CN115374570A - Multi-source weighted training set construction method for deformation prediction of engineering tunnel crossing - Google Patents

Abstract

The invention discloses a method for constructing a multi-source weighted training set for deformation prediction of a crossing project tunnel, comprising: step S1, acquiring design data of the crossing project and corresponding tunnel deformation; step S2, processing the design data to obtain initial input Set DatasetV1; step S3, determine the weight of the information source according to the source of the design data; step S4, determine the adaptive weight of different data according to the value of the key feature of the data; step S5, combine the weights of steps S3 and S4 and use it as an input set features; step S6, using the input set in step S5 as an input training set, and using the tunnel deformation value as an output training set to form a training set. In the present invention, through structured and weighted processing of multi-source engineering information, a training set that can reflect the characteristics of the traversing engineering and is beneficial to the use of machine learning algorithms is obtained, effectively supporting the intelligent prediction of tunnel deformation, and providing quantification for related engineering evaluation in accordance with.

Description

A Multi-source Weighted Training Set Construction Method for Deformation Prediction Through Engineering Tunnels

technical field

The invention belongs to the technical field of tunnel crossing engineering, and in particular relates to a method for constructing a multi-source weighted training set for deformation prediction of tunnel crossing engineering.

Background technique

With the continuous development of computer processing power and intelligent algorithms, machine learning has gradually become an effective means for solving practical problems in the industry. Supervised learning in machine learning methods can directly establish the potential connection between input and output based on existing cases, thus avoiding complex intermediate links, and can predict the required quantity simply, quickly and accurately, which is the realization of intelligence in various fields. General tools for automation and automation. This supervised learning model is highly dependent on the training set used by the learning algorithm. Using a structured and clearly characterized training set can help the learning algorithm achieve the learning goal more efficiently.

For tunnel deformation through engineering, a large number of engineering practices and scholars' research have accumulated a lot of valuable data, which provides excellent conditions for machine learning to solve deformation prediction problems. However, these data sources are diverse, including data with different characteristics obtained from various research methods such as field measurements, numerical simulations, theoretical analysis, and model tests. At the same time, there are many unstructured data such as design drawings and cloud images. These data are difficult to be directly utilized by relevant learning algorithms. On the other hand, there are differences in the credibility of various information sources. At the same time, the deformation prediction is highly related to the attributes of the project itself, and the degree of fit of the case has an important impact on the prediction. Each learning case is considered to have the same value, and the training set should be adjusted according to needs and prior knowledge.

Therefore, there is a need for a training set generation method that can start from existing data and adapt to the requirements of the use of relevant learning algorithms, so as to solve the defects in the prior art.

Contents of the invention

In order to solve the above technical problems, the present invention provides a method for constructing a multi-source weighted training set for deformation prediction of engineering tunnels. According to the characteristics of crossing engineering, the reshaping data is weighted to obtain a training set that can be directly applied to machine learning algorithms, providing strong support for efficient prediction of tunnel deformation. In addition, a method and system for predicting the deformation of a crossing engineering tunnel based on a machine learning algorithm are also provided.

In order to achieve the above object, the technical scheme adopted in the present invention comprises:

A method for constructing a multi-source weighted training set for deformation prediction of an engineering tunnel, characterized in that it comprises the following steps:

Step S1, obtaining the design data of the crossing project and the corresponding tunnel deformation, wherein the design data is the input part of the training set, including engineering geological information, tunnel structure information, and construction information, and the tunnel deformation is the output part of the training set;

Step S2, divide it into numerical data and non-numeric data according to the characteristics of the design data in step S1, form the data set ASet with the numerical data, process the non-numeric data to obtain the data set Bset, combine Aset and Bset, get the initial input set DatasetV1;

Step S3, according to the source of the design data in the initial input set DatasetV1, adopt a certain measurement strategy to determine the weight of the information source;

Step S4, according to the values of the key features of the design data in the initial input set DatasetV1, determine the adaptability weights of different data;

Step S5, combining the weights in steps S3 and S4 and using it as the feature of the input set to obtain the weighted input set DatasetV2;

Step S6, using the weighted input dataset DatasetV2 in step S5 as the input training set, and using the tunnel deformation value as the output training set to form a standard training set TrainDataset.

