A stress rotation freeze-thaw coupling frozen soil mechanics evolution intelligent prediction method
By constructing a multi-source hybrid database and an LSTM-based intelligent prediction proxy model, the problem of low computational efficiency and accuracy of traditional frozen soil mechanical prediction methods under the coupled conditions of stress rotation and freeze-thaw cycles is solved, and efficient and accurate prediction of the evolution of frozen soil mechanical properties is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GANNAN UNIV OF SCI & TECH
- Filing Date
- 2026-05-09
- Publication Date
- 2026-07-31
AI Technical Summary
Traditional methods for analyzing and predicting the mechanical properties of frozen soil suffer from low computational efficiency, high cost, and difficulty in accurately describing the evolution of mechanical properties with strong nonlinearity and strong time-dependent characteristics when dealing with coupled stress rotation and freeze-thaw cycles. These methods cannot meet the needs of rapid prediction in engineering projects.
An intelligent prediction agent model based on LSTM is adopted to construct a multi-source hybrid database. Data is collected through a hollow cylindrical torsion shear tester and field monitoring equipment. Data preprocessing and fusion are performed, and an intelligent prediction model for the evolution of frozen soil mechanical properties is constructed using a long short-term memory network (LSTM). The model predicts the evolution of properties and outputs the results. The prediction accuracy is improved through iterative optimization.
It enables accurate prediction of the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternating coupled conditions, improves computational efficiency and prediction accuracy, and meets the rapid prediction needs of engineering sites.
Smart Images
Figure CN122491029A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of geotechnical engineering and frozen soil mechanical property prediction technology in cold regions, and in particular to an intelligent prediction method for the mechanical evolution of frozen soil through stress rotation freeze-thaw coupling. Background Technology
[0002] With the rapid advancement of infrastructure construction in cold regions, major projects such as the Qinghai-Tibet Railway, the Qinghai-Tibet Highway, the China-Russia East Route Natural Gas Pipeline, and wind and photovoltaic power generation in cold regions have been successively implemented. The long-term stability control of projects in permafrost regions has become a core technical challenge. Permafrost is a geological body extremely sensitive to temperature. The alternating freeze-thaw cycle triggers the growth and melting of ice crystals within the permafrost, as well as the reorganization of soil particle structure, leading to a decrease in permafrost strength and cumulative deformation, severely impacting the service safety of projects in cold regions.
[0003] In actual cold-region engineering, under dynamic loads such as traffic loads, wind loads, seismic loads, and wave loads, the principal stress axes of the foundation soil will continuously rotate, rather than the unidirectional loading mode in traditional frozen soil mechanics research. The continuous rotation of the principal stress axes and the repeated freeze-thaw phase transitions have a significant synergistic effect. The coupling of the two will trigger the irreversible evolution of the internal structure of the frozen soil, and its mechanical property evolution path is fundamentally different from that of unidirectional loads and static freeze-thaw cycles.
[0004] Existing methods for analyzing and predicting the mechanical properties of frozen soil have significant limitations. Traditional methods based on laboratory tests require sophisticated equipment such as hollow cylindrical torsion-shear apparatus to conduct multi-condition coupled tests, which are time-consuming, costly, and difficult to cover all parameter combinations. Traditional numerical simulation methods require the establishment of complex elastoplastic constitutive models, which are difficult to accurately describe the evolution of mechanical properties under stress rotation-freeze-thaw alternation, which are highly nonlinear and time-dependent, and have low computational efficiency, failing to meet the needs of rapid prediction in engineering fields. Summary of the Invention
[0005] The purpose of this invention is to provide an intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling. Based on an intelligent prediction proxy model constructed using LSTM, it effectively overcomes the shortcomings of traditional methods in processing time-series coupled data.
[0006] This invention provides an intelligent prediction method for the mechanical evolution of frozen soil through stress-rotation freeze-thaw coupling, comprising the following steps: Step S1: Construct a multi-source hybrid database, including the collection, preprocessing, and fusion of stress rotation-freeze-thaw alternating coupled working condition parameter data and the evolution data of frozen soil mechanical properties under its action; Step S2: Intelligent predictive agent model construction and optimization, including model structure building, model training and testing, and model parameter optimization and selection; Step S3: Intelligent prediction of the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternating coupling, including property evolution prediction and result output; Step S4: Evaluation of the prediction effect of mechanical property evolution, including time series comparison, frequency domain spectrum analysis, calculation of result loss and comprehensive evaluation of prediction effect; Step S5: Iterative optimization of the mechanical property evolution prediction model, including data return, model re-optimization, and result output.
