and to generate a set of initial foundation treatment suggestion schemes by combining historical engineering big data.

[0073] The parameter optimization unit is used to embed the key parameters of the foundation treatment scheme into a differentiable deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through a gradient optimization algorithm to obtain the optimal foundation treatment scheme.

[0074] The solution output and iteration unit is used to output the optimal solution and prediction evaluation results, dynamically adjust parameters according to actual on-site needs, and trigger the parameter optimization unit to perform iterative optimization.

[0075] Reference Figure 3 The third aspect of the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor being configured to acquire the computer program and execute the above-described method for intelligent prediction and optimization of road soft soil settlement in time and space.

[0076] It should be noted that the processor, by reading and executing the computer program, utilizes the hardware resources of the computer device and, following the preset steps in the program, sequentially completes a series of operations, including multi-source data preprocessing, network construction and training, settlement prediction and zoning, scheme parameter optimization, and scheme output and adjustment, thereby achieving intelligent spatiotemporal prediction and optimization of road soft soil settlement. By transforming the aforementioned intelligent spatiotemporal prediction and optimization method for road soft soil settlement into an engineering-implementable computer program, and automating the execution of the method through the hardware resources of the computer device, the execution efficiency and ease of operation are significantly improved, reducing errors caused by manual intervention. This allows the method to be quickly applied to actual engineering projects, providing timely and reliable technical support for engineering design and construction.

[0077] In some embodiments, the spatiotemporal intelligent prediction and optimization processing method for road soft soil settlement in the above embodiments can be implemented by a computer device, which includes at least one processor, a communication bus, a memory, and at least one communication interface.

[0078] A processor can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0079] A communication bus can be used to transmit information between the aforementioned components.

[0080] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a communication bus. The memory can also be integrated with the processor.

[0081] The memory stores program code for executing the solution of this application, and its execution is controlled by a processor. The processor executes the program code stored in the memory. The program code may include one or more software modules. In the above embodiments, the spatiotemporal intelligent prediction and optimization processing method for road soft soil settlement can be implemented by the processor and one or more software modules in the program code in the memory.

[0082] A communication interface is a device that uses any transceiver or similar device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0083] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0084] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0085] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described method for intelligent spatiotemporal prediction and optimization of road soft soil settlement.

[0086] To provide a clearer understanding of the invention, the invention is further described below:

[0087] Reference Figure 1 This invention provides a method for spatiotemporal intelligent prediction and optimization of road soft soil settlement, comprising the following steps:

[0088] S1: Obtain multi-source settlement-related data for the entire construction cycle of the regional surface, and perform spatiotemporal alignment, noise reduction, and standardization preprocessing on the multi-source settlement-related data;

[0089] Specifically, the multi-source settlement-related data includes regional surface deformation field data acquired using time-series InSAR technology and concurrent settlement time-series data from monitoring points within the engineering area. The time-series InSAR technology employs Persistent scatterer interferometry (PS-InSAR) or small baseline set interferometry (SBAS-InSAR) to acquire high-precision deformation field data at multiple time points before, during, and after construction.

[0090] ,

[0091] Where T is the number of time points, and H×W is the spatial grid size; the settlement time series data of the monitoring points are as follows: N represents the number of monitoring points.

[0092] The preprocessing of multi-source data includes: first, aligning the InSAR deformation field data and monitoring point data in time and space based on GPS coordinates and timestamps to ensure that the data correspond one-to-one in spatial location and time node; then, using wavelet transform denoising method to remove random noise and systematic errors in the data; finally, using Z-score normalization method to convert the data into standardized data with a mean of 0 and a variance of 1 to eliminate the impact of dimensional differences on subsequent model training.

[0093] S2: Construct and train a spatiotemporal physical information fusion deep learning network. The network takes preprocessed settlement data as input, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, and outputs the settlement prediction values ​​of each monitoring point at multiple times.

[0094] The structure of the spatiotemporal physical information fusion deep learning network includes a convolutional neural network (CNN) layer, a long short-term memory neural network (LSTM) layer, and a physical information constraint module. The specific implementation process is as follows:

[0095] First, the preprocessed settlement data is constructed as a joint spatial-temporal input. For any time t, settlement data from N monitoring points over the previous L time steps are selected to construct a two-dimensional matrix:

[0096] ,

[0097] in, Let be the settlement data vector of all monitoring points at time t.

