A method for predicting the strength of in-situ mixed soil by coupling multiple environmental factors based on a deep learning hybrid model

Abstract

This invention discloses a multi-environmental factor coupled prediction method for the strength of on-site mixed soil based on a deep learning hybrid model, belonging to the field of engineering construction quality control technology. The method includes: S1, on-site acquisition of multi-source parameters; S2, data preprocessing and construction of time-varying functions, as well as multi-dimensional feature extraction and reconstruction; S3, construction and training of the Attention-CNN-LSTM coupled prediction model; S4, dynamic prediction of on-site mixed soil strength; and S5, prediction result verification and model optimization. By systematically collecting core environmental factors such as temperature, humidity, and rainfall, along with construction parameters, a multi-environmental factor time-varying function is constructed. Combined with the Attention-CNN-LSTM coupled model, multi-domain feature extraction, temporal pattern learning, and adaptive weight allocation of key environmental factors are achieved, ultimately realizing accurate and dynamic prediction of on-site mixed soil strength. This aims to improve the accuracy, real-time performance, and generalization ability of the prediction results, providing technical support for construction quality control and long-term performance evaluation of on-site mixed soil engineering.

Description

Technical Field

[0001] This invention relates to the field of engineering construction quality control technology, specifically to a method for predicting the strength of on-site mixed soil by coupling multiple environmental factors based on a deep learning hybrid model. Background Technology

[0002] The strength of in-situ mixed soil is a core indicator for construction quality control, structural design, and long-term performance evaluation. It is influenced by key environmental factors such as temperature, humidity, and rainfall, as well as construction factors such as ash content, curing time, and mixing process. Currently, the methods for predicting the strength of in-situ mixed soil are mainly divided into traditional indoor test prediction methods and simplified in-situ fitting prediction methods.

[0003] Traditional indoor testing methods for predicting soil strength involve collecting soil samples from the field and curing them in a static environment with simulated temperature and humidity in a laboratory. This ignores the dynamic impact of rainfall on soil moisture content. After measuring soil strength indicators at different ages, simple algorithms such as linear regression and grey models are used to fit the strength variation patterns, which are then used as the basis for predicting soil strength in the field. However, traditional indoor testing methods have the following core drawbacks: First, the laboratory only simulates static, constant temperature and humidity, failing to reproduce the actual dynamic changes in temperature, humidity, and rainfall over time in the field, especially neglecting the real-time impact of rainfall on soil moisture content, resulting in significant deviations between predicted and actual field conditions. Second, it fails to systematically integrate multiple environmental factors such as temperature, humidity, and rainfall, considering only the influence of a single or few factors and ignoring the coupling effect between environmental and construction factors, leading to incomplete characterization. Third, the prediction model algorithms are simple and cannot uncover the deep nonlinear correlation between multiple environmental factors such as temperature, humidity, and rainfall and soil strength, resulting in low prediction accuracy.

[0004] The simplified on-site fitting prediction method collects limited temperature and humidity data by setting up a small number of monitoring points at the construction site, without systematically monitoring key environmental factors such as rainfall. It then combines soil strength test results to construct a fitting formula for a single or a few factors, achieving a rough prediction of the strength of the mixed soil. Compared to traditional indoor test prediction methods, this method is more closely aligned with on-site conditions, but it still does not fully consider the coupling effect of multiple environmental factors such as temperature, humidity, and rainfall with construction factors. The simplified on-site fitting prediction method has the following core shortcomings: First, it only collects temperature and humidity data sporadically, without quantitative monitoring and modeling of rainfall, and without constructing a function model of multiple environmental factors changing over time, thus failing to capture the real-time impact of dynamic environmental changes on soil strength. Second, it does not extract features and assign weights to multiple environmental factors such as temperature, humidity, and rainfall, making it unable to identify the critical influence of each environmental factor on soil strength, resulting in insufficient specificity and accuracy in the prediction. Third, the model has weak generalization ability, making it difficult to adapt to the needs of predicting the strength of mixed soil under different climatic regions and construction conditions with varying combinations of temperature, humidity, and rainfall. Summary of the Invention

[0005] This invention provides a method for predicting the strength of field-mixed soil by coupling multiple environmental factors based on a deep learning hybrid model. It aims to improve the accuracy, real-time performance, and generalization ability of the prediction results, and provide technical support for the construction quality control and long-term performance evaluation of field-mixed soil engineering.

