A method for predicting lateral displacement of a pile foundation of a high-pile wharf

Abstract

The application discloses a kind of high-pile wharf pile foundation lateral displacement prediction methods, comprising the following steps: obtaining high-pile wharf pile foundation lateral displacement depth learning sample set;Establish high-pile wharf pile foundation lateral displacement depth learning prediction model;Using depth learning sample set to train prediction model to obtain the high-pile wharf pile foundation lateral displacement depth learning prediction model of optimized training;Real-time acquisition high-pile wharf pile measured data is input to the high-pile wharf pile foundation lateral displacement depth learning prediction model of optimized training and obtains model output prediction result.The application provides rich sample data for establishing high-pile wharf pile foundation lateral displacement depth learning prediction model by the mixing and revision of high-pile wharf pile foundation lateral displacement measured data set and finite element simulation data set, and guarantees the accuracy and reliability of lateral displacement prediction.

Description

Technical Field

[0001] This invention relates to the field of deep learning and simulation prediction, and in particular to a method for predicting the lateral displacement of pile foundations for high-pile wharves. Background Technology

[0002] High-pile wharves are a major type of coastal port wharf structure, widely used in wharf construction on soft soil foundations. They are prevalent in the economically vibrant coastal areas of my country, including the Bohai Bay, northern Jiangsu, the Yangtze River estuary, Hangzhou Bay, the Minjiang River estuary, the Pearl River estuary, and northwestern Hainan Island. Due to the complex and harsh marine environment of seaports, wharf structures bear significant service loads while also enduring loads from soil, waves, currents, sea ice, tides, typhoons, earthquakes, and ships. Under these harsh conditions, wharf structures are highly susceptible to damage, leading to a reduction in overall structural strength. Furthermore, the natural aging of concrete materials and human factors further exacerbate the safety and service life of the wharf structure. Meanwhile, during the operation of high-pile wharves, the deformation of soft soil foundations is significant, especially excessive lateral deformation, which leads to excessive horizontal loads on the wharf pile foundations. This excessive horizontal displacement can cause damage to the pile-beam joints, thus affecting the wharf's bearing capacity. As soil creep intensifies, the lateral displacement of the wharf further increases. Under service loads, the wharf structure will bear additional bending moments caused by lateral displacement, further exacerbating the damage to structural components and joints, and even causing the entire structure to overturn, seriously affecting the safe operation of the wharf. Existing technologies often use real-time measurement of pile foundation displacement without considering future displacement prediction. In fact, predictive data is more reflective of the future safety of the pile foundation. Real-time displacement measurement only reflects the safety risks corresponding to existing displacements, which is not very meaningful for risk prediction. For example, 201910703778.1 cannot solve the problem of displacement prediction.

[0003] Therefore, it is imperative to carry out research and construction on intelligent monitoring and early warning of high-pile wharf structures and break through the key technologies of intelligent monitoring and early warning of port infrastructure. There is currently no relevant literature available on this technology. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for predicting the lateral displacement of pile foundations of high-pile wharves. The method is based on a combination of deep learning and finite element simulation to predict the lateral displacement, providing quantitative prediction data for wharf management.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a method for predicting the lateral displacement of pile foundations for high-pile wharves, comprising the following steps:

[0006] Obtain a deep learning sample set of lateral displacements of high-pile wharf foundation piles;

[0007] Establish a deep learning prediction model for the lateral displacement of the pile foundation of a high-pile wharf;

[0008] The prediction model was trained using a deep learning sample set to obtain an optimized deep learning prediction model for the lateral displacement of the high-pile wharf foundation.

[0009] The real-time measured data of the high-pile wharf piles are input into the optimized and trained deep learning prediction model of the lateral displacement of the high-pile wharf pile foundation to obtain the model output prediction results.

[0010] The deep learning sample set for the lateral displacement of pile foundations of high-pile wharves includes the measured dataset of high-pile wharves and the finite element simulation dataset of high-pile wharves.

[0011] The establishment of the measured dataset for high-pile wharves includes:

[0012] The measured lateral displacements of the top and bottom of the piles in typical high-pile wharf projects over the years were collected and arranged in chronological order as output variables of the deep learning prediction model.

[0013] The structural parameters of the wharf, soil properties, wave height, tidal conditions, typhoon data affecting the wharf, and tonnage data of ships docked at the high-pile wharf are used as input variables for the deep learning prediction model to form a measured dataset of lateral displacement of the high-pile wharf.

