CN115130734A - Method and system for predicting construction influence of penetration project based on LightGBM and deep learning algorithm - Google Patents
Method and system for predicting construction influence of penetration project based on LightGBM and deep learning algorithm Download PDFInfo
- Publication number
- CN115130734A CN115130734A CN202210635344.4A CN202210635344A CN115130734A CN 115130734 A CN115130734 A CN 115130734A CN 202210635344 A CN202210635344 A CN 202210635344A CN 115130734 A CN115130734 A CN 115130734A
- Authority
- CN
- China
- Prior art keywords
- construction
- deep learning
- influence
- lightgbm
- data set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000010276 construction Methods 0.000 title claims abstract description 55
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 42
- 238000013135 deep learning Methods 0.000 title claims abstract description 38
- 230000035515 penetration Effects 0.000 title abstract description 3
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 8
- 238000004140 cleaning Methods 0.000 claims abstract description 6
- 238000012360 testing method Methods 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 238000013528 artificial neural network Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 238000012216 screening Methods 0.000 claims description 5
- 238000006073 displacement reaction Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 230000002860 competitive effect Effects 0.000 claims description 2
- 239000000463 material Substances 0.000 claims description 2
- 230000009897 systematic effect Effects 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- 230000014509 gene expression Effects 0.000 claims 1
- 238000012502 risk assessment Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 238000009933 burial Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000005056 compaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000002689 soil Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 238000005406 washing Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/08—Construction
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Human Resources & Organizations (AREA)
- Physics & Mathematics (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- General Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Health & Medical Sciences (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Primary Health Care (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method and a system for predicting construction influence of a penetration project based on a LightGBM and a deep learning algorithm, wherein the method comprises the following steps: step S1, collecting the information data related to the crossing project; step S2, digitalizing the information data, cleaning the initial data, and establishing input and output parameter data sets; s3, establishing a regression model through a LightGBM algorithm, training by using a data set and obtaining a key feature set; step S4, carrying out standardization processing on the data set; s5, constructing an initial deep learning network, and optimizing a network model hyper-parameter to enable the error of the network model to be smaller than a threshold value; and step S6, inputting the target engineering parameters to the deep learning network model, and predicting the construction influence value of the target engineering. The method can select key contents, directly establish mathematical and physical relations between the engineering information and the crossing influence quantity, efficiently and quickly realize prediction, and provide powerful support for risk assessment and scheme comparison.
Description
Technical Field
The invention belongs to the technical field of tunnel crossing engineering, and particularly relates to a method and a system for predicting crossing engineering construction influence based on LightGBM and a deep learning algorithm.
Background
With the acceleration of urbanization progress in China, the working conditions of mutual crossing among the existing subway tunnels are more frequent, the crossing engineering often causes deformation and deflection of the existing tunnels, settlement and deformation of stratums at crossing sections and changes of soil bodies and structural stress, and when the construction influence is strong, the normal use of the tunnels is influenced, and even safety accidents are generated when the construction influence is serious. Therefore, the influence of the construction on the existing engineering structure is always considered at the beginning of the design of the new engineering
In the existing design process, the influence of the traversing engineering is generally evaluated by methods such as engineering analogy, numerical calculation, expert demonstration and the like, the related evaluation is guided by experience, the existing engineering is taken as reference, the method highly depends on the engineering experience and the engineering cognitive level of designers and expert teams, an intuitive, quantitative and reproducible guiding method is difficult to generate, and the method is also more and more difficult to adapt to the construction current situation that the engineering complexity is increased more and the environmental control requirement is stricter. Therefore, a new technical method capable of solving the defects in the prior art from the past engineering data is needed.
Disclosure of Invention
In order to solve the technical problems, the invention provides a method and a system for predicting the construction influence of the traversing engineering based on a LightGBM and a deep learning algorithm, which aim to screen key contents in massive engineering information, construct a potential relation between the engineering information and the influence information by using a machine learning means, and create an end-to-end calculation model of the engineering information and the influence information, thereby efficiently and quickly predicting the construction influence of the traversing engineering and providing powerful support for risk assessment and scheme selection in the later period.