[0007] Preferably, step S1 includes the following steps: Step S11: Collect parameter data of stress rotation-freeze-thaw alternation coupling condition and corresponding frozen soil mechanical property evolution data, simulate coupling conditions with different principal stress axis rotation amplitude, rotation rate, number of freeze-thaw cycles, temperature amplitude, confining pressure, and deviatoric stress parameters, and calculate the evolution law of frozen soil mechanical properties under coupling effect by combining different frozen soil types and physical property parameter combinations, and obtain numerical simulation data. Using a hollow cylindrical torsion shear tester, temperature control system, strain sensor, stress sensor, and pore water pressure sensor, the stress path, temperature change, strain evolution, strength decay, and pore water pressure change data of frozen soil under coupled working conditions are monitored and recorded in real time to form indoor test data; and using field monitoring equipment, the stress state, freeze-thaw cycle process, deformation and settlement data of frozen soil at the cold region engineering site are collected to form field measured data. Step S12: Process the collected coupled working condition parameter data and their corresponding mechanical property evolution data. Use the mean filtering method to remove noise interference in the data. Based on the data mean and standard deviation, statistically identify features and remove outliers. Use linear spline interpolation to fill missing values. Use normalization method to unify data of different types and ranges to the same scale. Step S13: Using association rule mining algorithm and cluster analysis algorithm, the correlation between numerical simulation data, indoor test data and field measured data is mined, and the coupled working condition parameter data and the corresponding frozen soil mechanical property evolution data under its action are integrated to form a multi-source hybrid database containing multi-source information.
[0008] Preferably, step S2 includes the following steps: Step S21: Construct an intelligent prediction proxy model for the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternation coupling based on the Long Short-Term Memory (LSTM) network, including an input layer, a hidden layer, and an output layer; The input layer receives time-series and statistical parameters of the stress rotation-freeze-thaw coupling condition, as well as time-series data of frozen soil mechanical properties under the coupling effect. The hidden layer extracts the time-series features of the input data and captures the synergistic evolution law of the principal stress axis rotation and freeze-thaw coupling effect. The output layer predicts the evolution parameters of frozen soil mechanical properties, including cumulative plastic strain, shear strength, elastic modulus, Poisson's ratio, void ratio, cohesion, internal friction angle, and pore water pressure. Step S22: Randomly select a portion of data from the multi-source hybrid database as the training set to initialize the training of the intelligent prediction agent model; during the training process, adjust the model parameters to minimize the prediction error; Step S23: Test the training effect under different combinations of prediction parameters, compare the training results under different parameter combinations, and select the parameter combination that makes the model achieve the best test effect as the initial parameters of the final LSTM intelligent prediction agent model.
[0009] Preferably, step S3 includes the following steps: Step S31: Input the pre-processed and fused stress rotation-freeze-thaw alternating coupling data into the intelligent prediction agent model that has been built, trained and optimized to predict the evolution of the mechanical properties of frozen soil under the coupling effect, including cumulative plastic strain, shear strength attenuation, elastic modulus evolution, cohesion and internal friction angle changes, and pore water pressure dissipation and accumulation law. Step S32: Visualize the predicted evolution results of frozen soil mechanical properties and store the prediction results in the database.
[0010] Preferably, step S4 includes the following steps: Step S41: Compare the predicted time series data of frozen soil mechanical property evolution with the actual experimental / measured time series data point by point, analyze the numerical differences between the two at the same time / number of cycles, and evaluate the accuracy of the prediction results in the time series. Step S42: Perform frequency domain analysis on the predicted results and actual measurement results, convert the time domain signal to the frequency domain, and plot the spectrum of the response signal; analyze the distribution and amplitude of different frequency components in the spectrum, compare the characteristic differences between the predicted results and actual results in the frequency domain, and evaluate the model's ability to predict the evolution of mechanical properties under different cycle periods. Step S43: Calculate the loss between the predicted result and the actual result, and use the mean square error, root mean square error and mean absolute error as loss functions to quantify the accuracy of the prediction result. Step S44: Based on the accuracy ratio of time series comparison, the consistency ratio of frequency domain spectrum analysis, and the calculation ratio of result loss, comprehensively evaluate the prediction effect of mechanical property evolution; assign reasonable weights to each evaluation indicator according to the application scenario and requirements, and determine whether the prediction effect meets the requirements.
