A method and device for analyzing the robustness of a foundation pit

Abstract

The present invention discloses the technical field of foundation pit design analysis, and specifically discloses a method and device for analyzing the robustness of a foundation pit. The present invention uses a convolutional neural network surrogate model that can consider spatial variability as the basis for robustness evaluation, which can fully reflect the randomness of soil properties, and thus can predict accurate and real physical and mechanical design response values, further ensuring the accuracy and reliability of the analysis results. At the same time, through weight fusion calculation, the relationship between various design response values is comprehensively considered, providing a scientific evaluation method for the robustness evaluation of the foundation pit, providing effective technical support for engineering practice, providing a reliable decision-making basis for the design and construction of the foundation pit, and further ensuring the safety and stability of the foundation pit project from the design perspective.

Description

Technical Field

[0001] The present invention relates to the technical field of foundation pit design and analysis, and particularly to a method and device for analyzing the robustness of a foundation pit. Background Art

[0002] In existing foundation pit designs, a deterministic design method is usually adopted, which means that according to the determined geotechnical mechanical parameters (mean value assumption), a combination of geometric parameters that meet the safety requirements of anti-heave, anti-overturning, and lateral displacement deformation of the support structure is obtained. Due to the influence of the variability of soil parameters, on the basis of deterministic design, the analysis and design of the robustness of the foundation pit should also be considered, that is, when the uncertainty factors change within a given range, the stability of the foundation pit support system must be ensured not to be affected.

[0003] Due to the action of load history conditions, physical and chemical processes, etc. on natural soil, the parameters such as the strength and modulus of the soil show spatial variability. In traditional foundation pit robustness analysis, the soil parameters (physical and mechanical property values) at different positions in the soil layer cannot be accurately determined, and only the local mean value can be used to represent the shear strength of the entire soil layer, thus lacking the consideration of the influence of the variability of soil parameters on the analysis results. Summary of the Invention

[0004] In order to solve the problems that the spatial variability of soil parameters is not considered in the existing foundation pit robustness analysis, resulting in the fact that the stability analysis of the foundation pit cannot fully utilize various parameters of the soil to obtain reliable results, etc., the present invention provides a method and device for analyzing the robustness of a foundation pit.

[0005] In order to achieve the above invention purpose, the present invention provides the following technical solutions:

[0006] In the first aspect, the present invention provides a method for analyzing the robustness of a foundation pit, and the method includes:

[0007] S1. After performing random field simulation on the soil property values of the target foundation pit, map them into a numerical model to obtain a number of first foundation pit samples; the first foundation pit samples include the distribution image of the soil property values, as well as the mean value and variance of the soil property values;

[0008] S2. Calculate the mean value of the design response corresponding to each first foundation pit sample through the numerical model, and label the first foundation pit samples according to the mean value of the design response to generate a data set;

[0009] S3. Train a surrogate model according to the data set to obtain a trained surrogate model;

[0010] S4. Repeat S2 to S3 to obtain multiple trained surrogate models corresponding to the mean values of the design responses;

[0011] S5. Repeat S1 to obtain a number of second foundation pit samples, and then input them into multiple trained surrogate models to obtain multiple predicted mean values of design responses;

[0012] S6. Perform weighted fusion calculation based on the standard deviation and covariance of each predicted mean value of design responses to obtain the robustness index of the target foundation pit.

[0013] According to a specific implementation manner, in the above analysis method, the random field simulation includes:

[0014] Construct a random field by using the Monte Carlo sampling method based on the random parameters of the soil property values; wherein, the random parameters include the mean value, variance, coefficient of variation, vertical fluctuation distance, and horizontal fluctuation distance of the soil property values.

[0015] According to a specific implementation manner, in the above analysis method, the soil property values are selected from any three of the physical and mechanical properties of the soil of the target foundation pit.