In order to achieve the purpose, the technical scheme adopted by the invention comprises the following steps:
step S1, collecting relevant data of the crossing project, including information such as construction elements, geological conditions, structural attributes and construction influence values, wherein the construction elements include information such as construction methods, control measures and tunnel location relations, the geological conditions include information such as stratum distribution, engineering geological parameters and hydrological conditions, the structural attributes include information such as structure dimensions and material attributes, and the construction influence values include information such as tunnel displacement, tunnel deformation, joint opening amount, structural stress and stratum displacement;
step S2, performing digital processing on the information data and cleaning the initial data, and establishing a first input parameter data set InputV1 and a first output parameter data set OutputV1 of the construction influence value corresponding to the first input parameter data set InputV 1;
step S3, establishing a regression model ModelA through a LightGBM algorithm, training by using a first input parameter data set InputV1 and a first output parameter data set OutputV1 in S2, and obtaining a second input parameter data set InputV2 only containing key features;
step S4, carrying out standardization processing on a second input parameter data set InputV2 and a first output parameter data set OutputV1 to respectively obtain a third input parameter data set InputV3 and a second output parameter data set OutputV2, and dividing the third input parameter data set InputV3 and the second output parameter data set OutputV2 into a training set and a testing set according to a certain division strategy;
step S5, constructing an initial deep learning network model ModelB, and optimizing the hyper-parameters of the deep network model to enable the error of the network model to be smaller than a threshold value;
and step S6, inputting relevant parameters of the target project into the deep network model, and predicting the construction influence value of the target project.
According to the embodiment of the present invention, in the step S2, the digitizing process is to digitally represent non-numerical parameters, such as the type of construction method, control measures, etc., so as to have a discrete or continuous digital representation.
According to an embodiment of the present invention, in the step S2, the data cleansing means performing a certain mathematical transformation on the initial data to make it meet the use requirement of the related algorithm, and specifically includes the following steps:
s2.1: filling missing data, and filling the missing values in the original data by using empirical values, statistical quantities or guesses obtained according to other data;
s2.2: unifying the numerical apertures, unifying and converting the measurement units in the data set;
s2.3: removing abnormal values, defining effective ranges of all variables in the data set, and deleting abnormal data beyond the ranges; and
s2.4: and removing repeated values, compressing all data information of each case into a one-dimensional array, comparing each case, and deleting completely repeated data.
According to an embodiment of the present invention, the step S3 includes the following specific steps:
s3.1: setting a LightGBM model hyper-parameter according to the number of data set samples and the number of data lumped characteristics, and establishing a regression model A;
s3.2: randomly selecting 90% of samples as a training set, and selecting 10% of samples as a testing set, and training ModelA;
s3.3: sorting the feature importance of the input parameters according to a model of ModelA, and taking the features of a ranking preamble as key features according to a certain screening rule; and
s3.4: the parameter data set InputV1 is divided to obtain an input parameter data set InputV2 which only contains key features.
According to an embodiment of the present invention, in the step S4, the normalization refers to normalizing the data magnitude value to be within the interval of [0,1], and the specific operation can be expressed as:
in the above formula, x i And
sample data before and after normalization, x max And x min Respectively the maximum and minimum values in all sample data.
According to the embodiment of the present invention, in the step S4, the dividing strategy is to adopt hierarchical sampling when there is a significant proportion difference in a certain characteristic according to the distribution characteristics of the samples, and adopt simple random sampling or systematic sampling when there is no significant difference, so as to sample a certain number of samples as a training set, and the rest of samples that are not sampled are used as a test set.