[0011] Preferably, in step S44, determining whether the prediction effect meets the requirements includes: calculating the comprehensive evaluation score using a weighted average method, and setting the comprehensive evaluation result as follows: E p Set the evaluation threshold as M e ;like E p Not less than M e If the prediction effect is deemed satisfactory, the prediction result will be output; if E p Less than M e If the prediction effect is not satisfactory, the model will be further optimized.
[0012] Preferably, in step S5, if the prediction effect does not meet the requirements, all types of data involved in the evaluation of the mechanical property evolution prediction effect, including prediction results, actual measurement results, and loss calculation data, are returned to the multi-source hybrid database; based on the original input data and the returned data, and the results of the prediction effect evaluation, the intelligent prediction agent model is optimized again to improve the prediction accuracy of the model; the re-optimized model is used to predict the mechanical property evolution of frozen soil under the stress rotation-freeze-thaw alternating coupling action again, and the process of time-domain sequence comparison, frequency-domain spectrum analysis, result loss calculation, and prediction effect evaluation is repeated to verify the prediction effect of the optimized model.
[0013] Preferably, in step S5, if the prediction effect meets the requirements, the prediction result is output, including time domain and frequency domain comparison data, evaluation data and the final determined initial parameter settings of the model.
[0014] Therefore, the present invention adopts the above-mentioned intelligent prediction method for the mechanical evolution of frozen soil with stress rotation freeze-thaw coupling, and the intelligent prediction surrogate model based on LSTM effectively overcomes the shortcomings of traditional methods in processing time-series coupled data.
[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0016] Figure 1 This is an overall flowchart of the intelligent prediction method for the mechanical evolution of frozen soil by stress rotation freeze-thaw coupling according to the present invention. Figure 2This is a schematic diagram of the LSTM intelligent prediction surrogate model structure of the intelligent prediction method for stress rotation freeze-thaw coupling frozen soil mechanical evolution of the present invention. Figure 3 This is a flowchart illustrating the multi-source hybrid database construction and data fusion process of an intelligent prediction method for the mechanical evolution of frozen soil through stress rotation freeze-thaw coupling, as described in this invention. Detailed Implementation
[0017] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.
[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.
[0019] The terms "first," "second," and similar words used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0020] Example 1 like Figures 1-3 As shown, the present invention provides an intelligent prediction method for the mechanical evolution of frozen soil through stress-rotation freeze-thaw coupling, comprising the following steps: Step S1: Construct a multi-source hybrid database, including the collection, preprocessing, and fusion of stress rotation-freeze-thaw alternating coupled working condition parameter data and the evolution data of frozen soil mechanical properties under its action.
[0021] Step S1 includes the following steps: Step S11: Collect parameter data of stress rotation-freeze-thaw alternation coupling condition and corresponding evolution data of frozen soil mechanical properties. On the one hand, based on the frozen soil elastoplastic constitutive model, finite element numerical simulation software is used to simulate the coupling condition with different principal stress axis rotation amplitude, rotation rate, number of freeze-thaw cycles, temperature amplitude, confining pressure, and deviatoric stress parameters. Combined with different frozen soil types, initial water content, dry density and other physical property parameters, the evolution law of frozen soil mechanical properties under coupling action is calculated to obtain numerical simulation data.
[0022] On the other hand, indoor coupled tests were conducted using a hollow cylindrical torsion-shear apparatus, a high- and low-temperature environmental chamber, a temperature control system, strain sensors, stress sensors, and pore water pressure sensors. Real-time monitoring and recording of stress path, temperature changes, strain evolution, strength attenuation, and pore water pressure changes in the frozen soil under coupled conditions were used to generate indoor test data. Field monitoring equipment was used to collect data on the stress state, freeze-thaw cycle process, and deformation and settlement of the frozen soil at the cold-region engineering site, generating field measured data.
[0023] The parameter range for the stress rotation-freeze-thaw alternating coupled condition in actual cold-region engineering was determined, and the number of freeze-thaw cycles was set as follows. N f The temperature amplitude is Δ T The principal stress axis rotation angle is θ The rotational speed is ω The confining pressure is σ 3. The deviatoric stress is σ d H sets of coupled operating condition parameter combinations are set; a multi-source input mode combining numerical simulation data, indoor test data, and field measured data is adopted, with the numerical data sequence accounting for a certain percentage. P 1. The proportion of indoor test data sequences is: P 2. The proportion of the on-site measured data sequence is: P 3. Obtain P 1+ P 2+ P 3 = 100%.
