Light soil subgrade filling construction method and system close to pier column

Abstract

The application discloses a lightweight soil roadbed filling construction method and system close to a pier column, and the method comprises the following steps: inputting small-range test filling data in the early stage into BP neural network training to output adaptive reference parameters; performing construction according to the reference parameters, and first completing early-stage treatment around the pier column; when performing layer-by-layer filling, pressure, flow and pouring surface elevation data are collected through a pumping equipment sensor, GNSS is used to monitor the displacement of the pier column, and radar detection or elastic wave testing is used to obtain compactness data; after data normalization, the network is operated to extract the correlation law, and the correlation law is compared with a target threshold value; if the compactness does not meet the standard, the weight is corrected reversely, and a pumping parameter adjustment instruction is output; if the displacement is close to a warning value, a layer-by-layer thickness and a foam particle content adjustment scheme are generated; the transition section parameters are determined according to the settlement prediction value at the joint, and the data after construction is used to optimize the parameters of the next section, so that a dynamic closed loop is formed; and the overall quality of the project is improved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent roadbed construction control technology, and more specifically, relates to a method and system for filling lightweight soil roadbeds adjacent to piers. Background Technology

[0002] In highway and bridge connection projects, the construction of roadbed filling adjacent to piers must simultaneously meet the roadbed bearing capacity requirements and the structural safety of the piers, which has always been a key and challenging area in engineering practice. Currently, construction in this scenario relies heavily on traditional experience parameters, such as determining the lightweight soil mix ratio, layer thickness, and pumping parameters based on similar engineering cases. This lacks targeted adaptation to the specific geological conditions around the piers. In some areas, due to differences in geological bearing capacity and changes in groundwater level, construction using fixed parameters can easily lead to insufficient roadbed compaction or uneven flow and diffusion of lightweight soil, resulting in excessive pier displacement. This not only affects the long-term stability of the roadbed but may also cause structural damage to the piers.

[0003] During the layered filling process, traditional construction methods rely heavily on intermittent manual data collection for monitoring key indicators. This includes manually recording pumping equipment operating parameters, periodically measuring pier displacement, and sampling to test the compaction of lightweight soil. However, data acquisition is often delayed. When compaction fails to meet standards or pier displacement approaches warning levels, construction must be halted and the plan adjusted. This not only extends the construction period but may also lead to repeated trial and error due to a lack of data support for parameter adjustments, increasing construction costs. Furthermore, the data from manual monitoring is highly discrete, making it difficult to establish a continuous construction quality control chain. It cannot provide real-time feedback on the correlation between pumping parameters and quality and safety indicators, and it is also ill-suited to the dynamic compaction characteristics of lightweight soil during the fluid filling process.

[0004] Furthermore, controlling differential settlement at the junction of lightweight soil subgrade and traditional subgrade is another core challenge. Due to the fundamental differences in material properties and hardening processes between lightweight soil and traditional subgrade, subgrade cracking and pavement damage are prone to occur during the operational period due to inconsistent settlement. In existing technologies, transition section construction parameters are mostly determined based on static design specifications, failing to fully consider the dynamic changes in geological conditions and the flow and diffusion effects of lightweight soil during actual construction. Parameters such as the pouring boundary slope, which are adapted to the characteristics of lightweight soil, are also not specifically incorporated. This makes it difficult to accurately match the needs of differential settlement control, resulting in the junction section becoming a weak link in the subgrade engineering, leading to high maintenance costs and impacting the overall service life and traffic safety of the road. Summary of the Invention

[0005] This invention addresses the challenge of constructing lightweight soil subgrades adjacent to piers by integrating trial filling data using a BP neural network to determine benchmark values ​​such as mix proportions and pumping parameters. It dynamically adjusts parameters in real time by collecting pumping, displacement, and compaction data, and optimizes the transition section by combining differential settlement prediction. This collaborative approach controls quality, ensures pier safety, and improves efficiency and stability.

[0006] In view of the above-mentioned defects or improvement needs of the existing technology, as a first aspect of the present invention, the present invention provides a method for constructing a lightweight soil subgrade adjacent to a pier, comprising: S1. Collect geological data around the pier and determine the construction target. Incorporate the measured data of the small-scale trial filling in the early stage of the project into the training sample of the BP neural network and output the appropriate benchmark parameters, including the lightweight soil mix ratio, layer thickness, and pumping parameters. S2. Based on the model reference parameters, the preliminary treatment around the pier is completed first; when filling lightweight soil in layers, the pumping equipment sensors collect data on pumping pressure, flow rate and elevation changes of the pouring surface, GNSS equipment monitors the displacement of the pier, and non-destructive testing methods including radar detection or elastic wave testing are used to obtain lightweight soil compaction data. The three types of data are transmitted to the BP neural network simultaneously. The S3.BP neural network first normalizes the pumping parameters, displacement, and compaction data, then passes them through the input layer to the hidden layer for computation. It extracts the correlation patterns among the three and compares them with the construction target threshold. If the compaction does not meet the standard, the network corrects the neuron connection weights through backpropagation, recalculates, and outputs instructions to adjust the pumping parameters. If the pier displacement is close to the warning threshold, a scheme to adjust the layer thickness and foam particle content is generated in the same way. After each parameter adjustment, the relevant data is automatically added to the training set as new samples to iteratively optimize the model accuracy. S4. At the junction of lightweight soil and traditional subgrade, the construction parameters of the transition section are determined based on the differential settlement prediction value output by the BP neural network. After construction, the settlement data of the transition section is incorporated into the model to optimize the parameters of the next section, forming a closed loop of dynamic optimization throughout the entire process.

[0007] Furthermore, the geological data in S1 includes: detailed geological survey results within a 3m radius around the pier, specifically covering the physical and mechanical parameters of each soil layer, groundwater level elevation and seasonal variation, soil layer distribution morphology, and special adverse geological conditions.