[0006] This invention provides the following technical solution: a method for predicting the strength of in-situ mixed soil by coupling multiple environmental factors based on a deep learning hybrid model, comprising:

[0007] S1. Set up environmental monitoring equipment and soil performance testing points in the on-site soil mixing construction area, and build a multi-source parameter acquisition system to monitor the three major environmental factors of temperature, humidity and rainfall, and simultaneously collect dynamic environmental parameters, core construction parameters and soil strength indicators.

[0008] S2. Perform data cleaning and standardization on the collected multi-source parameters, construct the model input feature set, extract the feature set from two dimensions: spatial features and temporal features, and construct a two-dimensional feature matrix adapted to the Attention-CNN-LSTM model.

[0009] S3. Construct an Attention-CNN-LSTM coupled prediction model. Divide the two-dimensional feature matrix obtained in S3 into a training set and a test set, and input them into the model for training.

[0010] S4. Input the real-time multi-dimensional environmental influencing factor data of the mixed soil to be predicted into the trained model, and the model outputs the strength prediction value of the mixed soil at the target age.

[0011] S5 uses real-time environmental data and measured strength values ​​from on-site monitoring to incrementally update and dynamically correct the model, achieving rolling optimization prediction of the strength of the mixed soil.

[0012] The present invention has the following beneficial effects:

[0013] 1. By adopting a strategy of systematically collecting core environmental factors such as temperature, humidity, and rainfall and constructing coupled time-varying functions, the actual working conditions of multiple environmental factors changing dynamically over time are accurately reproduced. In particular, the real-time impact of rainfall on soil moisture is quantified. At the same time, the coupling effect of multiple environmental factors with construction factors such as ash content and age is fully considered. This breaks through the limitations of traditional prediction methods in terms of environmental simulation distortion and incomplete feature representation, and can better reflect the influence law of soil strength in accordance with the actual site conditions.

[0014] 2. An Attention-CNN-LSTM coupled prediction model was constructed. Feature extraction and temporal learning were performed for scenarios involving multiple environmental factors such as temperature, humidity, and rainfall. This effectively uncovered the deep nonlinear correlation between various environmental factors, environmental and construction factors, and soil strength, significantly improving the prediction accuracy of on-site mixed soil strength.

[0015] 3. Based on real-time monitoring data of temperature, humidity, and rainfall, dynamic prediction of the strength of on-site mixed soil is realized. It can quickly output the predicted value of soil strength at different ages, providing data support for real-time control of engineering construction quality and solving the problem of lag in traditional prediction methods.

[0016] 4. The model is continuously corrected and self-optimized by using measured data and corresponding dynamic data on temperature, humidity and rainfall, which improves the model's generalization ability under different climate regions, different combinations of environmental factors and different geological conditions, and can be adapted to the strength prediction needs of various on-site mixed soil engineering projects.

[0017] 5. The generated soil strength change trend curve and the weighted report of key environmental factors such as temperature, humidity and rainfall can provide targeted suggestions for optimizing on-site construction technology and adjusting maintenance measures, reduce engineering quality risks, and improve the long-term stability of on-site mixed soil engineering. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this specification will be explained and described below with reference to the accompanying drawings. However, the following embodiments are only preferred embodiments of this specification and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments in the implementation methods without creative effort are all within the protection scope of this specification.

[0020] Please see Figure 1 As shown, a method for predicting the strength of in-situ mixed soil based on a deep learning hybrid model, involving the coupling of multiple environmental factors, is characterized by comprising:

[0021] S1. Set up environmental monitoring equipment and soil performance testing points in the on-site soil mixing construction area to build a multi-source parameter acquisition system. The system will focus on monitoring three major environmental factors: temperature, humidity, and rainfall, and simultaneously collect dynamic environmental parameters, core construction parameters, and soil strength indicators.

[0022] The dynamic environmental parameters mentioned above are: real-time monitoring of air temperature, soil surface / internal humidity, and cumulative rainfall / hourly rainfall, with a sampling frequency of once per hour to accurately capture the dynamic changes of multiple environmental factors.

[0023] The core construction parameters are: the amount of soil and ash to be mixed, the mixing speed, the mixing time, the number of compaction passes, and the gradation of the fill material.

[0024] The soil strength indicators are as follows: the compressive strength, shear strength, and unconfined compressive strength of the mixed soil are measured by in-situ testing at different ages, and these indicators are used as label data for model training.

[0025] S2. Perform data cleaning and standardization on the collected multi-source parameters, construct the model input feature set, extract the feature set from two dimensions: spatial features and temporal features, and construct a two-dimensional feature matrix adapted to the Attention-CNN-LSTM model.