[0098] Will The input is fed into a CNN layer, where spatial correlation features between monitoring points are extracted through sliding convolutional kernel operations, generating a spatial feature time series. This process effectively captures the spatial interactions between different monitoring points, overcoming the limitations of single-point prediction.

[0099] The spatial feature time series is input into the LSTM layer, and the long-term time dependence of subsidence development is learned by using the gating mechanism of the LSTM unit (input gate, forget gate, output gate), thus characterizing the evolution trend of subsidence over time.

[0100] During network training, a physical information loss term is introduced as a constraint. This physical information loss term is based on Terzaghi's one-dimensional consolidation equation. Constructing the discrete form of the large deformation consolidation equation, combined with the settlement increment formula considering creep:

[0101]

[0102] in, The initial thickness of the soil layer. The initial void ratio, The compression index is... The effective stress at time t-1 This is for additional effective stress (related to the foundation treatment parameter p). The creep coefficient is 1. For reference only.

[0103] By constructing residual terms using the aforementioned physical laws, and minimizing the residuals between the network prediction results and the physical mechanisms through iterative training, the prediction results not only conform to the statistical laws of data, but also follow the basic physical principles of soil consolidation and creep, thereby improving the model's generalization ability and extrapolation reliability.

[0104] The network training uses the Adam optimizer, which combines the mean square error (MSE) between the predicted value and the actual settlement data with the physical information loss term as the total loss function. The training is iterated until the loss function converges, and finally the predicted settlement values ​​of each monitoring point at multiple subsequent time points are output.

[0105] S3: Based on the recursive prediction results of the network, the regional settlement impact is divided into zones, and a set of initial foundation treatment suggestions is generated by combining historical engineering big data.

[0106] First, a pre-trained spatiotemporal physical information fusion deep learning network is used to obtain regional subsidence field distribution data at different future times through a recursive prediction method. During the recursive prediction process, the prediction result of the previous time step is used as the input data for the next time step, achieving multi-step continuous prediction.

[0107] Then, based on the predicted regional settlement field distribution data, settlement contour lines are extracted or settlement gradients are calculated. Combined with the settlement control thresholds set by engineering specifications, the region is automatically classified and zoned. Specifically, it is divided into four levels: major impact zone, large impact zone, general impact zone, and minor impact zone. The zoning result map is output to clarify the degree of settlement risk in different areas.

[0108] The historical engineering big data includes settlement impact zoning records from historical engineering cases within the region, corresponding foundation treatment schemes (including reinforcement methods, design parameters, and construction techniques), soil physical and mechanical parameters, construction cost data, carbon emission data, and post-construction settlement verification data. Based on the currently predicted settlement impact zoning type, similar zoning corresponding to engineering-verified foundation treatment schemes are matched from the historical engineering big data to generate an initial set of foundation treatment recommendation schemes with different probability levels.

[0109] The initial set of solutions is generated using a probability matching formula:

[0110]

[0111] in, This indicates that the settlement gradient is The k-th level processing scheme is recommended for partitioning. The probability of; and These are respectively based on historical data statistics and the proposed solutions. Matching standard settlement gradient thresholds and their standard deviations; This is an adjustable sensitivity coefficient related to the location of the affected partition; K is the total number of preset processing scheme levels. This probability level is used to initialize the design parameter vector and multi-objective weight allocation in the subsequent optimization process.

[0112] S4: Differentiablely embed the key parameters of the foundation treatment scheme into the deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through the gradient optimization algorithm to obtain the optimal foundation treatment scheme;

[0113] First, the key design parameters of the foundation treatment scheme are vectorized into... ,in The specific parameters for each foundation treatment scheme include, but are not limited to, drainage board spacing, drainage board length, vacuum preloading degree, preloading time, mixing pile layout, mixing pile spacing, mixing pile length and replacement rate, etc.

[0114] Then, the parameter vector is determined through differentiable function relations. By embedding a deep learning network that fuses spatiotemporal physical information, the entire settlement prediction model becomes more accurate. Differentiable. The differentiable function relationship is determined based on the working mechanism of different foundation treatment schemes. For example, in the vacuum preloading scheme, the vacuum degree parameter affects the additional effective stress. Embedded models, which in turn affect settlement prediction results.

