BIM-based tunnel soft rock deformation data processing method and system

Abstract

The application provides a tunnel soft rock deformation data processing method and system based on BIM. First, a plurality of tunnel soft rock monitoring data is acquired, and feature coding is performed thereon to convert the monitoring data into a first coding representation vector that can be used for machine learning model processing. Then, a first machine learning model including a plurality of deformation derivation networks and a feature transmission control network is used to perform deep analysis and derivation on the first coding representation vector, and output a deformation correlation vector associated with BIM deformation simulation conditions and monitoring data. Further, the deformation correlation vector is re-encoded by a second machine learning model to obtain a more refined third coding representation vector. Finally, the third coding representation vector is input into a deformation risk derivation model to accurately derive the deformation risk trend of the target tunnel section under different BIM deformation simulation conditions, thereby improving the efficiency and accuracy of tunnel soft rock deformation data processing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer technology, in particular to a tunnel soft rock deformation data processing method and system based on BIM. BACKGROUND

[0002] In the field of tunnel engineering, soft rock deformation is one of the important factors affecting the stability and safety of tunnels. Due to the low strength and large deformation of soft rock, tunnels often face the risk of soft rock deformation during construction and operation. In order to effectively control and predict soft rock deformation, it is necessary to continuously monitor the tunnel soft rock and analyze and predict the deformation trend based on the monitoring data.

[0003] Traditional tunnel soft rock deformation data processing methods mainly rely on manual experience and simple mathematical statistical models. These methods have many limitations in processing large-scale, multi-dimensional monitoring data. On the one hand, manual experience is difficult to fully and accurately capture the complex rules of soft rock deformation; on the other hand, simple mathematical statistical models often cannot effectively process nonlinear, high-dimensional data characteristics, resulting in limited accuracy and reliability of the prediction results.

[0004] With the rapid development of Building Information Modeling (BIM) technology, its application in the field of tunnel engineering is also increasingly widespread. BIM technology can realize three-dimensional visualization modeling and information integration of tunnel engineering, providing a new means for monitoring and analyzing tunnel soft rock deformation. However, how to effectively combine BIM technology with tunnel soft rock deformation data processing is still a problem to be solved. SUMMARY

[0005] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a tunnel soft rock deformation data processing method based on BIM, which comprises:

[0006] Obtaining a plurality of tunnel soft rock monitoring data, the tunnel soft rock monitoring data being used to derive the deformation risk trend of the target tunnel section to the soft rock deformation control scheme for a plurality of BIM deformation simulation conditions;

[0007] Feature coding representation is performed on the plurality of tunnel soft rock monitoring data, and a plurality of first coding representation vectors are outputted;

[0008] loading each of the first encoding representation vectors into a first machine learning model corresponding to the first encoding representation vectors, the first machine learning model comprising a plurality of first deformation deriving networks associated with a plurality of the BIM deformation simulation conditions and a plurality of first feature transfer control networks, the first feature transfer control networks outputting deformation associated vectors corresponding to the BIM deformation simulation conditions and the first encoding representation vectors according to derivation results of the first deformation deriving networks;

[0009] outputting a second encoding representation vector of the BIM deformation simulation condition according to a plurality of the deformation associated vectors corresponding to the BIM deformation simulation condition and a plurality of the first encoding representation vectors, loading the second encoding representation vector into a second machine learning model corresponding to the BIM deformation simulation condition, and outputting a third encoding representation vector corresponding to the BIM deformation simulation condition;

[0010] loading the third encoding representation vectors corresponding to the plurality of the BIM deformation simulation conditions into a deformation risk deriving model corresponding to the plurality of the BIM deformation simulation conditions, and outputting the deformation risk trends of the soft rock deformation control scheme for the target tunnel section under the plurality of the BIM deformation simulation conditions.

[0011] In still another aspect, the embodiments of the present application also provide a BIM-based tunnel soft rock deformation data processing system, comprising a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0012] Based on the above aspects, the embodiments of the present application realize efficient processing and analysis of tunnel soft rock monitoring data through effective combination of feature encoding representation and machine learning model. Firstly, a plurality of tunnel soft rock monitoring data are acquired and feature coded to convert into first encoding representation vectors which can be processed by machine learning model. Then, the first encoding representation vectors are deeply analyzed and derived by a first machine learning model comprising a plurality of deformation deriving networks and feature transfer control networks to output deformation associated vectors associated with BIM deformation simulation conditions and monitoring data. Further, the deformation associated vectors are re-encoded by a second machine learning model to obtain more refined third encoding representation vectors. Finally, the third encoding representation vectors are input into a deformation risk deriving model to accurately derive deformation risk trends of the target tunnel section under different BIM deformation simulation conditions. This method not only improves the efficiency and accuracy of tunnel soft rock deformation data processing, but also provides data basis for soft rock deformation control scheme in tunnel engineering, effectively reducing the safety risk in tunnel construction and operation process. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 is the execution flow schematic diagram of the tunnel soft rock deformation data processing method based on BIM provided by an embodiment of the application.

[0014] Figure 2 is the hardware architecture schematic diagram of the tunnel soft rock deformation data processing system based on BIM provided by an embodiment of the application. DETAILED DESCRIPTION

[0015] The application will be described in detail below with reference to the accompanying drawings, Figure 1 is the flow schematic diagram of the tunnel soft rock deformation data processing method based on BIM provided by an embodiment of the application, and the tunnel soft rock deformation data processing method based on BIM will be described in detail below.

[0016] In step S110, a plurality of tunnel soft rock monitoring data are acquired, and the tunnel soft rock monitoring data are used to derive a deformation risk trend of a target tunnel section to a soft rock deformation control scheme for a plurality of BIM deformation simulation conditions.

[0017] In this embodiment, for a tunnel construction project in a large mountainous area, the target tunnel section is located in a complex geological structure area, and soft rock is widely distributed and the geological conditions are variable. In order to ensure the safety and stability of the tunnel construction, a plurality of monitoring points are arranged around the tunnel, and these monitoring points are equipped with advanced sensor equipment. For example, stress sensors can measure the stress conditions of soft rock at different positions, strain sensors can accurately record the strain data of soft rock, and displacement sensors are used to monitor whether the soft rock has displacement and the size of the displacement.

[0018] With the advancement of tunnel construction, these sensors continuously collect data. The stress sensor records the stress value of the soft rock every certain period of time (such as every hour), and these values may fluctuate due to factors such as vibration during tunnel excavation, changes in surrounding rock pressure, etc. The strain sensor records the stretching or compression of the soft rock in different directions in the form of strain values, and the displacement sensor can detect whether the soft rock has a slight movement, and if there is a movement, it records the direction and distance of the movement.

[0019] These stress, strain and displacement data obtained from different monitoring points are what we call tunnel soft rock monitoring data. At the same time, in this project, in order to comprehensively evaluate the deformation risk of soft rock, multiple BIM deformation simulation conditions are set. For example, the deformation of soft rock under different tunnel excavation speeds, different support structure installation times and different blasting intensities is simulated in the BIM model. The tunnel soft rock monitoring data obtained is used to deduce the deformation risk trend of the soft rock deformation control scheme under these different BIM deformation simulation conditions in the target tunnel section. For example, if the stress data continues to increase under a certain simulation condition, the strain data also shows that the soft rock has a large tensile trend, and the displacement data shows that the soft rock has a tendency to move towards the inside of the tunnel, then it can be preliminarily judged that the soft rock deformation risk is high under this simulation condition, and thus provide a basis for formulating a soft rock deformation control scheme.

[0020] Step S120, a plurality of tunnel soft rock monitoring data are encoded and represented, and a plurality of first encoding representation vectors are output.

[0021] For a large number of tunnel soft rock monitoring data obtained in step S110, feature encoding representation is needed. First, these monitoring data are preliminarily classified. For example, stress-related data are classified into one category, strain-related data are classified into another category, and displacement-related data are classified into another category, thus generating subsets of different types of data.

[0022] For the stress data subset, feature recognition is performed. The maximum stress value in the stress data is found, which may occur when the tunnel is excavated to a certain key position, and the surrounding soft rock is subjected to the maximum extrusion stress. The minimum stress value may occur in a relatively stable soft rock area with less interference. Stress change frequency is also an important feature, which reflects the number of stress fluctuations of soft rock in a period of time, for example, during the tunnel blasting operation, the stress change frequency will increase significantly.

[0023] For the strain data subset, the strain peak value represents the maximum stretching or compression degree of the soft rock at a certain moment, which may be caused by local stress concentration during tunnel excavation. The strain valley value is the strain value of the soft rock in a relatively relaxed state. The average change rate of strain reflects the overall change trend of the strain of soft rock in a period of time, for example, if the average change rate continues to increase, it means that the deformation of the soft rock is in an accelerating state.

[0024] After identifying these basic features (maximum stress value, minimum stress value, stress variation frequency, strain peak value, strain valley value, average rate of strain change, etc.), they are layered according to their importance and relevance. For example, the maximum stress value and the strain peak value are very important for judging the limit state of soft rock, and are classified into the main feature layer; while the stress variation frequency and the average rate of strain change are also important, but relatively secondary, and are classified into the secondary feature layer.

[0025] For the features of the main feature layer, a more complex and accurate conversion method is used, such as nonlinear transformation, to highlight their key role in soft rock deformation analysis. For the features of the secondary feature layer, a relatively simple linear conversion method is used. Then the converted features of different levels are combined according to the internal correlation between the features. For example, the stress-related main features and the strain-related secondary features are combined in a certain logical order to generate feature combination data with associated relationships.