[0014] The establishment of the finite element simulation dataset for high-pile wharves includes:

[0015] A typical high-pile wharf structure was selected for finite element refined simulation modeling. The relevant parameters of the finite element model were optimized and adjusted using measured datasets. Then, reasonable upper and lower limits and intervals were set for the input soil parameters, wave parameters, typhoon parameters, ship loads, etc. The finite element simulation output results of the lateral displacement of the high-pile wharf pile foundation were obtained by combining the simulated input parameters, forming a finite element simulation dataset of the lateral displacement of the high-pile wharf pile foundation.

[0016] The finite element simulation dataset of lateral displacement of pile foundation of high pile wharf is filtered and noisy data is removed. The remaining finite element simulation dataset is then merged with the measured dataset of high pile wharf to form a deep learning sample set of lateral displacement of pile foundation of high pile wharf.

[0017] A modified model for the finite element simulation dataset of lateral displacement of high-pile wharf foundation was established using a generative adversarial neural network (GAN). A generative adversarial neural network (CAN) was established and trained using the measured dataset of lateral displacement of high-pile wharf foundation. Then, the finite element simulation dataset of lateral displacement of high-pile wharf foundation was judged, and samples that did not conform to the rules of the measured dataset were removed.

[0018] Training the prediction model using a deep learning sample set includes:

[0019] The deep learning sample set for lateral displacement of high-pile wharf foundations was preprocessed by normalization and time series transformation. Cross-validation was used to optimize the deep learning prediction model: the sample set was divided into 5 parts, and 4 parts of the dataset were used for model training and optimization each time. The remaining 1 part of the dataset was used to test the model. By optimizing the hyperparameters such as the number of hidden layers and the number of neurons in the hidden layers of the deep learning model, the optimal deep learning neural network structure was obtained, and the deep learning prediction model for lateral displacement of high-pile wharf foundations was established.

[0020] The lateral displacement data of the top and bottom of the high-pile wharf are used as the output of the prediction model. The structural parameters of the high piles of the wharf, the physical and mechanical properties of the soil, the historical wave height and tide data of the wharf, the historical typhoon wind speed affecting the wharf, and the load data such as the tonnage of the ships at the wharf are used as the input to establish the prediction model. After the prediction model is trained and optimized, the real-time collected model input data is input into the model, and then the prediction model outputs the corresponding prediction results.

[0021] The prediction model is optimized using 5-fold cross-validation. The sample set is randomly divided into 5 parts. In each round, 4 parts are randomly selected as the training set and the remaining part is used as the test set. After this round, 4 parts are randomly selected again to train the data. After 4 rounds, the error between the measured value and the predicted value is used to evaluate the optimal model and parameters.

[0022] The advantages of this invention are as follows: By mixing and correcting the measured dataset of lateral displacement of high-pile wharf foundation piles and the finite element simulation dataset, rich sample data is provided for establishing a deep learning prediction model of lateral displacement of high-pile wharf foundation piles, ensuring the accuracy and reliability of lateral displacement prediction. Compared with traditional mechanical calculation analysis, the deep learning prediction model can consider more comprehensive influencing factors. For finite element numerical simulation, the deep learning prediction model saves a lot of repetitive modeling, mesh generation, and model parameter tuning processes, making it more efficient. The generative adversarial neural network model used to correct the sample set can make full use of the accuracy of the measured dataset of lateral displacement of high-pile wharf foundation piles and the richness of the finite element simulation dataset, exploring new ideas for intelligent early warning of high-pile wharves. The prediction model is optimized by using 5-fold cross-validation, ensuring optimal training of the model and improving the accuracy and reliability of the model prediction results. Attached Figure Description

[0023] The following is a brief explanation of the contents of each of the accompanying drawings and the markings in the drawings:

[0024] Figure 1 This is a flowchart illustrating the construction process of the deep learning sample set for the lateral displacement of the high-pile wharf foundation in this invention.

[0025] Figure 2This is a flowchart illustrating the construction and training process of the deep learning-based lateral displacement prediction model for high-pile wharf foundations according to the present invention. Detailed Implementation

[0026] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and the description of the preferred embodiments.

