A method and system for predicting the strength and modulus of all mix proportions of silt solidified soil

Abstract

The present invention discloses a method and a system for predicting the full mix ratio strength and modulus of silt solidified soil, which relates to the field of building materials. The method includes the following steps: obtaining relevant input data of the silt solidified soil; constructing a physics-informed neural network model, and embedding the physical equations describing the mechanical behavior of the silt solidified soil as constraints into the neural network model; training the physics-informed neural network model by using the input data so that the model parameters gradually converge, and comprehensively considering the experimental data error and the physical equation residual during the training process; and using the trained and converged physics-informed neural network model to predict the strength and modulus of the silt solidified soil. The system includes various modules for implementing the above respective steps. The present invention combines the traditional neural network with the physical equations describing the silt solidified soil, effectively overcoming the problems of the traditional prediction method being overly dependent on empirical formulas, having a large data demand, and poor generalization performance.

Description

Technical Field

[0001] The present invention relates to the field of building materials, and more specifically, to a method and system for predicting the strength and modulus of the full mix ratio of silt solidified soil. Background Art

[0002] Silt solidified soil refers to the soil material formed by solidifying dredged silt by adding curing agents such as cement, and is widely used in engineering such as soft foundation reinforcement, roadbed filling, and marine reclamation. Accurately predicting the mechanical properties (such as compressive strength and deformation modulus) of silt solidified soil is crucial for engineering design and safety assessment. However, there are many deficiencies in the existing prediction methods: traditional empirical methods usually fit empirical formulas based on limited indoor test data, and it is difficult to cover all factors affecting strength and modulus, resulting in limited prediction accuracy; although simple data-driven models (such as traditional BP neural networks or regression models) can use a large amount of test data for prediction, when the data volume is insufficient or beyond the training range, the model often lacks physical constraints and may have unreasonable prediction results; purely relying on numerical simulation (such as finite element consolidation analysis) also requires accurate acquisition of soil parameters, and the process is complex and time-consuming. Generally speaking, the existing technology lacks a prediction method that can integrate physical mechanisms and use measured data to improve the accuracy and generalization ability of predicting the strength and modulus of silt solidified soil. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for predicting the strength and modulus of the full mix ratio of silt solidified soil to solve the problems mentioned in the background art.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A method for predicting the strength and modulus of the full mix ratio of silt solidified soil, comprising the following steps:

[0006] Obtain relevant input data of silt solidified soil;

[0007] Construct a physics-informed neural network model, and embed the physical equations describing the mechanical behavior of silt solidified soil as constraints into the neural network model;

[0008] Use the input data to train the physics-informed neural network model so that the model parameters gradually converge, and comprehensively consider the experimental data error and the physical equation residual during the training process;

[0009] Use the physics-informed neural network model after the training converges to predict the strength and modulus of silt solidified soil.

[0010] Optionally, the relevant input data includes the initial physical and chemical parameters of silt solidified soil and the curing condition data;

[0011] The initial physicochemical parameters include water content, density, particle composition, and curing agent dosage;

[0012] The curing condition data include age, ambient temperature, and humidity.

[0013] Optionally, the physics-informed neural network model is a multi-layer deep neural network, including an input layer, an output layer, and at least two hidden layers;

[0014] Among them, the input layer receives the relevant input data of the silt-solidified soil, and the output layer outputs the predicted values of the strength and modulus of the silt-solidified soil.

[0015] Optionally, during the training process of the physics-informed neural network model, physical mechanism constraint equations related to the strength and modulus of the silt-solidified soil are introduced, including the control equation of soft soil consolidation and the empirical equation of the strength growth of the solidified soil.

[0016] Optionally, the control equation of soft soil consolidation includes Terzaghi's one-dimensional consolidation equation;

[0017] The empirical equation of the strength growth of the solidified soil includes:

[0018] ;

[0019] Among them, is the compressive strength of the solidified soil at the curing age t, is the strength stability limit value, and k is the strength growth rate constant.

[0020] Optionally, the method further includes constructing a loss function containing physical constraints to train the physics-informed neural network model;

[0021] The loss function is composed of a data fitting error term and a physical equation residual term;

[0022] Among them, the data fitting error term is used to measure the deviation between the predicted strength and modulus of the model and the measured values; the physical equation residual term is used to measure the degree of violation of the model prediction results against the relevant physical equations;

[0023] The neural network parameters are optimized by minimizing the loss function.