According to an embodiment of the present invention, in step S5, the deep learning network model is a multilayer fully-connected neural network, which specifically includes the following steps:
s5.1: an initial network structure is defined. The network model comprises an input layer, N hidden layers and an output layer, wherein the number of the initial hidden layers can be 3. The number M of the nodes of the input layer is equal to the number of the key features, the number of the nodes of the output layer is one, and the number of the nodes of the initial hidden layer can be 2M;
s5.2: a transfer algorithm is determined. The calculation method of each node of the hidden layer and the output layer comprises the following steps:
in the above formula, y is the current node output value, w i As a weight of the layer network, x i Taking the input value of the previous layer, n is the number of the input values of the previous layer, b is the offset of the layer, f is an activation function, and the functions of sigmoid, Relu, tanh and the like can be taken initially; and
s5.3: and calculating the error between the predicted value and the actual value of the network, and updating the network weight by adopting a back propagation algorithm until the specified iteration times are reached.
According to the embodiment of the present invention, in step S5, the hyper-parameters include the number of hidden layers, the number of hidden layer nodes, the types of activation functions, the learning rate, and the like, when optimizing the hyper-parameters, the possible values of each hyper-parameter should be listed, when the number of value combinations is small, network search is adopted for optimization, and when the number of values is large, methods such as random search, evolutionary algorithm, bayesian optimization, and the like are adopted for optimization.
According to an embodiment of the present invention, the filtering rule in step S5 may be formulated according to the following principles:
(1) selecting the characteristics of the O bits before the characteristic importance degree ranking, wherein O is one third to two thirds of the total characteristic number;
(2) selecting the characteristics with the characteristic importance degree larger than P, wherein P is one third to one half of the maximum characteristic importance degree;
(3) adopting a competitive bidding competition algorithm, randomly returning and extracting two characteristics from all the characteristics, comparing the importance degrees of the characteristics, selecting the characteristics with larger importance degrees as important characteristics, and repeating a Q round, wherein Q can be one half to two thirds of the total characteristic number;
according to another aspect of the present invention, there is provided a system for predicting influence of tunnel traversing engineering construction based on LightGBM and deep learning algorithm, comprising: one or more processors; and a memory storing instructions executable by the one or more processors to cause the automatic identification system to perform the method according to the present invention.
Compared with the prior art, the invention has the following advantages:
(1) the invention applies the deep learning algorithm and can fit the nonlinear relation with high precision, thereby constructing the mathematical relation between the engineering information and the influence information, reducing the time of manual calculation and analysis and simplifying and visualizing the influence prediction method.
(2) According to the method, the importance degree of all features is extracted by applying the characteristics of the tree algorithm in the LightGBM, key influence factors in engineering information are screened in advance, the prediction complexity of the deep neural network is reduced, and the number of potential invalid nodes is reduced, so that the requirements of the deep neural network on input parameters are met, the prediction precision is greatly enhanced, meanwhile, the related feature ordering is beneficial to mutual verification with the existing engineering cognition, mechanism analysis and the like, and the model reliability is improved.
(3) The algorithm used by the invention has high calculation speed and low calculation cost, and can predict the engineering influence at high speed according to the engineering information. Compared with the common methods such as numerical simulation, model test and the like, the prediction model can be called repeatedly and quickly after being trained, is beneficial to quick evaluation when the engineering parameters change and the construction scheme needs to be adjusted, and effectively promotes the development of influence prediction to the direction of automation and intellectualization.
Drawings
Fig. 1 is a schematic flow chart of a method for predicting influence of tunnel traversing engineering construction based on LightGBM and a deep learning algorithm according to an embodiment of the present invention;
fig. 2 is a schematic diagram illustrating a process of generating input and output data sets of a method for predicting influence of tunnel traversal engineering construction based on a LightGBM and a deep learning algorithm according to an embodiment of the present invention;
fig. 3 is a schematic diagram illustrating a key feature screening process of a method for predicting influence of tunnel cross-over engineering construction based on LightGBM and a deep learning algorithm according to an embodiment of the present invention; and
fig. 4 is a schematic diagram of a deep learning network model of a method for predicting influence of tunnel traversal engineering construction based on LightGBM and a deep learning algorithm according to an embodiment of the present invention.