[0008] Furthermore, the calculation process of the reference parameters in S1 is as follows: Let the geological feature vector be The test filling measured matrix is ​​as follows: ,in, Indicates the first The first group of filling tests Item index values; construct gradient coupling functions to quantify the intrinsic relationship between geological parameters and trial filling effects: , In the formula, For Hamiltonian operators; The parameter solution space encompasses geological parameters. With test filling indicators All possible values; This is a vector dot product operation, representing the intensity of the gradient influence of geological parameters on the trial filling effect; Construct a pre-defined performance surface based on the performance targets of lightweight soil. The mix proportions are determined by minimizing the gradient difference: , In the formula, , For cement admixture, For the amount of lightweight aggregate, This refers to the water-to-binder ratio; It is an L2 norm; The gradient vector of the performance target surface, which includes flowability, setting time, and strength indices; Based on the Jacobian matrix determinant characterizing parameter sensitivity, the layer thickness is derived. : , In the formula, for Jacobian matrix, ; This refers to matrix determinant operations. Given the inverse of the Jacobian matrix, the layer thickness suitable for the compaction equipment's performance is determined through parameter sensitivity inversion; based on the analysis of the viscous fluid motion equation, the correlation between pumping parameters and the flow characteristics of lightweight soil is established. , In the formula, The density of lightweight soil, The velocity vector is the flow velocity vector. For time, For pumping pressure, The dynamic viscosity coefficient of lightweight soil is calculated from the flowability and spread measured during trial filling. Let gravitational acceleration be the acceleration due to gravity. By solving the steady-state solution of this equation, the corresponding optimal pumping pressure can be obtained. With traffic .

[0009] Furthermore, the specific process of the pre-treatment around the pier in S2 is as follows: clear surface debris within a range of no less than 2m around the pier down to the original soil surface; level the base to ensure that the slope is no more than 1% to meet the flow and diffusion requirements of lightweight soil; lay a graded crushed stone cushion layer with a thickness of no less than 30cm and compact it according to the benchmark parameters; treat the surface of the cushion layer with laser leveling technology to ensure that the elevation deviation is controlled within ±5mm, so as to provide a uniform bearing base for lightweight soil; and attach a cement corrosion resistant polyethylene film around the base of the pier to form a protective layer. The height of the film should be no less than 50cm above the top surface of the cushion layer, and the joints should be sealed with hot-melt welding to prevent grout from seeping into the gap between the pier and the cushion layer during the flow and filling of lightweight soil.

[0010] Furthermore, the specific steps for extracting the correlation patterns among the three elements in S3 are as follows: Let the normalized pumping parameter vector be , This is the normalized value of the pumping pressure. The normalized value for flow rate is [value], and the displacement data is [value]. The density data is The three constitute the input space vector. Feature mapping is achieved through nonlinear transformation of the hidden layer. , In the formula, This is the connection matrix from the input layer to the hidden layer. This is the hidden layer bias vector. It is the Sigmoid activation function. The high-dimensional feature vector output by the hidden layer; Further, association patterns are constructed using feature space distance metrics: , In the formula, For the sample data space, For neighboring sample vectors, The similarity function characterizes the physical relationship between samples; As a feature space correlation index, by minimizing Extract the intrinsic coupling law of "compaction parameters-displacement-compaction degree" in high-dimensional space, and complete the quantitative characterization of the dynamic correlation among the three.

[0011] Furthermore, the specific process of correcting the neuron connection weights through backpropagation in S3 is as follows: Let the loss function between the output layer prediction and the target threshold be... For cases where the density does not meet the standard, , The target value for compaction, The predicted compaction value output by the network; During backpropagation, the partial derivatives of the loss function with respect to the output layer weights are calculated first: , In the formula, For the hidden layer The nth neuron to the output layer The connection weights of each neuron; For the output layer The input of each neuron, , For the hidden layer The output of each neuron For the output layer Bias of each neuron; Then, the partial derivatives of the weights in the hidden layer and the input layer are passed to each other using the chain rule: , In the formula, For the input layer The first neuron to the hidden layer The connection weights of each neuron; For the hidden layer The input of each neuron, , For the input layer The input of each neuron, For the hidden layer Bias of each neuron; Finally, update the weights using gradient descent: , In the formula, The learning rate is determined by the convergence characteristics of the data during the trial filling stage; For the updated weights, To update the weights, repeat the above process until... , The weights are corrected to the preset precision threshold.

[0012] Furthermore, the predicted differential settlement value in S4 is specifically as follows: Let the monitoring objects on both sides of the connection section be lightweight soil objects. and traditional roadbed side objects Define the following variables: the time variable is... , , The duration of post-construction operation; Target The settlement function is Output by a BP neural network, with input variables including geological parameters. Lightweight soil construction parameters and settlement data of the filled section ; object The settlement function is Similarly, the output of a BP neural network is the same as the input variable. Consistent; Differential settlement prediction values ​​at the same monitoring section and the same time point The expression is: , In the formula, the absolute value sign represents the magnitude of the settlement difference; the model learns the mapping relationship between the above variables and settlement, focusing on strengthening the influence weights of the flow characteristics and hardening process of lightweight soil on settlement development, and outputs the results at any time during the entire operation period. corresponding This provides a quantitative basis for optimizing the construction plan for the transition section.

[0013] Furthermore, the process for confirming the construction parameters of the transition section in S4 is as follows: Let the transition section parameter vector be... , The width of the step. The extension length of the geogrid. The spacing between blind drains, The boundary slope is to be constructed using lightweight soil; the predicted differential settlement value is... The control target is , To allow for the maximum differential settlement threshold, a functional extremum model of the parameter-settlement correlation is constructed: , In the formula, For the coupling operator of parameters and settlement gradient, For gradient operators, For variational operators, For the transition zone spatial domain, the intrinsic relationship between parameters and differential settlement is established through functional extremum conditions, while strengthening... Its regulatory effect on the flow settlement of lightweight soil; Constrained by control objectives, In the whole domain It must be established within the specified time and must meet the following requirements. The corresponding flow distance of the lightweight soil does not exceed the boundary of the transition section. Solving the variational equation yields the parameter solution set: , Introducing construction feasibility mapping The relationship between the characterization parameters and the field conditions is described, and the final parameters are: , In the formula, This is the feasible area under the on-site conditions. This is the inverse operation of the mapping, which uses set operations to select parameter combinations from the solution set that simultaneously satisfy settlement control and construction feasibility.