[0026] A multi-source data acquisition system was constructed, focusing on temperature, humidity, and rainfall, to monitor and quantify the dynamic changes of these three environmental factors, ensuring the comprehensiveness and relevance of the predicted data. Based on real-time monitoring data of temperature, humidity, and rainfall, dynamic prediction of the strength of the mixed soil was achieved. This system can quickly output predicted soil strength values ​​at different ages, providing data support for real-time control of construction quality and solving the problem of lag in traditional prediction methods.

[0027] S2 specifically includes data preprocessing and time-varying function construction, multi-dimensional feature extraction and reconstruction.

[0028] The data preprocessing and time-varying function construction specifically include: preprocessing the collected multi-source parameters to eliminate noise interference and data anomalies; firstly, the 3σ criterion is used to remove outliers in temperature, humidity, rainfall, and construction parameters, and then linear interpolation is used to complete the missing environmental monitoring data; for rainfall, hourly rainfall is converted into cumulative rainfall and correlated with soil moisture to quantify the actual impact of rainfall on soil moisture content. Based on the preprocessed dynamic data of temperature, humidity, and rainfall, continuous functions of each environmental factor changing with time are constructed through time series fitting, and then fused into a multi-environmental factor coupled time-varying function to accurately quantify the dynamic changes and mutual coupling relationships of temperature, humidity, and rainfall; at the same time, all parameters are normalized (using min-max normalization to the [0,1] interval) to eliminate dimensional differences.

[0029] The aforementioned construction of the time-varying function is the first to systematically integrate temperature, humidity, and rainfall to construct a coupled time-varying function, reproducing the actual working conditions of dynamic changes in multiple environmental factors on site. This breaks through the limitations of traditional static environmental simulation and single-factor modeling, achieving accurate quantitative characterization of multiple environmental factors. By employing a strategy of systematically collecting core environmental factors such as temperature, humidity, and rainfall and constructing coupled time-varying functions, the actual working conditions of dynamic changes in multiple environmental factors over time are accurately reproduced. In particular, the real-time impact of rainfall on soil moisture is quantified. Simultaneously, the coupling effect of multiple environmental factors with construction factors such as ash content and age is fully considered, overcoming the limitations of traditional prediction methods in terms of distorted environmental simulation and incomplete feature representation. This approach more closely reflects the influence of soil strength on actual site conditions.

[0030] Specifically, the multi-dimensional feature extraction and reconstruction includes: using normalized ash content, mixing process parameters, and other construction parameters, as well as the feature parameters of time-varying functions of multiple environmental factors such as temperature, humidity, and rainfall, and the number of days in age as the model input feature set, the feature set is extracted and reconstructed from two dimensions: spatial features and temporal features. Spatial features are the coupling and correlation features between various environmental factors such as temperature, humidity, and rainfall, and between environmental factors and construction factors. Temporal features are the dynamic features of temperature, humidity, rainfall, age, etc., which change over time, forming a two-dimensional feature matrix adapted to the Attention-CNN-LSTM model.

[0031] The reconstruction of multi-dimensional features extracts features from both spatial and temporal dimensions, fully exploring the coupling effects between multiple environmental factors such as temperature, humidity, and rainfall, as well as between environmental and construction factors, and the impact of dynamic changes on soil strength.

[0032] S3. Construct an Attention-CNN-LSTM coupled prediction model. Divide the two-dimensional feature matrix obtained in S3 into a training set and a test set, and input them into the model for training.

[0033] The construction steps of the Attention-CNN-LSTM coupled prediction model are as follows:

[0034] S301. Spatial features in the feature matrix are extracted using a CNN network to explore the coupling relationship between temperature, humidity, and rainfall, as well as the interaction between environmental and construction factors.

[0035] S302. Input the output of the CNN network into the LSTM network to learn the dynamic change law between temporal features and soil strength, and accurately capture the real-time impact of environmental dynamic changes on soil strength development.

[0036] S303. An attention mechanism is introduced to adaptively allocate weights to the features extracted by CNN-LSTM, automatically identify environmental and construction factors that play a key role in soil strength, and increase the weight ratio of key features.

[0037] S304. The mean squared error and coefficient of determination are used as performance evaluation indicators for the model. The hyperparameters of the model are optimized by grid search until the model converges and reaches the optimal performance.