[0026] Based on the pre-set encoding rules, which are determined according to the feature range of soft rock deformation data and the expected encoding representation requirements. For example, according to the possible value range of soft rock stress (from the minimum static pressure to the maximum blasting impact pressure) and the possible value range of strain (from the tiny elastic strain to the limit strain that may cause soft rock failure), an encoding space is determined. Then according to the division of this encoding space, an encoding mapping relationship is established for each feature combination data. According to this encoding mapping relationship, each feature combination data is mapped into a corresponding encoding space, in which different encoding segments represent different feature combination situations. Finally, the various encoding segments are combined in a pre-set logical order to construct an encoding representation vector. For example, if the encoding space is divided into stress-related encoding segments, strain-related encoding segments and displacement-related encoding segments, these encoding segments are combined in the order of stress first, then strain, and then displacement, to obtain a complete first encoding representation vector.

[0027] Step S130, load each of the first encoding representation vectors into a first machine learning model corresponding to the first encoding representation vector, the first machine learning model comprising a plurality of first deformation derivation networks associated with a plurality of BIM deformation simulation conditions and a plurality of first feature transfer control networks, the first feature transfer control network outputting a deformation association vector corresponding to the BIM deformation simulation condition and the first encoding representation vector according to the derivation result of the first deformation derivation network.

[0028] Assume that in this tunnel project, the first machine learning model is specifically built for analyzing the relationship between soft rock deformation and BIM deformation simulation conditions. Among them, multiple first deformation derivation networks are respectively used to analyze the relationship between different soft rock deformation mechanisms and BIM deformation simulation conditions. For example, there is a first deformation derivation network that is specifically used to study the relationship between soft rock under stress change (which is one of the BIM deformation simulation conditions, such as simulating different external stress loading conditions) and the actual deformation of soft rock, and another first deformation derivation network focuses on the relationship between soft rock strain and the deformation of soft rock under the constraint of different support structures simulated by BIM.

[0029] At the same time, multiple first feature transfer control networks play a role in coordinating and integrating the results of different first deformation derivation networks. When loading the first encoding representation vector into this first machine learning model, take the first deformation derivation network for stress change as an example. This network uses its internal algorithms and pre-trained parameters to derive the possible deformation trend of soft rock under specific BIM deformation simulation conditions (such as simulating stress changes under fast tunnel excavation speed) according to stress-related features in the first encoding representation vector (such as maximum stress value, stress change frequency, etc.). The derivation result may be a numerical range about the deformation degree of soft rock or a vector representation of the deformation trend of soft rock.

[0030] And the first feature transfer control network corresponding to this first deformation derivation network begins to play a role. If this first feature transfer control network is only associated with this first deformation derivation network, it will obtain the derivation result of the first collaborative network (assuming that the first collaborative network integrates part of the information of multiple first deformation derivation networks to provide more comprehensive background information). For example, the first collaborative network may give a general evaluation value of the overall stability of the soft rock according to the preliminary results of multiple first deformation derivation networks. Then, the first feature transfer control network obtains the first influence coefficient corresponding to the first collaborative network and this first deformation derivation network respectively according to the first encoding representation vector. The specific acquisition method is as follows:

[0031] First, the first network parameter array corresponding to the first cooperative network and each first deformation derivation network is obtained. The column quantity of this array is the network quantity of the first deformation derivation network and the first cooperative network, and the row quantity is the feature category quantity of the first encoding representation vector. Assuming that the feature categories of the first encoding representation vector include stress features, strain features and displacement features, and the first deformation derivation network and the first cooperative network have a total of five, then the first network parameter array is a 5*3 matrix. Then, the first dot product calculation is performed on the first network parameter array and the inverse arrangement vector of the first encoding representation vector. For example, the inverse arrangement vector of the first encoding representation vector is [displacement feature value, strain feature value, stress feature value], and the dot product calculation is performed on the first network parameter array to obtain an intermediate result. Finally, the intermediate result is processed by an activation function (such as using a ReLU function), and the first influence coefficient is output.

[0032] According to the first influence coefficient, the fusion result of the derivation result of the first cooperative network and the derivation result of the first deformation derivation network associated with the first feature transmission control network is performed. For example, if the first influence coefficient indicates that the result of the first deformation derivation network has a high weight, the result of this network will be more inclined in the fusion, so as to output the deformation associated vector corresponding to the BIM deformation simulation condition and the first encoding representation vector. The deformation associated vector contains the soft rock deformation associated information obtained by the first deformation derivation network and the first feature transmission control network under the specific BIM deformation simulation condition according to the soft rock monitoring data (reflected by the first encoding representation vector), which will be used for further analysis of the soft rock deformation risk in the subsequent steps.

[0033] If the first feature transmission control network is associated with the first cooperative network, it will fuse the derivation results of the first cooperative network and each first deformation derivation network, and output the first cooperative vector corresponding to the first encoding representation vector. For example, assuming that there are three first deformation derivation networks, which are respectively used to deduce the relationship between stress, strain and displacement and the soft rock deformation under the BIM deformation simulation condition. The first cooperative network gives a preliminary evaluation value of the overall stability of the soft rock. The first feature transmission control network fuses these results according to the first influence coefficient corresponding to each network (also according to the method of obtaining the first influence coefficient described above). If the first influence coefficient of the stress-related first deformation derivation network is high, the stress-related result will have a large proportion in the first cooperative vector in the fusion, and the first cooperative vector will also play an important role in the subsequent steps.

[0034] Step S140, according to the plurality of deformation associated vectors corresponding to the BIM deformation simulation condition and the plurality of first encoding representation vectors, output a second encoding representation vector of the BIM deformation simulation condition, load the second encoding representation vector to a second machine learning model corresponding to the BIM deformation simulation condition, and output a third encoding representation vector corresponding to the BIM deformation simulation condition.

[0035] In the previous steps, for each first encoding representation vector, a deformation associated vector corresponding to the BIM deformation simulation condition is obtained under the first machine learning model. Now, according to these deformation associated vectors, a second encoding representation vector of the BIM deformation simulation condition is constructed. For example, assume that there are three deformation associated vectors, each corresponding to the result of a different first encoding representation vector under a certain BIM deformation simulation condition. These deformation associated vectors contain deformation associated information of soft rock related to the BIM simulation condition in terms of stress, strain and displacement. The relevant features in these deformation associated vectors are combined and encoded according to certain rules.

[0036] For example, deformation associated features of soft rock under different stress simulation conditions are extracted from stress-related deformation associated vectors, and corresponding key features are also extracted from strain and displacement-related deformation associated vectors, and then these features are recombined and encoded according to the requirements of the BIM deformation simulation condition. This encoding process may involve weighting, normalization and other operations on the features, and finally a second encoding representation vector of the BIM deformation simulation condition is obtained.

[0037] Load this second encoding representation vector into a second machine learning model corresponding to the BIM deformation simulation condition. This second machine learning model includes a second deformation derivation network associated with the BIM deformation simulation condition and a second feature transfer control network associated with the BIM deformation simulation condition.

[0038] Taking the second deformation derivation network as an example, when the second encoding representation vector is loaded, it will further derive the deformation of soft rock under the BIM deformation simulation condition according to its internal algorithm and pre-trained parameters. Assuming that this second deformation derivation network is specifically designed to analyze the deformation of soft rock under complex BIM simulation geological conditions and construction procedures, it can derive more in-depth deformation trends of soft rock under this specific BIM deformation simulation condition according to the geological condition-related features (such as the type of soft rock, the distribution of surrounding rocks, etc. Information embodied after encoding in the previous steps) and construction procedure-related features (such as the simulated excavation sequence, support installation time, etc.) in the second encoding representation vector. The derivation result may be a detailed description vector of soft rock deformation.

[0039] Then, the second feature transfer control network generates a third encoding representation vector corresponding to the BIM deformation simulation condition according to the derivation result of the second deformation derivation network. Specifically, first, a first collaborative encoding representation vector is output according to a plurality of first collaborative vectors corresponding to a plurality of first encoding representation vectors, and is loaded into a second collaborative network. For example, assuming that three first collaborative vectors have been obtained before, they are combined into a first collaborative encoding representation vector according to certain rules, and this vector contains the comprehensive information of the plurality of first encoding representation vectors after collaborative processing in the first machine learning model. After loading the first collaborative encoding representation vector into the second collaborative network, the second collaborative network will process the vector according to its own algorithm and pre-training parameters to obtain a comprehensive result related to soft rock deformation.

[0040] At the same time, a second influence coefficient corresponding to the second deformation derivation network associated with the second feature transfer control network and the second collaborative network is obtained according to the first combined encoding representation vector (this vector is obtained by concatenating a plurality of first encoding representation vectors). The specific acquisition process is similar to the acquisition of the first influence coefficient. After obtaining the second network parameter array corresponding to the second deformation derivation network and the second collaborative network (the column quantity of this array is the network quantity of the second deformation derivation network and the second collaborative network, and the row quantity is the feature category quantity of the first combined encoding representation vector), the second network parameter array and the inverse arrangement vector of the first combined encoding representation vector are subjected to second dot product calculation, and then the calculation result is subjected to activation function processing to obtain the second influence coefficient.

[0041] Finally, the derivation result of the second collaborative network and the derivation result of the second deformation derivation network associated with the second feature transfer control network are fused according to the second influence coefficient corresponding to the second deformation derivation network associated with the second feature transfer control network and the second collaborative network. For example, if the second influence coefficient indicates that the result of the second deformation derivation network has a higher weight in generating the third encoding representation vector, the result of this network will be considered more in the fusion, so as to output the third encoding representation vector corresponding to the BIM deformation simulation condition. This third encoding representation vector will be used for the final soft rock deformation risk trend derivation.

