Automatic identification method for seabed soil type based on naive bayes algorithm

Abstract

The application discloses a seabed soil type automatic identification method based on a naive Bayes algorithm. The method comprises the following basic steps: 1) CPT data acquisition; 2) attribute feature calculation; 3) normalized cone tip resistance fluctuation feature calculation; 4) feature parameter filtering; and 5) using a Bayes classifier to identify the seabed soil type. The method has the advantages of simplicity, small calculation amount, good real-time performance, saving of manpower and easiness in implementation, and is suitable for automatic identification of seabed soil types.

Description

Technical Field

[0001] This invention relates to the technical fields of marine engineering, marine surveying, marine oil exploration and development, and marine surveys, and specifically to an automatic identification method for seabed soil types based on the Naive Bayes algorithm. Background Technology

[0002] The types and characteristics of seabed soil are important aspects of marine engineering, marine surveys, and offshore oil exploration and development. They are also important reference indicators for offshore oil extraction and offshore wind power construction. A thorough understanding of seabed soil types and characteristics is a prerequisite for the exploitation of marine resources.

[0003] Static cone penetration tests (CPTs) obtain signals from the CPT probe after it contacts the seabed soil, including data on three properties: cone tip resistance, sidewall friction, and pore water pressure. These data reflect soil properties at the measurement point. Soil classification maps are typically created using two of these three properties as the basis for soil classification. Furthermore, through formula derivation, these three properties are modified to create normalized attribute characteristics, which are then applied to soil classification. Compared to the original three properties, using modified and normalized attribute data for soil classification eliminates the influence of the detection equipment and soil depth on soil classification, resulting in a more reasonable data structure and distribution.

[0004] Traditional soil classification typically relies on soil property data and classification criteria provided by classification charts, with manual judgment required. In recent years, research has gradually shifted towards automated soil classification methods using machine learning. These methods utilize soil property data to create a training dataset, and then train a machine learning model to achieve automatic identification of soil lithology. This approach offers advantages such as simplicity and high efficiency, enabling online, real-time, and automated prediction of seabed soil types.

[0005] While some research has been conducted on soil classification methods, most of this research focuses on terrestrial soil type identification, with very little attention paid to seabed soil type identification. Furthermore, the existing methods still require improvement in terms of algorithm simplicity and classification accuracy. Summary of the Invention

[0006] The purpose of this invention is to automatically identify seabed soil types using CPT data. This invention has advantages such as simple method, low computational load, good real-time performance, labor saving, and ease of implementation. It is suitable for the automatic identification of seabed soil types.

[0007] This invention includes the following steps:

[0008] (1) CPT data acquisition:

[0009] The cone tip resistance q was obtained through a static cone penetration test. c Sidewall friction force f s Pore ​​water pressure u2 (unit: MPa), penetration depth d at the detection point (unit: m);

[0010] (2) Calculate attribute characteristics:

[0011] according to The calculated pore pressure parameter ratio B q ,according to The normalized cone tip resistance Q was calculated. t ,according to The normalized friction ratio F was calculated. r Where, Δu=u2-γ w ×d represents excess pore water pressure, γ w q represents the bulk density of seawater or mud. t =q c +(1+α)u2 is the corrected cone tip resistance, α is the cone tip specification; σ' vo =γ'×d is the effective overburden pressure, γ' is the average effective unit weight of the overburden soil layer, γ w γ' is set manually according to the characteristics of the area, and α is the CPT equipment parameter, which is determined according to the CPT equipment;

[0012] (3) Calculate Q t Fluctuation characteristics:

[0013] Q t Perform mean filtering to obtain the filtered normalized cone end resistance Q. t ', then according to ΔQ t =|Q t -Q t The normalized cone end resistance deviation ΔQ was calculated. t If ΔQ t_j If ΔQ < 0.5, then ΔQ t_j =ΔQ t_j-1 , where ΔQ t_j-1 and ΔQ t_j These are the normalized cone tip resistance deviations at penetration depths j-1 and j, respectively, where j = 2, 3, ..., n, and n is the maximum penetration depth;

[0014] (4) Feature parameter filtering:

[0015] For B q Q t F r ΔQ t Perform a sliding median filter with a filter window width of w1, where w1 is an odd number.

