Deep learning-based high-efficiency simulation method and system for grouting at different temperatures

Abstract

The present application relates to grouting simulation technical field, particularly to different temperature grouting efficient simulation method and system based on deep learning, wherein the method comprises: constructing the relationship of the change of rock-soil medium and grouting material under different temperature conditions; the relationship of the change of rock-soil medium and grouting material under different temperature conditions comprises: the relationship of temperature and porosity, the relationship of temperature and stress-strain, and the relationship of temperature and slurry viscosity; based on the actual stratum data of grouting area, fitting the to-be-determined parameters in the above various relationship formulas; constructing an improved recurrent neural network; training the improved recurrent neural network to obtain the trained improved recurrent neural network; inputting the temperature of the to-be-predicted grouting area into the trained improved recurrent neural network to obtain the slurry flow speed and slurry penetration and diffusion distance of the to-be-predicted grouting area. The efficient simulation of the whole process of grouting under different temperatures is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of grouting simulation, in particular to a different-temperature grouting efficient simulation method and system based on deep learning. BACKGROUND

[0002] At present, the construction of underground projects is in a period of rapid development, and more and more underground spaces and tunnel projects are advancing towards extremely complex areas. The complex geological sections and frequent geological disasters encountered during the construction process pose great challenges to underground engineering construction. Among them, the influence of ground temperature as one of the problems encountered in underground engineering construction in recent years, compared with the normal temperature environment, sudden geological disasters such as high-temperature water and high-temperature steam spouting in cracks often produce greater destructive effect under the influence of temperature effect, and grouting as one of the effective means to govern such sudden geological disasters often has good plugging and reinforcing effect. However, under the influence of temperature effect, the physical parameter properties of grouting material often change significantly, further affecting the plugging and reinforcing effect of grouting construction, and making it difficult to effectively treat the geological disasters in front.

[0003] The existing grouting research methods mainly include physical experiments, theoretical analysis and numerical simulation, among which grouting numerical simulation can well visualize the whole grouting process and is widely used. However, the existing research on grouting simulation considering temperature effect is not perfect, and the non-single function change of rheological properties caused by heat interaction between grouting and surrounding environment during grouting process is not considered, which makes it difficult to effectively simulate the plugging and reinforcing effect of grouting slurry penetration and diffusion considering temperature effect. SUMMARY

[0004] In order to solve the problems of the prior art, the present application provides a different-temperature grouting efficient simulation method and system based on deep learning. The present application considers the time-varying nature of the hydration reaction speed and viscosity of grouting caused by temperature change, and also considers the stress-strain change of rock-soil medium itself caused by temperature effect and the expansion and shrinkage properties caused by temperature change. In the model training, the temperature is considered as a learning parameter, the heat interaction between the grouting itself and the medium environment is considered, the mapping of similar temperature conditions at different time sequences is carried out to accelerate the simulation solving process, the corresponding relationship between the physical parameter change of grouting-rock-soil medium and grouting diffusion under different temperature conditions is established, and the efficient simulation of the whole grouting process under different temperatures is realized.

[0005] In one aspect, a different-temperature grouting efficient simulation method based on deep learning is provided, comprising:

[0006] The relationship formula of the change of the rock-soil medium and the grouting material under different temperature conditions is constructed, and the relationship formula of the change of the rock-soil medium and the grouting material under different temperature conditions includes a relationship formula of temperature and porosity, a relationship formula of temperature and stress-strain, and a relationship formula of temperature and slurry viscosity; based on the actual stratum data of the grouting area, the to-be-determined parameters in the above relationship formulas are fitted out;

[0007] An improved recurrent neural network is constructed, and the improved recurrent neural network includes a recurrent layer, the recurrent layer includes a first design layer, a second design layer and a third design layer connected in sequence, wherein the output end of the second design layer is also connected with the input end of the first design layer; the relationship formula of temperature and porosity, the relationship formula of temperature and stress-strain, and the relationship formula of temperature and slurry viscosity are set in the third design layer;

[0008] The temperature of the to-be-predicted grouting area is input into the trained improved recurrent neural network to obtain the slurry flow speed and the slurry penetration and diffusion distance of the to-be-predicted grouting area.