[0115] A multi-objective optimization reward function is constructed, which comprehensively considers three optimization objectives: settlement control, engineering cost, and carbon emissions. The specific form of this function is as follows:

[0116]

[0117] in, , , This is a weighting coefficient, which can be adjusted according to the actual needs of the project (such as focusing on controlling settlement or prioritizing economic efficiency); Let be the objective function for settlement control. The predicted maximum post-construction settlement, This represents the difference between the actual settlement and the allowable settlement. Used to ensure that post-construction settlement meets the specifications; The cost function is constructed based on the mapping relationship between historical cost data and parameter vector θ, and is used to minimize the total project cost. This is a carbon emission function that calculates carbon emissions based on material consumption and construction energy consumption for different foundation treatment schemes, in order to achieve low-carbon and environmentally friendly goals.

[0118] Gradient optimization algorithms (such as gradient descent and Adam algorithm) are used to optimize the parameter vector. Optimization is performed. The reward function is calculated using automatic differentiation techniques. right The gradient is used to iteratively update the parameters along the ascending gradient until the reward function reaches its maximum value, at which point the corresponding parameter vector is obtained. This represents the optimal parameter combination, and the corresponding foundation treatment scheme is the optimal scheme. The foundation treatment scheme includes one or more combinations of vacuum preloading, surcharge preloading, and cement-soil mixing pile composite foundation methods.

[0119] S5: Output the optimal solution and prediction evaluation results. Dynamically adjust the parameters according to the actual needs on site and repeat step S4 to achieve iterative optimization of the solution.

[0120] Based on the optimal design parameters obtained through optimization, specific ground treatment method combinations, parameter details, and corresponding prediction and evaluation charts (including regional settlement prediction evolution curves, settlement impact zoning maps, cost-carbon emission-settlement relationship diagrams, etc.) are output.

[0121] Based on the actual conditions at the engineering site (such as changes in geological conditions and surrounding environmental constraints) and specific engineering requirements, the foundation treatment parameters are dynamically adjusted. During the adjustment process, the allowable deformation values ​​of adjacent buildings, underground pipelines, or roads are used as the dynamic adjustment threshold to ensure that the adjusted scheme will not adversely affect surrounding facilities. The design parameter vector is then updated after adjustment. Repeat step S4 to perform a new round of parameter optimization, realize iterative optimization of the solution, and ensure the practicality and safety of the solution.

[0122] To make the technical solution of the present invention clearer and easier to understand, the present invention will be described in detail below with reference to specific embodiments.

[0123] This embodiment takes a highway soft soil foundation treatment project in a coastal soft soil area as an example, and applies the spatiotemporal intelligent prediction and optimization treatment method for road soft soil settlement of the present invention. The specific implementation steps are as follows:

[0124] Data acquisition and preprocessing (step S1):

[0125] A 50km stretch along the highway was selected as the research object. SBAS-InSAR technology was used to acquire surface deformation field data at 36 time points from January 2023 to June 2024 (6 months before construction to 6 months during construction). The spatial resolution of the data was 10m × 10m (H=5000, W=100). Meanwhile, 80 settlement monitoring points were evenly distributed throughout the area, and settlement time-series data were collected using settlement meters. ( (Up to 80), with a sampling frequency of once a week.

[0126] The above multi-source data were preprocessed as follows: spatial alignment was achieved by matching the GPS coordinates (error ±2cm) of the monitoring points with the spatial grid of the InSAR data; db4 wavelet transform was used to remove atmospheric delay noise from the InSAR data and measurement errors from the monitoring data; and Z-score normalization was applied. All data are standardized, among which The mean of the data. This represents the standard deviation of the data.

[0127] Network construction and training (step S2):

[0128] A spatiotemporal physical information fusion deep learning network was constructed, in which the CNN layer adopted 2 convolutional layers (with convolutional kernel sizes of 3×3 and 2×2, and output channels of 32 and 64 respectively) and 1 pooling layer (with pooling kernel size of 2×2); the LSTM layer adopted 2 hidden layers, with 128 hidden units in each layer; the physical information constraint module embedded Terzaghi's one-dimensional consolidation equation and creep settlement increment formula.

[0129] The preprocessed data from the first 24 time points was selected as the training set, and the data from the last 12 time points was used as the validation set. The time step size was set to L=6, meaning that the settlement data from the first 6 time steps was used to predict the settlement values ​​for the subsequent 1-3 time steps. The total loss function for network training was the weighted sum of the MSE loss and the physical information loss (weight ratio 0.7:0.3), using the Adam optimizer with a learning rate of 0.001 and 200 iterations. After training, the root mean square error (RMSE) of the settlement prediction on the validation set was 3.2 mm, meeting the engineering accuracy requirement (≤5 mm).

