CN-115758845-B - Multi-scale soft soil stratum single pile P-Y curve correction method based on cyclic neural network

CN115758845BCN 115758845 BCN115758845 BCN 115758845BCN-115758845-B

Abstract

The invention relates to a multi-scale soft soil stratum single pile P-Y curve correction method based on a cyclic neural network, which can accurately predict deformation of a single pile head after horizontal load is applied, and particularly relates to the field of pile body horizontal load and deformation calculation. According to the invention, the single pile microcosmic P-Y curve is obtained through numerical simulation, the single pile mechanical P-Y curve is obtained through an empirical formula, and the multi-scale P-Y curve sequence data of the combined pile and stratum environment variables are constructed. And then, constructing a P-YLSTM model based on the LSTM cyclic neural network, wherein the model comprehensively considers the multi-scale mechanical deduction result and the multi-mode empirical data influence, greatly improves the accuracy of P-Y curve prediction, and can accurately calculate the accurate values of the load and pile body deformation of the single pile head in the soft soil stratum after the single pile head is subjected to horizontal load.

Inventors

SHI MINGXIA
ZHANG JUN
ZHOU QI
ZHANG DINGWEN
YANG TAO
WANG YU
XU CHAOYANG
XU JINGMIN

Assignees

宿迁市高速铁路建设发展有限公司
中铁四局集团有限公司
东南大学

Dates

Publication Date: 20260508
Application Date: 20221207

Claims (4)

1. A multi-scale soft soil stratum single pile P-Y curve correction method based on a cyclic neural network is characterized by comprising the following steps: S1, establishing a model and assigning parameters based on the working conditions of a single pile and surrounding soft soil stratum by adopting finite difference numerical simulation software, and acquiring a microscopic P-Y curve of the single pile by applying a horizontal load on a pile head and recording a horizontal displacement value of the pile head under the horizontal load; S2, correcting a part of pile side deformation exceeding 8Y 50 based on a single pile P-Y curve empirical formula in port engineering pile foundation standard to obtain a single pile mechanical P-Y curve; s3, carrying out multi-scale data fusion based on the microcosmic P-Y curve of the single pile determined in the step S1 and the P-Y curve of the single pile mechanics determined in the step S2, wherein the method specifically comprises the following steps: S31, taking the same displacement starting point as a sampling starting point, continuously sampling a single pile microscopic P-Y curve and a single pile mechanical P-Y curve for n times to obtain a 2-dimensional single pile multi-scale sequence (n, 2), and sampling a corresponding point of the actually measured P-Y curve to obtain a one-dimensional single pile actually measured P-Y sequence (n); s32, carrying out normalization treatment on soil intensity, soil moisture content and pile section rigidity of the single pile; S33, copying and expanding the data sequence with the normalized soil body strength, the normalized soil body water content and the normalized pile body section rigidity expressed as (1, 3) in a first dimension for n times to obtain a single pile environment variable sequence with the dimensions of (n, 3); S34, splicing the single-pile multi-scale sequence in the form of (n, 2) and the single-pile environment variable sequence in the form of (n, 3) in a second dimension to obtain a single-pile fusion sequence in the form of (n, 5); s35, the sampling starting point moves backwards along the displacement axis, the steps S31-S34 are repeated, m single pile fusion sequences are obtained, the description is that (m, n, 5) respectively represents (the number of single pile fusion sequences, the number of single pile fusion sequence data points and the number of data point characteristics), m single pile actual measurement P-Y sequences are obtained, and the description is that (m, n) respectively represents (the number of single pile actual measurement P-Y sequences and the number of actual measurement sequence data points); S4, carrying out cyclic neural network training by taking the single pile fusion sequence as shown in (m, n, 5) as input and taking the single pile actually measured P-Y sequence as shown in (m, n) as output, and constructing a P-YLSTM model based on the cyclic neural network.
2. The method for correcting a P-Y curve of a single pile according to claim 1, wherein in step S2, the corrected P-Y curve of the single pile has an empirical formula of Wherein p is the horizontal load of the pile head, N, p u is the limit soil resistance of the pile periphery, N, y 50 is the horizontal displacement of the pile side when the limit soil resistance of the pile periphery reaches half of the limit soil resistance, and m.
3. The method for correcting the single pile P-Y curve according to claim 1, wherein in step S4, the P-YLSTM model comprises a P-Y data encoding structure and a P-Y data decoding structure, the P-Y data encoding structure is composed of continuous 5 LSTM layers with n units, the P-Y data encoding structure and the P-Y data decoding structure are connected by a dimension reduction layer, the P-Y data decoding structure is composed of an MLP module composed of continuous 3 full connection layers, and a ReLU activation function layer is connected after the P-Y data decoding structure and mapped through line output.
4. The method for correcting the single pile P-Y curve according to claim 3, wherein the P-YLSTM model is built by using a Tensorflow2.0 frame, the training round is 80, the training batch size is 4, the training process uses MSE as a loss function, and the optimization algorithm selects Adam.