[0024] Optionally, the physics-informed neural network model adopts a multi-stage training optimization method: the training process is divided into multiple stages, including the early stage of rapid strength growth, the middle stage of stable strength growth, and the late stage of strength tending to be stable; the weight of the physical equation residual term in the loss function is dynamically adjusted in different training stages.

[0025] Optionally, an iterative optimization algorithm based on gradient descent is used to train the physical information neural network model. During the training process, a weight coefficient is set for the physical equation residual term to balance its contribution with the data error term. The training is terminated when the loss function converges to a preset threshold or reaches the maximum number of training iterations.

[0026] Optionally, the method uses multi-source data for training, including on-site sensor monitoring data and laboratory test data.

[0027] Among them, the on-site sensor monitoring data includes the measured data of soft soil foundation settlement, pore water pressure or water content.

[0028] The laboratory test data includes the data obtained from the unconfined compressive strength and deformation modulus tests on the silt solidified soil samples taken from the site. The data diversity and reliability for model training are improved through data preprocessing and fusion.

[0029] Optionally, the method is used for any of the following scenarios:

[0030] Foundation treatment engineering, predicting the strength and modulus of silt solidified soil in the soft soil foundation after reinforcement.

[0031] Soft soil foundation reinforcement in road construction, predicting the strength and modulus of silt solidified soil in the subgrade.

[0032] Dredger fill silt foundation treatment in ocean engineering, predicting the strength and modulus of the dredger fill silt after solidification treatment.

[0033] Accordingly, the present invention also discloses a prediction system for the full mix ratio strength and modulus of silt solidified soil, including a data acquisition module, a physical information neural network calculation module, and a result output module.

[0034] Among them, the data acquisition module is used to collect the initial parameters and curing condition data of the silt solidified soil.

[0035] The physical information neural network calculation module is used to call the trained physical information neural network model according to the initial parameters and curing conditions to calculate the strength and modulus prediction results.

[0036] The result output module is used to output the prediction results.

[0037] The advantages of the present invention over the prior art are as follows. By adopting the method and system for predicting the strength and modulus of silt-solidified soil based on the physics-informed neural network, the data-driven characteristics of the traditional neural network are combined with the consolidation equation and strength growth equation describing the physical process of silt-solidified soil, effectively overcoming the problems of the traditional prediction method being overly dependent on empirical formulas, requiring a large amount of data, and having poor generalization performance. The present invention significantly improves the prediction accuracy and generalization ability of the model through physical constraints. Even when the experimental data is relatively scarce or the prediction conditions exceed the experimental range, it can still obtain prediction results that conform to the actual laws of the soil body. In addition, the present invention further adopts a multi-stage training optimization method. According to the stage characteristics of the rapid growth in the early stage, stable growth in the middle stage, and gradually stable strength in the late stage of silt-solidified soil, the model training strategy is dynamically optimized stage by stage, significantly improving the accuracy of the prediction model in the long-term prediction scenario, effectively overcoming the problem of poor long-term prediction effect of the traditional data-driven model, and significantly enhancing the reliability and applicability in engineering practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is the overall flowchart of the method of the present invention;

[0039] Figure 2 is a comparison chart of the compressive strength prediction of silt-solidified soil by the PINN model of the present invention and the traditional BP neural network;

[0040] Figure 3 is a comparison chart of the relative error of the PINN model of the present invention and the traditional BP neural network in predicting the strength and deformation modulus of silt-solidified soil;

[0041] Figure 4 is a comparison of the extrapolation prediction charts of the PINN model of the present invention and the traditional BP neural network when the curing age reaches 90 days. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following describes the specific embodiments of the present invention with reference to the drawings.

[0043] The present invention provides a method and system for predicting the strength and modulus of the full mix ratio of silt-solidified soil based on the physics-informed neural network (i.e., PINN, Physics-Informed Neural Network), which overcomes the deficiencies of the traditional method by integrating physical constraints into the neural network model.

[0044] As Figure 1 shown, the overall process of the method of the present invention is as follows:

[0045] Obtain relevant input data of silt-solidified soil;

[0046] Construct a physics-informed neural network model and embed the physical equations describing the mechanical behavior of silt-solidified soil as constraints into the neural network model;

[0047] Use the input data to train the physics-informed neural network model to gradually converge the model parameters. During the training process, comprehensively consider the experimental data error and the physical equation residual;

[0048] Use the trained and converged physics-informed neural network model to predict the strength and modulus of silt-solidified soil.