Detailed Description
The present invention will now be described in further detail by way of specific embodiments with reference to the accompanying drawings, which are included to illustrate and not to limit the invention in its broadest sense.
The method for predicting the influence of the traversing engineering construction based on the LightGBM and the deep learning algorithm is explained in detail by taking the prediction of the influence of the traversing construction of the newly built shield tunnel on the existing shield tunnel as an example. As shown in fig. 1, the method of the present invention is performed as follows:
step S1, collecting crossing project related data including necessary information such as construction elements, geological conditions, structural attributes and construction influence values:
for the embodiment, relevant design, calculation, numerical simulation results and monitoring data of a plurality of subway crossing projects are collected, and construction elements in the data comprise the buried depth of a newly-built tunnel, the stratum loss rate generated by the construction of the newly-built tunnel, the horizontal direction angle, the vertical direction angle and the clear distance between the newly-built tunnel and an existing tunnel; the geological conditions comprise main stratum information related to the crossing engineering, and the specific information of each stratum comprises stratum thickness, dry weight, saturated weight, secant rigidity of a triaxial test, tangential rigidity of a main compaction test, unloading rigidity, a friction angle, cohesive force, an expansion angle and water level burial depth; the structural attributes comprise the outer diameter, the thickness, the lining weight, the lining elastic modulus and the Poisson ratio of the newly-built tunnel and the existing tunnel; the construction influence values comprise maximum stratum settlement above the existing tunnel, convergence deformation of the tunnel and maximum pressure stress of the existing tunnel structure.
Step S2, digitalizing the information and cleaning the initial data, establishing an input and output parameter data set:
all data are digitized and collected into a table form, each row of the table represents a sample, and each column represents a feature information. The flow of processing data on the initial table is shown in fig. 3, wherein the specific steps of cleaning data are as follows:
s2.1: filling missing data; filling missing values in the original data with empirical values, statistical quantities or guesses derived from other data;
s2.2: the numerical apertures are unified; unifying and converting the measurement units in the data set;
s2.3: removing abnormal values; defining effective ranges of all variables in a data set, and deleting abnormal data beyond the ranges;
s2.4: removing the repetition value; compressing all data information of each case into a one-dimensional array, comparing each case, and deleting completely repeated data.
The original data sample size was 1143, and 994 valid samples were obtained after washing. Taking engineering information as input information, obtaining an input parameter data set InputV1 which comprises 5 construction elements, 30 geological conditions and 10 structural attributes and has 45 characteristics in total; the construction influence value is used as output information to obtain 3 output parameter data sets, and in the embodiment, one of the parameters is taken for subsequent description, that is, the maximum stratum settlement above the existing tunnel is used as an output data set OutputV 1. It should be understood that other output parameters may be used to obtain the corresponding prediction model as follows.
Step S3, establishing a regression model through a LightGBM algorithm, training by using a data set, and obtaining a key feature set according to a training result:
the input parameter data set InputV1 is a matrix of 994 × 39, and the output parameter data set OutputV1 is a matrix of 994 × 1, so that key features are screened by using the LightGBM algorithm according to a data structure, and the flow is shown in fig. 3, and the specific steps are as follows:
s3.1: and setting a LightGBM model hyperparameter according to the number of the data set samples and the number of the data set features, and establishing a regression model ModlA. In this embodiment, the number of samples is less than 1000, the number of features is less than 50, and the hyper-parameters may be set according to default values without special setting.
S3.2: and randomly selecting 90% of samples as a training set and 10% of samples as a testing set, and training the Model A.
S3.3: and sequencing the feature importance of the input parameters according to a model of ModelA, and taking the features of the ranking preambles as key features according to a certain screening rule. The top 30 bits are ranked according to feature importance as key features.
S3.4: according to the key features, the parameter data set InputV1 is divided, for example, a data set including the key features is selected, and an input parameter data set InputV2 including only the key features is obtained.