[0024] Using finite element numerical simulation software, in H Batch generation under group coupling operating conditions R One set of numerical simulation samples was generated. N One set of coupled working condition parameters and corresponding numerical data on the evolution of frozen soil mechanical properties, with the numerical sequence numbered sequentially as [1,2,…, N 1], there are N 1= H × R 1.
[0025] Based on the indoor hollow cylinder torsion-shear coupled test system, the parameters were obtained under the H group of coupled working conditions. R Two groups of experimental samples were obtained. N Two sets of coupled working condition parameters and corresponding indoor test data on the evolution of frozen soil mechanical properties, the test sequence numbers are as follows: [ N 1+1, N 1+2,…, N 1+ N 2], obtained N 2= H × R 2.
[0026] Based on the long-term on-site monitoring system for cold-region engineering projects, respectively in H Under typical working conditions, field monitoring samples of group R3 were obtained, and a total of N3 coupled working condition parameters and corresponding field measured data on the evolution of frozen soil mechanical properties were obtained.
[0027] The measured sequence numbers are as follows: [ N 1+ N 2+1, N 1+ N 2+2,…, N 1+ N 2+ N 3], obtained N 3= H × R 3.
[0028] The initial collection of mixed data on the evolution of coupled operating parameters and mechanical properties yielded a total sample size of [number missing]. N = N 1+ N 2+ N 3. The mixed data sequence numbers are [1,2,…, N Each sample group contains a coupled working condition time series parameter sequence and the evolution time series sequence of frozen soil mechanical properties under its action.
[0029] The study selected silty clay from the Qinghai-Tibet Plateau as the research object, and provided a detailed description. The initial moisture content of the frozen soil sample was 18%, and the dry density was 1.7 g / cm³. 3 The freeze-thaw cycle temperature range is -20℃ to +20℃, the principal stress axis rotation angle range is 0° to 360°, the rotation speed is 5° / s to 60° / s, and the confining pressure range is 100kPa to 500kPa. Numerical simulations were conducted using ABAQUS finite element software, indoor coupled tests were carried out using a hollow cylindrical torsion shear apparatus, and field measured data were obtained from permafrost monitoring stations along the Qinghai-Tibet Railway. A preliminary multi-source hybrid database containing 2... N Data sequences, divided into N Group.
[0030] Step S12: Process the collected coupled working condition parameter data and its corresponding mechanical property evolution data to improve data quality.
[0031] Mean filtering is used to remove noise interference from the data. Mean filtering is also used for cleaning mixed data. This is particularly relevant for the initially collected data. N Group data, set the filter window size to Q For each data point, the average value of the data within the window surrounding each data point is calculated to obtain a sequence of data of the same duration, which is then used as the filtered value for that point to remove interference such as sensor noise and acquisition errors from the data.
[0032] Based on the data mean and standard deviation, statistical features are identified and outliers are removed. Linear spline interpolation is used to fill in missing values. Normalization methods are used to unify data of different types and ranges to the same scale.
[0033] Based on the statistical characteristics of the mixed data, the mean and standard deviation of each dataset are calculated. Data sequences that deviate from the mean by more than three times the standard deviation are identified as outliers and removed from the mixed database.
[0034] For data sequences with missing values, linear spline interpolation is used. S (x), fills in missing values based on the values of adjacent data sequences, using a linear spline interpolation function. S (x), as shown in the following formula: S (x)= y i +( y i +1- y i )×( x - x i ) / ( x i +1- x i ); in, x i and x i +1 represents the x-coordinate of two adjacent data sequences. y i and y i +1 is its corresponding ordinate. x It is an interval [ x i , x i Any point within [+1] S (x) is the value of the linear spline function at x within the interval.
[0035] According to this method N The mixed database is used to remove outlier data and fill in missing values.
[0036] The Z-score data normalization method is implemented using the StandardScaler class in the sklearn library of Python programming software. After cleaning, data of different types and ranges are input into the object for fitting and transformation, and the data are unified to the same scale with a mean of 0 and a standard deviation of 1, eliminating the influence of dimensions and facilitating subsequent model processing.
[0037] Step S13: Using association rule mining algorithm and cluster analysis algorithm, the correlation between numerical simulation data, indoor test data and field measured data is mined, and the coupled working condition parameter data and the corresponding frozen soil mechanical property evolution data under its action are integrated to form a multi-source hybrid database containing multi-source information.
[0038] Data fusion: Read numerical simulation data, indoor test data and field measurement data, and convert them into a unified format for data fusion.
[0039] The Apriori association rule mining algorithm was used to mine the association relationships between three types of data. By setting minimum support and minimum confidence, frequent itemsets and association rules were identified, and the intrinsic relationship between coupled working condition parameters and the evolution of frozen soil mechanical properties was clarified.