[0049] More specific embodiments are as follows:

[0050] First, collect the key parameters affecting the strength and modulus of silt-solidified soil, including the initial water content, density, particle size composition of silt, the type and dosage of solidifying agent, and the solidification and curing conditions such as age, environmental temperature and humidity, etc.

[0051] In some embodiments, multi-source data can be used for training, including on-site sensor monitoring data and laboratory test data; among them, the on-site sensor monitoring data includes the measured data of soft soil foundation settlement, pore water pressure or water content; the laboratory test data includes the data obtained from the unconfined compressive strength and deformation modulus tests on the silt-solidified soil samples taken from the site.

[0052] Then, construct a physics-informed neural network model, that is, fuse the physical equations reflecting the mechanical behavior of silt-solidified soil as constraints in the deep neural network architecture.

[0053] Specifically, the neural network can adopt a fully connected feedforward network structure, for example, including several hidden layers to capture the non-linear relationship between the input parameters and the output mechanical indexes. To ensure that the model prediction conforms to the engineering laws, physical mechanism constraints such as soft soil consolidation and strength development are introduced during the network training process: on the one hand, the control equations of consolidation theory are used as constraint conditions to ensure that the mechanical response of the soil predicted by the model conforms to the trend of the consolidation process (for example, the law that the strength increases and the void ratio decreases with the increase of the degree of consolidation);

[0054] On the other hand, introduce the empirical model of the strength of solidified soil increasing with age (such as the curve relationship that the strength gradually increases with time and tends to be stable) as a supplementary physical constraint. These physical constraints are added to the network training through the loss function, so that the network not only has to fit the existing data, but also needs to make the prediction results physically reasonable.

[0055] A commonly used physical formula is the Terzaghi one-dimensional consolidation equation, which describes the change of pore water pressure in the soil during the consolidation process. The expression is:

[0056] ;

[0057] where is the excess pore water pressure, is the consolidation coefficient, t is the time, and z is the depth coordinate.

[0058] This equation can reflect the physical laws of the gradual drainage of water and the compression of soil during consolidation. When embedding it into the PINN, the partial derivatives can be calculated through the automatic differentiation of the network output. For example, and , and then the residual term is constructed. This residual will be added to the loss function as part of the physical constraint, prompting the network prediction results to not only fit the data but also satisfy the consolidation equation as much as possible.

[0059] In addition to the consolidation equation, the empirical formula for the strength increasing with time is also a physical constraint worth introducing. For solidified silt soil, the compressive strength usually gradually increases with the curing age and can be described by the following expression:

[0060] ;

[0061] where is the compressive strength of the solidified soil at the curing age t, is the strength stability limit value, and k is the strength growth rate constant.

[0062] This formula captures the trend of the strength tending to be stable during the curing process. The method of embedding this formula into the PINN is to compare the strength value predicted by the network with the value calculated by the formula, calculate the difference between the two as the residual, and incorporate it into the loss function. In this way, the model will consider both the experimental data and the physical laws of strength growth during training, thereby improving the rationality of the prediction. If the type of curing agent or environmental conditions are different, parameters in the formula such as k can also be set as learnable variables and optimized automatically by the network according to the data.

[0063] Another physical formula that can be introduced is the description based on Biot's consolidation theory, which is more complex than Terzaghi's equation and is applicable to cases considering the nonlinear deformation of soil or the three-dimensional stress state. Biot's theory involves the coupled equations of displacement and pore water pressure. The embedding method of this theory is similar to that of Terzaghi's equation, calculating the relevant derivatives through automatic differentiation and adding the equation residual to the loss function. However, due to the high computational complexity, it is usually necessary to decide whether to adopt this more refined model according to the specific requirements of the problem.

[0064] The embedding of these physical formulas is usually achieved through the loss function mentioned later. In actual operation, the selection and embedding method of physical formulas need to be flexibly adjusted according to the specific characteristics of the silt-solidified soil. If the consolidation process shows obvious nonlinearity, or the chemical reaction of the curing agent significantly affects the strength growth, it may be necessary to introduce more complex formulas or modify the parameters of the existing formulas. By setting some parameters to be learnable, PINN can automatically adapt to these changes during training. This flexibility combined with physical constraints makes the model accurate and reliable when predicting strength and modulus, and at the same time, it can evaluate the degree of compliance with physical laws by analyzing the residual terms.

[0065] During the model training stage, the above-collected experimental data is used for supervised learning to construct a composite loss function containing physical constraints.