Step S4, carrying out standardization processing on the data set, and dividing the data set into a training set and a test set according to a dividing strategy:
the key characteristic data set InputV2 and the output parameter data set OutputV1 are subjected to standardization processing, the data magnitude is normalized to be within a [0,1] interval, and the concrete operation is as follows:
in the above formula, x i And
sample data before and after normalization, x max And x min Respectively the maximum and minimum values in all sample data.
In the embodiment, the data has no obvious hierarchical grading, so that 90% of samples can be selected as a training set and the rest 10% can be selected as a test set according to a simple random sampling method. Of course, other suitable partitioning strategies may be employed.
Step S5, constructing an initial deep learning network model ModelB, and optimizing a network model hyper-parameter to enable a network model error to be smaller than a threshold value:
more specifically, a multilayer fully-connected neural network can be constructed according to a training set and a test set, and the specific steps are as follows:
s5.1: an initial network structure is defined. The network model includes an input layer, N hidden layers and an output layer, and in this embodiment, the number of the initial hidden layers is 3. The number M of the nodes of the input layer is equal to the number of the key features, the number of the nodes of the output layer is one, and the number of the nodes of the initial hidden layer in the embodiment is 30;
s5.2: a transfer algorithm is determined. The calculation method of each node of the hidden layer and the output layer comprises the following steps:
in the above formula, y is the current node output value, w i As a weight of the layer network, x i Taking a previous layer input value, n is the number of the previous layer input values, b is the layer offset, f is an activation function, and functions such as sigmoid, Relu and tanh can be taken initially, and the Relu function is taken in the embodiment;
s5.3: and calculating the error between the predicted value and the actual value of the network, and updating the network weight by adopting a back propagation algorithm until the specified iteration times are reached. In the embodiment, the average absolute error is taken as the error, and 5000 times are taken as the iteration times.
Through the training, the error of the initial deep learning network model ModelB on the test set is 3.59; in order to meet the requirement of engineering precision, a threshold value can be set to be 1.5 (absolute error, mm), then network search is carried out on the hyper-parameters, the set hyper-parameters comprise the number of hidden layers, the number of hidden layer nodes, the types of activation functions, regularization coefficients, learning rates and the like, and the search values are shown in table 1.
TABLE 1 hyper-parameter value range for deep learning network
| Hyper-parameter | Value range |
| Number of hidden layers | [3,5] |
| Number of hidden layer nodes | [5,40] |
| Class of activation function | {‘sigmoid’,‘Relu’,‘tanh’} |
| Regularization coefficients | [0.0001,0.001] |
| Learning rate | [0.0001,0.001] |
Through hyper-parameter optimization, the training effect is best when the network structure is as shown in fig. 4, the number of hidden layers of the deep learning network model is 3, the hidden layers are respectively provided with 15, 10 and 5 nodes, the activation function is tanh, and the regularization coefficient and the learning rate are both 0.0001. The average mean square error of the optimized deep learning network model on the test set is 1.29, which means that the average absolute error of the model for predicting the ground surface settlement above the tunnel is 1.29mm and is smaller than a threshold value, the effect is good, the engineering precision requirement is met, and good influence prediction can be provided for the construction of the crossing engineering. It should be appreciated that the above threshold values may be readily determined on a project-specific basis, with reference to relevant engineering criteria.
Step S6, after the model is established and verified, the model may be used for prediction, that is, the target engineering parameters may be input into the deep learning network model to predict the construction influence value of the target engineering. In the embodiment, the deep learning network model obtained in the preamble step can be used for predicting the stratum settlement above the tunnel according to the key characteristics of other penetrating projects.
The embodiment of the invention also provides a system for predicting the construction influence of tunnel crossing engineering based on the LightGBM and the deep learning algorithm, which can comprise one or more processors; and a memory storing instructions executable by the one or more processors to cause the automatic identification system to perform the automatic identification method according to the present invention.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. It will be readily apparent to those skilled in the art that various modifications to these embodiments may be made, and the generic principles described herein may be applied to other embodiments without the use of the inventive faculty. Therefore, the present invention is not limited to the embodiments described herein, and those skilled in the art should make improvements and modifications within the scope of the present invention based on the disclosure of the present invention.