[0009] In another aspect, a deep learning-based different-temperature grouting efficient simulation system is provided, which includes:

[0010] The relationship formula construction module is configured to construct a relationship formula of the change of the rock-soil medium and the grouting material under different temperature conditions; the relationship formula of the change of the rock-soil medium and the grouting material under different temperature conditions includes a relationship formula of temperature and porosity, a relationship formula of temperature and stress-strain, and a relationship formula of temperature and slurry viscosity; based on the actual stratum data of the grouting area, the to-be-determined parameters in the above relationship formulas are fitted out;

[0011] The network construction module is configured to construct an improved recurrent neural network; the improved recurrent neural network includes a recurrent layer, the recurrent layer includes a first design layer, a second design layer and a third design layer connected in sequence, wherein the output end of the second design layer is also connected with the input end of the first design layer; the relationship formula of temperature and porosity, the relationship formula of temperature and stress-strain, and the relationship formula of temperature and slurry viscosity are set in the third design layer;

[0012] The output module is configured to input the temperature of the to-be-predicted grouting area into the trained improved recurrent neural network to obtain the slurry flow speed and the slurry penetration and diffusion distance of the to-be-predicted grouting area.

[0013] The above technical solution has the following advantages or beneficial effects:

[0014] A grouting parameter database under different temperature conditions is established, grouting diffusion data under different temperature conditions are collected to form a data set, and the rock-soil medium corresponding to the temperature change under the limited temperature condition and the change data of the physical parameters of the slurry, the divided three-dimensional grid and the center point data are stored.

[0015] The change of the physical parameters of the slurry and the rock-soil medium caused by the temperature effect is considered, and the temperature change of the rock-soil medium is comprehensively determined by combining the heat transfer of fluid and solid and the environmental heat transfer. In the model training, the temperature is considered as a learning parameter, the mapping of similar temperature conditions on different time sequences is considered to accelerate the simulation solving process, and the more the number of regions mapped in this process, the more the trend type of the slurry diffusion trend, which assists in determining the overall slurry diffusion effect.

[0016] At the same time, the gain degree of different parameters on the liquidity of the slurry and the actual penetration and diffusion distance is considered, the actual flow velocity and penetration and diffusion distance of the slurry are determined by nonlinear gain superposition, the viscosity properties and penetration and diffusion form of the slurry at each spatial position on different time sequences are obtained in real time, the diffusion effect of the slurry in the rock-soil medium is analyzed, the actual plugging and reinforcement of grouting is judged, and the grouting strategy is further optimized for efficient treatment of grouting disasters under different temperature conditions. BRIEF DESCRIPTION OF DRAWINGS

[0017] The drawings constituting a part of this application are used to provide a further understanding of the application, the illustrative embodiments of the application and the description thereof are used to explain the application, and do not constitute an improper limitation on the application.

[0018] Figure 1 The method flowchart of example one is shown in Figure 1.

[0019] Figure 2 The internal structure diagram of the improved recurrent neural network of example one is shown in Figure 2. DETAILED DESCRIPTION

[0020] It should be pointed out that the following detailed description is exemplary and is intended to provide further explanation of the application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as generally understood by those skilled in the art to which the application belongs.

[0021] Example one, as shown in Figure 1, provides a deep learning-based efficient grouting simulation method under different temperatures, which includes: Figure 1

[0022] ​S101: Construct a relationship between changes of the rock-soil medium and the grouting material under different temperature conditions; the relationship between changes of the rock-soil medium and the grouting material under different temperature conditions includes: a relationship between temperature and porosity, a relationship between temperature and stress-strain, and a relationship between temperature and slurry viscosity; based on actual stratum data of a grouting area, determine parameters in the above relationships;

[0023] S102: Construct an improved recurrent neural network; as shown in the figure, the improved recurrent neural network includes a recurrent layer, and the recurrent layer includes a first design layer, a second design layer and a third design layer connected in sequence, wherein an output end of the second design layer is also connected to an input end of the first design layer; the third design layer is provided with the relationship between temperature and porosity, the relationship between temperature and stress-strain, and the relationship between temperature and slurry viscosity; Figure 2

[0024] S103: Input the temperature of a grouting area to be predicted into the trained improved recurrent neural network to obtain the slurry flow speed and the slurry penetration and diffusion distance of the grouting area to be predicted.

[0025] Further, the relationship between temperature and porosity has a specific expression as follows:

[0026] φ(T) = φ0(1 + α T (T-T0));

[0027] wherein T represents temperature, φ(T) represents porosity varying with temperature, φ0 represents porosity under a reference temperature T0, α T is a volume expansion coefficient varying with temperature.

[0028] Further, the relationship between temperature and stress-strain has a specific expression as follows:

[0029]

[0030] wherein T represents temperature, σ(T) represents stress varying with temperature, E0 represents elastic modulus under a reference temperature T0, β T is an exponential temperature attenuation coefficient of elastic modulus, and ε is strain.