[0027] This solution provides a rapid and intelligent prediction method for the lateral displacement of high-pile wharf foundations. Utilizing historical deformation data and related parameter datasets of the high-pile wharf foundations, the measured deformation dataset is expanded using finite element numerical simulation. A generative adversarial neural network (GAN) is incorporated to improve the reliability of the finite element simulation results. The measured dataset and the corrected simulation dataset are combined to form an intelligent prediction sample set for the lateral displacement of high-pile wharf foundations. A deep learning neural network model is established, and cross-validation is used to optimize the prediction model, achieving intelligent and rapid prediction of the lateral displacement of high-pile wharf foundations. The use of a GAN combined with finite element simulation to expand the measured dataset for the lateral displacement of high-pile wharf foundations ensures data reliability while obtaining a sufficient prediction sample set. The deep learning prediction model solves the problems of long finite element modeling time and difficult model parameter tuning, providing data support for the monitoring and early warning of lateral displacement of high-pile wharf foundations.

[0028] To predict the lateral displacement of pile foundations in high-pile wharves, a deep learning prediction sample set for lateral displacement is constructed by combining finite element numerical simulation with generative adversarial neural networks. The deep learning prediction model is optimized by cross-validation, and a deep learning prediction model for the lateral displacement of pile foundations in high-pile wharves is established. This enables accurate and rapid prediction of the lateral displacement of pile foundations in high-pile wharves, thereby providing data support for the monitoring and early warning of the structural health of high-pile wharves.

[0029] A method for predicting the lateral displacement of high-pile wharf foundations by combining deep learning and finite element simulation includes:

[0030] (1) The goal is to establish a deep learning prediction model for the lateral displacement of the high-pile wharf foundation, and to realize the prediction of the wharf's lateral displacement using measured high-pile wharf foundation data and finite element simulation data.

[0031] (2) Object: Data of high-pile wharves, including but not limited to: measured lateral displacement data of the top and bottom of the high-pile wharves over the years, structural parameters of the high piles of the wharf, physical and mechanical properties of the soil, wave height and tide data of the wharf over the years, typhoon wind speed data affecting the wharf over the years, and load data such as the tonnage of the wharf vessels.

[0032] (1) Process:

[0033] a) Establishment of the measured dataset for high-pile wharves: Measured lateral displacements at the top and bottom of piles in typical high-pile wharf projects over the years were collected and arranged in chronological order, serving as output variables for the deep learning prediction model. Wharf structural parameters, such as pile length, structural type, and insertion ratio; soil property parameters, such as unit weight, deformation modulus, and lateral resistance coefficient; wave height and tidal conditions, arranged in chronological order; data on the number of typhoons affecting the wharf, typhoon wind speed, and duration, also arranged in chronological order; and data on the tonnage of ships berthed at the high-pile wharves, etc., served as input variables for the deep learning prediction model, forming the measured dataset of lateral displacements of high-pile wharves (Data_monitoring).

[0034] b) Establishment of finite element simulation dataset for high-pile wharf: Select a typical high-pile wharf structure for refined finite element simulation modeling. Optimize and adjust the relevant parameters of the finite element model using measured datasets. Then, set reasonable upper and lower limits and intervals for the input soil parameters, wave parameters, typhoon parameters, ship loads, etc. Obtain multiple sets of lateral displacement calculation results for the high-pile wharf pile foundation using different combinations of input parameters, forming a finite element simulation dataset (Data_simulation) for the lateral displacement of the high-pile wharf pile foundation.

[0035] c) Establishment of a deep learning sample set for the lateral displacement of high-pile wharf foundations: A modified model for the finite element simulation dataset of the lateral displacement of high-pile wharf foundations is established using a generative adversarial network (GAN). The GAN is then built and trained using the measured lateral displacement dataset of high-pile wharf foundations (Data_monitoring). The finite element simulation dataset of the lateral displacement of high-pile wharf foundations (Data_simulation) is then evaluated, removing samples that do not conform to the rules of the measured dataset. The remaining finite element calculation samples are merged with the measured dataset to form a deep learning sample set for the lateral displacement of high-pile wharf foundations (Data_set_DL). The measured values ​​of the lateral displacement curves of the pile foundations exhibit several typical patterns. If the simulated displacement curve values ​​do not significantly exceed the range of these deformation patterns, they are considered not to conform to the rules of the measured data.

[0036] d) Establishment of a deep learning prediction model for the lateral displacement of high-pile wharf foundations: The deep learning sample set for the lateral displacement of high-pile wharf foundations is preprocessed by normalization and time series transformation. The sample set is divided into 5 parts, and 4 parts of the dataset are used for model training and optimization each time. The remaining 1 part of the dataset is used to test the model. By optimizing the hyperparameters such as the number of hidden layers and the number of neurons in the hidden layers of the deep learning model, the optimal deep learning neural network structure is selected, and a deep learning prediction model for the lateral displacement of high-pile wharf foundations is established. Using newly obtained data such as sea waves, typhoons, and ship tonnage as input, the model predicts the development and changes of the lateral displacement of high-pile wharf foundations in the future and assesses the health status of the high-pile wharf structure.