[0130] Settlement zoning and initial scheme generation (step S3):

[0131] The trained network was used to recursively predict the regional settlement field 12 and 24 months after construction, obtaining settlement distribution data at different times. Based on the post-construction settlement control threshold specified in the "Technical Specifications for Design and Construction of Highway Embankments on Soft Soil Foundations" (post-construction settlement of highway embankments ≤ 30cm), and combined with the predicted settlement contour lines, the area was divided into a major impact zone (settlement > 30cm), a large impact zone (20cm < settlement ≤ 30cm), a general impact zone (10cm < settlement ≤ 20cm), and a minor impact zone (settlement ≤ 10cm).

[0132] The region's historical engineering database is accessed, containing data on 20 similar soft soil road projects from the past 10 years. Based on the current settlement zoning results, a probability matching formula is used... Generate an initial set of foundation treatment schemes: The recommended scheme for major impact areas is the combination of "cement-soil mixing piles + vacuum preloading" (probability 0.85); the recommended scheme for large impact areas is vacuum preloading (probability 0.78); the recommended scheme for general impact areas is surcharge preloading (probability 0.82); and the recommended scheme for minor impact areas is shallow replacement (probability 0.90).

[0133] Parameter optimization (step S4):

[0134] Taking a major impact area as an example, determine the key parameter vector of the foundation treatment scheme. ,in The spacing between the mixing piles is in meters (m). The length of the mixing pile is in meters. Vacuum degree (kPa) Pre-compression time (d). The replacement rate of the mixing piles (%). This is calculated using a linearly differentiable function. Embedded deep learning networks, where vacuum degree By influencing the additional effective stress Implement embedding.

[0135] Set multi-objective optimization weight coefficients , , Construct a reward function The Adam optimization algorithm is used to optimize the performance of the target system. After 50 iterations of optimization, the optimal parameter vector is obtained. The corresponding optimal solution is "mixing pile spacing 1.5m, length 18m, vacuum degree 85kPa, preloading time 180d, replacement rate 25%". At this time, the predicted post-construction settlement is 28.5cm, the project cost is reduced by 12% compared with the initial solution, and the carbon emissions are reduced by 8%.

[0136] Solution output and iteration (step S5):

[0137] Output the above optimal solution and related prediction and evaluation charts, including the settlement evolution curve 24 months after construction, the settlement impact zoning map, a cost-carbon emission-settlement road soft soil foundation settlement spatiotemporal intelligent prediction, optimization treatment method, and system three-dimensional relationship diagram. On-site investigation revealed an underground gas pipeline within the significant impact area, with an allowable deformation value of 25cm. Therefore, this value was used as the dynamic adjustment threshold to adjust the optimal parameters: the length of the mixing piles was increased to 20m, the replacement rate was increased to 28%, and step S4 was re-executed for optimization to obtain the final optimal parameter vector. The predicted post-construction settlement is 24.8cm, which meets the allowable deformation requirements for underground pipelines, and the cost and carbon emission increases are controlled within 5%.

[0138] This embodiment achieves high-precision spatiotemporal prediction of soft soil settlement and intelligent optimization of foundation treatment schemes for highways by applying the method of the present invention. It effectively solves the problems of insufficient prediction accuracy and low optimization efficiency of traditional methods, and provides scientific and reliable technical support for engineering construction.

[0139] Compared with the prior art, the present invention has the following advantages:

[0140] This invention constructs a spatiotemporal physical information fusion deep learning network, which integrates the spatial correlation characteristics and temporal evolution laws of multi-source monitoring data and embeds soil consolidation physical constraints, thereby achieving high-precision spatiotemporal prediction of regional settlement fields. It breaks through the limitations of traditional methods in terms of spatiotemporal fragmentation and weak physical mechanisms, and improves the prediction accuracy by more than 30% compared with traditional methods.

[0141] This invention embeds the differentiable parameters of the foundation treatment scheme into the prediction model, and combines multi-objective optimization functions and gradient optimization algorithms to realize a closed loop from settlement prediction to scheme optimization, replacing the traditional "simulation-trial calculation" mode. The optimization efficiency is improved by more than 80%, and it can simultaneously meet the multi-objective requirements of settlement control, cost saving and low carbon environmental protection.

[0142] This invention generates an initial set of solutions based on historical engineering big data, and combines dynamic threshold adjustment to achieve iterative optimization of the solutions, thereby enhancing the practicality and compatibility of the solutions. It can adapt to different geological conditions and surrounding environmental requirements, and provides intelligent and personalized solutions for road engineering in soft soil areas.