According to an embodiment of the present invention, in step S2, the processing of the non-numeric data comprises adopting the One-Hot method to encode the non-numeric data, and the encoding process includes the following steps:

Step S2.1, filter data with non-numeric features and form a dataset FeatureSet. The total number of features in the dataset FeatureSet is denoted as m, and the dataset corresponding to each feature is denoted as FeatureSet _i (i=1, 2, ... , m);

Step S2.3. For each feature data set FeatureSet _i , divide n _i state features, and according to the value of each case, obtain the data set ConditionSet _i under the state feature representation, where n _i is the state feature under the feature the number of all discrete values; and

Step S2.4, combining m data sets ConditionSet _i represented by state features to obtain a non-numeric data state feature representation data set Bset.

According to an embodiment of the present invention, the sources of design data in the step S3 include theoretical analysis, numerical simulation, model test, and field monitoring.

According to an embodiment of the present invention, the weight of the information source in the step S3 is an indicator for measuring the importance of different data sources, and the specific measurement strategy may include:

(1) Expert scoring method, directly determine the score of each data source according to expert evaluation, so that data from the same source have the same weight;

(2) Interval random generation method, which stipulates the weight interval of each data source, and randomly generates weight within the weight interval according to its source for each case;

(3) Sorting and comparison method, sorting the importance of data sources, extracting multiple cases from the data set without replacement and sorting them according to the importance, and assigning weights from high to low according to the ranking, until all cases are extracted; as well as

(4) The comprehensive method, using the weights obtained by two or more methods respectively, to obtain the final weight by synthesis.

According to an embodiment of the present invention, the adaptive weight in the step S4 is to measure the closeness of the data to the typical working conditions, including the following steps:

Step S4.1, determine the important judgment feature value y _i (i=1, 2, ..., n) of typical working conditions, where n is the number of important judgment features (ie key features);

Step S4.2. Calculate the Jacquard distance D ₁ of the discrete features among the important decision features, the formula is:

In the above formula, x and y are the sets of discrete feature values of the case and typical working conditions respectively;

Step S4.3. Calculating the relative Euclidean distance D ₂ of the continuous features among the important decision features, the formula is:

In the above formula, x _i and y _i are the values of the continuous features of item i in the case and typical working conditions respectively, and l is the number of continuous features in the important judgment features; and

Step S4.4, combine the distances D ₁ and D ₂ to calculate the adaptive weight of each case, the formula is:

ω _fit =f(D ₁ )+g(D ₂ )

In the formula, both f(x) and g(x) are distance conversion functions, and the inverse proportional function whose inverse proportional coefficient is a positive number, the linear function whose slope is a negative number and the intercept is a positive number, or other monotonically decreasing values with a positive value range function.

According to an embodiment of the present invention, the weight combination in step S5 is based on the information source weight and adaptive weight of each case, and the comprehensive weight of the case is obtained through combination strategies such as scaling and distribution. The specific combination strategy includes:

(1) Accumulation method, which stipulates the proportion of sub-item weights to combined weights, and the result of cumulative addition in proportion is taken as combined weights;

(2) Cumulative multiplication, multiplying all sub-item weights, and taking the result of the product as the combined weight;

(3) Take the small method, select the minimum value in all sub-item weights as the combined weight;

(4) Take the big method, select the maximum value in all sub-item weights as the combined weight; and

(5) Random method, the minimum value of all sub-item weights is used as the lower boundary of the interval, the maximum value is used as the upper boundary of the interval, and the weights are randomly generated within the interval as the combination weight.

In addition, the present invention also provides a deformation prediction method and system for crossing engineering tunnels based on machine learning algorithms.

Compared with the prior art, the present invention has the following advantages:

(1) The present invention starts from the existing engineering data that is easy to obtain, and through digital coding transformation means, the multi-source data has a unified structured format, and performs One-hot processing on non-numerical features, avoiding non-numerical features. The data additivity loss problem greatly enhances the availability of the training set, thus supporting the use of related machine learning algorithms and providing a basis for rapid prediction.

(2) According to the characteristics of the traversing project, the present invention weights the samples in the training set under the premise of meeting the requirements of the learning algorithm, so as to incorporate the key characteristics of the traversing project at the data level in a targeted manner, so as to provide relevant learning algorithms A large number of key learning features.

Description of drawings

Fig. 1 is a schematic flow chart of a method for constructing a multi-source weighted training set for deformation prediction of a tunnel through an engineering tunnel according to an embodiment of the present invention; and

Fig. 2 is a schematic diagram of the generation process of the initial input set DatasetV1 in the method for constructing the multi-source weighted training set for deformation prediction of tunnel crossing according to the embodiment of the present invention.

Detailed ways

The present invention will be further described in detail through specific embodiments below in conjunction with the accompanying drawings, and the shown content is used to fully illustrate the content of the present invention, but not to limit the present invention.