Description

Multi-scale soft soil stratum single pile P-Y curve correction method based on cyclic neural network Technical Field The invention relates to a multi-scale soft soil stratum single pile P-Y curve correction method based on a cyclic neural network, which can accurately predict deformation of a single pile head after horizontal load is applied, and particularly relates to the field of pile body horizontal load and deformation calculation. Background The existing standard adopted single pile P-Y curve has errors with field actual measurement data in the actual application process. The method for predicting the P-Y curve at the present stage mainly comprises the following steps that (1) after a single pile head in a soft soil stratum is subjected to horizontal load, deformation of a pile body is related to surrounding soil layer conditions, elastic modulus of the pile body and load, and the deformation characteristics of the pile body after the pile head is subjected to the horizontal load are difficult to accurately obtain according to a mechanical empirical method of the single pile horizontal load and displacement given in Port engineering pile foundation Specification). Especially when the horizontal displacement of the single pile exceeds 8y 50, the difference between the calculation result of the mechanical empirical method and the on-site actual measurement value is larger and larger. (2) The P-Y curve of the microscopic pile body dimension obtained by adopting the finite difference method has larger error with the deformation condition of the pile body after horizontal load in actual engineering. (3) The correction technology for the P-Y curve stays in single-scale correction, or corrects microscopic data by introducing a plurality of parameters, or corrects a macroscopic mechanical calculation formula, and the condition of combining the macroscopic and the microscopic is not considered. (4) Correction techniques for P-Y curves fail to introduce practical empirical data, lacking targeted correction for specific environments. Therefore, accurately predicting the true deformation curve of the pile body under horizontal load is an urgent need in the art. Disclosure of Invention In order to solve the key problem of accurately predicting the true deformation curve of the pile body under the horizontal load, the invention corrects the existing P-Y curve based on a circulating neural network means, and builds a P-Y curve correction model which accurately reflects the relationship between the horizontal load and the deformation of the pile top. Specifically, the invention adopts a multi-scale soft soil stratum single pile P-Y curve correction method based on a cyclic neural network, not only considers microscopic scale but also considers mechanical characteristics, and compared with a single scale P-Y curve calculation method, the method greatly improves the accuracy and rapidness of calculation, and specifically comprises the following steps: S1, establishing a model and assigning parameters based on the working conditions of a single pile and surrounding soft soil stratum by adopting finite difference numerical simulation software, and acquiring a microscopic P-Y curve of the single pile by applying a horizontal load on a pile head and recording a horizontal displacement value of the pile head under the horizontal load; S2, based on a single pile P-Y curve empirical formula in Port engineering pile foundation Specification, correcting the single pile P-Y curve empirical formula to obtain a single pile mechanical P-Y curve; wherein the single pile P-Y curve correction empirical formula is as follows Wherein p is the horizontal load of the pile head, N, p u is the limit soil resistance of the pile periphery, N, y 50 is the horizontal displacement of the pile side when the limit soil resistance of the pile periphery reaches half of the limit soil resistance, and m. Further, only the pile side deformation exceeding 8y 50 is partially corrected. S3, carrying out multi-scale data fusion based on the microcosmic P-Y curve of the single pile determined in the step S1 and the P-Y curve of the single pile mechanics determined in the step S2, wherein the method specifically comprises the following steps: S31, taking the same displacement starting point as a sampling starting point, continuously sampling a single pile microscopic P-Y curve and a single pile mechanical P-Y curve for n times to obtain a 2-dimensional single pile multi-scale sequence (n, 2), and sampling a corresponding point of the actually measured P-Y curve to obtain a one-dimensional single pile actually measured P-Y sequence (n); s32, carrying out normalization treatment on soil intensity, soil moisture content and pile section rigidity of the single pile; S33, copying and expanding the data sequence with the normalized soil body strength, the normalized soil body water content and the normalized pile body section rigidity expressed as (1, 3) in a firs