[0031] Further, the relationship between temperature and slurry viscosity has a specific expression as follows:

[0032] μ(T,t) = μ0(T)·e (α(T)t) +μ1(T);

[0033] ​Wherein, mu (T, t) represents the slurry viscosity function changing with temperature and time, T represents temperature, t represents time, mu0 (T) is a coefficient changing with temperature, alpha (T) is a viscosity change coefficient changing with temperature, mu1 (T) is a coefficient changing with temperature.

[0034] It should be understood that the porosity refers to the percentage of the total volume of the rock-soil body pores and natural state.

[0035] Stress refers to the internal force between the parts of the rock-soil body when it is affected by external force, self-gravity, humidity, temperature and other factors. Strain refers to the relative deformation of the local object under the action of external force and non-uniform temperature field and other factors.

[0036] Slurry viscosity refers to the rheological property of nonlinear growth over time, which is reflected in the phase transition of slurry from liquid-solid two-phase.

[0037] It should be understood that the actual stratum of the grouting area is considered first, and the original rock-soil body of the area is collected for analysis. The constant temperature curing box is used to maintain its constant temperature, and the porosity and stress-strain relationship of the medium in the original state are determined by physical and mechanical tests. The original sample analysis of different temperature regions of the stratum is selected in turn, and the porosity and stress-strain relationship data of different temperature regions are obtained by nonlinear fitting. The porosity change and stress-strain relationship curve of the rock-soil medium considering the temperature effect are fitted according to the obtained data.

[0038] The basic properties of the grouting material are collected, including the selection of the material and the type of the slurry. The selected material of the grouting includes different types of cement and water glass. The type of the slurry includes cement single slurry, cement-water glass double slurry, chemical slurry, etc. The size of the selected grouting cement particles is determined, the time-varying data of the slurry viscosity of each slurry type under different temperature conditions are obtained by rheological test, and the strength characteristics after solidification are determined by data fitting into viscosity time-varying function.

[0039] Further, the to-be-determined parameters in the above relationship are fitted based on the actual stratum data of the grouting area, and polynomial curve fitting is used to realize it.

[0040] Further, the improved recurrent neural network, the basic framework of the recurrent neural network includes an input layer, a hidden layer and an output layer. In the calculation of the hidden layer, the mutual influence on different time sequences is considered, and the value of the current time hidden layer depends not only on the input of the current time, but also on the value of the previous hidden layer.

[0041] Further, the improved recurrent neural network comprises: an input layer, a hidden layer and an output layer connected in sequence, an output end of the hidden layer is further connected with an input end of a recurrent layer, and an output end of the recurrent layer is connected with an input end of the hidden layer; in the calculation of the hidden layer of the improved recurrent neural network, a recurrent step for temperature judgment is added on each time sequence, before considering the input received by the hidden layer of the current time sequence and the value of the hidden layer of the previous time sequence, a neighbor temperature judgment of the current time sequence and the previous time sequence is added, when the judgment result is met, the hidden layer calculation of the current time sequence is directly skipped, and the temperature of the previous time sequence is directly mapped to the current time sequence through the comparison relationship of the temperatures, meanwhile, the temperature data is stored on each time sequence for the neighbor temperature judgment and mapping of the next time sequence.

[0042] Further, the first design layer comprises:

[0043] If T j,t = nβ j,k,t T j,k,t-1 , then T r,t = nT r,k,t-1 is executed.

[0044] Wherein, j represents that the neighbor different region of the k region is calculated first, T j,t represents the temperature of the different j region at t time, n represents the integral multiple relationship, β j,k,t represents the similarity coefficient under the condition that the different j region of the k region neighbor of the t time sequence is in the n multiple relationship with the current calculation region; T j,k,t-1 represents the temperature of the different j region at t-1 time, T r,k,t-1 represents the temperature value of the calculation region at the previous time step, T r,t is the temperature data used in the actual calculation process, k represents the k region which is the neighbor of the current calculation region of the previous time sequence, β j,k is the similarity coefficient under the condition that the different j region of the k region neighbor is in the n multiple relationship with the current calculation region, by combining the time representation of the trigonometric function, the numerical limit close to 1 under different recurrent iteration steps is set, and the own cycle and the discontinuous period are set, so as to approach the actual calculation temperature change value.

