KR20200058258A - System and method for predicting ground layer information, and a recording medium having computer readable program for executing the method - Google Patents

Abstract

A system for predicting ground layer information includes a drilling area information input unit, a borehole classification unit, a parameter optimization unit, a non-drilling area information input unit, and a layer information prediction unit. The drilling area information input unit receives input information including coordinate information on a plurality of boreholes of a drilled area and layer information measured in a borehole. The borehole classification unit classifies a plurality of boreholes into boreholes for learning and boreholes for test. A test information calculation unit calculates data measured in the boreholes for learning and layer information of the boreholes for test using a preset layer information calculation algorithm. The parameter optimization unit compares the calculated layer information with measured layer information of the boreholes for test to calculate an optimization value of a parameter in the layer information calculation algorithm. The non-drilling area information input unit receives input information on a non-drilled area. The layer information prediction unit calculates layer information of the non-drilled area using a parameter to which an optimization value is applied.

Description

Ground layer information prediction system, method, and storage medium recording computer readable program for executing the method {SYSTEM AND METHOD FOR PREDICTING GROUND LAYER INFORMATION, AND A RECORDING MEDIUM HAVING COMPUTER READABLE PROGRAM FOR EXECUTING THE METHOD}

The present invention relates to a ground information prediction system and method, and more particularly, to a system and method for predicting the layer information of an undrilled section by using the layer information of the surrounding borehole.

In order to accurately predict the amount of earthwork, it is important to input attribute information such as geological column and soil characteristics in 3D terrain information, but it is difficult to secure sufficient data. It is difficult to distinguish continuous ground properties of spatial units from point-based drilling information on uncertainty and volatile grounds, and the method of determining ground properties by the engineer's engineering judgment is a predictive method based on empirical and statistical methods. A lot of errors occur with the condition of, and this often leads to a significant increase and decrease in construction volume, which often requires redesign.

As such, estimating the layer information of the ground is important for accurate soil volume prediction, and in order to predict the ground layer information, it is necessary to interpolate against the area or distance and to use geo-statistical analysis methods. Ground layer information can be predicted.

In the interpolation method, the Inverse Distance Weighting (IDW) method is widely used, and the representative method in the geostatistical method is the kriging technique. In both methods, it is possible to predict the ground layer information of the undrilled section through the layer information of the drilled section.

However, it is impossible to accurately classify continuous ground properties in spatial units from point-based drilling information, and if there is not enough data when entering the attribute information of the terrain, errors may occur due to the situation applied to the design depending on the engineer's experience. have.

The method of determining the ground property by the engineer's engineering judgment is a predictive method based on the empirical method, and many errors occur depending on the conditions of the actual site, and as a result, the amount of construction is greatly increased, and cases requiring redesign frequently occur. . Therefore, in order to predict ground information similar to the conditions of the actual site, it is essential to secure accuracy through a theoretical approach in addition to the existing empirical approach.

In addition, the existing geostatistical methods such as Kriging assume weak immutability and predict the unknown value through modeling considering the existence of covariance. There are limits to finding it.

KR 101927659 B1

The present invention has been devised to solve the above-mentioned conventional problems, and avoids a method based on past personal experiences and more effectively predicts the layered information of the undrilled section by using the layered information of the surrounding borehole through theoretical analysis. It is an object to provide a system, and method for.

In addition, an object of the present invention is to provide a system and method for more effectively predicting the layered information of an undrilled section by utilizing not only the layered information of the drilled ball but also various additional information.

To achieve the above object, the ground layer information prediction system according to the present invention includes a drilling area information input unit, a borehole classification unit, a parameter optimization unit, a non-drilling area information input unit, and a layered information prediction unit.

The borehole information input unit receives input information including coordinate information for a plurality of boreholes in the drilled region and the layered information measured by the borehole, and the borehole classification unit classifies the plurality of boreholes into a learning borehole and a test borehole, and tests The information calculation unit calculates the layer information of the test borehole using data measured in the borehole for learning and a preset layer information calculation algorithm, and the parameter optimizer compares the calculated layer information with the measured layer information of the test borehole to measure the layer information. The optimization value of the parameter in the calculation algorithm is calculated, and the non-drilling region information input unit receives input information for the undrilled region, and the layered information prediction unit uses the parameter to which the optimization value is applied to obtain the layered information of the undrilled region. Calculate.

According to such a configuration, it is possible to more effectively predict the layer information of the undrilled section from the layer information of the surrounding borehole by engineering judgment rather than the designer's experience.

At this time, the layer information calculation algorithm may be an artificial neural network based algorithm.

Also, a normalization unit that performs normalization on input information may be further included. According to this configuration, it is possible to control the influence on the prediction result even when the absolute size of different input data has a large difference.