[0016] (5) Identifying seabed soil types using a Bayesian classifier:

[0017] according to The characteristic parameter x is calculated. k The estimated conditional probability G corresponding to the i-th soil type i (x k ),according to The probability that the soil at the point to be identified belongs to the i-th soil type is calculated, max{GP i The corresponding i represents the soil type of the point to be identified, where p i μ is the prior probability of the i-th soil type, determined offline based on the training set data. i and σ i The feature parameters x are respectively the feature parameters of the i-th soil type in the training set. k The mean and standard deviation are determined offline based on the training set data; k = 1, 2, ..., 4, x1, x2, x3, x4 are respectively B q Q t F r ΔQ t The value of the point to be identified; i represents the soil type number, i = 1, 2, ..., 8. Attached Figure Description

[0018] Figure 1 For well CPT-02-01 q c f s , u2 variation with penetration depth;

[0019] Figure 2 CPT-02-01 well B q Q t F r Plot showing variation with penetration depth;

[0020] Figure 3 The graph shows the variation of normalized cone resistance deviation with penetration depth in well CPT-02-01.

[0021] Figure 4 The curve of characteristic parameters of well CPT-02-01 before median filtering;

[0022] Figure 5 This is a curve of the characteristic parameters of well CPT-02-01 after median filtering;

[0023] Figure 6 To obtain the prior probabilities, mean and standard deviation of feature parameters for each soil type in the training dataset;

[0024] Figure 7 A comparison chart of the automatic soil identification results and the manual classification results for well CPT-02-01; Detailed Implementation

[0025] Based on the fundamental data obtained from the static cone penetration test, this embodiment calculates the pore pressure parameter ratio, normalized cone tip resistance, normalized friction ratio, and normalized cone tip resistance deviation, and uses the Naive Bayes method to achieve automatic identification of seabed soil types.

[0026] The specific steps are as follows:

[0027] (1) CPT data acquisition:

[0028] The cone tip resistance q was obtained through a static cone penetration test. c Sidewall friction force f s Pore ​​water pressure u2 (unit: MPa), penetration depth d at the detection point (unit: m);

[0029] In this embodiment, taking well CPT-02-01 in a certain sea area of ​​my country as an example, the collected q c f s The curves showing the variation of u2 with penetration depth d are as follows: Figure 1 As shown.

[0030] (2) Calculate attribute characteristics:

[0031] according to The calculated pore pressure parameter ratio B q ,according to The normalized cone tip resistance Q was calculated. t ,according to The normalized friction ratio F was calculated. r Where, Δu=u2-γ w ×d represents excess pore water pressure, γ w q represents the bulk density of seawater or mud. t =q c +(1+α)u2 is the corrected cone tip resistance, α is the cone tip specification; σ' vo =γ'×d is the effective overburden pressure, γ' is the average effective unit weight of the overburden soil layer, γ w γ' is set manually according to the characteristics of the area, and α is the CPT equipment parameter, which is determined according to the CPT equipment;

[0032] In this embodiment, α = 0.75, γ w =10.05KN / m 3 γ'=0.01MN / m 3 Calculate B in well CPT-02-01 q Q t F r Attribute characteristics, B q Q t F r Changes with penetration depth, such as Figure 2 As shown.

[0033] (3) Calculate Q t Fluctuation characteristics:

[0034] Q t Perform mean filtering to obtain the filtered normalized cone end resistance Q. t ', then according to ΔQ t =|Q t -Q t The normalized cone end resistance deviation ΔQ was calculated. t If ΔQ t_j If ΔQ < 0.5, then ΔQ t_j =ΔQ t_j-1 , where ΔQ t_j-1 and ΔQ t_j These are the normalized cone tip resistance deviations at penetration depths j-1 and j, respectively, where j = 2, 3, ..., n, and n is the maximum penetration depth;

[0035] In this embodiment, the window width of the mean filter is 30, and the normalized cone resistance deviation ΔQ of well CPT-02-01 is calculated. t ΔQ t Changes with penetration depth, such as Figure 3 As shown.

[0036] (4) Feature parameter filtering:

[0037] For B q Q t F r ΔQ t Perform a sliding median filter with a filter window width of w1, where w1 is an odd number.

[0038] In this embodiment, w1 = 31, which represents the characteristic parameter B of well CPT-02-01. q Q t F r ΔQ t The curves showing the changes in characteristic parameters with penetration depth before and after sliding median filtering are as follows: Figure 4 , Figure 5 As shown.