[0045] It should be understood that the temperature mapping of the adjacent sequence is considered under the current time sequence: the temperature mapping relationship under the adjacent time sequence is added in the self-loop part of the hidden layer, the adjacent area of the target area is searched before the target area temperature change solving of the current time step, and the temperature condition of the region of the next time sequence of the current time sequence is similar to the temperature condition of the surrounding region in the adjacent time dimension. The calculated temperature condition of the previous time sequence is directly mapped to accelerate the calculation and prediction of the temperature condition of each region at different time steps.

[0046] Further, the second design layer comprises:

[0047]

[0048] wherein T i,t is the temperature change after the slurry and the region produce heat interaction, T K,t is the temperature data in the surrounding three-dimensional direction, K represents the current time sequence of the present calculation region surrounding calculation region, N represents the number of the current time sequence of the present calculation region surrounding calculation region, ε K,t is the temperature difference reduction coefficient, which is smaller in the direction of the grouting slurry flow, and larger in the other two directions, T r,t-1 represents the temperature data of the region at t-1 time. The temperature difference reduction coefficient refers to the temperature attenuation coefficient existing in the environmental heat interaction, which can be set as different constants according to the actual situation.

[0049] It should be understood that the temperature effect of the current time sequence: the temperature data of different grids in the first time sequence and the time sequence calculation without mapping relationship are considered at the same time under the influence of temperature effect, that is, the fluid-solid heat transfer and environmental heat interaction in the grouting process under the influence of temperature effect, that is, the grouting slurry will interact with the surrounding rock-soil medium to produce temperature change during the process of penetration and diffusion, and the surrounding rock-soil medium will also be affected by the surrounding environment to produce temperature change. In the process of temperature change caused by heat exchange, the viscosity of the slurry first follows the time-varying function relationship corresponding to the initial temperature in the database, and with the passage of time, the temperature change of the slurry itself leads to different viscosity time-varying properties at each time sequence, and the viscosity time-varying properties of the slurry under different temperature conditions at each time sequence are obtained.

[0050] For the change of the temperature of the slurry, the fluid-solid coupling heat transfer method is used for calculation, and the calculated temperature change is the change of the temperature of the slurry itself.

[0051] For the rock-soil medium of the region, fluid-solid coupling heat transfer method is adopted in combination with heat exchange with the surrounding environment. The surrounding environment is considered to transfer heat to the region, the direction of grouting slurry flow is taken as the gradient descending direction of temperature change, different three-dimensional space points and temperature data are taken as basic data, based on the temperature value of slurry and current grid region after heat exchange, the temperature of the rock-soil medium of the region still needs to add the temperature difference multiplied by the coefficient value with the surrounding environment, and the actual temperature value of the rock-soil medium of the region is updated to the next time sequence.

[0052] Further, the third design layer comprises:

[0053] Y(l p ) t =W q,t α q,t m q,t -W q,t-1 α q,t-1 m q,t-1 +b(T r,t -T r,t-1 )

[0054] Wherein, α q,t represents the vector gain value of m q to l p at t time sequence, W q,t represents the weight coefficient of different vector gain values at t time sequence; m q,t represents the physical parameter value of slurry and rock-soil medium at t time sequence, including the viscosity μ(T r,t ,t) of slurry, the porosity φ(T r,t ) t , stress and strain σ(T r,t ) t of rock-soil medium; T r,t-1 represents the temperature data of the region at t-1 time sequence; α q,t-1 represents the vector gain value of m q to l p at t-1 time sequence, W q,t-1 represents the weight coefficient of different vector gain values at t-1 time sequence; m q,t-1 represents the physical parameter value of slurry and rock-soil medium at t-1 time sequence, including the viscosity μ(T r,t-1 ,t-1) of slurry, the porosity φ(T r,t-1 ) t-1 , stress and strain σ(T r,t-1 ) t-1 of rock-soil medium; T r,t-1 represents the temperature data of the region at t-1 time sequence; Y represents the output data of each time sequence, l prepresents the vector growth of the slurry flow velocity and the penetration diffusion distance under the current time sequence condition, m q represents the normalized physical parameter value of the slurry and the geotechnical medium under different temperature conditions, a q represents the vector gain value of the slurry under different m q represents the vector gain value of the slurry under different m p represents the vector gain value of the slurry under different m q The physical parameters include: the viscosity of the slurry, the porosity of the geotechnical medium, and the stress and strain.