[0042] In step S150, the third encoding representation vectors corresponding to the plurality of BIM deformation simulation conditions are loaded into the deformation risk derivation model corresponding to the plurality of BIM deformation simulation conditions, and the derived deformation risk trend of the target tunnel section to the soft rock deformation control scheme for the plurality of BIM deformation simulation conditions is output.

[0043] In this tunneling project, the deformation risk derivation model is specifically designed to derive the deformation risk trends of the target tunnel segment for the soft rock deformation control scheme under different BIM deformation simulation conditions based on the third encoding representation vectors obtained from the previous steps.

[0044] Assuming there are multiple BIM deformation simulation conditions, such as different excavation speeds, different support structure types, and different blasting intensities, each condition has a corresponding third encoding representation vector. These third encoding representation vectors are loaded into the deformation risk derivation model.

[0045] This deformation risk derivation model contains complex algorithms and pre-trained parameters inside, which are trained based on a large amount of historical tunneling project data (including soft rock deformation data, effect data under different control schemes, etc.). For example, for the third encoding representation vector corresponding to the BIM deformation simulation condition of fast excavation speed, the deformation risk derivation model will analyze the relevant features of the soft rock (such as stress, strain, displacement, etc., which are contained in the third encoding representation vector after encoding and processing) and combine the historical data of similar excavation speeds and the effects of different deformation control schemes to derive the deformation risk trend of the soft rock deformation control scheme under this excavation speed.

[0046] If the third encoding representation vector shows that the stress of the soft rock changes greatly, the strain is close to the limit, and the displacement has a clear trend of moving towards the tunnel, and the historical data shows that the existing soft rock deformation control scheme (such as a certain support structure and drainage measures) may not be able to effectively control the soft rock deformation under such conditions, then the deformation risk derivation model will judge that the deformation risk trend of the soft rock deformation control scheme under the BIM deformation simulation condition of this excavation speed is high.

[0047] For different BIM deformation simulation conditions, such as different support structure types, the deformation risk derivation model will also analyze the relevant features in the corresponding third encoding representation vector. If the third encoding representation vector corresponding to a certain support structure type shows that the stress distribution of the soft rock is relatively uniform, the strain is within a controllable range, and the displacement is small, and the historical data shows that this support structure has good control effect on soft rock deformation under similar conditions, then it will be judged that the deformation risk trend of the soft rock deformation control scheme under the BIM deformation simulation condition of this support structure type is low.

[0048] In this way, the deformation risk derivation model outputs the deformation risk trend of the target tunnel segment for the soft rock deformation control scheme under each BIM deformation simulation condition. These deformation risk trends will provide important reference for engineers to develop reasonable soft rock deformation control schemes, helping them adjust the construction strategy in time during the tunnel construction process, ensuring the safety and stability of the tunnel construction.

[0049] Based on the above steps, the embodiment of the application realizes efficient processing and analysis of tunnel soft rock monitoring data through effective combination of feature coding representation and machine learning model. The method first acquires a plurality of tunnel soft rock monitoring data, and performs feature coding on the monitoring data to convert the monitoring data into a first coding representation vector that can be processed by a machine learning model. Then, a first machine learning model including a plurality of deformation derivation networks and a feature transmission control network is used to perform deep analysis and derivation on the first coding representation vector, and output a deformation correlation vector associated with BIM deformation simulation conditions and monitoring data. Further, the second machine learning model is used to re-code the deformation correlation vector to obtain a more refined third coding representation vector. Finally, the third coding representation vector is input into a deformation risk derivation model to accurately derive the deformation risk trend of the target tunnel section under different BIM deformation simulation conditions. The method not only improves the efficiency and accuracy of tunnel soft rock deformation data processing, but also provides data basis for soft rock deformation control scheme in tunnel engineering, effectively reducing the safety risk in the process of tunnel construction and operation.

[0050] In a possible implementation, the second machine learning model includes a second deformation derivation network associated with the BIM deformation simulation condition and a second feature transmission control network associated with the BIM deformation simulation condition.

[0051] Step S140 includes:

[0052] Step S141, loading the second coding representation vector into the second deformation derivation network corresponding to the BIM deformation simulation condition, and outputting the derivation result of the second deformation derivation network.

[0053] Step S142, calling the second feature transmission control network to generate the third coding representation vector corresponding to the BIM deformation simulation condition according to the derivation result of the second deformation derivation network.

[0054] In a possible implementation, the first machine learning model further includes a first collaborative network, and the second machine learning model further includes a second collaborative network.

[0055] Step S130 includes:

[0056] Step S131, if the first feature transmission control network is associated with one of the first deformation derivation networks, the first feature transmission control network fuses the derivation result of the first collaborative network and the derivation result of the first deformation derivation network associated with the first feature transmission control network, and outputs the deformation correlation vector corresponding to the BIM deformation simulation condition and the first coding representation vector.

[0057] If the first feature transmission control network is associated with the first collaborative network, the first feature transmission control network fuses the inference result of the first collaborative network and the inference result of each first deformation inference network, and outputs a first collaborative vector corresponding to the first encoding representation vector.

[0058] Step S142 includes:

[0059] Step S1421, according to a plurality of first collaborative vectors corresponding to a plurality of first encoding representation vectors, a first collaborative encoding representation vector is outputted and loaded into the second collaborative network.

[0060] Step S1422, according to the second feature transmission control network, the inference result of the second collaborative network and the inference result of the second deformation inference network associated with the second feature transmission control network are fused, and the third encoding representation vector corresponding to the BIM deformation simulation condition is outputted.

[0061] In the large mountainous tunnel construction project scenario mentioned earlier, for the operation related to the second machine learning model, the second machine learning model includes the second deformation inference network associated with the BIM deformation simulation condition and the second feature transmission control network associated with the BIM deformation simulation condition. When the second encoding representation vector obtained in the previous step is loaded into the second deformation inference network corresponding to the BIM deformation simulation condition, this process is based on the algorithm and parameters preset in the second deformation inference network. For example, in this tunnel project, the BIM deformation simulation condition sets the influence of different blasting methods on soft rock. The second encoding representation vector contains various information related to the blasting method in the previous processing process, such as the encoded representation of the stress, strain, displacement, and other related characteristics of soft rock under different blasting intensities. After the second deformation inference network receives this vector, it infers according to the algorithm logic in its internal for the deformation mechanism of soft rock under the blasting method. If the historical data shows that the stress concentration area of soft rock under a certain blasting intensity has a specific distribution rule, the second deformation inference network will search for similar stress distribution characteristics according to the information in the second encoding representation vector, and then infer the deformation trend of soft rock under this blasting method. This inference result may be a result set containing information such as deformation amount prediction and deformation direction judgment of different parts of soft rock, which is the inference result of the second deformation inference network.

[0062] Then, a second feature transfer control network is called to generate a third encoding representation vector corresponding to the BIM deformation simulation condition according to the derivation result of the second deformation derivation network. The second feature transfer control network plays a role of integrating and optimizing information in this process. It needs to consider the relationship between the derivation result of the second deformation derivation network and the whole soft rock deformation analysis system. For example, the derivation result of the second deformation derivation network shows that the strain of a certain local area of the soft rock has a rapid increasing trend under a certain blasting mode. The second feature transfer control network will adjust and optimize this result by combining the overall stability requirements of the soft rock, the deformation conditions of other parts, and the safety standards set in the project, and other factors, so as to generate a third encoding representation vector corresponding to the BIM deformation simulation condition. This vector will more accurately reflect the comprehensive deformation characteristics of the soft rock under the BIM deformation simulation condition, so as to carry out more accurate deformation risk assessment in the subsequent steps.

[0063] For the first collaborative network in the first machine learning model and related operations, the following situations exist in the previous tunnel project scenario. The first machine learning model also includes a first collaborative network, and the second machine learning model also includes a second collaborative network.

[0064] When the first feature transfer control network is associated with a first deformation derivation network, for example, in the first deformation derivation network related to soft rock stress analysis. The first collaborative network has comprehensively evaluated various deformation-related factors of the whole soft rock in the previous processing process and obtained a preliminary result. This result may include the overall stability evaluation of the soft rock, the correlation between the deformations of different areas of the soft rock, and other information. The first deformation derivation network related to soft rock stress analysis focuses on the relationship between soft rock stress changes and soft rock deformation under BIM deformation simulation conditions (such as different tunnel excavation depths corresponding to different stress loading conditions), and its derivation result is the deformation caused by the change of soft rock stress based on the soft rock stress monitoring data through complex algorithms. The first feature transfer control network fuses the derivation result of the first collaborative network and the derivation result of the first deformation derivation network associated with the first feature transfer control network. Specifically, it will perform weighted processing according to the importance of the two results in the soft rock deformation analysis. If the result of the first collaborative network is more valuable for judging the overall stability of the soft rock, it will be given a higher weight in the fusion; and if the result of the first deformation derivation network is more accurate in the analysis of soft rock deformation caused by specific stress changes, it will also be given a suitable weight accordingly. Through such a fusion process, a deformation correlation vector corresponding to the BIM deformation simulation condition and the first encoding representation vector is output. This deformation correlation vector integrates the deformation information of the whole soft rock and under specific stress analysis, providing comprehensive soft rock deformation correlation data for subsequent steps.