Taking the prediction of the vertical deformation of an existing tunnel in the case of crossing construction as an example, the method for constructing a multi-source weighted training set for deformation prediction of a crossing engineering tunnel according to the present invention is described in detail. As shown in accompanying drawing 1, the scheme of the present invention may comprise the following steps:

Step S1. Obtain the design data of the crossing project and the corresponding tunnel deformation. The design data is the input part of the training set, which can include engineering geological information, tunnel structure information, construction information, etc., and the tunnel deformation is the output part of the training set:

For this embodiment, multi-channel data such as engineering design documents, published academic documents, and related monitoring reports were collected, and a total of 187 design data cases and corresponding tunnel vertical deformation values were obtained after sorting and screening.

Step S2, divide it into numerical data and non-numeric data according to the characteristics of the design data in step S1, form the data set ASet with the numerical data, process the non-numeric data to obtain the data set Bset, combine Aset and Bset, get the initial input set DatasetV1 (see Figure 2):

According to the case information, the common or transformable design data characteristics are sorted out, specifically including 30 engineering geological characteristics, 10 tunnel structure characteristics, and 5 construction information, a total of 45 characteristics, and the above data are divided into numerical and non-numerical type data:

The step of forming the data set ASet from numerical data includes filtering the data whose characteristic is numerical and forming the data set ASet. In the data of this embodiment, 42 features are numerical data, which together form the data set ASet.

Use the One-Hot method to encode non-numeric data. The encoding process includes:

Step S2.1, filter data with non-numeric features and form a dataset FeatureSet. The total number of features in the dataset FeatureSet is denoted as m, and the dataset corresponding to each feature is denoted as FeatureSet _i (i=1, 2, ... , m). In the data of this embodiment, there are 3 features of "lining type", "construction method" and "control measures" which are non-numeric data, thus forming a FeatureSet data set with the total number of features m=3, and the data sets of the three features are respectively Corresponding to FeatureSet ₁ , FeatureSet ₂ , and FeatureSet ₃ , as shown in Table 1.

Table 1: FeartureSet dataset

case number FeatureSet₁ FeatureSet₂ FeatureSet₃ 1 Shield segments shield method micro-disturbance grouting 2 shotcrete mine law none ... 187 Shield segments TBM method End reinforcement

Step S2.2. For each feature data set FeatureSet _i , divide n _i state features, and according to the value of each case, obtain the data set ConditionSet _i represented by the state feature, where n _i is the state feature under the feature The number of all discrete values. In this embodiment, taking FeatureSet ₂ composed of "construction methods" as an example, the feature values in 187 cases include four discrete values of "shield tunneling method", "TBM method", "mine method" and "open cut method". Value, based on which n ₂ = 4 state features are divided and correspond to the original discrete values respectively. When the case belongs to the discrete value, the feature value is 1, and when it does not belong to the feature value, the feature value is 0, thus generating the data set under the state feature representation ConditionSet ₂ , as shown in Table 2 for details.

Table 2: ConditionSet ₂ dataset

Step S2.3, combining m data sets ConditionSet _i represented by state features to obtain a non-numeric data state feature representation data set Bset.

After that, the initial input set DatasetV1 is further processed.

Step S3, according to the source of the design data, use a certain measurement strategy to determine the weight of the information source; the weight of the information source is an index to measure the importance of different data sources, and can be determined by expert scoring method, interval random generation method, sorting comparison method and comprehensive method, etc. Sure.

Sources of design data may include theoretical analysis, numerical simulation, model tests, and field monitoring. For this embodiment, after statistics, there are 15 cases based on theoretical analysis, 136 cases based on numerical simulation, 7 cases based on model test, and 29 cases based on field monitoring. In this embodiment, according to the expert scoring method, the importance weights of theoretical analysis, numerical simulation, model test, and on-site inspection are respectively 0.07, 0.24, 0.12, and 0.57 after weighing.