[0066] For example, the loss function can consist of two parts:

[0067] One is the data error loss, which is used to measure the difference between the strength and modulus output by the network and the actual experimental values;

[0068] The other is the physical equation loss, which is used to measure the degree to which the network output violates the relevant physical equations (such as the consolidation equation, strength-age model). By assigning a certain weight coefficient to the physical equation loss, the roles of physical constraints and data fitting in the training process can be balanced.

[0069] The formula is as follows:

[0070] ;

[0071] L is the total loss function; L d is the data loss term, which is the mean square error between the predicted value and the actual measured value; L p is the physical equation residual term, which is used to measure the degree to which the model prediction violates the above physical equations;

[0072] is the weight coefficient, which is used to balance the two losses. The value range of usually is between 0 and 1. When the data is sufficient and reliable, can take a smaller value (such as 0.01 to 0.1) to give priority to fitting the data; when the data is sparse or noisy, should take a larger value (such as 0.5 to 1) to strengthen the role of physical constraints.

[0073] In some embodiments, considering the obvious stage law in the development of the strength and modulus of the silt-solidified soil, it is particularly manifested as:

[0074] The strength grows rapidly in the early stage (0 - 14 days);

[0075] The growth rate of strength slows down in the middle stage (14 - 28 days) and gradually tends to be stable;

[0076] The strength grows slowly in the late stage (after 28 days) and gradually tends to the stable limit value.

[0077] The neural network or data - driven model with the traditional single - stage training method in the prior art cannot fully utilize and reflect this phased change characteristic, resulting in poor model generalization, especially the obvious decline in the accuracy of long - age prediction.

[0078] The present invention can propose a multi - stage PINN training optimization method for this special property of silt solidified soil, which specifically includes:

[0079] The first stage (rapid training stage of the initial model): Use the test data of the early age period (such as 0 - 14 days) to train the PINN model preferentially, establish a preliminary model, so that the model can accurately capture the rapid strength growth law; Since the data is relatively sufficient and accurate, we hope that the model can fit the actual data characteristics of the rapid early - stage strength growth as soon as possible. Therefore, the weight of the physical constraint term in this stage is relatively small, and more emphasis is placed on the data fitting error term, allowing the model to learn features from the data more freely;

[0080] The second stage (strengthened training in the stable stage): When the model converges to a certain extent, add the data of the middle age period (such as 14 - 28 days), and adjust the weight of the physical constraint term in the loss function to emphasize the progressive law of strength growth and gradually converge to the trend of the stable stage; Since the model needs to reflect the physical law that the strength of silt solidified soil gradually grows and tends to be stable, appropriately increase the weight of the physical constraint term at this time, so that the model more strictly follows the trend of stable strength growth, rather than relying entirely on limited data fitting, thus making the model more in line with physical reality;

[0081] The third stage (long - age generalization training): In the subsequent stage, gradually add a small amount of test data of the late age period (28 - 90 days or even longer age periods), and at the same time further strengthen the constraint of the physical equation residual term, so that the generalization ability of the PINN model is significantly improved; Because the late - stage test data is often sparser, there is a greater risk of generalization if the model directly relies on data fitting. At this time, it is necessary to further increase the weight of the physical constraint term, even significantly higher than the previous two stages. This strengthened physical constraint forces the model to more strictly follow the theoretical trend that the long - term strength of solidified soil tends to be stable. Even if it is beyond the range of the original training data, the model can give reasonable predictions and significantly improve the generalization performance.

[0082] During the training process, the gradient descent optimization algorithm (such as the combination of Adam optimizer and L-BFGS, etc.) is used to iteratively adjust the network parameters to gradually reduce the total loss. To improve the training effect, hyperparameters such as the learning rate, batch size, and number of training epochs can be adjusted, and an early stopping strategy can be adopted to avoid overfitting. When the loss function converges to a predetermined threshold, the training ends, and a physically calibrated neural network model is obtained.

[0083] After the PINN model obtained by the method of the present invention is trained, the strength and modulus can be predicted for new input conditions.

[0084] The user only needs to input the initial state parameters of the silt-solidified soil and the predetermined curing conditions (for example, hoping to predict the strength at 28 days of age), and the model can output the corresponding predicted values of strength and deformation modulus. Since the model integrates physical laws, even if the input parameters exceed the range of the original data, the model can still give results that conform to engineering common sense, improving the credibility of the prediction.