[0055] It should be understood that the slurry diffusion of different time sequences: after updating the temperature data of each time sequence, the slurry diffusion data is updated, the gain degree of each parameter of the slurry and the geotechnical medium to the flowability of the slurry and the actual penetration diffusion distance is considered, the actual flow velocity and the penetration diffusion distance of the slurry are determined by the nonlinear gain superposition value, as the initial data of the spatio-temporal distribution of the slurry in the next time sequence.

[0056] Further, the S103: training the improved recurrent neural network, comprising:

[0057] A first training set is constructed; the first training set is the known slurry flow velocity and slurry penetration diffusion distance under different temperatures; the first training set is input into the improved recurrent neural network, the input value of the improved recurrent neural network is temperature, and the output value is the slurry flow velocity and the slurry penetration diffusion distance, which is trained, and the training is stopped when the loss function value of the recurrent neural network no longer decreases, to obtain the trained recurrent neural network, and the loss function adopts the mean square error loss function.

[0058] It should be understood that the slurry flow velocity refers to the flow velocity of the slurry in the injected medium at each time during the grouting process; and the slurry penetration diffusion distance refers to the penetration diffusion distance of the slurry in the injected medium at each time during the grouting process.

[0059] During the training of the model, the predicted data calculated by the contrast gradient transmission are compared with the true values at each time set, and the gain value a q and the weight value corresponding to different temperature conditions.

[0060] In each training period, the data are divided into multiple batches for training in turn. The predicted values of the model are calculated by using the forward propagation, the loss values are calculated in the loss function, and then the gain values and the weight values of the model are updated by the backward propagation and the optimizer until the loss value is within the threshold range, and the model training is completed.

[0061] Embodiment two

[0062] The embodiment provides a deep learning-based efficient grouting simulation system under different temperatures, which comprises:

[0063] a relationship construction module configured to construct relationships of changes of the rock-soil medium and the grouting material under different temperature conditions; the relationships of changes of the rock-soil medium and the grouting material under different temperature conditions include relationships of temperature and porosity, relationships of temperature and stress-strain, and relationships of temperature and slurry viscosity; based on actual stratum data of the grouting area, to-be-determined parameters in the above relationships are fitted;

[0064] a network construction module configured to construct an improved recurrent neural network; the improved recurrent neural network includes a recurrent layer, the recurrent layer includes a first design layer, a second design layer and a third design layer connected in sequence, wherein an output end of the second design layer is further connected with an input end of the first design layer; the relationships of temperature and porosity, relationships of temperature and stress-strain, and relationships of temperature and slurry viscosity are set in the third design layer;

[0065] an output module configured to input the temperature of the to-be-predicted grouting area into the trained improved recurrent neural network to obtain slurry flow speed and slurry penetration and diffusion distance of the to-be-predicted grouting area.

[0066] The above merely describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A high-efficiency simulation method for grouting at different temperatures based on deep learning, characterized by: include: The relationships between soil and rock media and grouting materials under different temperature conditions are constructed, including: the relationship between temperature and porosity, the relationship between temperature and stress-strain, and the relationship between temperature and grout viscosity; based on actual geological data of the grouting area, the parameters to be determined in each of the above relationships are fitted. An improved recurrent neural network is constructed. The improved recurrent neural network includes: a recurrent layer, which comprises: a first design layer, a second design layer, and a third design layer connected sequentially, wherein the output of the second design layer is also connected to the input of the first design layer; the third design layer contains the relationships between temperature and porosity, temperature and stress-strain, and temperature and slurry viscosity; the improved recurrent neural network also includes: an input layer, a hidden layer, and an output layer connected sequentially, wherein the output of the hidden layer is also connected to the input of the recurrent layer, and the output of the recurrent layer is connected to the input of the hidden layer; in the hidden layer calculation, considering the mutual influence at different time series, the value of the hidden layer at the current moment depends not only on the input at the current moment but also on the value of the previous hidden layer; The first design layer includes: if If established, then execute. ;in, This indicates that the calculation of the area surrounding the previous time series should be performed first. The surrounding different areas, Indicates in Different times The temperature of the area, Indicates integer multiples. This indicates that at time t, the series... Different areas around the region Region and current calculation region Similarity coefficient under the condition of multiple relationship; Indicates in Different times The temperature of the area, This represents the temperature value of the calculation region at the previous time step. These are the temperature data used in the actual calculation process; The second design layer includes: ;in, This refers to the temperature change resulting from the heat interaction between the slurry and the area. Temperature data in the surrounding three dimensions. This represents the surrounding computing region of the current time series in the current computing region. This represents the number of surrounding computation regions of the current time series in the current computation region. This is the temperature difference reduction factor. This value is smaller in the direction of grout flow and larger in the other two directions. express Temperature data for that area at any given time; The third design layer includes: ;in, express Different time series right The vector gain value, The weighting coefficients represent the different vector gain values ​​of the time series t; express Time series values ​​of various physical parameters of slurry and soil media; express Time series temperature data for this region; This represents the output data for each time series. This represents the normalized physical parameter values ​​of the slurry and soil media under different temperature conditions. This represents the vector growth of the slurry flow velocity and the penetration diffusion distance under the current time series conditions. This indicates the bias under different time series conditions, which varies accordingly with different temperatures; The temperature of the grouting area to be predicted is input into the trained improved recurrent neural network to obtain the grout flow velocity and grout penetration diffusion distance of the grouting area to be predicted.