Claims (10)
1. A method for predicting influence of tunnel crossing engineering construction based on LightGBM and a deep learning algorithm is characterized by comprising the following steps:
step S1, collecting the information data related to the crossing engineering, including construction elements, geological conditions, structural attributes and construction influence values, wherein the construction elements include construction methods, control measures and tunnel location relations, the geological conditions include stratum distribution, engineering geological parameters and hydrologic conditions, the structural attributes include structural dimensions and material attributes, and the construction influence values include tunnel displacement, tunnel deformation, joint opening, structural stress and stratum displacement;
step S2, digitalizing the information data and cleaning the initial data, and establishing a first input parameter data set InputV1 and a first output parameter data set OutputV1 of the construction influence value corresponding to the first input parameter data set InputV 1;
step S3, establishing a regression model ModelA through a LightGBM algorithm, training by using a first input parameter data set InputV1 and a first output parameter data set OutputV1 in S2, and obtaining a second input parameter data set InputV2 only containing key features;
step S4, carrying out standardization processing on a second input parameter data set InputV2 and a first output parameter data set OutputV1 to respectively obtain a third input parameter data set InputV3 and a second output parameter data set OutputV2, and dividing a third input parameter data set InputV3 and a second output parameter data set OutputV2 into a training set and a testing set according to a certain division strategy;
step S5, constructing an initial deep learning network model ModelB, and optimizing the hyper-parameters of the deep network model to enable the error of the network model to be smaller than a threshold value;
and step S6, inputting the relevant parameters of the target project into the deep learning network model, and predicting the construction influence value of the target project.
2. The method for predicting influence of tunnel crossing engineering construction based on LightGBM and deep learning algorithm as claimed in claim 1, wherein: the digitization processing in step S2 is to digitally characterize the non-numerical parameters so that they have discrete or continuous numerical expressions.
3. The method of claim 1, wherein the method for predicting the influence of the construction of the tunnel crossing engineering based on the LightGBM and the deep learning algorithm comprises the following steps: the data cleaning in step S2 is to perform mathematical transformation on the initial data to make the initial data meet the use requirements of the related algorithm, and specifically includes the following steps:
s2.1: filling missing data, and filling missing values in original data by using empirical values, statistical quantities or guessing quantities;
s2.2: unifying the numerical apertures, unifying and converting the measurement units in the data set;
s2.3: removing abnormal values, defining effective ranges of all variables in the data set, and deleting abnormal data beyond the ranges; and
s2.4: and removing repeated values, compressing all data information of each case into a one-dimensional array, comparing each case, and deleting completely repeated data.
4. The method of claim 1, wherein the method for predicting the influence of the construction of the tunnel crossing engineering based on the LightGBM and the deep learning algorithm comprises the following steps: the step S3 includes the following specific steps:
s3.1: setting a LightGBM model hyper-parameter according to the number of data set samples and the number of data lumped characteristics, and establishing a regression model A;
s3.2: randomly selecting 90% of samples as a training set, and 10% of samples as a testing set, and training ModelA;
s3.3: sorting the feature importance of the input parameters according to ModelA, and taking the features of the ranking preambles as key features according to a certain screening rule; and
s3.4: based on the key features, the first input parameter dataset InputV1 is divided, resulting in a second input parameter dataset InputV2 that only contains key features.
5. The method of claim 1, wherein the method for predicting the influence of the construction of the tunnel crossing engineering based on the LightGBM and the deep learning algorithm comprises the following steps: the normalization in step S4 is to normalize the data magnitude to the interval [0,1], and the specific operation can be expressed as:
in the above formula, x i And
sample data before and after normalization, x max And x min Respectively the maximum and minimum values in all sample data.