[0014] As a second aspect of the present invention, a lightweight soil subgrade filling construction system adjacent to the pier is also provided, comprising: The benchmark parameter confirmation unit is used to collect geological data around the pier and determine the construction target. It incorporates the measured data of the small-scale trial filling in the early stage of the project into the training sample of the BP neural network and outputs the appropriate benchmark parameters, including the lightweight soil mix ratio, layer thickness, and pumping parameters. The construction data acquisition unit is used for construction based on model reference parameters. First, the pre-treatment around the pier is completed. When filling lightweight soil in layers, the pumping equipment sensors collect data on pumping pressure, flow rate and elevation changes of the pouring surface, GNSS equipment monitors the displacement of the pier, and non-destructive testing methods including radar detection or elastic wave testing are used to obtain lightweight soil compaction data. The three types of data are transmitted to the BP neural network simultaneously. The parameter dynamic adjustment unit is used by the BP neural network to first normalize the pumping parameters, displacement, and compaction data, then pass them through the input layer to the hidden layer for computation, extract the correlation between the three and compare them with the construction target threshold; if the compaction does not meet the standard, the network corrects the neuron connection weights through backpropagation, recalculates, and outputs instructions to adjust the pumping parameters; if the pier displacement is close to the warning threshold, the same approach is used to generate a scheme to adjust the layer thickness and foam particle content; after each parameter adjustment, the relevant data is automatically added to the training set as new samples to iteratively optimize the model accuracy; The transition section parameter confirmation unit is used to determine the construction parameters of the transition section at the junction of lightweight soil and traditional subgrade based on the differential settlement prediction value output by the BP neural network. After construction, the settlement data of the transition section is incorporated into the model to optimize the parameters of the next section, forming a closed loop of dynamic optimization throughout the entire process.

[0015] As a third aspect of the invention, a computer-readable storage medium is also provided, on which a computer program is stored, which is executed by a processor, any step of the described method for constructing a lightweight soil subgrade adjacent to a pier.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: 1. This invention discloses a method for constructing lightweight soil subgrade adjacent to piers. Detailed geological data within a 3-meter radius of the pier is collected, and a backpropagation (BP) neural network is trained using experimental filling data. The network outputs suitable benchmark parameters, including lightweight soil mix proportions, layer thicknesses, and pumping parameters. The geological data encompasses soil physical and mechanical parameters and groundwater level changes, while the experimental filling data includes flow diffusion effects and compaction under different parameters. This process quantifies the correlation between geological conditions and construction effects using gradient coupling functions and determines pumping parameters based on the viscous fluid motion equation. This ensures that the benchmark parameters accurately match the on-site geological characteristics, solving the problem of poor adaptability of traditional empirical parameters. This provides a scientific initial basis for subsequent construction and ensures the uniformity and compaction of the lightweight soil flow filling.

[0017] 2. This invention discloses a method for constructing lightweight soil subgrade adjacent to piers, which involves real-time collection of three types of data during the layered filling stage and transmission to a BP neural network for dynamic parameter control. Specifically, this includes the pressure and flow rate of the pumping equipment, pier displacement monitored by GNSS, and the compaction of the lightweight soil detected by radar or elastic waves. After normalization, the network extracts the correlation between these three data points. If the compaction is insufficient, the weights are adjusted, and pumping parameter adjustment commands are output. If the pier displacement approaches a warning value, the layer thickness and material content are optimized. This dynamic control mechanism replaces traditional intermittent manual monitoring, eliminates data lag, avoids repeated trial and error, achieves a real-time balance between quality and safety indicators during construction, and improves construction efficiency and pier structural safety.

[0018] 3. This invention provides a method for constructing a lightweight soil subgrade adjacent to piers. It utilizes a BP neural network to output predicted differential settlement values ​​at the transition section between the lightweight soil and the traditional subgrade, thereby optimizing the construction parameters of the transition section. The predicted values ​​are determined based on the difference in settlement functions on both sides. Parameters include step width, geogrid length, blind drain spacing, and lightweight soil pouring boundary slope. A functional extremum model is used to establish the correlation between parameters and settlement, and the final parameters are selected using a construction feasibility mapping. After construction, actual settlement data is fed back into the model, forming a dynamic optimization closed loop. This process specifically addresses the problem of differential settlement at the transition section, ensuring that parameters are adapted to the flow characteristics of lightweight soil and site conditions, reducing the risk of subgrade cracking during operation, lowering later maintenance costs, and improving the overall stability and service life of the road. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a construction method for filling a lightweight soil subgrade adjacent to a pier, according to an embodiment of the present invention. Figure 2 This is a system unit diagram of an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] Example 1 Please refer to Figure 1 This embodiment 1 provides a method for constructing a lightweight soil subgrade adjacent to a pier, including: S1. Collect geological data around the pier and determine the construction target. Incorporate the measured data of the small-scale trial filling in the early stage of the project into the training sample of the BP neural network and output the appropriate benchmark parameters, including the lightweight soil mix ratio, layer thickness, and pumping parameters. S2. Based on the model reference parameters, the preliminary treatment around the pier is completed first; when filling lightweight soil in layers, the pumping equipment sensors collect data on pumping pressure, flow rate and elevation changes of the pouring surface, GNSS equipment monitors the displacement of the pier, and non-destructive testing methods including radar detection or elastic wave testing are used to obtain lightweight soil compaction data. The three types of data are transmitted to the BP neural network simultaneously. The S3.BP neural network first normalizes the pumping parameters, displacement, and compaction data, then passes them through the input layer to the hidden layer for computation. It extracts the correlation patterns among the three and compares them with the construction target threshold. If the compaction does not meet the standard, the network corrects the neuron connection weights through backpropagation, recalculates, and outputs instructions to adjust the pumping parameters. If the pier displacement is close to the warning threshold, a scheme to adjust the layer thickness and foam particle content is generated in the same way. After each parameter adjustment, the relevant data is automatically added to the training set as new samples to iteratively optimize the model accuracy. S4. At the junction of lightweight soil and traditional subgrade, the construction parameters of the transition section are determined based on the differential settlement prediction value output by the BP neural network. After construction, the settlement data of the transition section is incorporated into the model to optimize the parameters of the next section, forming a closed loop of dynamic optimization throughout the entire process.

