Double-wheel milling construction geology identification method based on random forest and support vector machine

Abstract

The invention discloses a double-wheel milling construction geology identification method based on a random forest and a support vector machine. The method comprises the following steps: S1, data preprocessing: data acquisition, sample marking, filtering processing and correlation analysis and feature selection; s2, geological recognition model training: S21, overburden and rock stratum recognition model training; s22, training a rock stratum identification model; during model training, taking a plurality of slot segment data in all slot segment data as a test set, and taking the data of the remaining slot segments as a training set; and S3, real-time geological recognition: after the model is obtained through training in the S2, real-time recognition of the residual groove section double-wheel milling grooving construction geology is carried out. According to the method, real-time recognition of the double-wheel milling construction geology can be realized, the geological judgment efficiency and accuracy are improved, and beneficial help is provided for quality improvement and efficiency improvement of underground diaphragm wall grooving construction.

Description

Technical Field

[0001] The invention relates to double-wheel milling and underground continuous wall construction, and in particular to a double-wheel milling construction geological two-level identification method combining random forest and support vector machine. Background Art

[0002] As the most advanced underground continuous wall construction equipment today, the hydraulic double-wheel milling machine has the advantages of high troughing construction efficiency, high precision, safety and environmental protection, and wide adaptability to strata. Its advantages are more prominent in hard rock strata with greater strength. The double-wheel milling machine will encounter different geological conditions during the downward construction, and the geological conditions have a great impact on the operation of the construction personnel. Different geological conditions require different control methods and equipment parameters. In traditional construction, it is more dependent on existing survey reports or judge the geological conditions based on the slag discharge. There may be problems such as incorrect geological surveys and delayed on-site observations, which have a great impact on the accurate judgment of geology and the correct operation of the equipment. Therefore, it is very necessary to find a method that can accurately judge the geology of double-wheel milling construction in real time. Summary of the invention

[0003] The purpose of the present invention is to solve the problem of inaccurate understanding of geological conditions when double-wheel milling is used for ground-connected wall trenching construction in the above background. To address this problem, a two-level identification method for double-wheel milling construction geology combining random forest and support vector machine is proposed.

[0004] The technical solution adopted by the present invention to solve this technical problem is: a double-wheel milling construction geological identification method based on random forest and support vector machine, comprising the following steps: S1. Data preprocessing: data collection, sample labeling, filtering, correlation analysis and feature selection; wherein, the sample labeling includes: in order to train the recognition model, the construction data of the ground-connected wall trench section near the exploration hole is selected as a sample in the early stage of construction, and geological labeling is performed according to the geological survey column chart and the depth in the sample data. A total of four types of strata are marked in the geological labeling: overburden, strongly weathered rock layer, moderately weathered rock layer and slightly weathered rock layer; S2, geological identification model training, including S21, overburden and rock layer identification model training; S22, rock layer identification model training; when performing model training, some slot segment data among all slot segment data are used as test sets, and the remaining slot segment data are used as training sets; S3, real-time geological identification, after the model is obtained through S2 training, real-time identification of the geology of the remaining slot section double-wheel milling construction is carried out.

[0005] As a further solution of the present invention, the collected data are characteristic parameters of the double-wheel milling equipment, including: delivery pump flow, right milling wheel hydraulic oil pressure, right milling wheel speed, ambient pressure, right milling wheel gear oil pressure, right milling wheel gear oil temperature, leakage oil pressure, weight on wire rope, slot milling machine depth, feed speed, additional load weight, left milling wheel gear oil temperature, left milling wheel gear oil pressure, left milling wheel speed, delivery pump speed, left milling wheel hydraulic oil pressure and delivery pump hydraulic oil pressure.

[0006] As a further solution of the present invention, the filtering process is specifically: using a Savitzky-Golay filter to filter the data, using a 15-period window for smoothing filtering, using a quadratic polynomial as shown in the following formula for fitting, and using local 15-period data to calculate the polynomial coefficients; (1) In the formula, n is the sequence number in the window, n = -7, -6, …, 0, …, 6,7, and p is the characteristic parameter value corresponding to each period in the window.

[0007] As a further solution of the present invention, the correlation analysis and feature selection are specifically as follows: the overburden layer, strongly weathered rock layer, moderately weathered rock layer and slightly weathered rock layer are represented by parameters 1, 2, 3 and 4 respectively, the correlation coefficients are calculated between the geological parameters and the characteristic parameters of the double-wheel milling equipment respectively, and the characteristic parameters of the double-wheel milling equipment with a high degree of correlation are selected.

[0008] As a further solution of the present invention, the correlation coefficient is calculated using the Pearson correlation coefficient formula, as shown below: (2) In the formula, x and y are two groups of variables, and x ̅ and y ̅ are the average values ​​of the two groups of variables.