[0040] The K-means clustering algorithm was used to cluster the data, dividing it into datasets with different permafrost types and different coupling strengths.
[0041] Based on association rules and clustering results, the coupled working condition parameter data and the corresponding frozen soil mechanical property evolution data are fused to form a multi-source hybrid database containing multi-source information for subsequent model training and prediction.
[0042] Step S2: Intelligent prediction agent model construction and optimization, including model structure building, model training and testing, and model parameter optimization and selection.
[0043] Step S2 includes the following steps: Step S21: Construct an intelligent prediction proxy model for the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternation coupling based on a long short-term memory network (LSTM), including an input layer, a hidden layer, and an output layer.
[0044] The input layer receives the time series and statistical parameters of the stress rotation-freeze-thaw coupling condition, as well as the time series data of frozen soil mechanical properties under the coupling effect. The hidden layer extracts the time series features of the input data and captures the synergistic evolution law of the principal stress axis rotation and freeze-thaw coupling effect. The output layer predicts the evolution parameters of frozen soil mechanical properties, including cumulative plastic strain, shear strength, elastic modulus, Poisson's ratio, void ratio, cohesion, internal friction angle, and pore water pressure.
[0045] An input-hidden-output surrogate model is constructed using the TensorFlow / Keras deep learning framework in Python. First, the input layer dimensions are defined, and the number of input layer nodes is determined based on the coupled working condition parameter sequences and the frozen soil mechanical property evolution data sequences from the hybrid database. Q Next, the hidden layer is constructed, and the number of LSTM units is set to [value missing]. M e ,andM e =128, using the `tf.keras.layers.LSTM` function to construct a two-layer LSTM hidden layer, adding a Dropout layer to prevent overfitting, with the Dropout rate set to 0.2. Then, the output layer is constructed, with the number of output layer nodes determined based on the number of frozen soil mechanical property parameters to be predicted. M p Number of predicted mechanical property parameters D =8, the number of output layer nodes is M p =8, and the output layer is constructed using the tf.keras.layers.Dense fully connected layer function.
[0046] Step S22: Randomly select a portion of data from the multi-source hybrid database as the training set to initialize the training of the intelligent prediction agent model; during the training process, adjust the model parameters to minimize the prediction error.
[0047] A proportion of P is randomly selected from a multi-source hybrid database. t The data is used as the training set. P t 80% of the data was used as the training set, and the remaining 20% was used as the test set. The LSTM intelligent prediction agent model was initialized and trained using the training set, with training parameters set and an initial learning rate selected. R l The value is 0.01, representing the number of training rounds. R r 300 rounds, batch size B s The learning rate is set to 64, and a learning rate decay strategy is adopted, with the learning rate decaying to 0.8 every 50 rounds.
[0048] During training, mean squared error (MSE) is used as the loss function, and the Adam optimization algorithm is used to adjust the model parameters to minimize the loss function value, that is, to minimize the prediction error, so that the model learns the inherent nonlinear relationship between the stress rotation-freeze-thaw alternating coupled working condition and the evolution of frozen soil mechanical properties.
[0049] After each round of training, the model is tested using a test set, and the loss value of the test set is recorded. M l Quantitatively evaluate the training effect of the model.
[0050] Step S23: Test the training effect under different combinations of prediction parameters, compare the training results under different parameter combinations, and select the parameter combination that makes the model achieve the best test effect as the initial parameters of the final LSTM intelligent prediction agent model.
[0051] Design parameter optimization experiments and select the number of LSTM units. M e Learning rate R l Training Wheel R r Different prediction parameters, such as the number of hidden layers, were used to test the training performance under different combinations of prediction parameters using a grid search algorithm. The evaluation metrics of the model on the test set were recorded in each experiment, and the coefficient of determination R0 was selected. 2 Mean absolute error (MAE) and root mean square error (RMSE) were used as evaluation metrics. The training results under different parameter combinations were compared, and the parameter combination that achieved the best test performance was selected as the initial parameters for the final LSTM intelligent prediction surrogate model.
[0052] Step S3: Intelligent prediction of the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternating coupling, including property evolution prediction and result output.
[0053] Step S31: Input the pre-processed and fused stress rotation-freeze-thaw alternating coupling data into the intelligent prediction agent model that has been built, trained and optimized. Based on the learned coupling evolution law, the model calls the model's predict function to predict the evolution of the mechanical properties of frozen soil under coupling action, including cumulative plastic strain, shear strength attenuation, elastic modulus evolution, cohesion and internal friction angle changes, and pore water pressure dissipation and accumulation law.