[0022] This embodiment 1 further elaborates on the above steps.

[0023] (1) Confirmation of benchmark parameters In the construction of lightweight soil subgrade adjacent to piers, the traditional method of relying on experience parameters is prone to quality and safety problems due to insufficient adaptation to geological conditions. Moreover, the lack of accurate data support for the determination of parameters in the early stage can create hidden dangers for subsequent construction. Therefore, it is necessary to complete the geological data collection and benchmark parameter calculation first to lay the foundation for the entire construction process.

[0024] Construction personnel first collected detailed geological data within a 3-meter radius of the pier, including the physical and mechanical parameters of each soil layer, groundwater level elevation and seasonal variation, and soil layer distribution. They also investigated any unusual geological conditions. This data directly relates to the suitability of subsequent construction parameters and is the core basis for defining construction objectives, including the uniformity of lightweight soil flow and diffusion, and meeting compaction requirements. Combined with measured data from small-scale trial filling in the early stages of the project—including the flowability, spread, and corresponding compaction of lightweight soil under different mix proportions—this data was incorporated into the training samples of a BP neural network. The network computation outputs baseline parameters including the lightweight soil mix proportion, layer thickness, and pumping parameters. The network computation process includes gradient coupling functions to quantify the correlation between geology and trial filling effects, and viscous fluid motion equations to analyze pumping characteristics, ensuring that the parameters initially match the site geological conditions, the characteristics of lightweight soil filling without compaction based on flowability, and construction requirements.

[0025] Specifically, the calculation process for the baseline parameters is as follows: Let the geological feature vector be The test filling measured matrix is ​​as follows: ,in, Indicates the first The first group of filling tests Item index values; construct gradient coupling functions to quantify the intrinsic relationship between geological parameters and trial filling effects: , In the formula, For Hamiltonian operators; The parameter solution space encompasses geological parameters. With test filling indicators All possible values; This is a vector dot product operation, representing the intensity of the gradient influence of geological parameters on the trial filling effect; Construct a pre-defined performance surface based on the performance targets of lightweight soil. The mix proportions are determined by minimizing the gradient difference: , In the formula, , For cement admixture, For the amount of lightweight aggregate, This refers to the water-to-binder ratio; It is an L2 norm; The gradient vector of the performance target surface, which includes flowability, setting time, and strength indices; Based on the Jacobian matrix determinant characterizing parameter sensitivity, the layer thickness is derived. : , In the formula, for Jacobian matrix, ; This refers to matrix determinant operations. Given the inverse of the Jacobian matrix, the layer thickness suitable for the compaction equipment's performance is determined through parameter sensitivity inversion; based on the analysis of the viscous fluid motion equation, the correlation between pumping parameters and the flow characteristics of lightweight soil is established. , In the formula, The density of lightweight soil, The velocity vector is the flow velocity vector. For time, For pumping pressure, The dynamic viscosity coefficient of lightweight soil is calculated from the flowability and spread measured during trial filling. Let gravitational acceleration be the acceleration due to gravity. By solving the steady-state solution of this equation, the corresponding optimal pumping pressure can be obtained. With traffic .

[0026] The benchmark parameters determined through this series of calculations not only solve the problem of blindness in traditional empirical parameters, but also provide a precise initial basis for subsequent construction, data collection and dynamic parameter adjustment.

[0027] (2) Construction data collection Construction is carried out under the guidance of benchmark parameters. First, preliminary treatment is carried out around the pier. Centered on the pier, surface debris, including vegetation remnants, gravel, and loose topsoil, within a radius of at least 2 meters around the pier is cleared until the original, structurally intact soil surface is exposed. During this process, a level is used to calibrate the flatness of the base, ensuring the slope does not exceed 1%, creating conditions for subsequent flow and diffusion of lightweight soil. When laying the graded crushed stone cushion layer, the particle size distribution of the cushion layer is adjusted according to the differences in soil bearing capacity in the geological data, and the thickness is controlled to be no less than 30 centimeters. Then, the design is carried out according to the benchmark parameters. The vibration method and duration of the operation are determined, and laser leveling technology is used to treat the surface of the cushion layer to ensure that the elevation deviation is controlled within ±5 mm, forming a flat and stable transition bearing layer. Before pasting the polyethylene film around the base of the pier, the surface of the pier is cleaned and decontaminated. A film resistant to cement corrosion is selected and is laid to ensure that it is tightly attached to the surface of the pier. The height exceeds the top surface of the cushion layer by no less than 50 cm. The joints are sealed with hot melt welding to form a continuous and gapless protective layer, which prevents the grout from seeping into the gap between the pier and the cushion layer during the subsequent flow and filling of lightweight soil.

[0028] During the layered filling of lightweight soil, a multi-dimensional collaborative data acquisition system was established. Dedicated sensors on the pumping equipment collected real-time pumping pressure and flow rate during lightweight soil delivery, simultaneously recording changes in the elevation of the pouring surface. The data underwent noise reduction and filtering preprocessing by the onboard data processing unit before being stably transmitted via a wireless transmission module. GNSS monitoring equipment installed on the top of the piers continuously collected three-dimensional displacement information of the piers, synchronously recording the time node of each data acquisition to ensure accurate correspondence between displacement changes and construction procedures. Non-destructive testing equipment, such as radar detection or elastic wave testing, was used to determine the distribution of detection points according to the filling area and layer thickness. Built-in sensing components collected the internal structural response signals of the lightweight soil, rapidly generating quantitative data on compaction. All three types of collected data underwent time synchronization calibration to eliminate time differences between devices, and then were transmitted in batches at second-level intervals to the data receiving layer of the BP neural network, forming a real-time monitoring data stream covering the entire filling process and with coherent data logic, providing comprehensive data support for subsequent parameter adjustments.