In addition, the classification of the borehole for learning and the borehole for testing may be performed repeatedly at random. According to this configuration, it is possible to obtain more effective prediction results by variously configuring a combination of a learning borehole and a test borehole.

In addition, the parameter optimization unit includes an error value calculation unit that calculates an error value between the calculated layer information and the layer information of the measured borehole for testing, a parameter value calculation unit that calculates a new parameter value to make the error value smaller, and a calculated value. It may include a parameter update unit for updating the previous parameter with the parameter value. According to such a configuration, it is possible to secure more accurate prediction results by repeatedly updating the parameters of the prediction algorithm.

Also, the coordinate information may further include elevation information of the ground floor information calculation area. According to such a configuration, unlike conventional 3D kriging, which is difficult to easily adopt due to high algorithm complexity, it is possible to more effectively predict layer information of the ground using 3D coordinates.

Further, the input information may further include underground water level information in the ground layer information calculation area. According to this configuration, by using not only the layer information of the drilled ball but also various additional information, it is possible to more effectively predict the layer information of the undrilled section.

In addition, the invention embodying the system in the form of a method and a storage medium recording a computer readable program for executing the method are also disclosed.

According to the present invention, it is possible to more effectively predict the layer information of the undrilled section from the layer information of the surrounding borehole by engineering judgment rather than the designer's experience.

In addition, even if the absolute size of different input data has a large difference, it is possible to control the influence on the prediction result.

In addition, it is possible to obtain more effective prediction results by variously configuring a combination of a learning borehole and a test borehole.

In addition, it is possible to obtain a more accurate prediction result of the layer information of the ground by repeatedly updating the parameters of the prediction algorithm.

In addition, unlike conventional 3D kriging, which is difficult to easily adopt due to high algorithm complexity, it is possible to more effectively predict layer information of the ground using 3D coordinates.

In addition, by using not only the layer information of the drilled ball but also various additional information, it is possible to more effectively predict the layer information of the undrilled section.

1 is a schematic block diagram of a ground layer information prediction system according to the present invention.
2 is a table showing an example of input data input to the drilling area information input unit.
3 is a view showing the general structure of the neural network.
4 is a diagram for explaining a learning and verification method performed in the ground layer information prediction system of FIG. 1.
5 to 7 are diagrams for comparison of measured layer information and predicted layer information.
8 is a diagram showing an example of the structure of an artificial neural network.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

1 is a schematic block diagram of a ground layer information prediction system according to the present invention. In FIG. 1, the ground layer information prediction system 100 includes a drilling area information input unit 110, a borehole classification unit 120, a test information calculation unit 130, a parameter optimization unit 140, and a non-drilling area information input unit ( 150), a layer information prediction unit 160, and a normalization unit 170, the parameter optimization unit 140 is an error value calculation unit 142, a parameter value calculation unit 144, and a parameter update unit 146 ) Again.

In FIG. 1, each component of the ground layer information prediction system 100 may be implemented only by hardware, but it may be generally implemented together with hardware and software operating on hardware.

The drilling area information input unit 110 receives input information including coordinate information for a plurality of boreholes in the drilled area and layer information measured in the borehole. At this time, the coordinate information may further include elevation information of the ground floor information calculation area. According to such a configuration, unlike conventional 3D kriging, which is difficult to easily adopt due to high algorithm complexity, it is possible to more effectively predict layer information of the ground using 3D coordinates.

Further, the input information may further include underground water level information in the ground layer information calculation area. According to this configuration, by using not only the layer information of the drilled ball but also various additional information, it is possible to more effectively predict the layer information of the undrilled section.

2 is a table showing an example of input data input to the drilling area information input unit. In FIG. 2, it is possible to confirm an example in which a three-dimensional position of a borehole, underground water level information, and layered information measured in a borehole are input.

The normalization unit 170 normalizes the input information. In the table of FIG. 2, the z-coordinate represents a large difference between the x and y coordinates and the absolute size. According to the normalization unit 170, in this way, even if the absolute size of the input data has a large difference, the effect on the prediction result Can be adjusted. At this time, normalization may be performed by various methods, for example, by the following equation.

The borehole classification unit 120 classifies a plurality of boreholes into a borehole for learning and a borehole for testing. At this time, the classification of the borehole for learning and the borehole for testing may be performed repeatedly at random. According to this configuration, it is possible to obtain more effective prediction results by variously configuring a combination of a learning borehole and a test borehole.

The test information calculation unit 130 calculates the layer information of the test borehole using data measured in the borehole for learning and a preset layer information calculation algorithm. At this time, the layer information calculation algorithm may be an artificial neural network based algorithm.

Recently, technologies based on Artificial Neural Network (ANN) are widely used with high performance in various fields. An artificial neural network is a method of machine learning that mimics the structure of a human neuron. It creates several layers of neural networks, multiplies the weights of signals in neurons in each neural network, multiplies them by weights, and then transmits them to the next neuron.