[0009] As a further solution of the present invention, the characteristic parameters of the double-wheel milling equipment with a high degree of correlation are selected as 10 parameters: left wheel speed, right wheel speed, left hydraulic oil pressure, right hydraulic oil pressure, delivery pump pressure, delivery pump flow, delivery pump speed, feed speed, additional load weight, and weight on the wire rope.

[0010] As a further solution of the present invention, when the overburden and rock strata identification model is trained in S21, a support vector machine model is used to classify and identify whether it is an overburden or a rock strata, and the rock strata are a general term for strongly weathered rock strata, moderately weathered rock strata and slightly weathered rock strata.

[0011] As a further solution of the present invention, random forest is used as a classifier in the training of the S22 rock formation identification model, and strongly weathered, moderately weathered, and slightly weathered rock formation samples are selected in equal proportions from the rock formation samples in the training set as a training set to train the rock formation identification model.

[0012] As a further solution of the present invention, the S3 real-time geological identification specifically includes: S31, for real-time data, select characteristic parameters of double-wheel milling equipment with high correlation as features, and use 15-stage SG filter for filtering; S32, using the trained cover layer and rock layer recognition model to identify whether the current stratum is a cover layer, if it is a cover layer, obtaining a final geological recognition result belonging to the cover layer, if it is a rock layer, entering S32; S33. Use the trained rock formation identification model to train and identify the weathering degree of the current rock formation, and then obtain the final geological identification result.

[0013] The present invention includes at least the following beneficial effects: First, in the training stage, according to the geological survey data, the data of the adjacent slot sections of the geological exploration holes constructed in the initial stage of the project are selected for accurate stratigraphic marking to form data samples, and the features are filtered to reduce noise, and the geological correlation coefficient is calculated and feature selection is performed. Then, according to the two-level recognition scheme from the overburden layer to the rock layer, the overburden layer and the rock layer SVM classification model are first trained, and then the strong, medium and micro rock layer random forest classification model is trained. After the model is trained, for the double-wheel milling grooving construction process of the remaining slot sections, feature selection and filtering are first performed, and then SVM is used to distinguish whether it is an overburden layer or a rock layer. If it is identified as a rock layer, the random forest model is further used to identify the degree of weathering of the strong, medium and micro. The present invention is based on the double-wheel milling construction data in the initial stage of the project. Through a two-level geological recognition method for overburden layers and rock layers combined with multiple feature processing methods and multiple models, it can realize the real-time recognition of the double-wheel milling construction geology, improve the efficiency and accuracy of geological judgment, and provide useful help for improving the quality and efficiency of the ground-connected wall grooving construction.

[0014] Other advantages, objectives and features of the present invention will be embodied in part through the following description, and in part will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a two-level geological identification flow chart of double-wheel milling construction of the present invention. DETAILED DESCRIPTION

[0016] The present invention is described in detail and completely below in conjunction with the accompanying drawings. A person of ordinary skill in the art will be able to implement the present invention based on these descriptions. Before describing the present invention in conjunction with the accompanying drawings, it should be particularly pointed out that the technical solutions and technical features provided in each part of the present invention, including the following description, can be combined with each other without conflict.

[0017] In addition, the embodiments of the present invention involved in the following description are generally only a part of the embodiments of the present invention, rather than all the embodiments. Therefore, based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0018] The present invention is further described in detail below in conjunction with the accompanying drawings and implementations, and its specific implementation process is as follows: like Figure 1 As shown, the present invention provides a double-wheel milling construction geological identification method based on random forest and support vector machine, (1) data preprocessing 1) Sample labeling The double-wheel milling construction data is exported in real time through the equipment gateway. The collected data includes the following 17 parameters: delivery pump flow, right milling wheel hydraulic oil pressure, right milling wheel speed, ambient pressure, right milling wheel gear oil pressure, right milling wheel gear oil temperature, leakage oil pressure, weight on wire rope, slot milling machine depth, feed speed, additional load weight, left milling wheel gear oil temperature, left milling wheel gear oil pressure, left milling wheel speed, delivery pump speed, left milling wheel hydraulic oil pressure and delivery pump hydraulic oil pressure.

[0019] The equipment status of the double-wheel milling machine during the underground continuous wall trenching construction is affected by many factors such as geological conditions, construction technology, and ground-connected wall design requirements. The same type of geology in different regions will also have specific differences. Therefore, it is not realistic to try to establish a reliable relationship between the double-wheel milling construction geology and construction status parameters through a few project data. Each project will conduct preliminary geological surveys, and the geological survey will obtain clear geological conditions, but geological exploration holes are generally only drilled at a certain distance apart in accordance with the specifications. For ground-connected wall construction, most construction areas and trench sections do not have more accurate geological information. Therefore, the present invention is based on a small amount of geological exploration holes in the early stage of construction. The adjacent trench section construction data is used to train the double-wheel milling construction geological recognition model, and the double-wheel milling construction geology is identified in real time during the construction of the remaining vast majority of trench sections.