[0143] The technical solutions provided by the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the embodiments of the present invention. The descriptions of the embodiments above are only for helping to understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for spatiotemporal intelligent prediction and optimization of road soft soil settlement, characterized in that, Includes the following steps: S1: Obtain multi-source subsidence related data of the regional surface over a period of time, and perform spatiotemporal alignment, denoising and standardization preprocessing on the multi-source subsidence related data; S2: Construct and train a spatiotemporal physical information fusion deep learning network. The deep learning network takes preprocessed multi-source settlement-related data as input, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, and outputs settlement prediction values ​​of each monitoring point at multiple times. S3: Based on the recursive prediction results of the deep learning network, the regional settlement impact is divided into zones, and a set of initial foundation treatment suggestions is generated by combining historical engineering big data. S4: Differentiablely embed the key parameters of the foundation treatment scheme into the deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through the gradient optimization algorithm to obtain the optimal foundation treatment scheme; S5: Output the optimal solution and prediction evaluation results. Adjust the parameters dynamically according to the actual needs on site and repeat step S4 to achieve iterative optimization of the solution.

2. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 1, characterized in that: In step S1, the multi-source settlement-related data includes regional surface deformation field data obtained through persistent scatterer interferometry or small baseline set interferometry, as well as synchronous settlement time series data of monitoring points within the engineering area.

3. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 1, characterized in that: In step S2, the physical constraints of soil consolidation are constructed based on the discrete forms of Terzaghi's one-dimensional consolidation equation and large deformation consolidation equation. By introducing physical information loss terms, the network prediction results are constrained to conform to the soil consolidation and creep mechanism.

4. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 1, characterized in that: In step S3, the settlement impact zone is automatically divided into four levels: major impact zone, large impact zone, general impact zone, and minor impact zone, based on the predicted settlement contour lines or settlement gradient and the settlement control threshold of the engineering specifications.

5. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 1, characterized in that: The historical engineering big data includes a library of engineering-verified foundation treatment schemes corresponding to similar settlement impact zones within the region, soil physical and mechanical parameters, construction records, and post-construction settlement monitoring data, providing data support for the generation of initial schemes.

6. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 1, characterized in that: In step S4, the key parameter vector of the foundation treatment scheme ,in This includes the spacing and length of drainage boards, vacuum degree, preloading time, arrangement, spacing, length and replacement rate of mixing piles.

7. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 6, characterized in that: In step S4, the multi-objective optimization function is: ， in, , , These are the weighting coefficients. Let be the objective function for settlement control. The predicted maximum post-construction settlement, This represents the difference between the actual settlement and the allowable settlement. For cost function, This is a carbon emission function.

8. The method for intelligent spatiotemporal prediction and optimization of road soft soil settlement according to claim 1, characterized in that: In step S4, the foundation treatment scheme includes one of the following methods: replacement method, vacuum preloading method, surcharge preloading method, cement-soil mixing pile composite foundation method, and pipe pile composite foundation method, using different parameter values, or a combination of multiple methods using different parameter values.

9. A spatiotemporal intelligent prediction and optimization system for road soft soil settlement, implementing the spatiotemporal intelligent prediction and optimization method for road soft soil settlement as described in any one of claims 1 to 8, characterized in that, include: The data acquisition and preprocessing unit is used to acquire multi-source settlement-related data throughout the entire construction cycle of the regional surface, and to perform spatiotemporal alignment, noise reduction, and standardization preprocessing. The network construction and training unit is used to build a spatiotemporal physical information fusion deep learning network. It inputs preprocessed multi-source settlement-related data into the deep learning network, integrates spatial correlation feature extraction, time-dependent law learning and soil consolidation physical constraints, completes the training of the deep learning network and outputs settlement prediction values. The settlement zoning and initial scheme generation unit is used to zon the regional settlement impact based on the prediction results of deep learning network, and to generate a set of initial foundation treatment suggestion schemes by combining historical engineering big data. The parameter optimization unit is used to embed the key parameters of the foundation treatment scheme into a differentiable deep learning network, construct a multi-objective optimization function, and optimize the parameter combination through a gradient optimization algorithm to obtain the optimal foundation treatment scheme. The solution output and iteration unit is used to output the optimal solution and prediction evaluation results, dynamically adjust parameters according to actual on-site needs, and trigger the parameter optimization unit to perform iterative optimization.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the spatiotemporal intelligent prediction and optimization processing method for road soft soil settlement as described in any one of claims 1 to 8.

Images