[0029] (3) Dynamic adjustment of parameters The BP neural network first normalizes the collected pumping parameters, pier displacement, and compaction data to unify the data scale before feeding them into the input layer, and then into the hidden layer for computation. In the hidden layer, nonlinear transformations are used to achieve feature mapping, converting the input data into high-dimensional feature vectors. Then, by calculating the distance and similarity between samples in the feature space, a correlation index is constructed to extract the inherent coupling patterns of the three data points in the high-dimensional space. The pumping parameters include pressure and flow data during the lightweight soil transportation process, and the compaction data is obtained through non-destructive testing methods such as radar detection or elastic wave testing, ensuring that the extracted patterns are suitable for the filling characteristics of lightweight soil that is uncompacted and relies on fluidity for filling.

[0030] Specifically, the relationship between the three is extracted as follows: Let the normalized pumping parameter vector be , This is the normalized value of the pumping pressure. The normalized value for flow rate is [value], and the displacement data is [value]. The density data is The three constitute the input space vector. Feature mapping is achieved through nonlinear transformation of the hidden layer. , In the formula, This is the connection matrix from the input layer to the hidden layer. This is the hidden layer bias vector. It is the Sigmoid activation function. The high-dimensional feature vector output by the hidden layer; Further, association patterns are constructed using feature space distance metrics: , In the formula, For the sample data space, For neighboring sample vectors, The similarity function characterizes the physical relationship between samples; As a feature space correlation index, by minimizing Extract the intrinsic coupling law of "compaction parameters-displacement-compaction degree" in high-dimensional space, and complete the quantitative characterization of the dynamic correlation among the three.

[0031] Subsequently, the network compares the extracted correlation patterns with the construction target threshold. If the compaction degree does not meet the standard, the network initiates a backpropagation mechanism to correct the neuron connection weights. First, a loss function is constructed using the difference between the target compaction degree and the predicted value. The partial derivative of this function with respect to the output layer weights is calculated. Then, the derivative is passed to the weights of the hidden and input layers using the chain rule. Finally, all weights are updated using the gradient descent method until the loss function value is lower than the preset accuracy threshold. After recalculation, an instruction to adjust the pumping parameters is output. The pumping parameters include the pressure and flow rate during lightweight soil transportation to ensure that the parameter adjustment can adapt to the characteristics of lightweight soil relying on fluidity for filling, thereby promoting the achievement of the compaction degree standard.

[0032] Specifically, the process of correcting neuron connection weights through backpropagation is as follows: Let the loss function between the output layer prediction and the target threshold be... For cases where the density does not meet the standard, , The target value for compaction, The predicted compaction value output by the network; During backpropagation, the partial derivatives of the loss function with respect to the output layer weights are calculated first: , In the formula, For the hidden layer The nth neuron to the output layer The connection weights of each neuron; For the output layer The input of each neuron, , For the hidden layer The output of each neuron For the output layer Bias of each neuron; Then, the partial derivatives of the weights in the hidden layer and the input layer are passed to each other using the chain rule: , In the formula, For the input layer The first neuron to the hidden layer The connection weights of each neuron; For the hidden layer The input of each neuron, , For the input layer The input of each neuron, For the hidden layer Bias of each neuron; Finally, update the weights using gradient descent: , In the formula, The learning rate is determined by the convergence characteristics of the data during the trial filling stage; For the updated weights, To update the weights, repeat the above process until... , The weights are corrected to the preset precision threshold.

[0033] If the pier displacement approaches the warning threshold, the network employs a similar backpropagation correction mechanism to generate schemes that adjust the layer thickness and lightweight aggregate content. Layer thickness adjustment must be adapted to the flow and diffusion range of the lightweight soil to avoid uneven flow and localized pressure concentration caused by excessive thickness. Lightweight aggregate content adjustment reduces the additional load on the foundation and pier by altering the overall material density. After each parameter adjustment, relevant data is automatically added to the training set. Through continuous iteration and optimization, the model's accuracy in capturing the correlation between construction parameters and quality and safety indicators is improved, ensuring that the adjustment scheme always adapts to the characteristics of lightweight soil—uncompacted and reliant on fluidity for filling.

[0034] (4) Confirmation of transition section parameters At the junction of lightweight soil and traditional subgrade, the monitoring targets on both sides of the junction should be clearly defined, namely the lightweight soil side and the traditional subgrade side. Lightweight soil and traditional subgrade have fundamentally different material properties (such as density and compression modulus) and filling mechanisms. Lightweight soil relies on fluidity for filling without compaction, while traditional subgrade relies on compaction. During filling and operation, they are prone to differential settlement due to differences in settlement rate and final settlement amount. Differential settlement is the core cause of pavement cracking and structural damage at the junction. Clearly defining the independent monitoring targets on both sides avoids the problem of inaccurately distinguishing the settlement contributions of both sides when the junction is monitored as a whole. This allows for more precise capture of their respective settlement patterns—facilitating subsequent quantification of the settlement difference between the two and providing data support for tracing the source of differential settlement (such as the shrinkage of lightweight soil during hardening or secondary consolidation of the traditional subgrade). This provides a targeted basis for the parameter design of the transition section from the source, avoiding the inability of parameter design to accurately match settlement control requirements due to ambiguity in the monitoring targets.

[0035] The BP neural network takes geological parameters, construction parameters (including lightweight soil mix proportions, pumping parameters, and layer thickness), and settlement data of the filled section as inputs. It outputs settlement functions for both sides of the object. By calculating the absolute value of the difference in settlement values ​​between the two sides at the same monitoring section and time point, a differential settlement prediction value is obtained. This prediction value covers the settlement differences at any time point throughout the entire operation period. Furthermore, the modeling process emphasizes the influence weights of the lightweight soil flow diffusion effect and the hardening process on its settlement function, ensuring that the prediction results accurately reflect the settlement difference patterns under the lightweight soil filling characteristics. This provides a precise basis for subsequent parameter optimization of the transition section. The specific calculation process for the differential settlement prediction value is as follows: Let the monitoring objects on both sides of the connection section be lightweight soil objects. and traditional roadbed side objects Define the following variables: the time variable is... , , The duration of post-construction operation; Target The settlement function is Output by a BP neural network, with input variables including geological parameters. Lightweight soil construction parameters and settlement data of the filled section ; object The settlement function is Similarly, the output of a BP neural network is the same as the input variable. Consistent; Differential settlement prediction values ​​at the same monitoring section and the same time point The expression is: , In the formula, the absolute value sign represents the magnitude of the settlement difference; the model learns the mapping relationship between the above variables and settlement, focusing on strengthening the influence weights of the flow characteristics and hardening process of lightweight soil on settlement development, and outputs the results at any time during the entire operation period. corresponding This provides a quantitative basis for optimizing the construction plan for the transition section.