[0020] In order to train the recognition model, the construction data of the ground-connected wall trench section near the exploration hole are first selected as samples in the early stage of construction, and geological marking is performed according to the geological survey column chart and the depth in the sample data.

[0021] In general projects, double-wheel milling trenching construction will pass through different geological strata from top to bottom, such as plain fill, clay, sand layer, strongly weathered rock layer, moderately weathered rock layer, slightly weathered rock layer, etc. In strata other than rock layers, double-wheel milling can maintain a relatively high construction efficiency, and geological changes have little effect on trenching construction efficiency. Therefore, strata other than rock layers are uniformly marked as one stratum, and a total of four strata are marked: overburden, strongly weathered rock layer, moderately weathered rock layer, and slightly weathered rock layer.

[0022] 2) Filtering During double-wheel milling construction, the characteristic parameters fluctuate violently and are almost unusable without processing, so filtering is required.

[0023] The present invention utilizes a Savitzky-Golay filter (SG filter) to filter the data. The SG filter is a filtering method based on local polynomial least squares fitting in the time domain. The greatest feature of this filter is that it can ensure that the shape and width of the signal remain unchanged while filtering out noise, and will not cause a large amount of information loss while achieving filtering.

[0024] The present invention adopts a 15-period window for smoothing filtering, selects a quadratic polynomial as shown in the following formula for fitting, and uses local 15-period data to calculate the polynomial coefficients.

[0025] (1) In the formula, n is the sequence number in the window, n = -7, -6, …, 0, …, 6,7, and p is the characteristic parameter value corresponding to each period in the window.

[0026] The specific implementation method is to take the data of 15 moments before and after each moment, and calculate the polynomial coefficients using the least squares method according to formula (1): a 0. a 1. a 2, and then calculate according to this formula p (0), replace the data at that moment with p (0).

[0027] 3) Correlation analysis and feature selection Not all of the 17 collected double-wheel milling parameters have a significant correlation with the stratum. The present invention uses the Pearson correlation coefficient to evaluate the correlation between the parameters and the geology. The calculation formula of the Pearson correlation coefficient R is shown in formula (2). The closer its absolute value is to 1, the higher the degree of correlation between the two groups of variables.

[0028] (2) In the formula, x , y There are two groups of variables, are the average values ​​of the two groups of variables.

[0029] The present invention represents the overburden layer, strongly weathered rock layer, moderately weathered rock layer and slightly weathered rock layer with parameters 1, 2, 3 and 4 respectively, calculates the correlation coefficients between the geological parameters and 17 double-wheel milling characteristic parameters respectively, and finally selects 10 parameters with better correlation, namely, left wheel speed, right wheel speed, left hydraulic oil pressure, right hydraulic oil pressure, delivery pump pressure, delivery pump flow, delivery pump speed, feed speed, additional weight and weight on the wire rope as characteristics.

[0030] (2) Geological identification model training The data of each slot section is a complete construction record, and the data of the same slot section at adjacent times are highly correlated. Therefore, in the process of model training and testing and repeated parameter adjustment, when obtaining sample data in this embodiment, data of a few slot sections are left as test sets, and data of other slot sections are used as training sets.

[0031] 1) Overburden and rock layer identification model training Compared with strongly weathered, moderately weathered and slightly weathered rock strata, the overburden layer is obviously different. Directly identifying the four strata will affect the effect. Therefore, the present invention adopts different methods for identifying the overburden layer and the strongly weathered, moderately weathered and slightly weathered rock strata. First, it is identified whether it is an overburden layer. If it is identified as a rock stratum, it is then determined whether it belongs to the strongly weathered, moderately weathered or slightly weathered rock strata.

[0032] The characteristic distribution of overburden and rock formations is obviously different, and it is suitable to use a binary classification model such as support vector machine (SVM) for classification. The present invention uses radial basis function as the kernel function of SVM, and selects overburden and rock formation samples in equal proportion in the training set to train the overburden and rock formation recognition SVM model.