6. The method for predicting influence of tunnel crossing engineering construction based on LightGBM and deep learning algorithm as claimed in claim 1, wherein: the dividing strategy in step S4 is to adopt hierarchical sampling when there is a significant proportion difference in a certain characteristic according to the distribution characteristics of the samples, and adopt simple random sampling or systematic sampling when there is no significant difference, so as to sample a certain number of samples as a training set, and the rest of the non-sampled samples as a test set.
7. The method of claim 1, wherein the method for predicting the influence of the construction of the tunnel crossing engineering based on the LightGBM and the deep learning algorithm comprises the following steps: the deep learning network model ModelB in the step S5 is a multilayer fully-connected neural network, and the construction of the multilayer fully-connected neural network comprises the following steps:
s5.1: defining an initial network structure, wherein a network model comprises an input layer, N hidden layers and an output layer; the number M of the input layer nodes is equal to the number of the key features, and the number of the output layer nodes is one;
s5.2: the calculation method for determining the transfer algorithm and each node of the hidden layer and the output layer comprises the following steps:
in the above formula, y is the current node output value, w i As a weight of the layer network, x i Is input for the previous layerThe value n is the number of input values of the previous layer, b is the offset of the layer, f is an activation function and is selected from sigmoid, Relu and tanh functions initially; and
s5.3: and calculating the error between the predicted value and the actual value of the network, and updating the network weight by adopting a back propagation algorithm until the specified iteration times are reached.
8. The method of claim 1, wherein the method for predicting the influence of the construction of the tunnel crossing engineering based on the LightGBM and the deep learning algorithm comprises the following steps: the hyper-parameters in the step S5 include the number of hidden layers, the number of hidden layer nodes, the types of activation functions, and the learning rate, the possible values of each hyper-parameter are listed during the hyper-parameters during optimization, network search is adopted for optimization when the number of value combinations is small, and a method selected from random search, evolutionary algorithm, and bayesian optimization is adopted for optimization when the number of value combinations is large.
9. The method of claim 4, wherein the method for predicting the influence of the construction of the tunnel crossing project based on the LightGBM and the deep learning algorithm comprises the following steps: the screening rule is formulated according to the following principles:
(1) selecting the characteristics of the O bits before the characteristic importance degree ranking, wherein O is one third to two thirds of the total characteristic number;
(2) selecting features with feature importance degrees larger than P, wherein P is one third to one half of the maximum feature importance degree; and
(3) and (3) adopting a competitive bidding competition algorithm, randomly returning and extracting two characteristics from all the characteristics, comparing the importance degrees of the characteristics, selecting the characteristics with larger importance degrees as the important characteristics, and repeating the Q round, wherein Q can take one half to two thirds of the total characteristic number.
10. A system for predicting tunnel crossing engineering construction influence based on LightGBM and a deep learning algorithm is characterized by comprising the following steps: one or more processors; and a memory storing instructions executable by the one or more processors to cause the automatic identification system to perform the method of any of claims 1-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210635344.4A CN115130734A (en) | 2022-06-06 | 2022-06-06 | Method and system for predicting construction influence of penetration project based on LightGBM and deep learning algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210635344.4A CN115130734A (en) | 2022-06-06 | 2022-06-06 | Method and system for predicting construction influence of penetration project based on LightGBM and deep learning algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115130734A true CN115130734A (en) | 2022-09-30 |
Family
ID=83378825