[0036] When determining the construction parameters for the transition section, a parameter vector is constructed using the step width, geogrid insertion length, blind drain spacing, and lightweight soil pouring boundary slope. The lightweight soil pouring boundary slope must be adapted to the flow and diffusion characteristics of the lightweight soil to avoid segregation or accumulation during pouring. Using the maximum allowable differential settlement as the control objective, a functional extremum model relating parameters and settlement is constructed. This model strengthens the coupling relationship between the lightweight soil pouring boundary slope and the differential settlement gradient, ensuring that parameter adjustments can specifically control settlement development during the lightweight soil flow and filling process. By establishing the intrinsic relationship between parameters and differential settlement through this model, it ensures that the predicted differential settlement value does not exceed the allowable threshold throughout the entire transition section, and that the lightweight soil flow distance does not exceed the transition section boundary without significant accumulation or segregation. This allows for the solution of a parameter set that satisfies both settlement control requirements and lightweight soil filling characteristics.

[0037] Subsequently, a construction feasibility mapping was introduced to consider the compatibility between parameters and on-site construction conditions. Specifically, this included the matching of step width with the operating radius of the lightweight soil pumping equipment, the compatibility of the lightweight soil pouring boundary slope with the flatness of the base, and the coordination of the geogrid extension length with the anchoring requirements of the lightweight soil before setting. From the parameter solution set, parameter combinations that simultaneously meet the settlement control target, lightweight soil filling characteristics, and on-site construction feasibility were selected as the final transition section construction parameters.

[0038] Specifically, in a preferred embodiment, the process for confirming the construction parameters of the transition section is as follows: Let the transition section parameter vector be... , The width of the step. The extension length of the geogrid. The spacing between blind drains, The boundary slope is to be constructed using lightweight soil; the predicted differential settlement value is... The control target is , To allow for the maximum differential settlement threshold, a functional extremum model of the parameter-settlement correlation is constructed: , In the formula, For the coupling operator of parameters and settlement gradient, For gradient operators, For variational operators, For the transition zone spatial domain, the intrinsic relationship between parameters and differential settlement is established through functional extremum conditions, while strengthening... Its regulatory effect on the flow settlement of lightweight soil; Constrained by control objectives, In the whole domain It must be established within the specified time and must meet the following requirements. The corresponding flow distance of the lightweight soil does not exceed the boundary of the transition section. Solving the variational equation yields the parameter solution set: Introducing construction feasibility mapping The relationship between the characterization parameters and the field conditions is described, and the final parameters are: , In the formula, This is the feasible area under the on-site conditions. This is the inverse operation of the mapping, which uses set operations to select parameter combinations from the solution set that simultaneously satisfy settlement control and construction feasibility.

[0039] After the transition section construction is completed, actual settlement data on both sides must be continuously collected at a preset monitoring frequency (e.g., 1 day, 3 days, 7 days after filling, and monthly). Environmental factors (such as rainfall and temperature changes) should be recorded simultaneously during data collection to avoid data deviation caused by environmental interference. These actual settlement data, including environmental variables, should be organized in a unified format, labeled with the corresponding transition section construction parameters (including the lightweight soil pouring boundary slope and mix proportion) and geological conditions, and added as new samples to the BP neural network training set. The network will automatically compare the new samples with historical samples, correcting the "parameter-settlement-environment" mapping relationship in the settlement prediction model, especially strengthening the weight of environmental factors on settlement during the lightweight soil hardening stage: if the actual settlement value deviates significantly from the predicted value, the weights of lightweight soil-related parameters under the corresponding geological conditions will be adjusted to optimize the differential settlement prediction algorithm; if the deviation is within the allowable range, the applicability of the current parameter combination to the lightweight soil filling characteristics will be strengthened. Through this cycle of "construction-monitoring-data supplementation-model optimization-guiding the next stage of construction", the entire filling construction process forms a dynamic closed loop. After each transition section is completed, the model's accuracy in controlling the settlement of lightweight soil subgrade under different geological conditions and environmental factors will be improved, ensuring that the construction parameters of subsequent connecting sections can more accurately match the on-site needs and continuously reduce the risk of differential settlement.

[0040] The method in Example 1 has broad application prospects in the field of roadbed construction connecting highways, railways, and bridges, especially suitable for scenarios such as the perimeter of urban elevated piers and approach roads of cross-river bridges, where there are stringent requirements for roadbed settlement and structural safety. Currently, such projects often face problems such as complex geological conditions, poor adaptability of traditional empirical parameters to lightweight soil, and difficulty in controlling differential settlement at connection sections. Example 1 uses a BP neural network to achieve dynamic optimization and precise control of relevant construction parameters for lightweight soil (such as pumping parameters and pouring boundary slope), which can ensure pier safety and roadbed compaction, reduce rework costs, and improve construction efficiency, effectively meeting the needs of modern transportation engineering for refined construction.

[0041] Meanwhile, with the advancement of intelligent transportation construction, the dynamic optimization closed-loop mode of this invention can be further integrated with engineering monitoring systems and intelligent pumping equipment to form an integrated intelligent construction system of "data acquisition - model calculation - parameter output - equipment execution," adapting to the construction process of lightweight soil without compaction and relying on fluidity for filling. In the future, in the reconstruction and expansion of existing roads and the subgrade construction of mountainous sections with complex geological conditions, its adaptability to the properties of lightweight soil under different geological conditions and its continuous optimization characteristics can provide technical support for more challenging projects, help reduce the later maintenance costs of transportation engineering, and promote the development of lightweight soil subgrade construction towards intelligence and precision.