[0054] Step S32: Plot the time / cycle number-mechanical property response curve using Matplotlib to visualize the predicted evolution of frozen soil mechanical properties. Different curve styles and colors are set to distinguish different mechanical property parameters. At the same time, the prediction results are stored in a database for subsequent engineering design and stability analysis applications.
[0055] Step S4: Evaluation of the prediction effect of mechanical property evolution, including time series comparison, frequency domain spectrum analysis, calculation of result loss, and comprehensive evaluation of prediction effect.
[0056] Step S4 includes the following steps: Step S41: Compare the predicted time series data of frozen soil mechanical property evolution with the actual experimental / measured time series data point by point, analyze the numerical differences between the two at the same time / number of cycles, and evaluate the accuracy of the prediction results in the time series.
[0057] Specifically, the difference between the predicted value and the actual value is calculated at each time point / cycle number, the numerical differences between the two at the same time / cycle number are analyzed, and the distribution of the difference is statistically analyzed to evaluate the accuracy of the prediction results on the time series.
[0058] Step S42: Perform frequency domain analysis on the predicted results and actual measurement results using Fast Fourier Transform (FFT). Set appropriate sampling frequency and signal length to convert the time domain signal to the frequency domain and plot the spectrum of the response signal. Analyze the distribution and amplitude of different frequency components in the spectrum, compare the characteristic differences between the predicted results and actual results in the frequency domain, and evaluate the model's ability to predict the evolution of mechanical properties under different cycle periods.
[0059] Step S43: Calculate the loss between the predicted result and the actual result, using mean square error, root mean square error and mean absolute error as loss functions to quantify the accuracy of the prediction result.
[0060] The Python programming software is used to calculate the loss between the predicted and actual results. The mean squared error (MSE), root mean square error (RMSE), and mean absolute error (MAE) are used as loss functions to quantify the accuracy of the prediction results, as shown in the following formula: MSE=(1 / n )×Σ( y i pred - y i true ) 2 .
[0061] RMSE = √[(1 / n )×Σ(( y i pred - y i true ) 2 ].
[0062] MAE=(1 / n )×Σ|y i pred -y i true |
[0063] in, n For the sample size, y i pred For predicted values, y i true This represents the actual value. The calculated loss results are stored in a database for subsequent analysis and comparison.
[0064] Step S44: Based on the accuracy percentage of time series comparisons CP 1. Consistency ratio of frequency domain spectrum analysis CP 2. Calculation of the proportion of the result loss CP 3. Conduct a comprehensive evaluation of the prediction effect on the evolution of mechanical properties. Establish an evaluation index system. CP 1+ CP 2+ CP 3 = 100%. CP 1=35%, CP 2 = 25%, CP 3=40%, assigning reasonable weights to each evaluation indicator based on the application scenario and requirements to determine whether the prediction effect meets the requirements.
[0065] In step S44, determining whether the prediction effect meets the requirements includes: calculating the comprehensive evaluation score using a weighted average method, and setting the comprehensive evaluation result as follows: E p Set the evaluation threshold as M e ;like E p Not less than M e If the prediction effect is deemed satisfactory, the prediction result will be output; if E p Less than M e If the prediction effect is not satisfactory, the model will be further optimized.
[0066] Step S5: Iterative optimization of the mechanical property evolution prediction model, including data return, model re-optimization, and result output.
[0067] In step S5, if the prediction effect does not meet the requirements, all types of data involved in the evaluation of the mechanical property evolution prediction effect, including prediction results, actual measurement results, and loss calculation data, are returned to the multi-source hybrid database. Based on the original input data, the returned data, and the prediction effect evaluation results, the intelligent prediction agent model is optimized again to improve the prediction accuracy of the model. The re-optimized model is used to predict the evolution of the mechanical properties of frozen soil under the stress rotation-freeze-thaw alternating coupling action again, and the process of time-domain sequence comparison, frequency-domain spectrum analysis, result loss calculation, and prediction effect evaluation is repeated to verify the prediction effect of the optimized model.
[0068] In step S5, if the prediction effect meets the requirements, the prediction result is output, including time domain and frequency domain comparison data, evaluation data and the final determined initial parameter settings of the model.