[0039] (5) Identifying seabed soil types using a Bayesian classifier:

[0040] according to The characteristic parameter x is calculated. k The estimated conditional probability G corresponding to the i-th soil type i (x k ),according to The probability that the soil at the point to be identified belongs to the i-th soil type is calculated, max{GP i The corresponding i represents the soil type of the point to be identified, where p i μ is the prior probability of the i-th soil type, determined offline based on the training set data. i and σ i The feature parameters x are respectively the feature parameters of the i-th soil type in the training set. k The mean and standard deviation are determined offline based on the training set data; k = 1, 2, ..., 4, x1, x2, x3, x4 are respectively B q Q t F r ΔQ t The value of the point to be identified; i represents the soil type number, i = 1, 2, ..., 8.

[0041] In this embodiment, the total number of data collected in a certain sea area of ​​my country is N = 34090. The training dataset is a matrix of size 4102 × 5, and the first 4 columns of the matrix are the feature parameters B. q Q t F r ΔQ t The data was randomly selected from all data collected by the CPT device. The 5th column of the matrix represents the soil type number corresponding to the data in the first 4 columns, obtained through laboratory analysis. The amount of data for the i-th soil type in the training dataset is... N i Let i be the amount of data for the i-th soil type in the total data volume collected, i = 1, 2, ..., 8, where 1 is clay, 2 is silt, 3 is fine sand, 4 is silty clay, 5 is silty sand, 6 is sandy silt, 7 is sandy clay, and 8 is clayey sand.

[0042] μ is used to determine the i-th soil type based on the training dataset. i and σ i ,according to The prior probability of the i-th soil type is calculated, and the specific value is as follows: Figure 6 As shown.

[0043] The seabed soil type identification results for well CPT-02-01 are as follows: Figure 7 As shown, it took 0.104 seconds and the recognition accuracy rate was 90.49%.

Claims

1. An automatic identification method for seabed soil types based on the Naive Bayes algorithm, characterized in that, Includes the following steps: (1) CPT data acquisition: The cone tip resistance q was obtained through static cone penetration test. c Sidewall friction force f s 1. Pore water pressure u2, unit: MPa; 2. Penetration depth of the detection point d, unit: m; (2) Calculate attribute characteristics: based on The calculated pore pressure parameter ratio B q ,according to The normalized cone tip resistance Q was calculated. t ,according to The normalized friction ratio F was calculated. r Where, Δu=u2-γ w ×d represents excess pore water pressure, γ w q represents the bulk density of seawater or mud. t =q c +(1+α)u2 is the corrected cone tip resistance, α is the cone tip specification; σ' vo =γ'×d is the effective overburden pressure, γ' is the average effective unit weight of the overburden soil layer, γ w γ' is set manually according to the characteristics of the area, and α is the CPT equipment parameter, which is determined according to the CPT equipment; (3) Calculate Q t Fluctuation characteristics: for Q t Perform mean filtering to obtain the filtered normalized cone end resistance Q. t ', then according to ΔQ t =|Q t -Q t The normalized cone end resistance deviation ΔQ was calculated. t If ΔQ t_j If ΔQ < 0.5, then ΔQ t_j =ΔQ t_j-1 , where ΔQ t_j-1 and ΔQ t_j These are the normalized cone tip resistance deviations at penetration depths j-1 and j, respectively, where j = 2, 3, ..., n, and n is the maximum penetration depth; (4) Feature parameter filtering: For B q Q t F r ΔQ t Perform a sliding median filter with a filter window width of w1, where w1 is an odd number. (5) Identifying seabed soil types using a Bayesian classifier: Based on The characteristic parameter x is calculated. k The estimated conditional probability G corresponding to the i-th soil type i (x k ),according to The probability that the soil at the point to be identified belongs to the i-th soil type is calculated, max{GP i The corresponding i represents the soil type of the point to be identified, where p i μ is the prior probability of the i-th soil type, determined offline based on the training set data. i and σ i The feature parameters x are respectively the feature parameters of the i-th soil type in the training set. k The mean and standard deviation are determined offline based on the training set data; k = 1, 2, ..., 4, x1, x2, x3, x4 are respectively B q Q t F r ΔQ t The value of the point to be identified; i represents the soil type number, i = 1, 2, ..., 8.

Images