2. The efficient simulation method for grouting at different temperatures based on deep learning as described in claim 1, characterized in that, The specific expression for the relationship between temperature and porosity is as follows: ； in, Indicates temperature. Porosity as a function of temperature Reference temperature Porosity under the conditions is the coefficient of volumetric expansion as a function of temperature.

3. The efficient simulation method for grouting at different temperatures based on deep learning as described in claim 1, characterized in that, The specific expression for the relationship between temperature and stress-strain is as follows: ； in, Indicates temperature. This represents the stress value as a function of temperature. Reference temperature elastic modulus under the condition, The elastic modulus is the exponential temperature decay coefficient. In response to the situation.

4. The efficient simulation method for grouting at different temperatures based on deep learning as described in claim 1, characterized in that, The specific expression for the relationship between temperature and slurry viscosity is as follows: ； in, The viscosity of the slurry as a function of temperature and time. T Indicates temperature. t Indicates time, The coefficient is the coefficient that varies with temperature. The viscosity coefficient varies with temperature. This is a coefficient that varies with temperature.

5. The efficient simulation method for grouting at different temperatures based on deep learning as described in claim 1, characterized in that, The improved recurrent neural network adds a loop step for temperature judgment in each time series during its hidden layer calculation. Before considering the input received by the hidden layer of the current time series and the hidden layer value of the previous time series, it adds a neighboring temperature judgment between the current time series and the previous time series. When the judgment effect is met, the hidden layer calculation of the current time series is skipped directly, and the temperature of the previous time series is directly mapped to the current time series by comparing it with the temperature of the previous time series. At the same time, temperature data is stored in each time series for neighboring temperature judgment and mapping in the next time series.

6. The efficient simulation method for grouting at different temperatures based on deep learning as described in claim 1, characterized in that, The third design layer, express Time-series physical parameters of the slurry and soil media, including the viscosity of the slurry. Porosity of soil and rock media Stress and strain .

7. The efficient simulation method for grouting at different temperatures based on deep learning as described in claim 1, characterized in that, The improved recurrent neural network after training includes the following training process: Construct a first training set; the first training set consists of the known slurry flow velocity and slurry penetration and diffusion distance at different temperatures; input the first training set into an improved recurrent neural network, the input value of which is temperature, and the output values ​​are slurry flow velocity and slurry penetration and diffusion distance, and train it. When the loss function value of the recurrent neural network no longer decreases, the training is stopped, and the trained recurrent neural network is obtained. The loss function adopts the mean squared error loss function.

8. A high-efficiency simulation system for grouting at different temperatures based on deep learning, characterized by: The method for efficient simulation of grouting at different temperatures based on deep learning, as described in any one of claims 1-7, includes: The relational construction module is configured to: construct relational expressions between the soil and rock medium and the grouting material under different temperature conditions; the relational expressions between the soil and rock medium and the grouting material under different temperature conditions include: the relationship between temperature and porosity, the relationship between temperature and stress-strain, and the relationship between temperature and grout viscosity; and fit the parameters to be determined in each of the above relational expressions based on the actual stratum data of the grouting area. A network construction module is configured to: construct an improved recurrent neural network; the improved recurrent neural network includes: a recurrent layer, the recurrent layer including: a first design layer, a second design layer and a third design layer connected in sequence, wherein the output of the second design layer is also connected to the input of the first design layer; the third design layer contains the relationship between temperature and porosity, the relationship between temperature and stress-strain, and the relationship between temperature and slurry viscosity; The output module is configured to input the temperature of the grouting area to be predicted into the trained improved recurrent neural network to obtain the grout flow velocity and grout penetration diffusion distance of the grouting area to be predicted.

Images