[0069] S51. Data Return and Model Re-optimization: If the prediction results do not meet the requirements, the prediction results are output, including time-domain and frequency-domain comparison data, evaluation data, and the final determined initial model parameter settings. Based on the Python data manipulation library, various data involved in the evaluation of the mechanical property evolution prediction results, including prediction results, actual measurement results, and loss calculation data, are returned to a multi-source hybrid database to expand the database sample size and provide more information for subsequent model re-optimization.
[0070] Based on the original input data, the returned data, and the results of the prediction performance evaluation, the intelligent prediction agent model is further optimized to improve the model's prediction accuracy. Further processing and model enhancement are performed by adjusting hyperparameters such as the number of neurons in the hidden layer, the number of hidden layers, the training set ratio, and the learning rate. A multi-input multi-output evaluation model is established, and model reconstruction and retraining are achieved by optimizing each step in steps S21, S22, and S23.
[0071] S52. Re-prediction and re-evaluation: Using the re-optimized model, the evolution of the mechanical properties of frozen soil under the coupled action of stress rotation and freeze-thaw cycles is predicted again according to steps S31 and S32. Steps S41, S42, S43 and S44 are repeated to compare the results of the re-prediction in the time domain, perform frequency domain spectrum analysis, calculate the loss amount, and evaluate the prediction effect. The prediction effect after model optimization is verified. If the prediction effect is still not ideal, the iterative process of data return and model re-optimization continues until the prediction effect meets the requirements of actual engineering applications.
[0072] S53. Final output of prediction results: If the prediction effect meets the requirements, the prediction results will be output, including time domain and frequency domain comparison data, comprehensive evaluation data and the final determined initial parameter settings of the model, to provide data support for the long-term stability design, operation and maintenance of frozen soil roadbeds, foundations and other structures in cold regions.
[0073] Therefore, this invention adopts the above-mentioned intelligent prediction method for the mechanical evolution of frozen soil with stress rotation and freeze-thaw coupling. Based on the intelligent prediction surrogate model constructed by LSTM, it effectively overcomes the shortcomings of traditional methods in processing time-series coupled data and accurately reveals the stress rotation-freeze-thaw alternation coupling mechanism. Through a comprehensive multi-dimensional prediction effect evaluation and closed-loop iterative optimization mechanism, the prediction accuracy and generalization ability of the model can be continuously improved.
[0074] The above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for intelligent prediction of the mechanical evolution of frozen soil through stress-rotation freeze-thaw coupling, characterized in that, Includes the following steps: Step S1: Construct a multi-source hybrid database, including the collection, preprocessing, and fusion of stress rotation-freeze-thaw alternating coupled working condition parameter data and the evolution data of frozen soil mechanical properties under its action; Step S2: Intelligent predictive agent model construction and optimization, including model structure building, model training and testing, and model parameter optimization and selection; Step S3: Intelligent prediction of the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternating coupling, including property evolution prediction and result output; Step S4: Evaluation of the prediction effect of mechanical property evolution, including time series comparison, frequency domain spectrum analysis, calculation of result loss and comprehensive evaluation of prediction effect; Step S5: Iterative optimization of the mechanical property evolution prediction model, including data return, model re-optimization, and result output.
2. The intelligent prediction method for the mechanical evolution of frozen soil through stress-rotation freeze-thaw coupling according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Collect parameter data of stress rotation-freeze-thaw alternation coupling condition and corresponding frozen soil mechanical property evolution data, simulate coupling conditions with different principal stress axis rotation amplitude, rotation rate, number of freeze-thaw cycles, temperature amplitude, confining pressure, and deviatoric stress parameters, and calculate the evolution law of frozen soil mechanical properties under coupling effect by combining different frozen soil types and physical property parameter combinations, and obtain numerical simulation data. Using a hollow cylindrical torsion shear tester, temperature control system, strain sensor, stress sensor, and pore water pressure sensor, the stress path, temperature change, strain evolution, strength decay, and pore water pressure change data of frozen soil under coupled working conditions are monitored and recorded in real time to form indoor test data; and using field monitoring equipment, the stress state, freeze-thaw cycle process, deformation and settlement data of frozen soil at the cold region engineering site are collected to form field measured data. Step S12: Process the collected coupled working condition parameter data and their corresponding mechanical property evolution data. Use the mean filtering method to remove noise interference in the data. Based on the data mean and standard deviation, statistically identify features and remove outliers. Use linear spline interpolation to fill missing values. Use normalization method to unify data of different types and ranges to the same scale. Step S13: Using association rule mining algorithm and cluster analysis algorithm, the correlation between numerical simulation data, indoor test data and field measured data is mined, and the coupled working condition parameter data and the corresponding frozen soil mechanical property evolution data under its action are integrated to form a multi-source hybrid database containing multi-source information.