Step S4, according to the values of the key features of the design data, determine the adaptability weight of different data, the adaptability weight is to measure the closeness of the data to the typical working conditions:

In this embodiment, shield tunnels in soft soil areas are taken as a typical case. The training set constructed is mainly aimed at predicting settlement in soft soil areas. At the same time, referring to the potential influence laws of other working conditions, the adaptive weight is calculated according to the following specific steps :

Step S4.1. Determine important judgment feature values y _i (i=1, 2, . . . , n) of typical operating conditions, where n is the number of important judgment features (key features). In this embodiment, the two features of "soil layer modulus" and "construction method" are used as key features, and 3000kPa and "shield tunneling method" are used as key feature values. It is worth noting that the feature of "construction method" has been carried out in step S2. Transformation, where the transformed state characteristics are used as key characteristics. Among them, a typical working condition refers to one or more representative cases in the crossing project, and there are one or more important judgment characteristic values in some characteristics, which can reflect the characteristics of this type of working condition; key features refer to To distinguish one or more features of different types of cases, the type of prediction target can be used as the basis for selection, and the engineering experience and classification metrics agreed in the industry can be used as key feature values, which are well known to those skilled in the art;

Step S4.2. Calculate the Jacquard distance D ₁ of the discrete features among the important decision features, the formula is:

In the above formula, x and y are the sets of discrete feature values of the case and typical working conditions, respectively.

Step S4.3. Calculating the relative Euclidean distance D ₂ of the continuous features among the important decision features, the formula is:

In the above formula, x _i and y _i are the values of the continuous features of the i item of the case and typical features respectively, and l is the number of continuous features in the important decision features.

Step S4.4, combine the distances D ₁ and D ₂ to calculate the adaptive weight of each case, the formula is:

ω _fit =f(D ₁ )+g(D ₂ )

In the formula, f(x) and g(x) are distance conversion functions. In this embodiment, the two distance conversion functions both take an inverse proportional function with an inverse proportional coefficient equal to 1.

Step S5, combine the weights in steps S3 and S4 and use them as the features of the input set to obtain the weighted input set DatasetV2:

More specifically, in this embodiment, the information source weight is determined according to the four types of sources in S3, and the adaptability weight is determined according to the soft soil shield working condition in S4, and the cumulative multiplication method is adopted, that is, all sub-item weights are multiplied to obtain The result of the product is used as the combination weight, and the weight is added as a new feature to the data set.

Step S6, using the weighted input set DatasetV2 in step S5 as the input training set, and using the tunnel deformation value as the output training set to form a standard training set TrainDataset.

In this embodiment, the training set used by the machine learning algorithm can be obtained by using the input training set obtained in the previous steps and combining the vertical deformation value of the tunnel collected in the data.

Further, the present invention also provides a machine learning algorithm-based deformation prediction method for crossing engineering tunnels, which may include constructing a multi-source weighted training set using the method described in the present invention; and then using the multi-source weighted training set to train machine learning algorithm. Machine learning algorithms can be, for example, algorithms such as neural networks and decision trees, and more specifically, can include ANN, CNN, XGboost, lightGBM, and so on. After training, the machine learning algorithm is used to predict the deformation of the engineering tunnel.

In addition, the present invention also provides a machine learning algorithm-based deformation prediction system for crossing engineering tunnels, the system may include one or more processors and memory, and the memory stores instructions executable by the one or more processors , the instructions enable the system to execute the method steps according to the present invention.

The above description of the implementation examples is to facilitate the understanding and application of the present invention by those of ordinary skill in the technical field. It is obvious that those skilled in the art can easily make various modifications to these embodiments, and apply the general principles described here to other embodiments without creative efforts. Therefore, the present invention is not limited to the embodiments herein. Improvements and modifications made by those skilled in the art according to the disclosure of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims (10)