3 is a view showing the general structure of the neural network. It consists of a total of three layers, consisting of an input layer, a hidden layer, and an output layer. Here, the hidden layer may be composed of several layers, and when it is composed of a large number of layers, it may be expressed as a deep neural network (DNN) when the neural network is expressed as deep.

The number of layers and the number of neurons must be determined as the hidden layer. In the example of FIG. 3, the number of units in the input layer is 6, the hidden layer is 4, and the output layer is 2. The input layer and the output layer determine the number of units depending on what data is received and what information is predicted. The number of layers and units of the hidden layer can be variably brought, and can be derived experimentally through multiple learnings. The relationship formula of the units in the neural network is as follows.

Here, i and j indicate an index of the number of units of each layer, and n denotes an index for a layer. After taking the linear combination of w ⁿ , it is connected to the neuron in the next layer through the activation function f (x).

Here, the activation function is taken after the linear combination in order to take a nonlinear function because the artificial neural network structure performs only a linear combination. In the present invention, the Rectified Linear Unit (ReLU) having the form f (x) = max (x, 0) Function was utilized.

The parameter optimization unit 140 compares the calculated layer information with the measured layer information of the test borehole to calculate an optimization value of the parameter in the layer information calculation algorithm.

To this end, the error value calculating unit 142 calculates an error value of the calculated layer information and the layer information of the measured borehole for testing, and the parameter value calculating unit 144 calculates a new parameter value to make the error value smaller. The parameter update unit 146 updates the previous parameter with the calculated parameter value. According to such a configuration, it is possible to secure more accurate prediction results by repeatedly updating the parameters of the prediction algorithm.

The neural network model is generally classified into regression and classification problems, and prediction of the ground layer information performed in the present invention belongs to regression because it is a problem of predicting continuous ground layer information. In this regression analysis, the cost function can be configured in the form of Minimum-Mean-Squared-Error (MMSE), and the formula is as follows.

와

는 지반 층상정보의 실측과 예측된 정보를 나타내며, w와 b는 인공신경망의 가중치와 바이어스, x는 입력 데이터를 각각 가리킨다. 또한, p는 norm의 차수를 나타내며, p=2일 경우에 l2-norm이라고도 불리는 MMSE 형태로 비용 함수가 결정된다. here,

Wow

Indicates the measured and predicted information of the ground layer information, w and b are the weight and bias of the artificial neural network, and x is the input data, respectively. In addition, p represents the order of norm, and when p = 2, the cost function is determined in the form of MMSE, also called l2-norm.

The process of determining the weight and bias in the direction in which the cost function of the above formula is minimized can be expressed as learning. Various methods exist for updating the weight, but in the present invention, an adaptive moment estimation (ADAM) technique is used. A batch normalization layer can be added between the affine layer and the activation function to accelerate the learning convergence rate.

Batch normalization reduces the dependence of the initial value, and also has the effect of suppressing the phenomenon of overfitting even without dropout or regularization. In the present invention, batch normalization was also performed, and batch normalization was performed before the activation function appeared in the affine layer.

The non-drilling area information input unit 150 receives input information for the undrilled area, and the ground layer information prediction unit 160 calculates the layered information of the undrilled area using the parameter to which the optimization value is applied. According to such a configuration, it is possible to more effectively predict the layer information of the undrilled section from the layer information of the surrounding borehole by engineering judgment rather than the designer's experience.

FIG. 4 is a diagram illustrating a learning and verification method performed in the ground layer information prediction system of FIG. 1. As shown in FIG. 4, the unit of the artificial neural network can be learned by using the layered information drilled. In the example of FIG. 4, to verify the performance of the trained artificial neural network model, the layer information of Yi Sun-Shin Bridge was used, 71 of the 84 ground layers were used as the learning data set, and the remaining 13 holes were verified data set. It was utilized as.

A total of 4 input information was used, and 3D location information and ground water level (GWL) information were used. 5 to 7 are diagrams for comparison of measured layer information and predicted layer information, respectively. 5 to 7 show the measured layer information for the 13 holes and the layer information predicted by using an artificial neural network and a normal kriging technique. Both methods show a pattern similar to the measured stratified information, but it is considered that more accurate prediction results can be shown by easily applying various additional information to the artificial neural network in the future.

8 is a view showing an example of the structure of an artificial neural network. Figure 8 An example of the structure of an artificial neural network for predicting ground layer information is shown. In the case of regular kriging, the complexity of the algorithm varies greatly depending on whether 2D or 3D kriging is used, and it is necessary to consider the cross variogram for using additional factors, but in the case of artificial neural networks, only one more unit of the input layer is added. It is easy to do in the form.