[0085] The method and system of the present invention can be implemented on a computer in software form, or integrated into a on-site monitoring system to predict the performance of foundation soil in real time. Its application range is wide, and it is applicable to various foundation engineering such as soft soil foundation treatment, road subgrade reinforcement, and marine reclamation soil reinforcement, providing reliable data support for engineering design.

[0086] To verify the effectiveness of the method of the present invention, experimental examples were carried out.

[0087] The silt soil samples of a coastal engineering project were selected for the experiment. The basic physical properties were a water content of about 60% and a natural density of 1.5 g / cm³. Cement was added to the silt as a curing agent, and the dosages were 5%, 10%, and 15% by mass ratio respectively. Standard specimens were prepared and cured indoors for different ages (7 days, 14 days, 28 days).

[0088] Unconfined compressive strength tests were carried out on the cured soil specimens at each age to obtain the compressive strength data, and the corresponding compression deformation modulus was calculated through the stress-strain curve. A number of groups of sample data were obtained (for example, the strength range was 0.5 - 2.0 MPa, and the modulus range was 50 - 200 MPa).

[0089] 80% of the data was used for training, and 20% of the data was used for verification and testing. A PINN model was constructed for the above data prediction: the network structure was an input layer - 3 hidden layers - an output layer. The input included parameters such as the curing agent dosage, initial water content, and age, and the output was the corresponding compressive strength and deformation modulus.

[0090] The ReLU activation function was used for the hidden layers, and the number of neurons in each layer was 32, 16, and 8 in sequence.

[0091] In terms of physical constraints, the relationship between the void ratio and time evolution in one-dimensional consolidation theory (such as Terzaghi's one-dimensional consolidation equation) and the empirical formula for strength growth are introduced as constraints.

[0092] In the loss function, the mean square error between the predicted and measured values of strength and modulus is used for the data error term, and the physical constraint term is defined according to whether the strength change with age output by the network conforms to the selected empirical model. The weight coefficient of the physical constraint term is set to 0.1, and the Adam optimization algorithm is used to train for 10,000 iterations with an initial learning rate of 0.001. During the training process, the learning rate is appropriately adjusted according to the performance of the validation set to accelerate convergence.

[0093] After training is completed, the model is applied to the test set data for prediction.

[0094] The results show that the predicted values of the PINN model established by the method of the present invention are in good agreement with the measured values: the average relative error of the predicted value of the compressive strength is about 5%, and the average relative error of the predicted value of the deformation modulus is about 8%. In contrast, when using a traditional BP neural network model (without introducing physical constraints) with the same training data for prediction, the average relative errors of strength and modulus reach 12% and 15% respectively. It can be seen that the neural network introducing physical information significantly improves the prediction accuracy. More importantly, when extrapolating and predicting situations beyond the training data range (such as predicting the strength after 90 days of curing), the model obtained by the method of the present invention still gives a reasonable strength growth trend and value; while the prediction results of the traditional neural network model fluctuate greatly and the physical rationality is poor when lacking corresponding data support. More specifically:

[0095] Figure 2 shows the comparison of the compressive strength prediction of the PINN model and the traditional BP neural network (Back Propagation Neural Network). It can be clearly seen from the figure that the prediction results of the PINN model are closer to the actual experimental data, indicating that the accuracy of the PINN model in predicting strength is significantly higher than that of the traditional BP neural network.

[0096] Figure 3 compares the relative errors of the PINN model and the traditional BP neural network in predicting the strength and deformation modulus of silt-solidified soil. It can be clearly seen from the figure that the strength prediction error of the PINN model is about 5%, and the modulus prediction error is about 8%, both of which are significantly lower than the prediction errors of the BP model, reflecting the advantage of the PINN model in prediction accuracy.

[0097] Figure 4 is the extrapolation prediction graph when the curing age reaches 90 days, showing the prediction performance of the PINN model and the traditional BP neural network model beyond the training data range. The PINN model gives a more reasonable and robust trend extrapolation prediction, while the BP model shows obvious unreasonable prediction fluctuations at long ages (such as 90 days), reflecting the stronger generalization performance of the PINN model due to the introduction of physical constraints.

[0098] It can be seen that the PINN-based prediction method for the strength and modulus of silt-solidified soil of the present invention has higher accuracy and robustness compared with the traditional method and will be more reliable in engineering applications. This experimental example verifies the effectiveness of the present invention, indicating that integrating physical constraints into the neural network can make full use of existing test data and follow the laws of soil mechanics, thereby realizing the accurate prediction of the mechanical properties of silt-solidified soil.

