CN-122020245-A - Island wall structure stability prediction construction method based on machine learning

CN122020245ACN 122020245 ACN122020245 ACN 122020245ACN-122020245-A

Abstract

The invention discloses a machine learning-based island wall structure stability prediction construction method which comprises the steps of constructing a multi-source heterogeneous real-time monitoring network, comprehensively collecting multi-source heterogeneous data affecting island wall stability, establishing a multi-parameter fusion dynamic prediction model, predicting a time sequence of island wall key point displacement and a structure stability coefficient in a future period based on the dynamic prediction model through historical monitoring data, establishing a dynamic construction optimization decision mechanism, comparing output of the dynamic prediction model with a preset safety control threshold value, dynamically optimizing construction parameters to form a real-time dynamic decision, and feeding back the regulated construction behavior and an actual monitoring result generated by the regulated construction behavior as a new data sample to the dynamic prediction model for training and optimization. The invention realizes the accurate prediction of the future state of the island wall structure, and a new method for dynamically and intelligently optimizing the construction scheme based on the prediction result so as to improve the construction safety, efficiency and economy.

Inventors

Lu Xiaoyou
ZHANG XI
XIE JINBO
LI YEXUN
MA ZHENJIANG
XIE LIZHAO
WANG XIAOJIAN
WANG XIU
ZHU HUI

Assignees

中交第三航务工程局有限公司

Dates

Publication Date: 20260512
Application Date: 20260122

Claims (9)