[0016] According to a specific implementation manner, in the above analysis method, the distribution image of the soil property values is an RGB image generated by converting a numerical model, including:

[0017] Convert the three soil property values of each unit body in the mapped numerical model into R-channel numerical value, G-channel numerical value, and B-channel numerical value respectively;

[0018] Assign the R-channel numerical values, G-channel numerical values, and B-channel numerical values of all unit bodies to the R-channel, G-channel, and B-channel respectively to generate an RGB image.

[0019] According to a specific implementation manner, in the above analysis method, the predicted mean values of design responses include the mean value of pile body stress, the mean value of pile body displacement, the mean value of bottom heave of the foundation pit, and the mean value of surrounding settlement.

[0020] According to a specific implementation manner, in the above analysis method, the calculation formula for the standard deviation of each predicted mean value of design responses is:

[0021]

[0022] wherein, σ i is the standard deviation of the i-th predicted mean value of design responses, x ik is the i-th predicted mean value of design responses corresponding to the k-th second foundation pit sample, is the i-th predicted mean value of design responses, and M is the number of second foundation pit samples.

[0023] According to a specific implementation manner, in the above analysis method, the calculation formula for the covariance of each predicted mean value of design responses is:

[0024]

[0025] Among them, Cov(X i , X j ) is the covariance between the predicted mean value of the i-th design response and the predicted mean value of the j-th design response, x ik is the predicted mean value of the i-th design response corresponding to the k-th first random field sample, and x jk is the predicted mean value of the j-th design response corresponding to the k-th second foundation pit sample. is the predicted mean value of the i-th design response, is the predicted mean value of the j-th design response, and M is the number of second foundation pit samples.

[0026] According to a specific implementation manner, in the above analysis method, the weight fusion according to the standard deviation and covariance of each predicted mean value of the design response includes:

[0027] Performing weight allocation according to the standard deviation of each predicted mean value of the design response, and the calculation formula is:

[0028]

[0029] Among them, ω i is the fusion weight of the predicted mean value of the i-th design response, σ i , σ j are the standard deviations of the predicted mean values of the i-th and j-th design responses respectively, and n is the number of types of predicted mean values of the design response;

[0030] Performing fusion calculation according to the weights of each predicted mean value of the design response, and the calculation formula is:

[0031]

[0032] Among them, σ p is the robustness index of the target foundation pit, and Cov(X i , X j ) is the covariance between the predicted mean value of the i-th design response and the predicted mean value of the j-th design response.

[0033] According to a specific implementation manner, in the above analysis method, the surrogate model adopts a convolutional neural network.

[0034] In a second aspect, the present invention provides an electronic device, including at least one processor, and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute a foundation pit robustness analysis method as described in any one of the above.

[0035] Advantages of the present invention compared with the prior art:

[0036] Based on the technical solution provided by the present invention, a convolutional neural network surrogate model that can consider spatial variability is used as the basis for robustness evaluation. Based on the random field simulation, the randomness of soil properties can be fully reflected, and then the accurate and real mean value of the physical and mechanical design responses can be predicted, ensuring the effectiveness of the analysis results. In addition, the way of adding a surrogate model in the present invention replaces the traditional numerical simulation analysis to obtain the mean values of various design responses of the foundation pit, significantly improving the analysis efficiency, and enabling each foundation pit project to have an applicable surrogate model, further improving the accuracy and reliability of the predicted mean value of the design response obtained. At the same time, the present invention calculates the predicted mean value of each design response obtained through weight fusion, comprehensively considering the relationship between the mean values of various design responses, providing a scientific evaluation method for the robustness evaluation of the foundation pit, providing effective technical support for engineering practice, providing a reliable decision-making basis for the design and construction of the foundation pit, and further ensuring the safety and stability of the foundation pit project from the design perspective. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a flowchart of a foundation pit robustness analysis method provided by an embodiment of the present invention;

[0038] Figure 2 It is an exemplary flowchart of the foundation pit robustness analysis method provided by an embodiment of the present invention;

[0039] Figure 3 It is a schematic diagram of a numerical model of a foundation pit provided by an embodiment of the present invention;

[0040] Figure 4 It is an RGB image of the foundation pit numerical model provided by an embodiment of the present invention;

[0041] Figure 5 It is a prediction schematic diagram of the surrogate model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0042] The present invention will be further described in detail below in conjunction with test examples and specific embodiments. However, it should not be understood that the scope of the above subject matter of the present invention is limited to the following embodiments. All technologies implemented based on the content of the present invention belong to the scope of the present invention.