[0042] Example 2 Please refer to Figure 2 This embodiment 2 provides a lightweight soil subgrade filling construction system adjacent to the pier, including: The benchmark parameter confirmation unit is used to collect geological data around the pier and determine the construction target. It incorporates the measured data of the small-scale trial filling in the early stage of the project into the training sample of the BP neural network and outputs the appropriate benchmark parameters, including the lightweight soil mix ratio, layer thickness, and compaction parameters. The construction data acquisition unit is used for construction based on model reference parameters. First, the preliminary treatment around the pier is completed. When filling lightweight soil in layers, the rolling frequency and speed are collected by the roller sensor, the pier displacement is monitored by the GNSS equipment, and the compaction degree is detected by the automatic drop hammer. The three types of data are transmitted to the BP neural network simultaneously. The parameter dynamic adjustment unit is used by the BP neural network to first normalize the rolling parameters, displacement, and compaction degree data, then pass them through the input layer to the hidden layer for computation, extract the correlation between the three and compare them with the construction target threshold; if the compaction degree does not meet the standard, the network corrects the neuron connection weights through backpropagation, recalculates, and outputs instructions to adjust the rolling parameters; if the pier displacement is close to the warning threshold, the same approach is used to generate a scheme to adjust the layer thickness and foam particle content; after each parameter adjustment, the relevant data is automatically added to the training set as new samples to iteratively optimize the model accuracy; The transition section parameter confirmation unit is used to determine the construction parameters of the transition section at the junction of lightweight soil and traditional subgrade based on the differential settlement prediction value output by the BP neural network. After construction, the settlement data of the transition section is incorporated into the model to optimize the parameters of the next section, forming a closed loop of dynamic optimization throughout the entire process.

[0043] Example 3 This embodiment 3 also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement any step of a construction method for filling a lightweight soil subgrade adjacent to a pier.

[0044] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0045] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0046] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements 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 method for constructing a lightweight soil subgrade adjacent to a pier, characterized in that, include: S1. Collect geological data around the pier and determine the construction target. Incorporate the measured data of the small-scale trial filling in the early stage of the project into the training sample of the BP neural network and output the appropriate benchmark parameters, including the lightweight soil mix ratio, layer thickness, and pumping parameters. S2. Based on the model reference parameters, the preliminary treatment around the pier is completed first; when filling lightweight soil in layers, the pumping equipment sensors collect data on pumping pressure, flow rate and elevation changes of the pouring surface, GNSS equipment monitors the displacement of the pier, and non-destructive testing methods including radar detection or elastic wave testing are used to obtain lightweight soil compaction data. The three types of data are transmitted to the BP neural network simultaneously. The S3.BP neural network first normalizes the pumping parameters, displacement, and density data, then passes them through the input layer to the hidden layer for computation, extracts the correlation between the three, and compares them with the construction target threshold. If the density does not meet the standard, the network corrects the neuron connection weights through backpropagation, recalculates, and outputs instructions to adjust the pumping parameters. If the pier displacement is close to the warning threshold, a scheme to adjust the layer thickness and foam particle content is generated in the same way. After each parameter adjustment, the relevant data will be automatically added to the training set as new samples to iteratively optimize the model accuracy; S4. At the junction of lightweight soil and traditional subgrade, the construction parameters of the transition section are determined based on the differential settlement prediction value output by the BP neural network. After construction, the settlement data of the transition section is incorporated into the model to optimize the parameters of the next section, forming a dynamic optimization closed loop throughout the entire process. The specific value of differential settlement prediction in S4 is as follows: Let the monitoring objects on both sides of the connection section be lightweight soil objects. and traditional roadbed side objects Define the following variables: the time variable is... , , The duration of post-construction operation; Target The settlement function is Output by a BP neural network, with input variables including geological parameters. Lightweight soil construction parameters and settlement data of the filled section ; object The settlement function is Similarly, the output of a BP neural network is the same as the input variable. Consistent; Differential settlement prediction values ​​at the same monitoring section and the same time point The expression is: ， In the formula, the absolute value sign represents the magnitude of the settlement difference; the model learns the mapping relationship between the above variables and settlement, focusing on strengthening the influence weights of the flow characteristics and hardening process of lightweight soil on settlement development, and outputs the results at any time during the entire operation period. corresponding This provides a quantitative basis for optimizing the construction plan of the transition section; The geological data in S1 includes: detailed geological survey results within a 3m radius around the pier, specifically covering the physical and mechanical parameters of each soil layer, groundwater level elevation and seasonal variation, soil layer distribution pattern and special adverse geological conditions; The calculation process of the reference parameters in S1 is as follows: Let the geological feature vector be The test filling measured matrix is ​​as follows: ,in, Indicates the first The first group of filling tests Item index values; construct gradient coupling functions to quantify the intrinsic relationship between geological parameters and trial filling effects: ， In the formula, For Hamiltonian operators; The parameter solution space encompasses geological parameters. With test filling indicators All possible values; This is a vector dot product operation, representing the intensity of the gradient influence of geological parameters on the trial filling effect; Construct a pre-defined performance surface based on the performance targets of lightweight soil. The mix proportions are determined by minimizing the gradient difference: ， In the formula, , For cement admixture, For the amount of lightweight aggregate, This refers to the water-to-binder ratio; It is an L2 norm; The gradient vector of the performance target surface, which includes flowability, setting time, and strength indices; Based on the Jacobian matrix determinant characterizing parameter sensitivity, the layer thickness is derived. : ， In the formula, for Jacobian matrix, ; This refers to matrix determinant operations. Given the inverse of the Jacobian matrix, the layer thickness suitable for the compaction equipment's performance is determined through parameter sensitivity inversion; based on the analysis of the viscous fluid motion equation, the correlation between pumping parameters and the flow characteristics of lightweight soil is established. ， In the formula, The density of lightweight soil, The velocity vector is the flow velocity vector. For time, For pumping pressure, The dynamic viscosity coefficient of lightweight soil is calculated from the flowability and spread measured during trial filling. Let gravitational acceleration be the acceleration due to gravity. By solving the steady-state solution of this equation, the corresponding optimal pumping pressure can be obtained. With traffic ; The specific process of the pre-treatment around the pier in S2 is as follows: clear surface debris within a range of no less than 2m around the pier down to the original soil surface; level the base to ensure that the slope is no more than 1% to meet the flow and diffusion requirements of lightweight soil; lay a graded crushed stone cushion layer with a thickness of no less than 30cm and compact it according to the benchmark parameters; treat the surface of the cushion layer with laser leveling technology to ensure that the elevation deviation is controlled within ±5mm, so as to provide a uniform bearing base for lightweight soil; and attach a cement corrosion resistant polyethylene film around the base of the pier to form a protective layer. The height of the film should be no less than 50cm above the top surface of the cushion layer, and the joints should be sealed with hot-melt welding to prevent grout from seeping into the gap between the pier and the cushion layer during the flow and filling of lightweight soil. The specific rules for extracting the correlation between the three elements in S3 are as follows: Let the normalized pumping parameter vector be , This is the normalized value of the pumping pressure. The normalized value for flow rate is [value], and the displacement data is [value]. The density data is The three constitute the input space vector. Feature mapping is achieved through nonlinear transformation of the hidden layer. ， In the formula, This is the connection matrix from the input layer to the hidden layer. This is the hidden layer bias vector. It is the Sigmoid activation function. This is the high-dimensional feature vector output by the hidden layer; Further, association patterns are constructed using feature space distance metrics: ， In the formula, For the sample data space, For neighboring sample vectors, The similarity function characterizes the physical relationship between samples; As a feature space correlation index, by minimizing Extract the intrinsic coupling law of "compaction parameters-displacement-compaction degree" in high-dimensional space, and complete the quantitative characterization of the dynamic correlation among the three.