[0038] The Attention-CNN-LSTM coupled model integrates the advantages of CNN spatial feature extraction, LSTM temporal pattern learning, and key factor weight allocation through the attention mechanism. It is customized for scenarios involving multiple coupled environmental factors such as temperature, humidity, and rainfall, achieving accurate prediction of soil strength. By constructing the Attention-CNN-LSTM coupled prediction model, feature extraction and temporal learning are performed for scenarios involving multiple coupled environmental factors such as temperature, humidity, and rainfall. This effectively uncovers the deep nonlinear correlation between various environmental factors, environmental and construction factors, and soil strength, significantly improving the prediction accuracy of on-site mixed soil strength.

[0039] S4. Input the real-time multi-dimensional environmental influencing factor data of the mixed soil to be predicted into the trained model, and the model outputs the strength prediction value of the mixed soil at the target age.

[0040] The workflow of the Attention-CNN-LSTM coupled prediction model is as follows:

[0041] The environmental parameters of temperature, humidity, and rainfall, which are collected and preprocessed in real time during construction, as well as construction parameters, are input into the trained Attention-CNN-LSTM coupled prediction model. The model automatically completes feature extraction, temporal learning, and key factor weight allocation, and outputs the strength prediction value of the on-site mixed soil at the corresponding age, realizing real-time dynamic prediction of soil strength; at the same time, it generates a soil strength change trend curve and a report on the weights of key environmental factors such as temperature, humidity, and rainfall.

[0042] S5. Based on real-time environmental data and measured strength values, the model is incrementally updated and dynamically corrected to achieve rolling optimization prediction of the strength of the mixed soil.

[0043] Among them, the rolling optimization prediction method for the strength of mixed soil includes:

[0044] The actual soil strength index is measured regularly using the in-situ testing method. The measured values ​​are compared with the model prediction values, and the deviation rate is calculated. If the deviation rate exceeds the preset threshold (e.g., 5%), the measured data and the corresponding dynamic monitoring data of temperature, humidity and rainfall are added to the model training set to fine-tune and correct the model, continuously improve the prediction accuracy of the model in the construction area, and realize the self-optimization of the prediction model.

[0045] By continuously correcting and self-optimizing the model using measured data and corresponding dynamic data on temperature, humidity, and rainfall, the model's generalization ability under different climatic regions, combinations of environmental factors, and geological conditions is improved, making it adaptable to the strength prediction needs of various on-site mixed soil engineering projects. The generated soil strength change trend curves and key environmental factor weight reports for temperature, humidity, and rainfall can provide targeted suggestions for optimizing on-site construction techniques and adjusting maintenance measures (such as adjusting ash content based on rainfall and optimizing maintenance time based on temperature and humidity changes), reducing engineering quality risks and improving the long-term stability of on-site mixed soil engineering projects.

[0046] Furthermore, in order to obtain more accurate prediction results, a sensitivity analysis step for intensity influencing factors is also included: based on the trained Attention-CNN-LSTM coupled prediction model, the attention weight coefficients output by the multi-head Attention dynamic weight allocation layer are statistically analyzed to quantify the influence of various environmental factors on the strength of the on-site mixed soil, and the ranking results of the core influencing factors are output, providing data support for the optimization of raw material mix ratio, adjustment of construction technology and formulation of maintenance plan for on-site mixed soil.

[0047] The embodiments described above are merely preferred embodiments of this specification and are not intended to limit the scope of this specification. Any modifications and improvements made by those skilled in the art to the technical solutions of this specification without departing from the spirit of this specification should fall within the protection scope defined by the claims of this specification.

Claims

1. A method for predicting the strength of in-situ mixed soil by coupling multiple environmental factors based on a deep learning hybrid model, characterized in that, include: S1. Set up environmental monitoring equipment and soil performance testing points in the on-site soil mixing construction area, and build a multi-source parameter acquisition system to monitor the three major environmental factors of temperature, humidity and rainfall, and simultaneously collect dynamic environmental parameters, construction core parameters and soil strength indicators. S2. Perform data cleaning and standardization on the collected multi-source parameters, construct the model input feature set, extract the feature set from two dimensions: spatial features and temporal features, and construct a two-dimensional feature matrix adapted to the Attention-CNN-LSTM model. S3. Construct an Attention-CNN-LSTM coupled prediction model. Divide the two-dimensional feature matrix obtained in S3 into a training set and a test set, and input them into the model for training. S4. Input the real-time multi-dimensional environmental influencing factor data of the mixed soil to be predicted into the trained model, and the model outputs the strength prediction value of the mixed soil at the target age. S5. Based on real-time environmental data and measured strength values, the model is incrementally updated and dynamically corrected to achieve rolling optimization prediction of the strength of the mixed soil.

2. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, The steps for constructing the two-dimensional feature matrix adapted to the Attention-CNN-LSTM model are as follows: First, the 3σ criterion is used to eliminate outliers in temperature, humidity, rainfall, and construction parameters; The missing environmental monitoring data were then supplemented using linear interpolation. For rainfall, hourly rainfall is converted into cumulative rainfall and corrected for it in relation to soil moisture. Based on the preprocessed dynamic data of temperature, humidity and rainfall, continuous functions of each environmental factor changing with time are constructed by time series fitting, and then fused into a time-varying function coupled with multiple environmental factors. At the same time, all the above parameters are normalized. Using normalized construction parameters, feature parameters of time-varying functions coupled with multiple environmental factors, and age in days as the model input feature set, the feature set is extracted and reconstructed from two dimensions: spatial features and temporal features, forming a two-dimensional feature matrix adapted to the Attention-CNN-LSTM model.

3. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors coupling, as described in claim 1 or 2, is characterized in that... Spatial characteristics are the coupling and correlation characteristics between various environmental factors such as temperature, humidity, and rainfall, and between environmental factors and construction factors; temporal characteristics are the dynamic characteristics of temperature, humidity, rainfall, and age changing over time.

4. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, The steps for constructing the Attention-CNN-LSTM coupled prediction model are as follows: Spatial features in the feature matrix are extracted using a CNN network to explore the coupling relationship between temperature, humidity, and rainfall, as well as the interaction between environmental and construction factors. The output of the CNN network is input into the LSTM network to learn the dynamic change law between temporal features and soil strength, and accurately capture the real-time impact of environmental dynamic changes on soil strength development. An attention mechanism is introduced to adaptively allocate weights to the features extracted by CNN-LSTM, automatically identify environmental and construction factors that play a key role in soil strength, and increase the weight ratio of key features. Mean squared error and coefficient of determination are used as performance evaluation indicators for the model. The hyperparameters of the model are optimized by grid search method until the model converges and reaches the optimal performance.

5. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, In S4, the workflow of the Attention-CNN-LSTM coupled prediction model is as follows: The environmental parameters of temperature, humidity, and rainfall, which are collected and preprocessed in real time during construction, as well as construction parameters, are input into the trained Attention-CNN-LSTM coupled prediction model. The model automatically completes feature extraction, temporal learning, and key factor weight allocation, and outputs the strength prediction value of the on-site mixed soil at the corresponding age, realizing real-time dynamic prediction of soil strength; at the same time, it generates a soil strength change trend curve and a report on the weights of key environmental factors such as temperature, humidity, and rainfall.

6. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors coupling, as described in claim 1, is characterized in that... In S5, the rolling optimization prediction method for the strength of mixed soil includes: The actual soil strength index is measured regularly using the in-situ testing method. The measured values ​​are compared with the model prediction values, and the deviation rate is calculated. If the deviation rate exceeds the preset threshold, the measured data and the corresponding dynamic monitoring data of temperature, humidity and rainfall will be added to the model training set to fine-tune and correct the model, continuously improve the prediction accuracy of the model in the construction area, and realize the self-optimization of the prediction model.

7. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, In S1, dynamic environmental parameters include: real-time monitoring of air temperature, soil surface / internal humidity, and cumulative rainfall / hourly rainfall, with a sampling frequency set to 1 time / hour.

8. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, In S1, the core construction parameters include at least: the amount of soil and ash to be mixed, the mixing speed, the mixing time, the number of compaction passes, and the gradation of the filler.

9. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, In S1, the soil strength index is: the compressive strength, shear strength, and unconfined compressive strength of the mixed soil are determined by in-situ testing at different ages, and used as label data for model training.

10. The method for predicting the strength of field-mixed soil based on a deep learning hybrid model with multiple environmental factors according to claim 1, characterized in that, It also includes a sensitivity analysis step for strength influencing factors: based on the trained Attention-CNN-LSTM coupled prediction model, the attention weight coefficients output by the multi-head Attention dynamic weight allocation layer are statistically analyzed to quantify the influence of various environmental factors on the strength of the on-site mixed soil, and the ranking results of the core influencing factors are output, providing data support for the optimization of raw material mix ratio, adjustment of construction technology and formulation of maintenance plan for on-site mixed soil.

Images