[0065] When the first feature transfer control network is associated with the first collaborative network, the first feature transfer control network fuses the inference results of the first collaborative network and the inference results of each first deformation inference network, and outputs a first collaborative vector corresponding to the first encoding representation vector. For example, there are multiple first deformation inference networks that respectively deduce the relationship between the strain, displacement, and different stress components (such as vertical stress and horizontal stress) of soft rock and the deformation of soft rock under the BIM deformation simulation condition. The inference result of the first collaborative network contains information such as the overall stability evaluation of the soft rock and the mutual relationship of the deformation of the soft rock in different regions. The first feature transfer control network fuses according to the importance of each result in the analysis of the deformation of the soft rock. Assuming that in a certain tunnel construction stage, the result of the first deformation inference network related to strain is very crucial for judging the current deformation state of the soft rock, the result related to displacement is of great significance for predicting the subsequent deformation trend of the soft rock, and the result of the first collaborative network is indispensable for overall understanding of the stability of the soft rock. The first feature transfer control network will determine the weight of each result according to these factors, and then perform a fusion operation. If the weight of the result related to strain is 0.4, the weight of the result related to displacement is 0.3, and the weight of the result of the first collaborative network is 0.3, the weighted sum of each result is performed according to this weight ratio, and a first collaborative vector corresponding to the first encoding representation vector is output. This first collaborative vector reflects the deformation-related information of the soft rock under the comprehensive consideration of multiple first deformation inference networks and the first collaborative network.

[0066] In the process of calling the second feature transfer control network to generate a third encoding representation vector corresponding to the BIM deformation simulation condition according to the inference results of the second deformation inference network, according to the multiple first collaborative vectors corresponding to the multiple first encoding representation vectors, a first collaborative encoding representation vector is output and loaded into the second collaborative network. For example, in the previous process of monitoring and analyzing the soft rock, each first encoding representation vector has been processed by the first machine learning model to obtain the corresponding first collaborative vector. These first collaborative vectors contain comprehensive deformation information of the soft rock at different monitoring points and from different angles of analysis. These first collaborative vectors are combined according to certain rules, which may be determined according to factors such as the spatial distribution characteristics of the soft rock and the importance of different monitoring points. For example, the first collaborative vectors corresponding to each monitoring point are sequentially combined in the order of the soft rock from the tunnel entrance to the exit to form a first collaborative encoding representation vector. Then this first collaborative encoding representation vector is loaded into the second collaborative network. The second collaborative network processes this vector according to its internal algorithm and pre-trained parameters. It will comprehensively consider factors such as the deformation correlation of the soft rock in the entire tunnel section and the cooperativity of the deformation of the soft rock in different regions to obtain a result related to the overall deformation of the soft rock.

[0067] Finally, according to the second feature transmission control network, the fusion result of the derivation result of the second collaborative network and the derivation result of the second deformation derivation network associated with the second feature transmission control network is outputted, and a third encoding representation vector corresponding to the BIM deformation simulation condition is outputted. For example, the derivation result of the second collaborative network shows that the soft rock has a certain overall deformation trend in the entire tunnel section, such as a slight overall displacement tendency in a certain direction, and the derivation result of the second deformation derivation network shows the local deformation of the soft rock under a specific BIM deformation simulation condition (such as a specific support structure installation sequence). The second feature transmission control network will determine the weight according to the importance of the overall and local deformation of the soft rock in the entire deformation analysis. If the weight of the overall deformation for judging the influence of the soft rock on the tunnel structure is 0.6, and the weight of the local deformation is 0.4, the derivation result of the second collaborative network and the derivation result of the second deformation derivation network are weighted and fused according to the weight ratio, and finally a third encoding representation vector corresponding to the BIM deformation simulation condition is outputted. The third encoding representation vector accurately reflects the comprehensive deformation characteristics of the soft rock under the BIM deformation simulation condition, and provides key data support for subsequent deformation risk derivation.

[0068] In a possible implementation, the step S131 comprises:

[0069] The step S1311 comprises: acquiring, according to the first encoding representation vector, a first influence coefficient corresponding to each of the first collaborative network and the first deformation derivation network.

[0070] The step S1312 comprises: fusing, according to the first influence coefficient of the first collaborative network and the first deformation derivation network associated with the first feature transmission control network, a derivation result of the first collaborative network and a derivation result of the first deformation derivation network associated with the first feature transmission control network, and outputting a deformation correlation vector corresponding to the BIM deformation simulation condition and the first encoding representation vector.

[0071] The step S132 comprises: fusing, according to the first influence coefficient corresponding to each of the first collaborative network and the first deformation derivation network, a derivation result of the first collaborative network and a derivation result of each of the first deformation derivation network, and outputting a first collaborative vector corresponding to the first encoding representation vector.

[0072] The step S1311 comprises:

[0073] Step S1311-1, obtain the first network parameter array corresponding to each of the first cooperative network and the first deformation derivation network respectively, the longitudinal quantity of the first network parameter array is the network quantity of the first deformation derivation network and the first cooperative network, and the horizontal quantity of the first network parameter array is the feature category quantity of the first encoding representation vector.

[0074] Step S1311-2, perform first dot product calculation on the first network parameter array and the inverse arrangement vector of the first encoding representation vector.

[0075] Step S1311-3, perform activation function processing on the calculation result of the first dot product calculation, and output the first influence coefficient.

[0076] In this embodiment, the first encoding representation vector contains various feature information of soft rock, such as stress characteristics, strain characteristics, displacement characteristics, etc. of soft rock. These feature information is obtained through a series of monitoring data processing and coding. The first cooperative network and each first deformation derivation network have their respective first network parameter array. Taking the first deformation derivation network for soft rock stress analysis as an example, the longitudinal quantity of the first network parameter array is the network quantity of the first deformation derivation network and the first cooperative network. Assuming that there are three first deformation derivation networks (corresponding to stress, strain, and displacement analysis respectively) and one first cooperative network, the longitudinal quantity is 4. The horizontal quantity is the feature category quantity of the first encoding representation vector. Assuming that the first encoding representation vector contains five soft rock feature categories, the first network parameter array is a 4x5 matrix. For the first deformation derivation network for soft rock stress analysis, the elements in the matrix can be determined according to a large amount of historical soft rock stress data and the stress change rule of soft rock under different BIM deformation simulation conditions. Similarly, the first network parameter array of the first cooperative network is determined based on the comprehensive consideration of various factors related to the overall deformation of soft rock.

[0077] Then, taking the first deformation derivation network for soft rock stress analysis as an example, the inverse arrangement vector of the first encoding representation vector is a vector obtained by inversely arranging the feature values originally arranged in a certain order (such as stress, strain, and displacement). Dot product calculation is performed on the inverse arrangement vector and the first network parameter array. This dot product calculation process is performed according to the rules of matrix multiplication, which multiplies each row element in the first network parameter array with the corresponding element in the inverse arrangement vector and adds them up to obtain an intermediate result. This intermediate result comprehensively considers the relationship between the network parameters in the first network parameter array and the soft rock feature information in the first encoding representation vector. The same operation is performed on the first cooperative network to obtain its corresponding intermediate result.

[0078] Then, the calculation result of the first dot product calculation is processed by an activation function, and the first influence coefficient is output. In this tunnel project, the activation function used may be the ReLU function. The intermediate result obtained in the foregoing is input into the ReLU function, which processes the input value. If the input value is less than 0, 0 is output; if the input value is greater than 0, the input value is output. The result obtained after the activation function processing is the first influence coefficient. This first influence coefficient reflects the relative importance of the first deformation derivation network for soft rock stress analysis in the entire fusion process based on the soft rock feature information contained in the first encoding representation vector. The same is true for the first collaborative network, and the first influence coefficient obtained by the first collaborative network reflects its relative importance in the fusion process.

[0079] According to the first influence coefficients of the first collaborative network and the first deformation derivation network associated with the first feature transmission control network, the fusion result of the derivation result of the first collaborative network and the derivation result of the first deformation derivation network associated with the first feature transmission control network is output, and the deformation associated vector corresponding to the BIM deformation simulation condition and the first encoding representation vector is output. In this tunnel project, the derivation result of the first collaborative network may be an evaluation value of the overall stability of soft rock, for example, the value indicates that the soft rock is in a relatively stable state as a whole at the current tunnel construction stage, but there is a risk of local instability. The derivation result of the first deformation derivation network for soft rock stress analysis may be the local deformation trend of soft rock under stress, for example, stress concentration at a certain position of the tunnel may cause soft rock to displace a certain distance into the tunnel. According to the first influence coefficient obtained in the foregoing, if the first influence coefficient of the first deformation derivation network for soft rock stress analysis is high, it indicates that the result of stress analysis is relatively more important in this fusion process. Therefore, in the fusion, the derivation result of the first deformation derivation network for soft rock stress analysis is given a greater weight. For example, the derivation result of the first collaborative network has a weight of 0.3, and the derivation result of the first deformation derivation network for soft rock stress analysis has a weight of 0.7. Through weighted summation and other fusion operations, the two results are fused together, and the result obtained is the deformation associated vector corresponding to the BIM deformation simulation condition and the first encoding representation vector. This deformation associated vector contains comprehensive information such as the overall stability evaluation of soft rock and the local deformation trend under stress, and can more comprehensively reflect the deformation associated situation of soft rock under the BIM deformation simulation condition.