3. The intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling according to claim 2, characterized in that, Step S2 includes the following steps: Step S21: Construct an intelligent prediction proxy model for the evolution of frozen soil mechanical properties under stress rotation-freeze-thaw alternation coupling based on the Long Short-Term Memory (LSTM) network, including an input layer, a hidden layer, and an output layer; The input layer receives time-series and statistical parameters of the stress rotation-freeze-thaw coupling condition, as well as time-series data of frozen soil mechanical properties under the coupling effect. The hidden layer extracts the time-series features of the input data and captures the synergistic evolution law of the principal stress axis rotation and freeze-thaw coupling effect. The output layer predicts the evolution parameters of frozen soil mechanical properties, including cumulative plastic strain, shear strength, elastic modulus, Poisson's ratio, void ratio, cohesion, internal friction angle, and pore water pressure. Step S22: Randomly select a portion of data from the multi-source hybrid database as the training set to initialize the training of the intelligent prediction agent model; during the training process, adjust the model parameters to minimize the prediction error; Step S23: Test the training effect under different combinations of prediction parameters, compare the training results under different parameter combinations, and select the parameter combination that makes the model achieve the best test effect as the initial parameters of the final LSTM intelligent prediction agent model.
4. The intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling according to claim 3, characterized in that, Step S3 includes the following steps: Step S31: Input the pre-processed and fused stress rotation-freeze-thaw alternating coupling data into the intelligent prediction agent model that has been built, trained and optimized to predict the evolution of the mechanical properties of frozen soil under the coupling effect, including cumulative plastic strain, shear strength attenuation, elastic modulus evolution, cohesion and internal friction angle changes, and pore water pressure dissipation and accumulation law. Step S32: Visualize the predicted evolution results of frozen soil mechanical properties and store the prediction results in the database.
5. The intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling according to claim 4, characterized in that, Step S4 includes the following steps: Step S41: Compare the predicted time series data of frozen soil mechanical property evolution with the actual experimental / measured time series data point by point, analyze the numerical differences between the two at the same time / number of cycles, and evaluate the accuracy of the prediction results in the time series. Step S42: Perform frequency domain analysis on the predicted results and actual measurement results, convert the time domain signal to the frequency domain, and plot the spectrum of the response signal; analyze the distribution and amplitude of different frequency components in the spectrum, compare the characteristic differences between the predicted results and actual results in the frequency domain, and evaluate the model's ability to predict the evolution of mechanical properties under different cycle periods. Step S43: Calculate the loss between the predicted result and the actual result, and use the mean square error, root mean square error and mean absolute error as loss functions to quantify the accuracy of the prediction result. Step S44: Based on the accuracy ratio of time series comparison, the consistency ratio of frequency domain spectrum analysis, and the calculation ratio of result loss, comprehensively evaluate the prediction effect of mechanical property evolution; assign reasonable weights to each evaluation indicator according to the application scenario and requirements, and determine whether the prediction effect meets the requirements.
6. The intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling according to claim 5, characterized in that, In step S44, determining whether the prediction effect meets the requirements includes: calculating the comprehensive evaluation score using a weighted average method, and setting the comprehensive evaluation result as follows: E p Set the evaluation threshold as M e ;like E p Not less than M e If the prediction effect is deemed satisfactory, the prediction result will be output; if E p Less than M e If the prediction effect is not satisfactory, the model will be further optimized.
7. The intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling as described in claim 6, characterized in that, In step S5, if the prediction effect does not meet the requirements, all types of data involved in the evaluation of the mechanical property evolution prediction effect, including prediction results, actual measurement results, and loss calculation data, are returned to the multi-source hybrid database. Based on the original input data, the returned data, and the prediction effect evaluation results, the intelligent prediction agent model is optimized again to improve the prediction accuracy of the model. The re-optimized model is used to predict the evolution of the mechanical properties of frozen soil under the stress rotation-freeze-thaw alternating coupling action again, and the process of time-domain sequence comparison, frequency-domain spectrum analysis, result loss calculation, and prediction effect evaluation is repeated to verify the prediction effect of the optimized model.
8. The intelligent prediction method for the mechanical evolution of frozen soil under stress-rotation freeze-thaw coupling according to claim 7, characterized in that, In step S5, if the prediction effect meets the requirements, the prediction result is output, including time domain and frequency domain comparison data, evaluation data and the final determined initial parameter settings of the model.