1. The island wall structure stability prediction construction method based on machine learning is characterized by comprising the following steps of: step S1, constructing a multi-source heterogeneous real-time monitoring network, and comprehensively collecting multi-source heterogeneous data affecting the stability of island walls; Step S2, a multi-parameter fusion dynamic prediction model is established, and a time sequence of island wall key point displacement and a structural stability coefficient in a period of time in the future is predicted based on the dynamic prediction model through historical monitoring data; step S3, a dynamic construction optimization decision mechanism is established, the output of the dynamic prediction model is compared with a preset safety control threshold value, construction parameters are dynamically optimized, and a real-time dynamic decision is formed; and S4, feeding back the adjusted construction behavior and the actual monitoring result generated by the construction behavior as a new data sample to the dynamic prediction model for training and optimizing.
2. The machine learning-based island wall structure stability prediction construction method according to claim 1, wherein in the step S1, a multi-source heterogeneous real-time monitoring network is constructed, comprising: High-precision and automatic monitoring equipment is distributed along key positions of an inner slope, an outer slope, a slope top and a slope foot of the island wall, and vertical displacement, horizontal displacement and deep displacement of the island wall are obtained in real time by adopting a distributed fiber Bragg grating sensor array and a GPS/GNSS receiver; Embedding sensors at different depths and positions inside island walls to monitor key soil mechanical parameters in real time; Monitoring the excess pore water pressure and dissipation condition of the excess pore water pressure in the soil filling consolidation process in real time by adopting a vibrating wire type or fiber bragg grating type pore water pressure meter; Recording and inputting the source, grading and compaction process parameters of backfill soil in real time by integrating construction quality control data, and taking the parameters as indirect input parameters for representing the density of the backfill soil; A GPS positioning system is arranged on the construction machine, the number, the position and the operation track of the large-scale machine are obtained in real time, and the square quantity of backfill soil and the load of preloading of each layer are accurately recorded through an electronic platform scale and a transportation record; And arranging a wave and tide level meter near the wave facing surface of the island wall, monitoring wave height, wave period and tide level change in real time, and quantifying the dynamic impact of wave load on the island wall.
3. The machine learning-based island wall structure stability prediction construction method according to claim 2, wherein in the step S1, all sensors are equipped with a 5G wireless communication technology, collected data are transmitted to a cloud data center in real time, and original data with different displacement, soil property, load and environment are denoised, aligned and fused to form a unified and high-dimensional multivariate time series input feature vector, so as to form a structured time series database.
4. The machine learning-based island wall structure stability prediction construction method according to claim 3, wherein in the step S2, a deep learning model specially designed for processing sequence data is selected as a dynamic prediction model, and a main body framework is an encoder-decoder framework, and optimization is performed on the basis of the deep learning model: the encoder adopts a plurality of layers of long-term and short-term memory networks as core components for inputting Encoding the multi-source feature vectors of the historical time steps, and extracting and memorizing the long-term time dependency relationship; A attention layer is introduced between the encoder and the decoder, different weights are distributed to input data of different monitoring points at different moments, and the model can focus on a key period and a key region which are most relevant to future instability; the decoder is also based on a multi-layer long-short-term memory network, and gradually generates the future according to the context vector and the attention weight output by the encoder Predicted sequence of time steps.
5. The machine learning-based island wall structure stability prediction construction method according to claim 4, wherein in the step S2, the input of the dynamic prediction model comprises: the structural response characteristics comprise a historical and real-time key point horizontal displacement rate, a vertical sedimentation rate and a deep displacement increment; The internal state characteristics of the rock and soil comprise pore water pressure values and pore pressure coefficients at different depths, and consolidation degree estimated values calculated based on pore pressure dissipation; The construction activity characteristics comprise accumulated filling height, filling rate and heavy machinery density on the current working surface; The environment disturbance characteristics comprise effective wave height, tide level and wave climbing estimated value calculated by combining the wave height and the tide level; Time-aligning the above feature vectors to construct training samples, each sample in the past Characteristic tensors for the individual time steps are used as inputs: ; In the formula, As a result of the feature tensor of the input, Is the current time step; Feature vector for each time step Four types of features are fused: ; In the formula, As a feature of the structural response, Is the characteristic of the internal state of the rock and soil, In order to construct the activity feature of the present invention, Is an environmental disturbance characteristic; The dynamic prediction model adopts a multi-task learning paradigm, and simultaneously outputs two target variables which have definite physical meaning and are related to each other, namely future displacement prediction and dynamic stability coefficients; Wherein, the Future displacement prediction, direct output of future Absolute displacement of each key monitoring point in each time step: ; In the formula, As a model of the output of the device, To the future (future) Absolute displacement predictions for each time; Dynamic stability coefficient (dynamic stability coefficient) of continuous island wall between 0 and 1 is output, which is a data-driven and probabilistic comprehensive safety state index, and the system is subjected to future learning of massive historical data Comprehensive assessment of probability of maintaining steady state for each period: ; In the formula, To be at future time Is used for the stability factor of (a), The function is to map the interval normalization between (0, 1), And For the weight matrix and bias term corresponding to the output layer, At time for model decoder State of (2); and when the coefficient continuously decreases and approaches a preset alarm threshold value, the characteristic instability risk is obviously increased.
6. The machine learning-based island wall structure stability prediction construction method according to claim 5, wherein in the step S2, training and verification of a dynamic prediction model are performed: Dividing a training set, a verification set and a test set from a structured time sequence database of a cloud data center, and enhancing the weight of a destabilizing sample by adopting an oversampling technology in order to ensure the prediction capability of a dynamic prediction model on extreme working conditions; a composite loss function is designed, which consists of two part weights: ; In the formula, And Is the weight of the two parameters, For the mean square error of displacement prediction, for ensuring the accuracy of displacement prediction, Is a cross entropy loss function of the stability coefficient and is used for optimizing the classification capacity of the stability coefficient.
7. The machine learning-based island wall structure stability prediction construction method according to claim 6, wherein in the step S3, a dynamic construction optimization decision mechanism is established: setting a multilevel early warning threshold, comprising: the stability coefficient threshold sets a three-level threshold for the output dynamic stability coefficient: the safety area (green) has a stability coefficient of more than 0.8, which indicates that the system state is good and the construction can be performed according to the original plan; the warning area (yellow) is 0.6 < stability coefficient less than or equal to 0.8, which indicates that the system state has unfavorable trend and needs to pay attention to and prepare intervention measures; The stability coefficient is less than or equal to 0.6, which means that the risk of system instability is high, and intervention measures need to be immediately taken; displacement and velocity threshold: Setting parallel three-level thresholds for the predicted displacement and the predicted displacement rate at the same time, and supplementing and verifying stability coefficients; real-time early warning and decision support, including: when the predicted result reaches the threshold value, the system automatically triggers hierarchical early warning: Yellow early warning, namely visually highlighting a digital twin interface of a management platform and pushing early warning notices to related engineers; Red alarm, namely, sending a locking or limited operation command to a construction machine control system of a related area automatically in addition to sending the highest-level alarm information to project management personnel, so as to forcedly pause high-risk operation technically; When the early warning is triggered, various preset regulation and control schemes are automatically evaluated, wherein the schemes comprise, but are not limited to, stopping filling, slowing down filling rate, starting precipitation, improving the future stability coefficient and recommending the expected control effect of the optimal regulation and control strategy.
8. The machine learning-based island wall structure stability prediction construction method according to claim 7, wherein in the step S3, the dynamically optimized construction process and parameters include: According to the generated regulation strategy, dynamically and accurately regulating the construction plan; Filling loading control, namely dynamically adjusting the thickness, the intermittent time and the daily filling rate of layered filling according to the pore water pressure prediction and the stability coefficient; Triggering and strengthening reinforcement measures, namely starting a preset reinforcement plan in a weak area in advance according to a deformation trend, or dynamically adjusting the construction density and depth of the gravel pile/vibroflotation pile and the interval of the drainage plates; The preloading management, namely optimizing the application and removal time of the preloading load and based on the predicted post-construction settlement and consolidation degree; and (3) dispatching the construction machinery, namely intelligently planning a walking path and a working area of the large-scale machinery according to the real-time stability map.
9. The machine learning-based island wall structure stability prediction construction method according to claim 8, wherein in the step S4, two main dynamic prediction model update trigger modes are established, and model update is triggered based on a fixed time interval or after a significant working condition change is encountered: the timing triggering is that based on a fixed time period, a model updating flow is automatically started; The event triggering is automatically triggered when the important change of the working condition of the site is monitored, and the specific conditions comprise that the site enters a brand new construction stage; the dynamic prediction model iteration comprises the following steps: the quality verification of the newly added data comprises the steps of automatically cleaning and marking the collected real-time data to form a newly added high-quality data set; And (3) incremental training and weight fine tuning, namely mixing the newly added data set with part of the historical core data set by adopting an incremental learning algorithm, and performing supervised fine tuning by taking the current model weight as an initial value.