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210635344.4A Pending CN115130734A (en) | 2022-06-06 | 2022-06-06 | Method and system for predicting construction influence of penetration project based on LightGBM and deep learning algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115130734A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116108547A (en) * | 2023-04-11 | 2023-05-12 | 北京城建设计发展集团股份有限公司 | Method and system for pile-up design and dynamic control in shield tunnel |
| CN116318172A (en) * | 2023-05-25 | 2023-06-23 | 山东华云三维科技有限公司 | Design simulation software data self-adaptive compression method |
| CN116992298A (en) * | 2023-09-28 | 2023-11-03 | 青岛东捷建设集团有限公司 | High-pressure jet grouting method and system based on seawater seepage |
| CN117522173A (en) * | 2024-01-04 | 2024-02-06 | 山东科技大学 | Natural gas hydrate depressurization exploitation productivity prediction method based on deep neural network |
-
2022
- 2022-06-06 CN CN202210635344.4A patent/CN115130734A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116108547A (en) * | 2023-04-11 | 2023-05-12 | 北京城建设计发展集团股份有限公司 | Method and system for pile-up design and dynamic control in shield tunnel |
| CN116318172A (en) * | 2023-05-25 | 2023-06-23 | 山东华云三维科技有限公司 | Design simulation software data self-adaptive compression method |
| CN116992298A (en) * | 2023-09-28 | 2023-11-03 | 青岛东捷建设集团有限公司 | High-pressure jet grouting method and system based on seawater seepage |
| CN116992298B (en) * | 2023-09-28 | 2023-12-22 | 青岛东捷建设集团有限公司 | High-pressure jet grouting method and system based on seawater seepage |
| CN117522173A (en) * | 2024-01-04 | 2024-02-06 | 山东科技大学 | Natural gas hydrate depressurization exploitation productivity prediction method based on deep neural network |
| CN117522173B (en) * | 2024-01-04 | 2024-04-26 | 山东科技大学 | Natural gas hydrate depressurization exploitation productivity prediction method based on deep neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115130734A (en) | Method and system for predicting construction influence of penetration project based on LightGBM and deep learning algorithm | |
| Ochmański et al. | Prediction of the diameter of jet grouting columns with artificial neural networks | |
| CN106405640B (en) | Microseismic signals based on depth conviction neural network then automatic pick method | |
| Leu et al. | Data mining for tunnel support stability: neural network approach | |
| Jan et al. | Neural network forecast model in deep excavation | |
| Zhang et al. | Comprehensive review of machine learning in geotechnical reliability analysis: Algorithms, applications and further challenges | |
| CN115659749B (en) | Foundation pit deformation prediction method and system, electronic equipment and storage medium | |
| CN115270635B (en) | Bayes-neural network high-rise building earthquake demand and vulnerability prediction method | |
| CN110362948B (en) | Karst ground subsidence treatment method based on cloud model | |
| Behnia et al. | Predicting crest settlement in concrete face rockfill dams using adaptive neuro-fuzzy inference system and gene expression programming intelligent methods | |
| Bashari et al. | Estimation of deformation modulus of rock masses by using fuzzy clustering-based modeling | |
| CN115310361B (en) | Underground coal mine dust concentration prediction method and system based on WGAN-CNN | |
| CN114548482A (en) | Creep type landslide kinetic energy change rate face-slip early warning method | |
| CN115293197A (en) | Borehole strain data anomaly detection method based on long-term and short-term memory network | |
| Jebur et al. | Settlement prediction of model piles embedded in sandy soil using the Levenberg–Marquardt (LM) training algorithm | |
| CN113792372A (en) | Ground continuous wall deformation dynamic prediction method based on CV-LSTM combined model | |
| Yang et al. | An intelligent procedure for updating deformation prediction of braced excavation in clay using gated recurrent unit neural networks | |
| Huynh et al. | Base resistance of super-large and long piles in soft soil: performance of artificial neural network model and field implications | |
| CN111859814A (en) | Rock aging deformation prediction method and system based on LSTM deep learning | |
| CN115049124A (en) | Deep and long tunnel water inrush prediction method based on Bayesian network | |
| CN112836789A (en) | Ground connection wall deformation dynamic prediction method based on composite neural network algorithm | |
| CN115495991A (en) | Rainfall interval prediction method based on time convolution network | |
| CN115221793A (en) | Tunnel surrounding rock deformation prediction method and device | |
| Elazouni et al. | Estimating resource requirements at conceptual design stage using neural networks | |
| CN108694816B (en) | Debris flow early warning method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |