A "pot lid effect" analysis method based on machine learning
A machine learning-based method using the Transformer architecture addresses the inaccuracy of existing 'pot lid effect' predictions by establishing a digital twin model, ensuring precise forecasting and early warning for infrastructure safety.
Patent Information
- Application Number
- CN202411422914.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-12
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2044-10-12
AI Technical Summary
Existing methods for predicting the spatiotemporal changes in moisture and temperature due to the 'pot lid effect' in unsaturated soil are inaccurate and inefficient, failing to effectively simulate and calculate this phenomenon, which leads to engineering disasters such as uneven foundation settlement and pavement cracks in arid and cold regions like Northwest and Northeast China.
A machine learning-based analysis method using the Transformer architecture, incorporating moisture and energy control equations, to establish a digital twin model for predicting the 'pot lid effect' by training on collected data and optimizing model parameters with MSE and RMS error, enabling accurate prediction and early warning.
The method achieves high accuracy in predicting and preventing engineering disasters by creating a digital twin of the 'pot lid effect', enhancing safety and efficiency in infrastructure operations.
Smart Images
Figure CN118966016B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of geotechnical engineering technology, and in particular to a "pot lid effect" analysis method based on machine learning. Background Art
[0002] The "pot lid effect" of an airport is a phenomenon in which, driven by the temperature gradient, water in deep, enclosed soil migrates from warm areas to cold areas in the form of vapor and condenses into water or sublimates into ice, resulting in an increase in the moisture content of the shallow soil. Northwest and Northeast my country are typical arid and cold regions, with groundwater levels as deep as tens of meters and large temperature differences between morning and evening. These factors lead to the occurrence of the "pot lid effect". The accumulated water caused by the "pot lid effect" affects the roadbed or pavement through the year-round freeze-thaw cycle. These factors induce disasters such as uneven foundation settlement, cracks in the pavement, and softening of concrete, posing a potential serious threat to people's lives and property safety. Therefore, a more accurate and efficient simulation and calculation of the "pot lid effect" is crucial to improving the safe operation of airports.
[0003] In the past, researchers calculated the water vapor migration of unsaturated soil by deriving empirical formulas through indoor and outdoor tests and combining relevant empirical parameters to analyze the changes in moisture and temperature of unsaturated soil. There are also numerical simulation analyses using hot water vapor multi-field coupling equations combined with numerical software. The prediction accuracy of the above methods for the spatiotemporal changes in moisture and temperature of unsaturated soil is quite different from the actual engineering practice, and the results are not satisfactory. In recent years, machine learning has played an important role in various fields, especially in solving engineering problems, demonstrating the advantages of artificial intelligence. At present, the "pot lid effect" mechanism has not been combined with machine learning to predict the spatiotemporal changes in moisture and temperature of unsaturated soil.
[0004] Therefore, it is necessary to provide a "pot lid effect" analysis method based on machine learning to solve the above problems. Summary of the invention
[0005] The purpose of the present invention is to provide a "pot lid effect" analysis method based on machine learning, to realize digital twinning of the entire "pot lid effect" process, and to apply it to the prevention and control of disasters of pavement structures such as airport runways and highways, so as to avoid losses caused by engineering disasters caused by the "pot lid effect".
[0006] To achieve the above object, the present invention provides a "pot cover effect" analysis method based on machine learning, comprising the following steps:
[0007] S1: Collect indoor and outdoor test data through experimental equipment;
[0008] S2: Use the Transformer architecture to establish a machine learning model framework for "pot lid effect" analysis, embed parameter equations into the machine learning model, the parameter equations include moisture control equations and energy control equations, and input the test data collected in step S1 into the machine learning model;
[0009] S3: Establish a loss function and determine the evaluation index based on the spatiotemporal change scale and the "pot cover effect" boundary conditions, train the machine learning model in step S2, and establish a specific digital twin model;
[0010] S4: Based on the digital twin model trained in step S3, the temporal and spatial changes of moisture and temperature are predicted to realize the intelligent digital twin of the "pot lid effect" of unsaturated soil.
[0011] Preferably, the test data in step S1 include unsaturated soil temperature and moisture content data collected outdoors, and unsaturated soil water output and pressure tube water level fluctuation data collected indoors.
[0012] Preferably, in step S2, the formula of the moisture control equation is as follows:
[0013] ;
[0014] In the formula and are the liquid water volume content and the equivalent gaseous water volume content (m 3 ·m -3 ), is the volume content of ice (m 3 ·m -3 ), is the ice phase density (kg·m -3 ), is the density of liquid water, h is the water head height (m), T is the temperature (K), z is the spatial coordinate, the direction is vertically upward (m), t is the time (s), (m·s -1 ) is the permeability coefficient of unsaturated frozen soil, (m 2 ·K -1 ·s -1 ) is the hydraulic conductivity caused by the temperature gradient, (m·s -1 )and (m 2 ·K -1 ·s -1 ) are the gas conductivities caused by the head gradient and temperature gradient, respectively;
[0015] The formula of the energy control equation is as follows:
[0016] ;
[0017] In the formula, C p is the total specific heat capacity (MJ·m -3 ·K -1 ), which is equal to the solid specific heat capacity C n (1.92 MJ·m -3 ·K -1 ), liquid relative heat capacity C L (4.18 MJ·m -3 ·K -1 ) and gas relative heat capacity C v (6.3 KJ·m -3 ·K -1 ) and the cumulative sum of the proportion scores of each component. L w is the latent heat of evaporation of water (kJ·kg -1 ), L i is the latent heat of freezing of water (kJ·kg -1 ), '( θ ) is the thermal conductivity of unsaturated frozen soil (W·m -1 ·K -1 ), q L (m·s -1 )and q V (m·s -1 ) are the flow rates of liquid water and gaseous water, respectively.
[0018] Preferably, in step S2, the machine learning model framework learns the spatiotemporal variation law of water vapor migration through a self-attention mechanism, and the specific steps are as follows:
[0019] S21: The test data collected in step S1 is encoded by the encoder and then input into the machine learning model framework as an input parameter;
[0020] S22: The input parameters are convolved with the input parameters through the multi-head attention mechanism to obtain convolution data;
[0021] S23: performing standardization processing on the convolution data to obtain standardized data;
[0022] S24: The standardized data and input parameters are convolved with the standardized data after the multi-head attention mechanism to obtain secondary convolution data;
[0023] S25: performing a normalization operation on the secondary convolution data to obtain secondary normalized data;
[0024] S26: Convolving the secondary normalized data with the secondary normalized data after forward feedback to obtain the third convolution data;
[0025] S27: The three-convolution data is normalized to obtain a prediction result, and the prediction result is output after passing through a decoder.
[0026] Preferably, in step S3, boundary conditions are introduced into the loss function as regularization terms, and the boundary conditions include temperature boundaries and flux boundaries. The flux boundary is that the flux at the top closed covering layer is 0; the temperature boundary is set to a negative temperature at the upper boundary and 15°C at the lower boundary according to the Dirichlet boundary condition.
[0027] Preferably, in step S3, the evaluation index adopts a mixed index of mean square error (MSE) and root mean square error (RMS) to train the model framework of S2, and the model parameters are optimized according to the training results to obtain a specific digital twin model.
[0028] Preferably, in step S4, the digital twin model outputs predicted values of the temporal and spatial variations of the moisture content and temperature of the unsaturated soil, and makes early warning judgments based on actual engineering parameters.
[0029] Therefore, the present invention adopts the above-mentioned "pot lid effect" analysis method based on machine learning, which has the following beneficial effects:
[0030] (1) The present invention can not only realize the full-process digital twin of the "pot lid effect" and achieve accurate prediction and early warning, but also realize efficient calculation.
[0031] (2) The present invention utilizes the powerful computing power of machine learning to replace complex human formula fitting, which is more suitable for use in actual engineering projects. It can achieve the purpose of early warning and prevention of engineering disasters such as uneven pavement subsidence of airports and highways caused by the "pot cover effect". It has the advantages of simple calculation and high prediction accuracy, and provides a strong guarantee for the safe operation of airports.
[0032] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is a flow chart of a "pot cover effect" analysis method based on machine learning of the present invention;
[0034] Figure 2 This is the structure diagram of the machine learning framework Transformer in the present invention. DETAILED DESCRIPTION
[0035] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.
[0036] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.
[0037] The words "include" or "comprises" and the like used in the present invention mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of also including other elements. The orientation or position relationship indicated by the terms "inside", "outside", "upper", "lower", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention. When the absolute position of the described object changes, the relative position relationship may also change accordingly. In the present invention, unless otherwise clearly specified and limited, the terms "attachment" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral body; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal connection of two elements or the interaction relationship between two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances.
[0038] like Figure 1 As shown, the present invention provides a "pot cover effect" analysis method based on machine learning, comprising the following steps:
[0039] S1: Collect indoor and outdoor test data through experimental equipment; the test data in step S1 includes unsaturated soil temperature and moisture content data collected outdoors, unsaturated soil water output and pressure tube water level fluctuation data collected indoors, and the moisture control equation, energy control equation and related parameter equations are modeled.
[0040] S2: Use the Transformer architecture to establish a machine learning model framework for "pot lid effect" analysis, embed parameter equations into the machine learning model, the parameter equations include moisture control equations and energy control equations, and input the test data collected in step S1 into the machine learning model; in step S2, the formula of the moisture control equation is as follows:
[0041] ;
[0042] In the formula and are the liquid water volume content and the equivalent gaseous water volume content (m 3 ·m -3 ), is the volume content of ice (m3 ·m -3 ), is the ice phase density (kg·m -3 ), is the density of liquid water, h is the water head height (m), T is the temperature (K), z is the spatial coordinate, the direction is vertically upward (m), t is the time (s), (m·s -1 ) is the permeability coefficient of unsaturated frozen soil, (m 2 ·K -1 ·s -1 ) is the hydraulic conductivity caused by temperature gradient. (m·s -1 )and (m 2 ·K -1 ·s -1 ) are the gas conductivities caused by the head gradient and temperature gradient, respectively.
[0043] The formula of the energy control equation is as follows:
[0044] ;
[0045] In the formula, C p is the total specific heat capacity (MJ·m -3 ·K -1 ), which is equal to the solid specific heat capacity C n (1.92 MJ·m -3 ·K -1 ), liquid relative heat capacity C L (4.18 MJ·m -3 ·K -1 ) and gas relative heat capacity C v (6.3 KJ·m -3 ·K -1 ) and the cumulative sum of the proportion scores of each component. L w is the latent heat of evaporation of water (kJ·kg -1 ), L i is the latent heat of freezing of water (kJ·kg -1 ), '( θ ) is the thermal conductivity of unsaturated frozen soil (W·m -1 ·K -1 ), q L (m·s -1 )andq V (m·s -1 ) are the flow rates of liquid water and gaseous water, respectively.
[0046] In step S2, the machine learning model framework learns the spatiotemporal variation of water vapor migration through the self-attention mechanism. Figure 2 The architecture shown in the figure is as follows:
[0047] S21: The test data collected in step S1 is encoded by the encoder and then input into the machine learning model framework as an input parameter;
[0048] S22: The input parameters are convolved with the input parameters through the multi-head attention mechanism to obtain convolution data;
[0049] S23: performing standardization processing on the convolution data to obtain standardized data;
[0050] S24: The standardized data and input parameters are convolved with the standardized data after the multi-head attention mechanism to obtain secondary convolution data;
[0051] S25: performing a normalization operation on the secondary convolution data to obtain secondary normalized data;
[0052] S26: Convolving the secondary normalized data with the secondary normalized data after forward feedback to obtain the third convolution data;
[0053] S27: The three-convolution data is normalized to obtain a prediction result, and the prediction result is output after passing through a decoder.
[0054] S3: Establish a loss function and determine the evaluation index according to the spatiotemporal change scale and the "pot lid effect" boundary conditions, and train the machine learning model in step S2 to obtain a specific model. If overfitting or underfitting problems are found in this process, the training results can be fed back to the framework of the machine learning model in step S2 to continue optimizing the machine learning model and establish a specific digital twin model.
[0055] In step S3, boundary conditions are introduced into the loss function as regularization terms. The boundary conditions include temperature boundary and flux boundary. The flux boundary is that the flux at the top closed cover layer is 0; the temperature boundary is set to a negative temperature at the upper boundary and 15°C at the lower boundary according to the Dirichlet boundary condition.
[0056] In step S3, the evaluation index adopts the mixed index of mean square error (MSE) and root mean square error (RMS) to train the model framework of S2, and optimizes the model parameters according to the training results to obtain a specific digital twin model.
[0057] S4: Based on the digital twin model trained in step S3, the temporal and spatial changes of moisture and temperature are predicted to realize the intelligent digital twin of the "pot cover effect" of unsaturated soil. In step S4, the digital twin model outputs the predicted values of the temporal and spatial changes of the moisture content and temperature of the unsaturated soil, and makes early warning judgments based on the actual engineering parameters to realize the intelligent digital twin of the "pot cover effect" of unsaturated soil, and apply it to the engineering to achieve the purpose of disaster prevention and mitigation.
[0058] Therefore, the present invention adopts the above-mentioned "pot lid effect" analysis method based on machine learning. The machine learning model is combined with the physical mechanism of the "pot lid effect", which not only realizes the digital twin of the whole process of the "pot lid effect" and has high prediction and early warning accuracy, but also uses the powerful computing power of machine learning to replace complex human formula fitting, thereby improving the fitting efficiency and providing effective early warning and monitoring means for the prevention and control of engineering disasters such as airport or road foundations.