2. The method for constructing a lightweight soil subgrade adjacent to a pier according to claim 1, characterized in that, The specific process of correcting neuron connection weights through backpropagation in S3 is as follows: Let the loss function between the output layer prediction and the target threshold be... For cases where the density does not meet the standard, , The target value for compaction, The predicted compaction value output by the network; During backpropagation, the partial derivatives of the loss function with respect to the output layer weights are calculated first: ， In the formula, For the hidden layer The nth neuron to the output layer The connection weights of each neuron; For the output layer The input of each neuron, , For the hidden layer The output of each neuron For the output layer Bias of each neuron; Then, the partial derivatives of the weights in the hidden layer and the input layer are passed to each other using the chain rule: ， In the formula, For the input layer The first neuron to the hidden layer The connection weights of each neuron; For the hidden layer The input of each neuron, , For the input layer The input of each neuron, For the hidden layer Bias of each neuron; Finally, update the weights using gradient descent: ， In the formula, The learning rate is determined by the convergence characteristics of the data during the trial filling stage; For the updated weights, To update the weights, repeat the above process until... , The weights are corrected to the preset precision threshold.

3. The method for constructing a lightweight soil subgrade adjacent to a pier according to claim 1, characterized in that, The specific process for confirming the construction parameters of the transition section in S4 is as follows: Let the transition section parameter vector be... , The width of the step. This refers to the length of the geogrid extension. The spacing between blind drains, The boundary slope is to be constructed using lightweight soil; the predicted differential settlement value is... The control target is , To allow for the maximum differential settlement threshold, a functional extremum model of the parameter-settlement correlation is constructed: ， In the formula, For the coupling operator of parameters and settlement gradient, For gradient operators, For variational operators, For the transition zone spatial domain, the intrinsic relationship between parameters and differential settlement is established through functional extremum conditions, while strengthening... Its regulatory effect on the flow and settlement of lightweight soil; Constrained by control objectives, In the whole domain It must be established within the specified time and must meet the following requirements. The corresponding flow distance of the lightweight soil does not exceed the boundary of the transition section. Solving the variational equation yields the parameter solution set: ， Introducing construction feasibility mapping The relationship between the characterization parameters and the field conditions is described, and the final parameters are: ， In the formula, This is the feasible area under the on-site conditions. This is the inverse operation of the mapping, which uses set operations to select parameter combinations from the solution set that simultaneously satisfy settlement control and construction feasibility.

4. A lightweight earthen roadbed filling construction system adjacent to piers, used to implement the lightweight earthen roadbed filling construction method adjacent to piers as described in claim 1, characterized in that, include: The benchmark parameter confirmation unit is used to collect geological data around the pier and determine the construction target. It incorporates the measured data of the small-scale trial filling in the early stage of the project into the training sample of the BP neural network and outputs the appropriate benchmark parameters, including the lightweight soil mix ratio, layer thickness, and pumping parameters. The construction data acquisition unit is used for construction based on model reference parameters. First, the pre-treatment around the pier is completed. When filling lightweight soil in layers, the pumping equipment sensors collect data on pumping pressure, flow rate and elevation changes of the pouring surface, GNSS equipment monitors the displacement of the pier, and non-destructive testing methods including radar detection or elastic wave testing are used to obtain lightweight soil compaction data. The three types of data are transmitted to the BP neural network simultaneously. The parameter dynamic adjustment unit is used by the BP neural network to first normalize the pumping parameters, displacement, and compaction data, and then pass them through the input layer to the hidden layer for calculation. The network extracts the correlation between the three and compares them with the construction target threshold. If the compaction does not meet the standard, the network corrects the neuron connection weights through backpropagation, recalculates, and outputs instructions to adjust the pumping parameters. If the pier displacement is close to the warning threshold, the network generates a scheme to adjust the layer thickness and foam particle content in the same way. After each parameter adjustment, the relevant data will be automatically added to the training set as new samples to iteratively optimize the model accuracy; The transition section parameter confirmation unit is used to determine the construction parameters of the transition section at the junction of lightweight soil and traditional subgrade based on the differential settlement prediction value output by the BP neural network. After construction, the settlement data of the transition section is incorporated into the model to optimize the parameters of the next section, forming a closed loop of dynamic optimization throughout the entire process.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor as described in any one of claims 1-3: a method for constructing a lightweight soil subgrade adjacent to a pier.

Images