As can be seen in FIG. 8, the total number of input units is 4, and x, y coordinates of spatial information, z coordinates that can be called elevation, and even a ground water level (GWL) were utilized. The hidden layer was composed of a total of 3 layers, and was configured to have a total of 16 hidden units. Since the output layer only needs to output one result of the ground layer information to be predicted, the number of units is one.

In summary, the present invention relates to a technique for estimating the layer information of an unexcavated section based on an artificial neural network based on the layered information of the drilled ball when calculating the amount of earthwork, and calculating the amount of earthwork based on this.

In order to learn the artificial neural network, a neural network model was constructed and the neural network model was trained by using ground 3D location information and check information. It can be seen that the ground layer information was predicted in a distribution similar to that of the normal kriging technique, and in the section with low ground layer information, the prediction was more accurate than the normal kriging technique.

In addition, according to the present invention, it is easy to utilize not only the layer information of the drilled ball but also various additional information. Accordingly, in addition to construction information, topographical information (altitude, slope, curvature, topographical index, wetness index, etc.), geological information (single-floor, water system, land use, etc.), geophysical exploration information, etc. are constructed based on GIS and then ground It is possible to secure refined data for the use of the neural network through analysis and quantification of the topographical and geological characteristics of the structure and sensitivity of the drilling column and physical exploration data.

Although the invention has been described in terms of some preferred embodiments, the scope of the invention should not be limited thereby, but should also extend to variations or improvements of the embodiments supported by the claims.

100: ground layer information prediction system
110: drilling area information input unit
120: borehole classification unit
130: test information calculation unit
140: parameter optimization unit
142: error value calculator
144: parameter value calculation unit
146: parameter update unit
150: Missing area information input unit
160: layered information prediction unit
170: normalization unit

Claims (15)

A drilling area information input unit for receiving input information including coordinate information for a plurality of boreholes in the drilled area and layered information measured in the borehole;
A borehole classifying unit classifying the plurality of boreholes into a learning borehole and a test borehole;
A test information calculation unit for calculating the layer information of the test borehole using data measured in the learning borehole and a preset layer information calculation algorithm;
A parameter optimization unit that compares the calculated layer information with the measured layer information of the test borehole to calculate an optimization value of a parameter in the layer information calculation algorithm;
An undrilled area information input unit that receives the input information for an undrilled area; And
And a layered information prediction unit for calculating layered information of the non-drilled area using the parameter to which the optimization value is applied.
The method according to claim 1,
The layered information calculation algorithm is an artificial neural network based algorithm, characterized in that the ground layer information prediction system.
The method according to claim 2,
And a normalization unit that performs normalization on the input information.
The method according to claim 3,
The classification of the borehole for learning and the borehole for testing is carried out randomly and repeatedly.
The method according to claim 4, The parameter optimization unit,
An error value calculator configured to calculate an error value between the calculated layer information and the measured layer information of the test borehole;
A parameter value calculator for calculating a new parameter value to make the error value smaller; And
And a parameter updating unit that updates a previous parameter with the calculated parameter value.
The method according to claim 5,
The coordinate information further includes elevation information of the ground layer information calculation area.
The method according to claim 6,
The input information further comprises underground water level information of the ground layer information calculation area.
A drilling region information input step of receiving input information including coordinate information for a plurality of boreholes in the drilled region and layered information measured in the borehole;
A borehole classification step of classifying the plurality of boreholes into a learning borehole and a test borehole;
A test information calculation step of calculating the layer information of the test borehole using data measured in the learning borehole and a preset layer information calculation algorithm;
A parameter optimization step of comparing the calculated layer information with the measured layer information of the test borehole to calculate an optimization value of a parameter in the layer information calculation algorithm;
A non-drilling area information input step of receiving the input information for an undrilled area; And
And a layer information prediction step of calculating layer information of the non-drilled region using the parameter to which the optimization value is applied.
The method according to claim 8,
The layer information calculation algorithm is an artificial neural network based algorithm, characterized in that the ground layer information prediction method.
The method according to claim 9,
And a normalizing step of performing normalization on the input information.
The method according to claim 10,
The classification of the borehole for learning and the borehole for testing is performed repeatedly and randomly.
The method according to claim 11, wherein the parameter optimization step,
An error value calculating step of calculating an error value between the calculated layer information and the measured layer information of the test borehole;
A parameter value calculating step of calculating a new parameter value to make the error value smaller; And
And a parameter updating step of updating a previous parameter with the calculated parameter value.
The method according to claim 12,
The coordinate information further comprises elevation information of the ground layer information calculation area.
The method according to claim 13,
The input information further comprises ground level information of the ground layer information calculation area.
A storage medium recording a computer readable program for executing the method of claim 8.

Images