[0037] The features of this solution include: This patent provides rich sample data for establishing a deep learning prediction model for the lateral displacement of high-pile wharf foundations by mixing and correcting the measured dataset of lateral displacement of the pile foundations and the finite element simulation dataset, ensuring the accuracy and reliability of the lateral displacement prediction. Compared with traditional mechanical calculation analysis, the deep learning prediction model can consider more comprehensive influencing factors. For finite element numerical simulation, the deep learning prediction model saves a lot of repetitive modeling, mesh generation and model parameter tuning processes, making it more efficient. The generative adversarial neural network model used to correct the sample set can make full use of the accuracy of the measured dataset of lateral displacement of the high-pile wharf foundations and the richness of the finite element simulation dataset, exploring new ideas for intelligent early warning of high-pile wharves.

[0038] One challenge of deep learning models lies in providing the datasets needed for training, such as... Figure 1 As shown, this application provides and expands the dataset in two ways to make the dataset sufficient to provide reliable basic data for training.

[0039] For high-pile wharves that require prediction of lateral displacement of the wharf pile foundation, the following data are collected: wharf structural parameters, physical and mechanical properties of the soil, historical wave height and tide data, historical typhoon wind speed data affecting the wharf, and load data such as wharf vessel tonnage. These data are then sorted by time to form a high-pile wharf lateral displacement measured dataset (Data_monitoring).

[0040] A refined numerical simulation model of a high-pile wharf was established using three-dimensional finite element simulation software. By reviewing literature and combining it with measured data of lateral displacement of high-pile wharves, the numerical simulation model was adjusted and optimized to ensure that the simulation results conformed to the requirements of the measured data.

[0041] Set upper and lower limits for soil mechanics parameters, historical wave height of the wharf, tidal data, typhoon wind speed data, and tonnage of ships at the wharf. Divide each data point at fixed intervals to form different combinations of finite element model input data. Calculate a set of lateral displacements at the top and bottom of the wharf piles based on each set of input data to form a finite element simulation dataset (Data_simulation) of the pile foundation lateral displacement.

[0042] A generative adversarial neural network (GAN) is established. The GAN model is trained using the measured dataset of lateral displacement of high-pile wharf (Data_monitoring). Then, the trained model is used to judge the finite element simulation dataset of lateral displacement of pile foundation (Data_simulation), removing samples that do not conform to the relevant rules of the measured dataset, thus forming a modified finite element simulation dataset of lateral displacement of pile foundation (Data_simulation_modified). The deep learning sample set of lateral displacement of high-pile wharf pile foundation (Data_set_DL) is constructed using the following formula.

[0043] Data_monitoring+Data_simulation_odified=Data_set_DL

[0044] In the deep learning sample set (Data_set_DL) of lateral displacement of pile foundation of high-pile wharf, the wharf structural parameters, soil mechanical parameters, historical wave height of the wharf, tide data, typhoon wind speed data, and tonnage of ships at the wharf are used as input data X of the deep learning prediction model, and the lateral displacement of the top and bottom of the piles of the high-pile wharf is used as output data Y of the deep learning prediction model.

[0045] like Figure 2 As shown, to ensure the prediction accuracy of the trained model, the data in the sample set is first preprocessed, including data normalization and time series transformation. Here, the input data is transformed to the [0, 1] interval by maximum and minimum value normalization. Specifically, for the input parameter x of the prediction model, where x is a type of parameter in the input data X, such as ship tonnage, x is transformed using the following formula, where x... max x represents the maximum value of this type of parameter. min This represents the minimum value of this type of parameter.

[0046]

[0047] The time series conversion process uses the input data X at times tn, ..., t-2, t-1 to predict the output data Y (lateral displacement of the high-pile wharf) at time t.

[0048] The prediction model is optimized using 5-fold cross-validation. The sample set is randomly divided into 5 parts. Each time, 4 parts are randomly selected as the training set, and the remaining part is used as the test set. After this round, 4 parts are randomly selected again for training. After 4 rounds, the error between the measured values ​​and the predicted values ​​is used to evaluate the optimal model and parameters. The optimization process of the prediction model involves setting different combinations of the number of neurons and the number of hidden layers, and using 5-fold cross-validation to observe the prediction performance of the model under different combinations. The model with the smallest prediction error is selected as the prediction model. The sample set is randomly divided into 5 parts, and each time 4 parts are randomly selected as the training set, and the remaining part is used as the test set. This process is repeated four times to determine if there are five model parameters. The model with the smallest error among the five trained models is then selected as the optimal model.