[0033] 2) Rock formation identification model training Using Random Forests as a classifier, strongly weathered, moderately weathered, and slightly weathered rock samples are selected in equal proportions from the rock samples in the training set as the training set to train the rock recognition model. During the training process, repeated attempts are required and tests are conducted on a separate slot test set. Finally, the training hyperparameters with better test results are determined. The training hyperparameters obtained by testing and optimizing the present invention in a certain project are as follows: Table 1 Random forest training hyperparameters Hyperparameters Value Maximum tree depth 100 Maximum feature ratio 0.7 Minimum number of samples for leaf nodes 30 Minimum number of samples for node branching 20 Minimum weight and 0.01 Maximum number of leaf nodes 210 Number of weak classifiers 100 (3) Real-time geological identification After the model is trained, it can be used for real-time identification of the geology of the remaining slot section during double-wheel milling. The identification steps are as follows: S1 For real-time data, the 10 parameters of the above feature selection are also selected as features, and the 15-stage SG filter is used for filtering.

[0034] S2 uses the trained radial basis kernel function SVM to identify whether the current stratum is an overburden layer. If it is an overburden layer, the final geological identification result belonging to the overburden layer is obtained. If it is a rock layer, enter S3.

[0035] S3 uses the trained random forest to identify the degree of weathering of the current rock layer and then obtains the final geological identification result.

[0036] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and the implementation modes. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and the embodiments shown and described herein.

Claims

1. A double-wheel milling construction geological identification method based on random forest and support vector machine, characterized in that: The following steps are involved: S1. Data preprocessing: data collection, sample marking, filtering, correlation analysis and feature selection; wherein, the sample marking includes: selecting the construction data of the ground-connected wall trench section near the exploration hole as a sample in the early stage of construction, and performing geological marking according to the geological survey columnar diagram and the depth in the sample data. A total of four types of strata are marked in the geological marking: overburden, strongly weathered rock layer, moderately weathered rock layer and slightly weathered rock layer; S2, geological identification model training, including S21, overburden and rock layer identification model training; S22, rock layer identification model training; when performing model training, some slot segment data among all slot segment data are used as test sets, and the remaining slot segment data are used as training sets; S3, real-time geological identification, after the model is obtained through S2 training, real-time identification of the geology of the remaining slot section double-wheel milling construction is carried out.

2. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 1, characterized in that, The collected data are characteristic parameters of the double-wheel milling equipment.

3. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 1, characterized in that, The filtering process is specifically as follows: using a Savitzky-Golay filter to filter the data, using a 15-period window for smoothing filtering, using a quadratic polynomial as shown in the following formula for fitting, and using local 15-period data to calculate the polynomial coefficients; （1） Where n is the sequence number in the window, n = -7, -6, …, 0, …, 6, 7, and p is the characteristic parameter value corresponding to each period in the window.

4. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 2, characterized in that, The correlation analysis and feature selection are as follows: the overburden layer, strongly weathered rock layer, moderately weathered rock layer and slightly weathered rock layer are represented by parameters 1, 2, 3 and 4 respectively, the correlation coefficients between the geological parameters and the characteristic parameters of the double-wheel milling equipment are calculated respectively, and the characteristic parameters of the double-wheel milling equipment with a high degree of correlation are selected.

5. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 4, characterized in that, The Pearson correlation coefficient formula was used to calculate the correlation coefficient, as shown below; （2） In the formula, x and y are two groups of variables, and x ̅ and y ̅ are the average values ​​of the two groups of variables.

6. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 5, characterized in that: The characteristic parameters of the double-wheel milling equipment with a high degree of correlation selected are left wheel speed, right wheel speed, left hydraulic oil pressure, right hydraulic oil pressure, delivery pump pressure, delivery pump flow, delivery pump speed, feed speed, additional load weight, and weight on the wire rope.

7. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 1, characterized in that: When the overburden and rock strata identification model is trained in S21, a support vector machine model is used to classify and identify whether it is an overburden or a rock strata. The rock strata are a general term for strongly weathered rock strata, moderately weathered rock strata and slightly weathered rock strata.

8. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 1 or 7, characterized in that, In the training of the S22 rock formation recognition model, random forest is used as a classifier, and strongly weathered, moderately weathered, and slightly weathered rock formation samples are selected in equal proportions from the rock formation samples in the training set as the training set to train the rock formation recognition model.

9. The double-wheel milling construction geological identification method based on random forest and support vector machine as claimed in claim 1, characterized in that: The S3 real-time geological identification specifically includes: S31, for real-time data, select characteristic parameters of double-wheel milling equipment with high correlation as features, and use 15-stage SG filter for filtering; S32, using the trained cover layer and rock layer recognition model to identify whether the current stratum is a cover layer, if it is a cover layer, obtaining a final geological recognition result belonging to the cover layer, if it is a rock layer, entering S32; S33. Use the trained rock formation identification model to train and identify the weathering degree of the current rock formation, and then obtain the final geological identification result.

Images