[0080] For the process of fusing the derivation results of the first collaborative network and the derivation results of each first deformation derivation network, and outputting the first collaborative vector corresponding to the first encoding representation vector. Similarly, according to the first influence coefficients corresponding to the first collaborative network and each first deformation derivation network respectively, the derivation results of the first collaborative network and the derivation results of each first deformation derivation network are fused, and the first collaborative vector corresponding to the first encoding representation vector is output. For example, there are three first deformation derivation networks, which are soft rock stress analysis, strain analysis and displacement analysis networks. The derivation result of the first collaborative network is the evaluation value of the overall stability of the soft rock, the derivation result of the first deformation derivation network of the soft rock stress analysis is the local deformation trend of the soft rock under the action of stress, the derivation result of the first deformation derivation network of the strain analysis is the strain distribution of the soft rock, and the derivation result of the first deformation derivation network of the displacement analysis is the displacement direction and size of the soft rock. According to the respective corresponding first influence coefficients, the fusion is carried out. Assuming that the first influence coefficient of the first deformation derivation network of the soft rock stress analysis is 0.3, the first influence coefficient of the first deformation derivation network of the strain analysis is 0.2, the first influence coefficient of the first deformation derivation network of the displacement analysis is 0.2, and the first influence coefficient of the first collaborative network is 0.3. According to these weights, the weighted sum and other fusion operations are carried out on each result. If the derivation result of the first collaborative network is a value representing the overall stability of the soft rock, the derivation result of the first deformation derivation network of the soft rock stress analysis is a vector representing the local deformation trend under the action of stress, the derivation result of the first deformation derivation network of the strain analysis is a matrix representing the strain distribution, and the derivation result of the first deformation derivation network of the displacement analysis is a vector representing the displacement direction and size, then through the weighted sum and other fusion operations, the result obtained is the first collaborative vector corresponding to the first encoding representation vector. This first collaborative vector comprehensively integrates the information of the overall stability of the soft rock, stress, strain and displacement, and provides comprehensive soft rock deformation related data for subsequent analysis.

[0081] In a possible implementation, the step S1422 comprises:

[0082] Step S1422-1, the plurality of first encoding representation vectors are combined in series to output a first combined encoding representation vector.

[0083] In this embodiment, during the tunnel construction process, the plurality of first encoding representation vectors are obtained by feature encoding representation of tunnel soft rock monitoring data. These vectors contain feature information of soft rock in different aspects. For example, each first encoding representation vector can contain stress, strain, displacement and other information of a specific area of soft rock, as well as the association information of these features with BIM deformation simulation conditions (such as different excavation depths, different support structures, etc.). These first encoding representation vectors are concatenated and combined in a certain order, which can be determined according to the spatial position of the soft rock in the tunnel or the numbering order of the monitoring points. Assuming that there are monitoring points numbered 1 to n from the entrance to the exit in the tunnel, the corresponding first encoding representation vectors are concatenated according to the numbering order of the monitoring points. The first combined encoding representation vector obtained in this way integrates various feature information of soft rock in different areas of the entire tunnel, providing a comprehensive data basis for subsequent acquisition of the second influence coefficient.

[0084] Step S1422-2, according to the first combined encoding representation vector, acquiring the second influence coefficient corresponding to the second collaborative network and the second deformation derivation network associated with the second feature transmission control network, respectively.

[0085] Step S1422-3, according to the second influence coefficient corresponding to the second collaborative network and the second deformation derivation network associated with the second feature transmission control network, respectively, fusing the derivation results of the second collaborative network and the derivation results of the second deformation derivation network associated with the second feature transmission control network, and outputting the third encoding representation vector corresponding to the BIM deformation simulation condition.

[0086] In the tunnel project, the derivation result of the second collaborative network can be an evaluation value of the overall deformation coordination of soft rock, for example, this value indicates that the overall deformation coordination of soft rock is good in the current tunnel construction stage, but there can be coordination problems in local areas. The derivation result of the second deformation derivation network of the stress change and deformation relationship of soft rock under a specific supporting structure can be the local deformation trend caused by the stress change of soft rock under the specific supporting structure, for example, due to the influence of the supporting structure, stress concentration can cause soft rock to produce a certain displacement towards the inside of the tunnel. According to the second influence coefficient obtained before, if the second influence coefficient of the second deformation derivation network of the stress change and deformation relationship of soft rock under a specific supporting structure is high, it means that the result of this network is relatively more important in this fusion process. Assuming that the derivation result weight of the second collaborative network is 0.4, and the derivation result weight of the second deformation derivation network of the stress change and deformation relationship of soft rock under a specific supporting structure is 0.6. By weighted summation and other fusion operations, the two results are fused together, and the result obtained is the third encoding representation vector corresponding to the BIM deformation simulation condition. This third encoding representation vector integrates the overall deformation coordination of soft rock and the local deformation trend under a specific supporting structure, and can more comprehensively reflect the characteristics of soft rock under the BIM deformation simulation condition, providing important data support for subsequent soft rock deformation risk assessment and other operations.

[0087] The step S1422-2 includes:

[0088] In step S1422-21, the second network parameter array corresponding to each second deformation derivation network and the second collaborative network is obtained. The longitudinal column quantity of the second network parameter array is the network quantity of the second deformation derivation network and the second collaborative network, and the horizontal row quantity of the second network parameter array is the feature category quantity of the first combined encoding representation vector.

[0089] For example, in the analysis of soft rock deformation, it is assumed that there are two second deformation derivation networks, one of which is dedicated to analyzing the stress change and deformation relationship of soft rock under a specific support structure, and the other is to analyze the strain and deformation relationship of soft rock under different drainage conditions, plus a second coordination network. The longitudinal quantity of the parameter array of this second network is the number of networks of the second deformation derivation network and the second coordination network, which is 3 here. And the horizontal quantity is the number of characteristic categories of the first combined code representation vector. Assuming that the first combined code representation vector contains 6 soft rock characteristic categories, the second network parameter array is a 3x6 matrix. For the second deformation derivation network of the stress change and deformation relationship of soft rock under a specific support structure, the elements in this matrix are determined according to a large number of historical stress deformation data of soft rock under similar support structures, physical properties of soft rock, and stress change rules under different BIM deformation simulation conditions. Similarly, for the second deformation derivation network and the second coordination network of the strain and deformation relationship of soft rock under different drainage conditions, their second network parameter arrays are also determined based on a variety of factors related to each other.

[0090] Step S1422-22, second dot product calculation is performed on the second network parameter array and the inverse arrangement vector of the first combined code representation vector.

[0091] The inverse arrangement vector of the first combined code representation vector is a vector obtained by inversely arranging the characteristic values originally arranged in a certain order. Taking the second deformation derivation network of the stress change and deformation relationship of soft rock under a specific support structure as an example, the inverse arrangement vector is dot product calculated with the second network parameter array. This dot product calculation follows the rules of matrix multiplication, multiplies each row element in the second network parameter array with the corresponding element in the inverse arrangement vector and adds them up to get an intermediate result. This intermediate result integrates the mutual relationship between the network parameters in the second network parameter array and the soft rock characteristic information in the first combined code representation vector. The same operation is also performed on the second deformation derivation network and the second coordination network of the strain and deformation relationship of soft rock under different drainage conditions to obtain their respective intermediate results.

[0092] Step S1422-23, the calculation result of the second dot product calculation is processed by an activation function, and the second influence coefficient is output.

[0093] In this tunnel project, the activation function that can be used is the sigmoid function. The intermediate result obtained in the foregoing is input into the sigmoid function, which processes the input value. The sigmoid function maps any real number input to the interval between 0 and 1. The result obtained after processing by the activation function is the second influence coefficient. For the second deformation derivation network of the stress change and deformation relationship of soft rock under a specific supporting structure, the second influence coefficient reflects the relative importance of the network in the entire fusion process based on the soft rock feature information contained in the first combination coding representation vector. For the second deformation derivation network and the second collaborative network of the strain and deformation relationship of soft rock under different drainage conditions, the second influence coefficient obtained also reflects their relative importance in the fusion process.

[0094] In one possible implementation, step S120 includes:

[0095] Step S121, the tunnel soft rock monitoring data is preliminarily classified to generate subsets of different types of data, feature recognition is performed for each type of data subset to generate basic features of each type of data subset, and the basic features include maximum stress value, minimum stress value, stress change frequency, strain peak value, strain valley value, and average change rate of strain.

[0096] In this embodiment, in the process of tunnel construction, the soft rock monitoring data is extensive and diverse in types. For example, a large amount of data is obtained through stress sensors, strain sensors, displacement sensors and other devices arranged inside and around the soft rock. These data contain various state information of the soft rock at different times and different positions. The monitoring data is classified according to the type of data, such as classifying all data obtained by stress sensors as a stress data subset, classifying data obtained by strain sensors as a strain data subset, and classifying data obtained by displacement sensors as a displacement data subset.

[0097] For example, in the stress data subset, the maximum stress value and the stress change frequency play a key role in judging the stress state of soft rock and predicting the deformation trend of soft rock under different BIM deformation simulation conditions, and they are classified into the main feature layer. The minimum stress value has relatively small influence on the overall stress state and can be classified into the secondary feature layer. For the features of the main feature layer, a more complex and accurate conversion method is adopted, such as using a nonlinear transformation function for conversion, which can highlight the importance of these key features in soft rock deformation analysis. For the features of the secondary feature layer, such as the minimum stress value, a relatively simple linear conversion method can be used. In the strain data subset, the strain peak value and the average change rate of strain are more critical for judging the strain state of soft rock and predicting the deformation trend, and they are classified into the main feature layer and converted using a complex nonlinear transformation similar to the main feature layer of the stress data; the strain valley value is classified into the secondary feature layer and converted using a linear conversion method.

[0098] In step S122, the basic features are layered according to their importance and relevance, and the features of the main feature layer and the secondary feature layer are generated. The features of the main feature layer and the secondary feature layer are converted using the corresponding conversion methods, respectively. The converted features of different levels are combined according to the internal correlation between the features to generate feature combination data with correlation.

[0099] For example, in the stress data subset, the maximum stress value and the stress change frequency play a key role in judging the stress state of soft rock and predicting the deformation trend of soft rock under different BIM deformation simulation conditions, and they are classified into the main feature layer. The minimum stress value has relatively small influence on the overall stress state and can be classified into the secondary feature layer. For the features of the main feature layer, a more complex and accurate conversion method is adopted, such as using a nonlinear transformation function for conversion, which can highlight the importance of these key features in soft rock deformation analysis. For the features of the secondary feature layer, such as the minimum stress value, a relatively simple linear conversion method can be used. In the strain data subset, the strain peak value and the average change rate of strain are more critical for judging the strain state of soft rock and predicting the deformation trend, and they are classified into the main feature layer and converted using a complex nonlinear transformation similar to the main feature layer of the stress data; the strain valley value is classified into the secondary feature layer and converted using a linear conversion method.

[0100] In the stress data subset, the converted maximum stress value and stress variation frequency (main feature layer feature) are combined with the converted minimum stress value (secondary feature layer feature) in a certain logical order. This logical order is determined based on the physical mechanism of soft rock stress variation, such as first considering the dominant influence of the main feature layer feature on the soft rock stress state, and then considering the supplementary influence of the secondary feature layer feature. Similarly, in the strain data subset, the converted strain peak value, average strain variation rate (main feature layer feature) are combined with the strain valley value (secondary feature layer feature) according to the internal logic of soft rock strain variation. Similar operations are also performed for the displacement data subset, combining displacement-related features at different levels together. Then, the feature combination data of different types of data subsets such as stress, strain and displacement are further combined according to the relationship between stress, strain and displacement in the soft rock deformation process, forming a feature combination data containing multiple soft rock features and having internal correlation.

[0101] Step S123, based on the pre-set encoding rule, an encoding mapping relationship is established for each feature combination data. The encoding rule is determined according to the feature range of soft rock deformation data and the expected encoding representation requirement.

[0102] The feature range of soft rock deformation data covers from the initial state of soft rock when it is not disturbed to the limit state after being affected by various factors during tunnel construction. For example, the stress value range may be from the static pressure of soft rock itself to the maximum stress value generated by construction operations such as tunnel excavation, blasting, etc.; the strain range from a small elastic strain to a limit strain that may cause soft rock failure; the displacement range from almost no displacement to the maximum allowable displacement that soft rock may appear. The expected encoding representation requirement is to accurately and comprehensively represent various feature combination situations of soft rock, so as to effectively analyze and process in the subsequent machine learning model. The encoding rule determined according to these factors establishes a unique encoding mapping relationship for each feature combination data.

[0103] Step S124, according to the encoding mapping relationship, each feature combination data is mapped into a corresponding encoding space, in which different encoding segments represent different feature combination situations.

[0104] Assuming that the encoding space is divided into multiple encoding segments, each encoding segment corresponds to a specific feature combination range. For example, a certain encoding segment may correspond to the feature combination situation of high stress, high strain peak value and large displacement, indicating that the soft rock is in a high deformation risk state; while another encoding segment may correspond to the feature combination situation of low stress, low strain peak value and small displacement, indicating that the soft rock is relatively stable. Each feature combination data is accurately mapped into the corresponding encoding segment according to its encoding mapping relationship.

[0105] Step S125, combining each encoding segment according to a preset logical order to construct an encoding representation vector.

[0106] In this embodiment, the preset logical order is determined based on the importance and mutual relationship between the soft rock characteristics. For example, the encoding segments are combined in the order of stress-related encoding segment, strain-related encoding segment, and displacement-related encoding segment. If the stress-related encoding segment represents that the stress state of the soft rock has the primary influence on the deformation of the soft rock, the strain-related encoding segment secondly, and the displacement-related encoding segment finally reflects the actual deformation result of the soft rock, then the encoding representation vector combined in this order can comprehensively and orderly represent various characteristic information of the soft rock. The encoding representation vector is the final output first encoding representation vector, which will be used for subsequent first machine learning model and other related operations, and provides a data basis for the derivation of the soft rock deformation risk trend.

[0107] In a possible implementation, after step S150, the method further includes:

[0108] Step S160, obtaining a key monitoring condition of the target tunnel section.

[0109] Step S170, obtaining a deformation risk trend for the key monitoring condition from the deformation risk trends for the plurality of BIM deformation simulation conditions.

[0110] Step S180, arranging the soft rock deformation control scheme according to the deformation risk trend for the key monitoring condition.

[0111] Step S190, applying the soft rock deformation control scheme to the target tunnel section according to the arrangement result.

[0112] Step S180 includes: if there are a plurality of key monitoring conditions, determining a global deformation risk trend according to the deformation risk trends of the plurality of key monitoring conditions. Arranging the soft rock deformation control scheme according to the global deformation risk trend.

[0113] In this tunnel project, there are some conditions that need to be focused on due to the special geological structure and complex construction environment of the target tunnel section. For example, the soft rock in some areas may have special geological structures, such as the presence of faults or alternating layers of soft and hard rock. The stability of soft rock in these areas is crucial to the safety of the entire tunnel construction, so the deformation of soft rock in these special geological structures becomes a key monitoring condition. In addition, some key processes during tunnel construction, such as the intensity and frequency of blasting operations, the installation time and quality of support structures, etc., also affect the deformation of soft rock. These process-related soft rock deformation conditions also belong to the key monitoring conditions. Through comprehensive analysis of geological exploration data, construction plans, and experience summaries from similar projects, etc., these key monitoring conditions are determined.

[0114] Next, from the deformation risk trends for multiple BIM deformation simulation conditions, the deformation risk trends for the key monitoring conditions are obtained. The deformation risk trends obtained through the deformation risk derivation model are comprehensive results under multiple BIM deformation simulation conditions. These results include the deformation risk of soft rock under different simulation conditions, such as different excavation speeds, different support structure types, etc. The deformation risk trends under the key monitoring conditions are the relevant parts extracted from the comprehensive results. For example, if the key monitoring condition is the deformation of soft rock in the fault area, then the soft rock deformation risk information related to the fault area is filtered from the overall deformation risk trend data. If the BIM deformation simulation conditions include different excavation speeds, then the deformation risk trends of soft rock in the fault area under different excavation speeds, such as the rapid increase in stress of soft rock in the fault area and the risk trend of strain exceeding the safety threshold, are the deformation risk trends obtained for this key monitoring condition.

[0115] Then, according to the deformation risk trends for the key monitoring conditions, the soft rock deformation control scheme is arranged. If there are multiple key monitoring conditions, the global deformation risk trend is determined based on the deformation risk trends of multiple key monitoring conditions. For example, in addition to the deformation of soft rock in the fault area, another key monitoring condition is the deformation of soft rock in a high humidity environment due to abundant groundwater. From the deformation risk trend data, it is obtained that the strength of soft rock in a high humidity environment decreases and is prone to large displacement. By combining the deformation risk trends of these two key monitoring conditions, the global deformation risk trend is determined. If the deformation risk trend of soft rock in the fault area indicates that soft rock has a high risk of damage under the existing soft rock deformation control scheme, and the deformation risk trend of soft rock in a high humidity environment shows that the stability of soft rock is also severely threatened, then the global deformation risk trend is that soft rock is in a high-risk state overall under these two key monitoring conditions.

[0116] According to this global deformation risk trend, the soft rock deformation control scheme is arranged. The soft rock deformation control scheme can include various measures, such as different types of support structures (such as bolt support, shotcrete support, etc.), drainage measures (such as setting drainage pipes, drainage holes, etc.), and adjustment of excavation process (such as controlling excavation speed, optimizing blasting parameters, etc.). If the global deformation risk trend is high risk, when arranging the soft rock deformation control scheme, those measures that can effectively reduce the risk of soft rock deformation will be given priority. For example, for high-risk soft rock deformation, measures such as increasing the density of anchor rods, lengthening the length of anchor rods, or increasing the strength of shotcrete are arranged in the front. At the same time, for the risk of soft rock deformation in a high-humidity environment, improving drainage measures (such as increasing the density of drainage pipes, increasing the diameter of drainage holes, etc.) is also arranged in a relatively high position. For the adjustment of the excavation process, if its effect on reducing the risk of soft rock deformation is relatively small, it is arranged in a relatively low position.

[0117] Finally, the soft rock deformation control scheme is applied to the target tunnel section according to the arrangement result. According to the arrangement order of the soft rock deformation control scheme determined in the foregoing, the schemes are gradually applied to the construction process of the target tunnel section. For example, the density and length of the anchor rods are first increased according to the design requirements, and then the improvement of the shotcrete support is carried out, and the improvement of the drainage measures is carried out. During the construction process, the deformation of the soft rock is continuously monitored, and the effect of the soft rock deformation control scheme is evaluated according to the actual monitoring data. If it is found that the risk of soft rock deformation has not been effectively controlled, it can be necessary to re-evaluate the key monitoring conditions, adjust the analysis of the deformation risk trend, or further optimize the soft rock deformation control scheme, so as to ensure that the soft rock deformation of the target tunnel section is in a controllable state, and the safety and smooth progress of the tunnel construction are ensured.

[0118] In a possible implementation, the plurality of tunnel soft rock monitoring data is derived from a sample learning data of the sample learning data sequence, the sample learning data carries a plurality of supervision data for a plurality of BIM deformation simulation conditions, and the supervision data represents deformation risk label data under the BIM deformation simulation condition for the sample learning data.

[0119] The training steps of the first machine learning model, the second machine learning model, and the deformation risk derivation model include:

[0120] In step S101, for each BIM deformation simulation condition, a training loss value of the BIM deformation simulation condition is determined according to the supervision data and the deformation risk trend for the BIM deformation simulation condition.

[0121] In step S102, a global loss value of the sample learning data is determined according to the training loss value.

[0122] Step S103, according to the global loss value, optimizing the first machine learning model, the second machine learning model and the deformation risk derivation model.

[0123] Step S102 includes: obtaining the condition influence coefficient of a plurality of BIM deformation simulation conditions, according to the condition influence coefficient, the training loss value of a plurality of BIM deformation simulation conditions is fused, and the global loss value is output.

[0124] In this embodiment, for each BIM deformation simulation condition, the training loss value of the BIM deformation simulation condition is determined according to the supervised data and the deformation risk trend for the BIM deformation simulation condition. In the tunnel construction scene, taking the soft rock under different excavation speeds as an example. The supervised data contains the actual deformation of the soft rock under the known excavation speed and the corresponding risk label, such as the actual monitoring data of the stress, strain, displacement and other parameters of the soft rock, and whether there is a crack, collapse and other risk marks. The deformation risk trend is the deformation risk trend of the soft rock under the excavation speed derived by the model. If the result derived by the model is greatly different from the actual deformation and risk label in the supervised data, the training loss value will be larger. For example, the model derives that the strain of the soft rock under a certain excavation speed should be within the safe range, but the supervised data shows that the actual strain has approached the failure strain and a collapse risk has occurred, which indicates that the prediction of the model under this BIM deformation simulation condition is inaccurate, resulting in an increase in the training loss value. This training loss value reflects the deviation of the prediction performance of the model under the BIM deformation simulation condition from the actual situation.

[0125] The global loss value of the sample learning data is determined according to the training loss value. In this process, first, the condition influence coefficients of multiple BIM deformation simulation conditions are obtained. In tunnel engineering, different BIM deformation simulation conditions have different degrees of influence on the overall soft rock deformation. For example, the excavation speed of soft rock may be more critical to the influence of soft rock deformation than the type of supporting structure, so the condition influence coefficient of the BIM deformation simulation condition corresponding to the excavation speed will be relatively high. Assuming that there are three BIM deformation simulation conditions, namely excavation speed, supporting structure type and blasting intensity, the condition influence coefficient of the excavation speed is determined to be 0.5 according to engineering experience and data analysis, the condition influence coefficient of the supporting structure type is 0.3, and the condition influence coefficient of the blasting intensity is 0.2. Then, according to these condition influence coefficients, the training loss values of multiple BIM deformation simulation conditions are fused to output the global loss value. If the training loss value corresponding to the excavation speed is 0.8, the training loss value corresponding to the supporting structure type is 0.6, and the training loss value corresponding to the blasting intensity is 0.4, then the global loss value is calculated as: 0.5×0.8 + 0.3×0.6 + 0.2×0.4 = 0.4 + 0.18 + 0.08 = 0.66. This global loss value comprehensively considers the performance of the model under different BIM deformation simulation conditions.

[0126] The first machine learning model, the second machine learning model and the deformation risk derivation model are optimized according to the global loss value. In the analysis of soft rock deformation in tunnel construction, when the global loss value is obtained, if the global loss value is large, it indicates that the overall prediction effect of the model is poor, and the model needs to be optimized. For the first machine learning model, the parameters of multiple first deformation derivation networks and first feature transmission control networks need to be adjusted. For example, in the first deformation derivation network of soft rock stress analysis, if the global loss value indicates that the model performs poorly under stress-related BIM deformation simulation conditions, the weight parameters, neuron connection methods, etc. in this network will be adjusted to improve its prediction accuracy of the relationship between soft rock stress and deformation. Similarly, the second deformation derivation network and the second feature transmission control network in the second machine learning model, and the parameters in the deformation risk derivation model will also be optimized according to the global loss value. For example, if the deformation risk derivation model deviates greatly from the supervision data when predicting the deformation risk trend of soft rock under multiple BIM deformation simulation conditions, the algorithm parameters used to analyze the relationship between soft rock features and deformation risk in the model will be adjusted, such as adjusting the weight coefficients of soft rock stress, strain, displacement, etc. in risk assessment. By continuously optimizing these models according to the global loss value, the prediction accuracy of the model for the deformation risk trend of soft rock can be improved, so as to better provide reliable decision basis for soft rock deformation control in tunnel construction.

[0127] In a possible implementation, the loading of each of the first encoding representation vectors into a first machine learning model corresponding to the first encoding representation vector comprises:

[0128] The loading of each of the first encoding representation vectors into a plurality of the first machine learning models corresponding to the first encoding representation vector outputs a plurality of reference deformation correlation vectors corresponding to the BIM deformation simulation condition and the first encoding representation vector.

[0129] According to a set rule, a deformation correlation vector corresponding to the BIM deformation simulation condition and the first encoding representation vector is determined according to the plurality of reference deformation correlation vectors.

[0130] In this embodiment, a plurality of first machine learning models are constructed for deformation analysis of soft rock during tunnel construction, each model having a specific function and focus. The first encoding representation vector contains various feature information of soft rock, such as stress characteristics, strain characteristics, displacement characteristics of soft rock, and information related to the BIM deformation simulation condition. When this first encoding representation vector is loaded into a plurality of first machine learning models, each model will process according to its own algorithm structure and pre-trained parameters.

[0131] Taking one of the first machine learning models as an example, this model may focus on the analysis of the relationship between soft rock stress change and soft rock deformation under the BIM deformation simulation condition. When the first encoding representation vector is loaded, the first deformation derivation network inside the model will derive according to the stress-related features in the vector, such as the maximum stress value, stress change frequency, etc., in combination with the BIM deformation simulation condition (such as the change of soft rock stress under different excavation speeds), through its internal algorithm. The first feature transfer control network will output a reference deformation correlation vector corresponding to the BIM deformation simulation condition and the first encoding representation vector according to the derivation result of the first deformation derivation network and other related information (such as part of the results of the first collaborative network, etc.). This reference deformation correlation vector contains the correlation information of soft rock deformation and stress-related factors under the analysis of this model and the BIM deformation simulation condition, such as the possible deformation direction and deformation range caused by stress change of soft rock under a specific excavation speed, etc.

[0132] For other first machine learning models, such as models focusing on soft rock strain analysis or displacement analysis and the relationship between soft rock deformation and the BIM deformation simulation condition, similar operations will be performed, respectively outputting their corresponding reference deformation correlation vectors. These reference deformation correlation vectors reflect the deformation correlation of soft rock under the BIM deformation simulation condition from different angles and focuses.

[0133] Then, according to a set rule, a deformation correlation vector corresponding to the BIM deformation simulation condition and the first encoding representation vector is determined according to the plurality of reference deformation correlation vectors. This set rule is determined based on the overall requirements of the soft rock deformation analysis and the importance of each reference deformation correlation vector. For example, the set rule can determine the final deformation correlation vector according to the weight of each first machine learning model in the entire soft rock deformation analysis system. If in the soft rock deformation analysis, the first machine learning model of stress analysis is considered more important, its weight can be set to 0.4, the weight of the first machine learning model of strain analysis is 0.3, and the weight of the first machine learning model of displacement analysis is 0.3.

[0134] In determining the deformation correlation vector, each reference deformation correlation vector is weighted according to these weights. Taking the reference deformation correlation vector output by the stress analysis model as an example, suppose this vector contains information such as the deformation direction and deformation amount of the soft rock under stress, if the weight of the stress analysis model is 0.4, then the proportion of information in this vector in the final deformation correlation vector will be calculated according to this weight. Similarly, the reference deformation correlation vectors output by the strain analysis and displacement analysis models are also calculated according to their respective weights. Through such weighted summation operations, multiple reference deformation correlation vectors are fused into a deformation correlation vector corresponding to the BIM deformation simulation condition and the first encoding representation vector. This final deformation correlation vector integrates soft rock deformation correlation information from different angles and can more comprehensively and accurately reflect the deformation of the soft rock under the BIM deformation simulation condition, providing an important data basis for subsequent soft rock deformation risk analysis and other operations.

[0135] Figure 2 The hardware structure of the BIM-based tunnel soft rock deformation data processing system 100 provided by the embodiment of the present application for implementing the above-mentioned BIM-based tunnel soft rock deformation data processing method is shown in FIG. 1. Figure 2 As shown in FIG. 1, the BIM-based tunnel soft rock deformation data processing system 100 can include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0136] The machine-readable storage medium 120 can store data and / or instructions. In some embodiments, the machine-readable storage medium 120 can store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 can store data and / or instructions used by the BIM-based tunnel soft rock deformation data processing system 100 to perform or use to complete the exemplary methods described in the present application.

[0137] In the implementation process, the one or more processors 110 execute the computer executable instructions stored in the machine readable storage medium 120, so that the processor 110 can execute the BIM-based tunnel soft rock deformation data processing method of the method embodiments as above. The processor 110, the machine readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the transceiving action of the communication unit 140.

[0138] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the BIM-based tunnel soft rock deformation data processing system 100 as above, which has similar implementation principles and technical effects, and the present embodiment will not be described here.

[0139] In addition, the embodiment of the present application also provides a readable storage medium, wherein the readable storage medium is pre-provided with computer executable instructions, and when the processor executes the computer executable instructions, the BIM-based tunnel soft rock deformation data processing method as above is realized.

[0140] It should be noted that, in order to simplify the description of the present application and help to understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A BIM-based method for processing tunnel soft rock deformation data, characterized in that, The method includes: Multiple tunnel soft rock monitoring data are acquired, and the tunnel soft rock monitoring data are used to deduce the deformation risk trend of the target tunnel section under multiple BIM deformation simulation conditions for the soft rock deformation control scheme. Multiple tunnel soft rock monitoring data are represented by feature encoding, and multiple first encoding representation vectors are output; Each of the first encoded representation vectors is loaded into a first machine learning model corresponding to the first encoded representation vector. The first machine learning model includes multiple first deformation derivation networks associated with multiple BIM deformation simulation conditions and multiple first feature transfer control networks. The first feature transfer control network outputs a deformation association vector corresponding to the BIM deformation simulation conditions and the first encoded representation vector based on the derivation results of the first deformation derivation network. Based on the BIM deformation simulation conditions and the multiple deformation association vectors corresponding to the multiple first coded representation vectors, output the second coded representation vector of the BIM deformation simulation conditions, load the second coded representation vector into the second machine learning model corresponding to the BIM deformation simulation conditions, and output the third coded representation vector corresponding to the BIM deformation simulation conditions; The third encoding representation vector corresponding to the multiple BIM deformation simulation conditions is loaded into the deformation risk derivation model corresponding to the multiple BIM deformation simulation conditions, and the derivation of the deformation risk trend of the target tunnel segment for the soft rock deformation control scheme for the multiple BIM deformation simulation conditions is output. The second machine learning model includes a second deformation derivation network associated with the BIM deformation simulation conditions and a second feature transfer control network associated with the BIM deformation simulation conditions; The step of loading the second encoded representation vector into the second machine learning model corresponding to the BIM deformation simulation conditions and outputting the third encoded representation vector corresponding to the BIM deformation simulation conditions includes: The second encoded vector is loaded into the second deformation derivation network corresponding to the BIM deformation simulation conditions, and the derivation result of the second deformation derivation network is output. The second feature transfer control network is invoked to generate the third coded representation vector corresponding to the BIM deformation simulation conditions based on the derivation results of the second deformation derivation network.