[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.
Claims
1. A "pot lid effect" analysis method based on machine learning, characterized by: The following steps are involved: S1: Collect indoor and outdoor test data through experimental equipment; S2: Use the Transformer architecture to establish a machine learning model framework for "pot lid effect" analysis, embed parameter equations into the machine learning model, the parameter equations include moisture control equations and energy control equations, and input the test data collected in step S1 into the machine learning model; S3: Establish a loss function and determine the evaluation index based on the spatiotemporal change scale and the "pot cover effect" boundary conditions, train the machine learning model in step S2, and establish a specific digital twin model; S4: Predict the spatiotemporal changes of moisture and temperature based on the digital twin model trained in step S3, and realize the intelligent digital twin of the "pot cover effect" of unsaturated soil; In step S2, the machine learning model framework learns the spatiotemporal variation of water vapor migration through the self-attention mechanism. The specific steps are as follows: S21: The test data collected in step S1 is encoded by the encoder and then input into the machine learning model framework as an input parameter; S22: The input parameters are convolved with the input parameters through the multi-head attention mechanism to obtain convolution data; S23: performing standardization processing on the convolution data to obtain standardized data; S24: The standardized data and input parameters are convolved with the standardized data after the multi-head attention mechanism to obtain secondary convolution data; S25: performing a normalization operation on the secondary convolution data to obtain secondary normalized data; S26: Convolving the secondary normalized data with the secondary normalized data after forward feedback to obtain the third convolution data; S27: The three-convolution data is normalized to obtain a prediction result, and the prediction result is output after passing through a decoder.
2. A method for analyzing the "pot lid effect" based on machine learning according to claim 1, characterized in that: The test data in step S1 include the unsaturated soil temperature and moisture content data collected outdoors, and the unsaturated soil water output and pressure tube water level fluctuation data collected indoors.
3. The method for analyzing the "pot cover effect" based on machine learning according to claim 1, characterized in that: In step S2, the moisture control equation is as follows: In the formula and are the liquid water volume content and the equivalent gaseous water volume content (m 3 ·m -3 ), is the volume content of ice (m 3 ·m -3 ), is the ice phase density (kg·m -3 ), is the density of liquid water, h is the water head height (m), T is the temperature (K), z is the spatial coordinate, the direction is vertically upward (m), t is the time (s), (m·s -1 ) is the permeability coefficient of unsaturated frozen soil, (m 2 ·K -1 ·s -1 ) is the hydraulic conductivity caused by the temperature gradient, (m·s -1 )and (m 2 ·K -1 ·s -1 ) are the gas conductivities caused by the head gradient and temperature gradient, respectively; The formula of the energy control equation is as follows: In the formula, C p is the total specific heat capacity (MJ·m -3 ·K -1 ), which is equal to the solid specific heat capacity C n , the parameter unit is 1.92MJ·m -3 ·K -1 , liquid relative heat capacity C L , the parameter unit is 4.18MJ·m -3 ·K -1 , and the relative heat capacity of gas C v , its parameter unit is 6.3KJ·m -3 ·K -1 , and the cumulative sum of the proportions of each component, L w is the latent heat of evaporation of water (kJ·kg -1 ), L i is the latent heat of freezing of water (kJ·kg -1 ), '( θ ) is the thermal conductivity of unsaturated frozen soil (W·m -1 ·K -1 ), q L (m·s -1 )and q V (m·s -1 ) are the flow rates of liquid water and gaseous water, respectively.
4. The method for analyzing the "pot lid effect" based on machine learning according to claim 1, characterized in that: In step S3, boundary conditions are introduced into the loss function as regularization terms. The boundary conditions include temperature boundary and flux boundary. The flux boundary is that the flux at the top closed cover layer is 0; the temperature boundary is set to a negative temperature at the upper boundary and 15°C at the lower boundary according to the Dirichlet boundary condition.
5. The method for analyzing the "pot lid effect" based on machine learning according to claim 1, characterized in that: In step S3, the evaluation index adopts the mixed index of mean square error (MSE) and root mean square error (RMS) to train the model framework of S2, and optimizes the model parameters according to the training results to obtain a specific digital twin model.
6. The method for analyzing the "pot lid effect" based on machine learning according to claim 1, characterized in that: In step S4, the digital twin model outputs the predicted values of the temporal and spatial variations of the moisture content and temperature of the unsaturated soil, and makes early warning judgments based on the actual engineering parameters.