[0049] After obtaining the deep learning prediction model for the lateral displacement of the high-pile wharf, newly collected data such as waves, typhoons, and ship tonnage, along with existing data such as wharf structure and soil parameters, can be used as input to predict the changes in the lateral displacement of the high-pile wharf pile foundation over a period of time, providing data support for assessing the service performance of the wharf pile foundation.

[0050] Obviously, the specific implementation of this invention is not limited to the above-described methods. Any non-substantial improvements made using the inventive concept and technical solution of this invention are within the protection scope of this invention.

Claims

1. A method for predicting the lateral displacement of pile foundations for high-pile wharves, characterized in that: Includes the following steps: Obtain a deep learning sample set of lateral displacements of high-pile wharf foundation piles; Establish a deep learning prediction model for the lateral displacement of the pile foundation of a high-pile wharf; The prediction model was trained using a deep learning sample set to obtain an optimized deep learning prediction model for the lateral displacement of the high-pile wharf foundation. The real-time measured data of the high-pile wharf piles are input into the optimized and trained deep learning prediction model of the lateral displacement of the high-pile wharf pile foundation to obtain the model output prediction results. The deep learning sample set for the lateral displacement of pile foundations of high-pile wharves includes the measured dataset of high-pile wharves and the finite element simulation dataset of high-pile wharves. The establishment of the measured dataset for high-pile wharves includes: The measured lateral displacements of the top and bottom of the piles in typical high-pile wharf projects over the years were collected and arranged in chronological order as output variables of the deep learning prediction model. The structural parameters of the wharf, soil properties, wave height, tidal conditions, typhoon data affecting the wharf, and tonnage data of ships docked at the high-pile wharf are used as input variables for the deep learning prediction model to form a measured dataset of lateral displacement of the high-pile wharf. The establishment of the finite element simulation dataset for high-pile wharves includes: A typical high-pile wharf structure was selected for finite element refined simulation modeling. The relevant parameters of the finite element model were optimized and adjusted using the measured dataset. Then, reasonable upper and lower limits and intervals were set for the input soil parameters, wave parameters, typhoon parameters and ship loads. Multiple sets of finite element simulation output results of the lateral displacement of the high-pile wharf pile foundation were obtained by combining the simulated input parameters, forming a finite element simulation dataset of the lateral displacement of the high-pile wharf pile foundation. The finite element simulation dataset of lateral displacement of pile foundation of high pile wharf is filtered and noisy data is removed. The remaining finite element simulation dataset is then merged with the measured dataset of high pile wharf to form a deep learning sample set of lateral displacement of pile foundation of high pile wharf. A modified model for the finite element simulation dataset of lateral displacement of high-pile wharf pile foundation was established using a generative adversarial neural network (GAN). A generative adversarial neural network (CAN) was established and trained using the measured dataset of lateral displacement of high-pile wharf pile foundation. Then, the finite element simulation dataset of lateral displacement of high-pile wharf pile foundation was judged, and samples that did not conform to the rules of the measured dataset were removed. Training the prediction model using a deep learning sample set includes: The deep learning sample set of lateral displacement of high-pile wharf pile foundation was normalized and time series transformed. The deep learning prediction model was optimized by cross-validation: the sample set was divided into 5 parts, 4 parts of the dataset were taken for model training and optimization each time, and the remaining 1 part of the dataset was used to test the model. By optimizing the number of hidden layers and the number of neurons in the hidden layer of the deep learning model, the optimal deep learning neural network structure was obtained, and the deep learning prediction model based on the lateral displacement of high-pile wharf pile foundation was established. The lateral displacement data of the top and bottom of the high-pile wharf piles are used as the output of the prediction model. The structural parameters of the high piles of the wharf, the physical and mechanical properties of the soil, the wave height and tide data of the wharf over the years, the typhoon wind speed data affecting the wharf over the years, and the tonnage data of the wharf ships are used as the input to establish the prediction model. After the prediction model is trained and optimized, the real-time collected model input data is input into the model, and then the prediction model outputs the corresponding prediction results. The prediction model is optimized using 5-fold cross-validation. The sample set is randomly divided into 5 parts. In each round, 4 parts are randomly selected as the training set and the remaining part is used as the test set. After this round, 4 parts are randomly selected again to train the data. After 4 rounds, the error between the measured value and the predicted value is used to evaluate the optimal model and parameters.

Images