[0099] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A prediction method for the strength and modulus of the full mix proportion of silt solidified soil, characterized in that Including the following steps: Obtain relevant input data of the silt solidified soil; Construct a physics-informed neural network model, and embed the physical equations describing the mechanical behavior of the silt solidified soil as constraints into the neural network model; Use the input data to train the physics-informed neural network model to gradually converge the model parameters. During the training process, comprehensively consider the experimental data error and the physical equation residual; Use the trained and converged physics-informed neural network model to predict the strength and modulus of the silt solidified soil; Introduce physical mechanism constraint equations related to the strength and modulus of the silt solidified soil during the training process of the physics-informed neural network model, including the control equation of soft soil consolidation and the empirical equation of the strength growth of the solidified soil; The control equation of soft soil consolidation includes Terzaghi's one-dimensional consolidation equation; The empirical equation of the strength growth of the solidified soil is as follows: ; among them, is the compressive strength of the solidified soil at the curing age t, is the strength stability limit value, and k is the strength growth rate constant.

2. The method according to claim 1, characterized in that: The relevant input data includes the initial physical and chemical parameters of the silt solidified soil and the curing condition data; The initial physical and chemical parameters include water content, density, particle composition, and curing agent dosage; The curing condition data includes age, environmental temperature, and humidity.

3. The method according to claim 1, wherein: The physics-informed neural network model is a multi-layer deep neural network, including an input layer, an output layer, and at least two hidden layers; Among them, the input layer receives the relevant input data of the silt solidified soil, and the output layer outputs the predicted values of the strength and modulus of the silt solidified soil.

4. The method according to claim 1, wherein The method further includes constructing a loss function containing physical constraints to train the physics-informed neural network model; The loss function is composed of a data fitting error term and a physical equation residual term; Among them, the data fitting error term is used to measure the deviation between the predicted strength and modulus of the model and the measured values; the physical equation residual term is used to measure the degree of violation of the model prediction results against the relevant physical equations; Optimize the neural network parameters by minimizing the loss function.

5. The method according to claim 1, wherein The physics-informed neural network model adopts a multi-stage training optimization method: divide the training process into multiple stages, including the early stage of rapid strength growth, the middle stage of stable strength growth, and the late stage of stable strength; dynamically adjust the weight of the physical equation residual term in the loss function at different training stages.

6. The method according to claim 5, characterized in that: Use an iterative optimization algorithm based on gradient descent to train the physics-informed neural network model. Set a weight coefficient for the physical equation residual term during the training process to balance its contribution with the data error term. Terminate the training when the loss function converges to a preset threshold or reaches the maximum number of training iterations.

7. The method according to claim 1, characterized in that: The method uses multi-source data for training, including on-site sensor monitoring data and laboratory test data; Among them, the on-site sensor monitoring data includes the measured data of soft soil foundation settlement, pore water pressure, or water content; The laboratory test data includes the data obtained from the unconfined compressive strength and deformation modulus tests on the silt solidified soil samples taken from the site.

8. The method according to claim 1, wherein: The method is used in any of the following scenarios: Foundation treatment engineering, predicting the strength and modulus of the silt solidified soil in the soft soil foundation after reinforcement; Soft soil foundation reinforcement in road construction, predicting the strength and modulus of the silt solidified soil in the subgrade; Treatment of dredger fill silt foundation in ocean engineering, predicting the strength and modulus of dredger fill silt after solidification treatment.

9. A prediction system for the strength and modulus of the full mix proportion of silt-solidified soil, characterized in that, It includes a data acquisition module, a physics-informed neural network calculation module, and a result output module; Among them, the data acquisition module is used to collect the initial parameters of the silt solidified soil and the data of the curing conditions, The physics-informed neural network calculation module is used to calculate the strength and modulus prediction results by calling the trained physics-informed neural network model according to the initial parameters and the curing conditions; The result output module is used to output the prediction results; During the training process of the physics-informed neural network model, physical mechanism constraint equations related to the strength and modulus of the silt solidified soil are introduced, including the control equation of soft soil consolidation and the empirical equation of the strength growth of the solidified soil; The control equation of the soft soil consolidation includes the Terzaghi one-dimensional consolidation equation; The empirical equation of the strength growth of the solidified soil is as follows: ; wherein, is the compressive strength of the solidified soil at the curing age t, is the strength stability limit value, and k is the strength growth rate constant.

Images