[0043] The terms "including" and "having" in the embodiments of the specification, claims and drawings of the present invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a series of steps or units included. A method, system, product or device is not necessarily limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0044] Please refer to Figure 1 , which shows a schematic flow chart of a foundation pit robustness analysis method provided by an embodiment of the present invention. The analysis method includes:

[0045] S1. After performing random field simulation on the soil property values of the target foundation pit, map them into the numerical model to obtain a number of first foundation pit samples.

[0046] S2. Calculate the mean value of the design response corresponding to each first foundation pit sample through the numerical model, and label the first foundation pit samples according to the mean value of the design response to generate a data set.

[0047] S3. Train a surrogate model according to the data set to obtain a trained surrogate model.

[0048] S4. Repeat S2 to S3 to obtain trained surrogate models corresponding to multiple mean values of the design response.

[0049] S5. After repeating S1 to obtain a number of second foundation pit samples, input them into multiple trained surrogate models to obtain multiple predicted mean values of the design response.

[0050] S6. Perform weighted fusion calculation according to the standard deviation and covariance of each predicted mean value of the design response to obtain the robustness index of the target foundation pit.

[0051] In the existing foundation pit robustness analysis, most of them adopt the method of numerical simulation analysis, and the most commonly used is FLAC software. Usually, for a foundation pit model with 2000 units, the calculation time for running the mean value of the design response by FLAC once is about 3 minutes, and the total calculation time for 10000 times is 500h. While using the method of the surrogate model, at most only 500 calculations are required by FLAC3D software to construct the training data set of the surrogate model, and then it takes 10 minutes to train the model, with a total time consumption of about 25h, and the calculation efficiency can be increased by 16 times. Moreover, the previous surrogate models usually cannot take into account the variability index of soil parameters. Therefore, an embodiment of the present invention provides a foundation pit robustness analysis method for numerical simulation analysis through a surrogate model, which introduces the variability of soil parameters into the analysis by random field simulation, and trains the surrogate model with this to obtain a surrogate model for foundation pit robustness analysis that can consider the spatial variability of soil parameters.

[0052] Furthermore, the geometric features (shapes), support structure settings, etc. of each foundation pit project are different, and it is often difficult for the surrogate model to take these features into account. Therefore, the surrogate model provided by the embodiment of the present invention only considers the variability of the geotechnical property values, so as to train a suitable surrogate model based on each foundation pit project, further improving the accuracy and reliability of the obtained predicted mean value of the design response.

[0053] The technical solutions provided by the embodiments of the present invention will be further introduced and described below through specific implementation manners.

[0054] Specifically, as described in S1 above, after performing random field simulation on the soil property values of the target foundation pit, they are mapped into the numerical model to obtain a number of first foundation pit samples.

[0055] In this step, a numerical model of the target foundation pit can be established through design parameters such as the exploration and design data of the target foundation pit. After establishment, based on the random parameters of the soil property values, Monte Carlo sampling method is used for random field simulation to construct a random field sample. Among them, the random parameters include the mean value, variance, coefficient of variation, vertical fluctuation distance, and horizontal fluctuation distance of the soil property values. Subsequently, the constructed random field sample is mapped into the numerical model, and the first foundation pit sample is obtained through the numerical model. The first foundation pit sample includes the distribution image of the soil property values and the mean value and variance of the soil property values.

[0056] It should be noted that since the distribution image of the soil property values is an RGB image generated by conversion through the numerical model, corresponding to the three channels in the RGB image, in the analysis method provided by the embodiments of the present invention, any three property values among the physical and mechanical properties of the soil of the target foundation pit are selected as the soil property values.

[0057] Further, the three soil property values of each unit body in the mapped numerical model are respectively converted into R-channel numerical values, G-channel numerical values, and B-channel numerical values;

[0058] The R-channel numerical values, G-channel numerical values, and B-channel numerical values of all unit bodies are respectively assigned to the R channel, G channel, and B channel to generate an RGB image.

[0059] Specifically, as described in S2 above, the mean value of the design response corresponding to each first foundation pit sample is calculated through the numerical model, and the first foundation pit sample is labeled according to the mean value of the design response to generate a data set.

[0060] This step is mainly to provide training samples for the training of the surrogate model through numerical simulation analysis.

[0061] Further, as described in S3 above, the surrogate model is trained according to the data set to obtain a trained surrogate model.

[0062] Specifically, as described in S4 above, S2 to S3 are repeated to obtain multiple trained surrogate models corresponding to the mean values of the design responses.

[0063] Under normal circumstances, a surrogate model can only obtain the corresponding mean value of one design response. Therefore, in order to obtain multiple mean values of design responses and provide higher reliability for robustness analysis, the embodiments of the present invention adopt trained different surrogate models to obtain different mean values of design responses. The mean value of the design response is a physical and mechanical response value based on the soil properties, that is, the design response that needs to be used and considered in the foundation pit design.

[0064] Further, the surrogate model adopts a convolutional neural network. In a possible implementation manner, the surrogate model provided by the embodiments of the present invention includes an input layer, a convolutional layer, an attention layer, a pooling layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. The learning rate of the surrogate model and the hyperparameters of the convolutional layer and the pooling layer are determined by Bayesian hyperparameter optimization.

[0065] Specifically, as described in S5 above, after repeating S1 to obtain a number of second foundation pit samples, they are input into multiple trained surrogate models to obtain multiple predicted mean values of design responses.

[0066] It can be understood that in S1 to S4 above, after performing numerical simulation analysis on the first foundation pit samples to provide training data for the surrogate model, the purpose is to obtain a trained surrogate model, so as to be able to replace the numerical simulation analysis, perform fast and accurate analysis, and obtain the mean values of each design response. And because the convolutional neural network has significantly high efficiency in data processing, the number of the second foundation pit samples can be much larger than the number of the first foundation pit samples. For example, 500 copies of the first foundation pit samples can be generated for training the surrogate model. The number of the second foundation pit samples input into the trained surrogate model can reach 10,000 or more, so as to obtain more accurate mean values of design responses.

[0067] Specifically, as described in S6 above, weight fusion calculation is performed according to the standard deviation and covariance of each predicted mean value of the design response to obtain the robustness index of the target foundation pit.

[0068] The embodiments of the present invention propose a method for weight-fusing the predicted mean values of each design response, so that the obtained robustness index can be more comprehensive and reliable.

[0069] Further, the calculation formula for the standard deviation of each predicted mean value of the design response is:

[0070]

[0071] where σ i is the standard deviation of the i-th predicted mean value of the design response, x ik is the i-th predicted mean value of the design response corresponding to the k-th second foundation pit sample, is the i-th predicted mean value of the design response, and M is the number of the second foundation pit samples.

[0072] The calculation formula for the covariance of the predicted mean of each design response is as follows:

[0073]

[0074] where Cov(X i , X j ) is the covariance between the predicted mean of the i-th design response and the predicted mean of the j-th design response, x ik is the predicted mean of the i-th design response corresponding to the k-th first random field sample, x jk is the predicted mean of the j-th design response corresponding to the k-th second foundation pit sample, is the predicted mean of the i-th design response, is the predicted mean of the j-th design response, and M is the number of second foundation pit samples.

[0075] The weight fusion based on the standard deviation and covariance of the predicted mean of each design response includes:

[0076] Performing weight assignment according to the standard deviation of the predicted mean of each design response, and the calculation formula is:

[0077]

[0078] where ω i is the fusion weight of the predicted mean of the i-th design response, σ, σ j are the standard deviations of the predicted means of the i-th and j-th design responses respectively, and n is the number of types of predicted means of design responses;

[0079] Performing fusion calculation according to the weights of the predicted means of each design response, and the calculation formula is:

[0080]

[0081] where σ p is the robustness index of the target foundation pit, Cov(X i , X j ) is the covariance between the predicted mean of the i-th design response and the predicted mean of the j-th design response.

[0082] Based on the technical solution provided by the present invention, a convolutional neural network surrogate model that can consider spatial variability is used as the basis for robustness evaluation. Based on the random field simulation, the randomness of soil properties can be fully reflected, and then the accurate and true mean value of the physical and mechanical design response can be predicted, ensuring the effectiveness of the analysis results. In addition, the way of adding a surrogate model in the present invention replaces the traditional numerical simulation analysis to obtain the mean value of each design response of the foundation pit, significantly improving the analysis efficiency, and enabling each foundation pit project to have an applicable surrogate model, further improving the accuracy and reliability of the predicted mean value of the obtained design response. At the same time, the present invention calculates the predicted mean value of each obtained design response through weight fusion, comprehensively considering the relationship between the mean values of each design response, providing a scientific evaluation method for the robustness evaluation of the foundation pit, providing effective technical support for engineering practice, providing a reliable decision-making basis for the design and construction of the foundation pit, and further ensuring the safety and stability of the foundation pit project from the design perspective.

[0083] Next, according to specific implementation manners, the technical solution provided by the embodiments of the present invention will be further introduced and described.

[0084] Please refer to Figure 2 , which shows an exemplary flowchart of the foundation pit robustness analysis method provided by the embodiments of the present invention.

[0085] In a possible implementation manner, in order to make the technical solution clearly described, in this embodiment, the above-mentioned soil properties are selected as three physical and mechanical properties of soil cohesion, friction angle, and elastic modulus, and the above-mentioned mean value of the design response and the predicted mean value of the design response are selected as three physical and mechanical response values of the displacement mean value of the support structure, the mean value of the soil heave at the bottom of the pit, and the mean value of the settlement around the foundation pit.

[0086] Specifically, in the above-mentioned foundation pit robustness analysis method, first, a numerical model of the foundation pit is established according to the exploration and design data, and then the spatial variability of soil cohesion, friction angle, and elastic modulus is considered. Random field samples are obtained through the random field simulation method and mapped into the numerical model of the foundation pit to obtain the magnitudes of soil cohesion, friction angle, and elastic modulus of each unit body. Due to the existence of randomness, the results of each random field simulation are different. Therefore, m groups of distribution data of soil cohesion, friction angle, and elastic modulus, that is, random field samples, can be generated by sampling.

[0087] Furthermore, a first foundation pit sample is obtained through the random field sample. The first foundation pit sample includes a distribution image of soil property values, as well as the mean and variance of the soil property values. Specifically, the distribution image of the soil property values is an RGB image generated by numerical model conversion. Considering that each unit in the numerical model has corresponding three property values (cohesion, friction angle, and elastic modulus), the three property values are respectively used as the RGB values of the numerical model image, and a color image is constructed for input. Simply put: If the cohesion, friction angle, and elastic modulus values of a certain grid unit in the numerical model are a, b, and c respectively, then the color of this grid unit is (R = a, B = b, G = c). Each cell in the foundation pit model has a corresponding RGB value. Based on this, a color image of the foundation pit model mapped with soil property values is constructed. The specific steps for generating the corresponding foundation pit model color image in each first foundation pit sample are as follows:

[0088] (1) Normalization: First, the three property values of cohesion, friction angle, and elastic modulus need to be normalized so that their range is between 0 and 1. This can be achieved by dividing each property value by the maximum value of this property.

[0089] (2) Scaling to the range of 0 - 255: Multiply the normalized value by 255 to convert it into an integer between 0 and 255, because the RGB values of the image need to be within this range.

[0090] (3) Generating the image: Corresponding the cohesion, friction angle, and elastic modulus values of each unit in the model to the RGB values respectively, and converting these data into an image format.

[0091] Based on the means and variances of the cohesion, friction angle, and elastic modulus of each first foundation pit sample above, a total of 6 numerical features, and 1 RGB image feature are used as the input of the CNN convolutional neural network.

[0092] Based on the above m groups of random field samples, numerical simulation analysis (annotation) is carried out through numerical software to obtain the mean value of the support structure displacement, the mean value of the soil heave at the bottom of the pit, and the mean value of the settlement around the foundation pit corresponding to each random field sample, which are used as the output of the surrogate model. Based on this, 3 surrogate model training data sets containing m samples respectively can be constructed for the mean value of the support structure displacement, the mean value of the soil heave at the bottom of the pit, and the mean value of the soil settlement around the foundation pit respectively. The construction process is shown in the following table:

[0093] Table 1 is a schematic table for the construction of the training data set.

[0094]

[0095] The structure of the proxy model is successively an input layer, a convolutional layer, an attention layer, a pooling layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. Hyperparameters such as the learning rate of the model, the sizes of the convolutional layer and the pooling layer, etc. are determined by Bayesian hyperparameter optimization. The above training data set containing m samples is divided into a training set, a test set, and a validation set according to the ratios of 80%, 10%, and 10%. The training set is used for model training. The model learns to establish the weights and parameters of the CNN by learning from the training set, so that it learns the mapping from the physical and mechanical parameter images of the foundation pit model to the numerical calculation response of the model; the validation set is used to adjust the model hyperparameters, and the best model is selected by evaluating the performance of different hyperparameters on the validation set; the test set is used to evaluate the model performance to check its generalization ability on unknown data. Among them, the training set and the validation set participate in model training, and the test set only participates in model evaluation and does not participate in model training.

[0096] Based on the above method, three proxy models are constructed for predicting the mean displacement of the support structure, the mean soil heave at the bottom of the pit, and the mean settlement around the foundation pit.

[0097] The trained proxy model above is used to carry out the robustness calculation of the foundation pit in combination with Monte Carlo sampling. When analyzing the stability of the foundation pit, not only the deformation of the support structure should be concerned, but also the displacement and deformation of the soil cannot be ignored, such as the soil heave at the bottom of the foundation pit and the settlement of the surrounding soil, etc. Therefore, when evaluating the robustness of the foundation pit, the fluctuation of the response value needs to be comprehensively considered. For this purpose, the embodiment of the present invention constructs the combined standard deviation of the physical and mechanical responses of the foundation pit support structure and the soil as the robustness index of the foundation pit system.

[0098] In a possible implementation manner, according to the above steps of generating m random field samples, M new random field samples are regenerated by Monte Carlo sampling, where M is much larger than m. Then, the second foundation pit samples corresponding to these M random field samples are obtained through the above steps. These M second foundation pit samples are respectively input into three proxy models for predicting the mean displacement of the support structure, the mean soil heave at the bottom of the pit, and the mean settlement around the foundation pit, and 3*M groups of data are obtained.

[0099] Finally, weight fusion is performed according to the standard deviation and covariance of the predicted mean value of each design response to obtain the robustness index of the target foundation pit.

[0100] Furthermore, the technical solutions provided by the embodiments of the present invention will be further described below in combination with specific implementation manners.

[0101] In a certain urban subway construction project, it is necessary to excavate a deep foundation pit for the construction of a subway station. The foundation pit is located in the city center, with dense surrounding buildings, a relatively high groundwater level, and complex geological conditions. To ensure the safety and stability during the construction of the foundation pit, it is necessary to consider the influence of the spatial variability of soil parameters on the stability of the foundation pit, and quantitatively analyze the robustness of the current foundation pit design scheme through the technical solution provided by the embodiments of the present invention.

[0102] First, according to geological exploration and design data, establish a numerical model of the foundation pit, including the foundation pit size, support structure, etc. Please refer to Figure 3 , which shows a schematic diagram of a numerical model of a foundation pit provided by an embodiment of the present invention.

[0103] Adopt random field simulation to simulate the spatial variability of the three physical and mechanical property values of cohesion, friction angle, and elastic modulus in the foundation pit model. Through multiple simulations, 500 groups of distribution data of cohesion, friction angle, and elastic modulus are generated, that is, random field samples.

[0104] Map the random field samples into the numerical model of the foundation pit. The distribution of soil physical and mechanical parameters in each sample is different. Then, through the training step of the surrogate model, generate a color image of the numerical model of the foundation pit mapped by the soil property values through the numerical model of the foundation pit, as shown in Figure 4 . It is the RGB image of the numerical model of the foundation pit provided by an embodiment of the present invention. At the same time, calculate the mean and variance of the cohesion, friction angle, and elastic modulus of each sample. Construct the RGB image and the mean and variance into the first foundation pit sample. Then, use the FLAC3D numerical analysis software to conduct numerical simulation analysis on each first foundation pit sample, and calculate the mean value of the support structure displacement, the mean value of the soil heave at the bottom of the pit, and the mean value of the settlement around the foundation pit corresponding to each first foundation pit sample, and label the first foundation pit sample as the sample label. In total, 500 labeled first foundation pit samples are generated as the dataset of the surrogate model for training. Among them, the RGB image, mean, and variance are used as the model input, and the sample label is used as the model output.

[0105] Use the above dataset to train 3 CNN convolutional neural network surrogate models respectively, so that they can respectively predict the predicted mean values of design responses such as the displacement of the foundation pit support structure, the soil heave at the bottom of the pit, and the settlement around the foundation pit. Please refer to Figure 5 , which shows a prediction schematic diagram of the surrogate model provided by an embodiment of the present invention.

[0106] The hyperparameters of the 3 models after Bayesian hyperparameter optimization are shown in the following table:

[0107] Table 2 is a schematic table of the hyperparameters of the surrogate model of the embodiment of the present invention

[0108]

[0109] Further, 10,000 new samples, i.e., the second foundation pit samples, are generated by Monte Carlo sampling, and the trained surrogate model is used for prediction to obtain the mean and standard deviation of each response value, and then the robustness index of the foundation pit is calculated. Note that the units of each response value should be unified during the calculation, and here meters are used as the unit. The calculation results are shown in Table 3:

[0110] Table 3 is a schematic table of the response value calculation results of the embodiment of the present invention

[0111]

[0112]

[0113] That is, in this embodiment, the robustness index of the foundation pit obtained is 0.009889.

[0114] On the other hand, the present invention also provides an electronic device, which includes a processor, a network interface, and a memory. The processor, the network interface, and the memory are interconnected. Among them, the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the above-mentioned method for analyzing the robustness of a foundation pit.

[0115] In the embodiment of the present invention, the processor may be an integrated circuit chip with signal processing capabilities. The processor may be a general-purpose processor, a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0116] It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention may be directly embodied as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The processor reads the information in the storage medium and combines its hardware to complete the steps of the above method.

[0117] The storage medium can be a memory, for example, it can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.

[0118] Among them, the non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable PROM (EEPROM), or a flash memory.

[0119] The volatile memory can be a Random Access Memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0120] The storage medium described in the embodiments of the present invention is intended to include but not limited to these and any other suitable types of memories.

[0121] It should be understood that the system disclosed in the present invention can be implemented in other ways. For example, the division of the modules is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the communication connection between the modules can be through some interfaces, and the indirect coupling or communication connection of the server or unit can be electrical or other forms.

[0122] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each module can exist physically alone, or two or more modules can be integrated in a processing unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0123] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0124] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A foundation pit robustness analysis method, characterized in that: The method comprises: S1. After performing random field simulation on the soil property values ​​of the target foundation pit, the soil property values ​​are mapped into the numerical model to obtain a number of first foundation pit samples; the first foundation pit samples include a distribution image of the soil property values ​​and a mean and variance of the soil property values; S2. Calculate the design response mean corresponding to each first foundation pit sample by using the numerical model, label the first foundation pit sample according to the design response mean, and generate a data set; S3, training a proxy model according to the data set to obtain a trained proxy model; S4, repeat S2 to S3 to obtain trained proxy models corresponding to multiple design response means; S5, repeat S1 to obtain a number of second foundation pit samples and then input them into a plurality of trained proxy models to obtain a plurality of design response prediction means; S6. Perform weighted fusion calculation based on the standard deviation and covariance of the predicted mean of each design response to obtain the robustness index of the target foundation pit; Among them, the weight fusion calculation is performed according to the standard deviation and covariance of the predicted mean of each design response, including: The weights are assigned based on the standard deviation of the predicted mean of each design response, calculated as: ， in, It is i The fusion weight of the predicted mean of the design responses, 𝜎 𝑖、 𝜎 j It is i , j The standard deviation of the predicted means of the design responses, n is the number of categories of the predicted mean of the design response; The fusion calculation is performed according to the weight of each design response prediction mean, and the calculation formula is: ， in, is the robustness index of the target foundation pit, It is i The predicted mean of the design response and the j The covariance between the predicted means of the design responses.

2. A foundation pit robustness analysis method according to claim 1, characterized in that: The random field simulation includes: A random field is constructed based on random parameters of the soil property values ​​using the Monte Carlo sampling method; wherein the random parameters include the mean, variance, coefficient of variation, vertical fluctuation distance and horizontal fluctuation distance of the soil property values.

3. A foundation pit robustness analysis method according to claim 1, characterized in that: The soil property values ​​are selected from any three property values ​​of the soil physical and mechanical properties of the target foundation pit.

4. A foundation pit robustness analysis method according to claim 3, characterized in that: The distribution image of the soil attribute value is an RGB image generated by numerical model conversion, including: The three soil property values ​​of each unit body in the mapped numerical model are converted into R channel values, G channel values ​​and B channel values ​​respectively; Assign the R channel values, G channel values ​​and B channel values ​​of all units to the R channel, G channel and B channel respectively to generate an RGB image.

5. A foundation pit robustness analysis method according to claim 1, characterized in that: The design response mean includes the pile body stress mean, the pile body displacement mean, the pit bottom uplift mean and the surrounding settlement mean.

6. A foundation pit robustness analysis method according to claim 1, characterized in that: The standard deviation of the predicted mean for each design response is given by: ， Among them, 𝜎 𝑖 It is i The standard deviation of the predicted means of the design responses, 𝑥 𝑖k It is k The predicted mean of the 𝑖th design response corresponding to the second foundation pit sample, is the predicted mean of the 𝑖th design response, M is the number of samples in the second foundation pit.

7. A foundation pit robustness analysis method according to claim 1, characterized in that: The covariance of the predicted mean for each design response is given by: ， in, It is i The predicted mean of the design response and the j The covariance between the predicted means of the design responses, 𝑥 𝑖k It is k The predicted mean of the 𝑖th design response corresponding to the first random field sample, 𝑥 jk It is k The second foundation pit sample corresponds to j The predicted mean of the design responses, is the predicted mean of the 𝑖th design response, It is j The predicted mean of the design responses, M is the number of samples in the second foundation pit.

8. A foundation pit robustness analysis method according to claim 1, characterized in that: The proxy model adopts a convolutional neural network.

9. An electronic device, characterized in that: It includes at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a foundation pit robustness analysis method as described in any one of claims 1 to 8.

Images