2. The BIM-based tunnel soft rock deformation data processing method according to claim 1, characterized in that, The first machine learning model further includes a first collaborative network, and the second machine learning model further includes a second collaborative network; Based on the derivation results of the first deformation derivation network, the first feature transfer control network outputs a deformation association vector corresponding to the BIM deformation simulation conditions and the first coded representation vector, including: If the first feature transfer control network is associated with a first deformation derivation network, the first feature transfer control network fuses the derivation result of the first collaborative network and the derivation result of the first deformation derivation network associated with the first feature transfer control network, and outputs the deformation association vector corresponding to the BIM deformation simulation conditions and the first coded representation vector. If the first feature transfer control network is associated with the first cooperative network, the first feature transfer control network fuses the derivation results of the first cooperative network and the derivation results of each of the first deformation derivation networks, and outputs the first cooperative vector corresponding to the first encoded representation vector. The step of calling the second feature transfer control network to generate the third encoded representation vector corresponding to the BIM deformation simulation conditions based on the derivation result of the second deformation derivation network includes: Based on the multiple first cooperative vectors corresponding to the multiple first encoded representation vectors, a first cooperative encoded representation vector is output and loaded into the second cooperative network; Based on the second feature transfer control network, the derivation results of the second collaborative network and the derivation results of the second deformation derivation network associated with the second feature transfer control network are fused together to output the third coded representation vector corresponding to the BIM deformation simulation conditions.

3. The BIM-based method for processing tunnel soft rock deformation data according to claim 2, characterized in that, The step of fusing the derivation results of the first collaborative network and the derivation results of the first deformation derivation network associated with the first feature transfer control network, and outputting the deformation association vector corresponding to the BIM deformation simulation conditions and the first coded representation vector, includes: Based on the first encoded representation vector, obtain the first influence coefficients corresponding to the first collaborative network and each of the first deformation derivation networks; Based on the first influence coefficient of the first collaborative network and the first deformation derivation network associated with the first feature transfer control network, the derivation results of the first collaborative network and the derivation results of the first deformation derivation network associated with the first feature transfer control network are fused together, and the deformation association vector corresponding to the BIM deformation simulation conditions and the first coded representation vector is output. The step of fusing the derivation results of the first collaborative network and the derivation results of each of the first deformation derivation networks to output a first collaborative vector corresponding to the first encoded representation vector includes: fusing the derivation results of the first collaborative network and the derivation results of each of the first deformation derivation networks according to the first influence coefficients corresponding to the first collaborative network and each of the first deformation derivation networks respectively, and outputting a first collaborative vector corresponding to the first encoded representation vector. The step of obtaining the first influence coefficients corresponding to the first cooperative network and each of the first deformation derivation networks based on the first encoded representation vector includes: Obtain the first network parameter arrays corresponding to the first collaborative network and each of the first deformation derivation networks. The number of columns in the first network parameter array is the number of networks in the first deformation derivation network and the first collaborative network. The number of rows in the first network parameter array is the number of feature categories in the first encoding representation vector. Perform a first dot product calculation on the inverse arrangement vector of the first network parameter array and the first encoded representation vector; The calculation result of the first dot product is processed by an activation function, and the first influence coefficient is output.

4. The BIM-based method for processing tunnel soft rock deformation data according to claim 2, characterized in that, The step involves fusing the derivation results of the second collaborative network and the derivation results of the second deformation derivation network associated with the second feature transfer control network, based on the second feature transfer control network, to output the third encoded representation vector corresponding to the BIM deformation simulation conditions, including: Multiple first-encoded representation vectors are concatenated and combined to output a first combined encoded representation vector; Based on the first combined encoding representation vector, obtain the second influence coefficients corresponding to the second cooperative network and the second deformation derivation network associated with the second feature transfer control network, respectively; Based on the second influence coefficients corresponding to the second collaborative network and the second deformation derivation network associated with the second feature transfer control network, the derivation results of the second collaborative network and the derivation results of the second deformation derivation network associated with the second feature transfer control network are fused together, and the third encoding representation vector corresponding to the BIM deformation simulation conditions is output. The step of obtaining the second influence coefficients corresponding to the second cooperative network and the second deformation derivation network associated with the second feature transfer control network based on the first combined encoding representation vector includes: Obtain the second network parameter arrays corresponding to the second collaborative network and each of the second deformation derivation networks. The number of columns in the second network parameter array is the number of networks in the second deformation derivation network and the second collaborative network. The number of rows in the second network parameter array is the number of feature categories in the first combined encoding representation vector. Perform a second dot product calculation on the inverse arrangement vector of the second network parameter array and the first combined encoded representation vector; The calculation result of the second dot product is processed by an activation function, and the second influence coefficient is output.

5. The BIM-based method for processing tunnel soft rock deformation data according to claim 1, characterized in that, The step of performing feature encoding representation on multiple tunnel soft rock monitoring data and outputting multiple first encoding representation vectors includes: The monitoring data of soft rock in the tunnel is initially classified to generate subsets of different types of data. Feature identification is performed on each type of data subset to generate basic features of each type of data subset. The basic features include maximum stress value, minimum stress value, stress change frequency, strain peak value, strain valley value, and average strain change rate. Based on the importance and relevance of the basic features, the basic features are layered to generate features of the primary feature layer and the secondary feature layer. The features of the primary feature layer and the secondary feature layer are then transformed using their respective transformation methods. The transformed features of different levels are combined according to the inherent relationship between the features to generate feature combination data with correlation. Based on pre-defined coding rules, a coding mapping relationship is established for each feature combination data; the coding rules are determined according to the feature range of soft rock deformation data and the expected coding representation requirements. According to the encoding mapping relationship, each feature combination data is mapped to a corresponding encoding space, in which different encoding segments represent different feature combinations; The coded segments are combined in a preset logical order to construct a coded representation vector.

6. The BIM-based method for processing tunnel soft rock deformation data according to claim 1, characterized in that, After loading the third coded representation vector corresponding to the multiple BIM deformation simulation conditions into the deformation risk derivation model corresponding to the multiple BIM deformation simulation conditions, and outputting the derived deformation risk trend of the target tunnel segment for the soft rock deformation control scheme under the multiple BIM deformation simulation conditions, the method further includes: Obtain the key monitoring conditions of the target tunnel section; From the deformation risk trends for multiple BIM deformation simulation conditions, obtain the deformation risk trend for the key monitoring conditions; Based on the deformation risk trends for the key monitoring conditions, the soft rock deformation control schemes are arranged in order; Based on the arrangement results, the soft rock deformation control scheme is applied to the target tunnel section; The step of arranging the soft rock deformation control schemes based on the deformation risk trends of the key monitoring conditions includes: If there are multiple key monitoring conditions, the global deformation risk trend is determined based on the deformation risk trends of the multiple key monitoring conditions. Based on the global deformation risk trend, the soft rock deformation control schemes are arranged.

7. The BIM-based method for processing tunnel soft rock deformation data according to claim 1, characterized in that, The multiple soft rock monitoring data of the tunnel are derived from a sample learning data sequence. The sample learning data carries multiple supervisory data for multiple BIM deformation simulation conditions. The supervisory data represents the deformation risk label data under the BIM deformation simulation conditions for the sample learning data. The training steps for the first machine learning model, the second machine learning model, and the deformation risk derivation model include: For each BIM deformation simulation condition, the training loss value of the BIM deformation simulation condition is determined based on the supervision data and the deformation risk trend for the BIM deformation simulation condition. Based on the training loss value, determine the global loss value for the sample learning data; Based on the global loss value, optimize the first machine learning model, the second machine learning model, and the deformation risk derivation model; The step of determining the global loss value of the sample learning data based on the training loss value includes: Obtain the conditional influence coefficients of multiple BIM deformation simulation conditions; Based on the conditional influence coefficient, the training loss values ​​of multiple BIM deformation simulation conditions are fused to output the global loss value.

8. The BIM-based method for processing tunnel soft rock deformation data according to claim 1, characterized in that, The step of loading each of the first encoded representation vectors into the first machine learning model corresponding to the first encoded representation vector includes: Each of the first encoded representation vectors is loaded into multiple first machine learning models corresponding to the first encoded representation vector, and multiple reference deformation associated vectors corresponding to the BIM deformation simulation conditions and the first encoded representation vectors are output. Based on the set rules and multiple reference deformation association vectors, a deformation association vector corresponding to the BIM deformation simulation conditions and the first coded representation vector is determined.

9. A BIM-based tunnel soft rock deformation data processing system, characterized in that, The BIM-based tunnel soft rock deformation data processing system includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the BIM-based tunnel soft rock deformation data processing method according to any one of claims 1-8.

Images