Description

Island wall structure stability prediction construction method based on machine learning Technical Field The invention belongs to the technical field of geotechnical engineering and artificial intelligence intersection, and particularly relates to a machine learning-based island wall structure stability prediction construction method. Background Sea-filling and land-building are important means for expanding urban space, building ports, airports and other important infrastructures. In the sea-filling engineering, island walls are used as enclosure structures, and the stability of the island walls is directly related to the safety, construction period and cost of the whole engineering. At present, island wall structures face extremely complex engineering and environmental challenges: (1) The geological conditions are complex and changeable, the bearing capacity of the deep and soft seabed silt foundation is low, uneven settlement and lateral displacement are easy to generate under the load action, and the island wall is threatened to be stable. (2) The construction load has obvious influence, and the construction activities such as large-scale backfilling operation, preloading, foundation reinforcement (such as gravel piles) and the like can continuously change the stress state of the foundation and the pore water pressure field, so that the construction load is a main factor for inducing the island wall instability. (3) The marine dynamic environment is bad, and periodic or random hydrodynamic loads such as waves, tides and ocean currents continuously act on island walls, so that not only is thrust directly generated, but also dynamic response of pore water pressure of a foundation soil body is caused, and the soil body strength is further weakened. Traditional island wall stability analysis methods rely primarily on limit balance methods or finite element numerical simulations. Although the method plays an important role in the design stage, the method has obvious limitations in the construction stage, namely firstly, a calculation model is often based on simplified geological parameters and load conditions, construction dynamics and environmental changes are difficult to reflect in real time, secondly, the calculation process is long in time, the requirement of a construction site on real-time decision making cannot be met, and finally, the on-site monitoring data is massive, but is mostly only used for post verification or passive early warning, and the prediction value of the on-site monitoring data cannot be fully mined to guide active and feedforward type construction control. Therefore, the island wall structure stability prediction construction method based on machine learning is provided. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a machine learning-based island wall structure stability prediction construction method, which is a novel method for realizing accurate prediction of the future state of an island wall structure and carrying out dynamic and intelligent optimization on a construction scheme based on a prediction result so as to improve construction safety, efficiency and economy. The technical scheme for achieving the purpose is as follows: A construction method for predicting the stability of an island wall structure based on machine learning comprises the following steps: step S1, constructing a multi-source heterogeneous real-time monitoring network, and comprehensively collecting multi-source heterogeneous data affecting the stability of island walls; Step S2, a multi-parameter fusion dynamic prediction model is established, and a time sequence of island wall key point displacement and a structural stability coefficient in a period of time in the future is predicted based on the dynamic prediction model through historical monitoring data; step S3, a dynamic construction optimization decision mechanism is established, the output of the dynamic prediction model is compared with a preset safety control threshold value, construction parameters are dynamically optimized, and a real-time dynamic decision is formed; and S4, feeding back the adjusted construction behavior and the actual monitoring result generated by the construction behavior as a new data sample to the dynamic prediction model for training and optimizing. Preferably, in the step S1, constructing a multi-source heterogeneous real-time monitoring network includes: High-precision and automatic monitoring equipment is distributed along key positions of an inner slope, an outer slope, a slope top and a slope foot of the island wall, and vertical displacement, horizontal displacement and deep displacement of the island wall are obtained in real time by adopting a distributed fiber Bragg grating sensor array and a GPS/GNSS receiver; Embedding sensors at different depths and positions inside island walls to monitor key soil mechanical parameters in real time; Monitoring the excess