1. A method for building a multi-source weighted training set for deformation prediction through engineering tunnels, comprising the following steps: Step S1, obtaining the design data of the crossing project and the corresponding tunnel deformation data, wherein the design data is used as the input part of the training set, including engineering geological information, tunnel structure information and construction information, and the tunnel deformation is used as the output part of the training set; Step S2, according to the characteristics of the design data in step S1, the design data is divided into numerical data and non-numeric data, and the numerical data is formed into a data set ASet, and the non-numeric data is processed to obtain a data set Bset , combine Aset and Bset to get the initial input set DatasetV1; Step S3, according to the source of the design data in the initial input set DatasetV1, adopt a certain measurement strategy to determine the weight of the information source; Step S4, according to the values of the key features of the design data in the initial input set DatasetV1, determine the adaptability weights of different data; Step S5, combining the weights in steps S3 and S4 and using it as the feature of the input set to obtain the weighted input set DatasetV2; Step S6, using the weighted input set DatasetV2 in step S5 as an input training set, and using the tunnel deformation data as an output training set to form a standard training set TrainDataset. 2. A kind of multi-source weighted training set construction method for deformation prediction of crossing engineering tunnels as claimed in claim 1, characterized in that: in step S2, the processing of non-numerical data includes adopting the One-Hot method to To encode non-numeric data, the encoding process includes the following steps: Step S2.1, screen the data whose design features are non-numerical and form a dataset FeatureSet. The total number of features in the dataset FeatureSet is denoted as m, and the dataset corresponding to each feature is denoted as FeatureSet _i (i=1, 2, ... ..., m); Step S2.2. For each feature data set FeatureSet _i , divide n _i state features, and according to the value of each case, obtain the data set ConditionSet _i represented by the state feature, where n _i is the state feature under the feature the number of all discrete values; and Step S2.3, combining m data sets ConditionSet _i represented by state features to obtain a non-numeric data state feature representation data set Bset. 3. A kind of multi-source weighted training set construction method that is used for deformation prediction of crossing engineering tunnels as claimed in claim 1, is characterized in that: the design data source in the described step S3 comprises theoretical analysis, numerical simulation, model test and On-site monitoring. 4. A method for constructing a multi-source weighted training set for deformation prediction of crossing engineering tunnels as claimed in claim 1, characterized in that: the weight of information sources in the step S3 is an index for measuring the importance of different data sources, and the weight of Method selected from: (1) Expert scoring method, directly determine the score of each data source according to expert evaluation, so that data from the same source have the same weight; (2) Interval random generation method, which stipulates the weight interval of each data source, and randomly generates weight within the weight interval according to its source for each case; (3) Sorting and comparison method, sorting the importance of data sources, extracting multiple cases from the data set without replacement and sorting them according to the importance, and assigning weights from high to low according to the ranking, until all cases are extracted; as well as (4) Comprehensive method, using the weights obtained by two or more methods respectively, and combining the weights to obtain the final weight. 5. A method for constructing a multi-source weighted training set for deformation prediction of crossing engineering tunnels as claimed in claim 1, characterized in that: in the step S4, the adaptive weight is to measure the closeness of the data to the typical working conditions, Its determination includes the following steps: Step S4.1, determining key feature values y _i (i=1, 2, ..., n) of typical operating conditions, where n is the number of key features; Step S4.2. Calculate the Jacquard distance D ₁ of the discrete features among the key features, the formula is:

In the above formula, x and y are the sets of discrete feature values of the case and typical working conditions respectively; Step S4.3. Calculate the relative Euclidean distance D ₂ of the continuous features among the key features, the formula is:

In the above formula, x _i and y _i are the value of the continuous feature of item i in the case and typical working conditions respectively, and l is the number of continuous features in the key features; and Step S4.4, combine the distances D ₁ and D ₂ to calculate the adaptive weight of each case, the formula is: ω _fit =f(D ₁ )+g(D ₂ ) In the formula, f(x) and g(x) are distance conversion functions. 6. A method for constructing a multi-source weighted training set for deformation prediction of crossing engineering tunnels as claimed in claim 5, wherein the distance conversion function is selected from an inverse proportional function whose inverse proportional coefficient is a positive number, and the slope is a negative intercept A linear function of positive numbers or a monotonically decreasing function of positive numbers. 7. A method for constructing a multi-source weighted training set for deformation prediction of crossing engineering tunnels as claimed in claim 1, characterized in that: the weight combination in step S5 is based on information source weights and adaptive weights, by combining To get the comprehensive weight, the combination strategy is selected from: (1) Accumulation method, which stipulates the proportion of sub-item weights to combined weights, and the result of cumulative addition in proportion is taken as combined weights; (2) Cumulative multiplication, multiplying all sub-item weights, and taking the result of the product as the combined weight; (3) Take the small method, select the minimum value in all sub-item weights as the combined weight; (4) Take the big method, select the maximum value in all sub-item weights as the combined weight; and (5) Random method, the minimum value of all sub-item weights is used as the lower boundary of the interval, the maximum value is used as the upper boundary of the interval, and the weights are randomly generated within the interval as the combination weight. 8. A method for predicting deformation of tunnels through engineering based on machine learning algorithms, comprising the steps of: Utilize the method described in any one in claim 1-7 to construct training set; training a machine learning algorithm using the training set; and Using machine learning algorithms to predict deformation through engineering tunnels. 9. The method of claim 8, wherein the machine learning algorithm is selected from ANN, CNN, XGboost, and lightGBM. 10. A deformation prediction system for crossing engineering tunnels based on machine learning algorithms, comprising: one or more processors; and a memory, the memory stores instructions executable by the one or more processors, and the instructions make The system performs a method according to any one of claims 8-9.

Images