A mine area geological landslide displacement prediction method based on a radial basis function neural network

Abstract

The application discloses a mine area geological landslide displacement prediction method based on a radial basis function neural network, relates to the technical field of mine area geological landslides, and comprises the following steps: acquiring time sequence parameter data of a mine area; extracting statistical features of the time sequence parameter data by adopting a time convolution network; inputting the statistical features into a trained radial basis function neural network model to obtain mine area geological landslide displacement; and training the radial basis function neural network model by taking statistical features of historical time sequence parameter data of the mine area as input and taking historical cumulative landslide displacement of the mine area as output. The application can realize accurate prediction and analysis of the mine area geological landslide situation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of mine geological landslide, and particularly relates to a mine geological landslide displacement prediction method and device based on a radial basis function neural network, a medium and a product. BACKGROUND

[0002] Mine geological landslide is one of the common geological disasters in mining production, which may lead to serious consequences such as personnel casualties, facility losses and environmental damage. Therefore, it is of great significance to accurately and timely analyze and predict the situation of mine geological landslide. The traditional geological landslide situation analysis method often relies on experience model or simple statistical method, which is limited by geological conditions, monitoring data quality and other factors, and is difficult to meet the accurate prediction demand in complex geological environment. SUMMARY

[0003] The purpose of the present application is to provide a mine geological landslide displacement prediction method and device based on a radial basis function neural network, a medium and a product, which realizes accurate prediction and analysis of the situation of mine geological landslide by using a radial basis function neural network.

[0004] To achieve the above purpose, the present application provides the following scheme:

[0005] A mine geological landslide displacement prediction method based on a radial basis function neural network, comprising:

[0006] obtaining time series parameter data of a mine area;

[0007] extracting statistical features of the time series parameter data by using a time convolution network;

[0008] inputting the statistical features into a trained radial basis function neural network model to obtain the displacement of mine geological landslide; the radial basis function neural network model is trained with the statistical features of the historical time series parameter data of the mine area as input and the historical cumulative landslide displacement of the mine area as output.

[0009] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0010] (1) The time convolution network (Temporal Convolutional Network, TCN) can capture long-term dependencies and local features in time series data, and is suitable for feature extraction of multi-dimensional monitoring data. The present application extracts statistical features of the time series parameter data by using a time convolution network, which can better capture the features in the time series parameter data. These features can effectively represent various patterns and trends of the time series parameter data;

[0011] (2) The radial basis function neural network (RBFNN) model can adapt to the change of the mining area monitoring condition by learning the features in the data, has good generalization ability, and thus improves the robustness and reliability of the model. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments will be briefly introduced below. Obviously, the drawings in the following description only represent some of the embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0013] Figure 1 The flowchart of the mining area geological landslide displacement prediction method based on the radial basis function neural network provided by the embodiments of the present application is shown in the figure.

[0014] Figure 2 The schematic diagram of the statistical feature and time sequence parameter data provided by the embodiments of the present application is shown in the figure.

[0015] Figure 3 The correlation analysis diagram of the statistical feature and time sequence parameter data provided by the embodiments of the present application is shown in the figure.

[0016] Figure 4 The schematic diagram of the change of the loss function of the radial basis function neural network model with the training round provided by the embodiments of the present application is shown in the figure.

[0017] Figure 5 The schematic diagram of the change of the loss function of the radial basis function neural network model with the verification round provided by the embodiments of the present application is shown in the figure.

[0018] Figure 6 The landslide situation prediction analysis diagram of the radial basis function neural network model on the test set provided by the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the scope of protection of the present application.

[0020] The purpose of this invention is to provide a method, device, medium, and product for predicting geological landslide displacement in mining areas based on radial basis function neural networks (RBNs). The invention extracts effective features from time-series parameter data through TCNs and inputs them into a trained RBN model for landslide displacement prediction, enabling accurate prediction and analysis of geological landslide trends in mining areas with higher accuracy, faster convergence speed, and stronger robustness.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] Example 1

[0023] This embodiment provides a method for predicting geological landslide displacement in mining areas based on radial basis function neural networks, such as... Figure 1 As shown, the method includes the following steps:

[0024] S1: Obtain time-series parameter data of the mining area.

[0025] Collect and organize time-series parameter data of the mining area, including information such as upper deep displacement (meters), upper shallow displacement (meters), middle deep displacement (meters), middle shallow displacement (meters), lower deep displacement (meters), lower shallow displacement (meters), vertical length (cm), total horizontal stress (kilopascals), effective horizontal stress (kilopascals), and hourly rainfall (mm / hour) at each time point.

[0026] S2: Use a temporal convolutional network to extract the statistical features of the time-series parameter data.

[0027] For each time point in the data, TCN is used to extract statistical features of each parameter, such as mean, variance, maximum, and minimum values. Compared to traditional feature extraction methods, such as manually designed features or statistically based feature extraction, TCN can automatically learn the temporal dependencies in the input sequence, thereby better capturing information from the data.

[0028] Statistical features are extracted using TCN, including:

[0029] S21: The temporal parameter data is convolved by the convolutional layer of the temporal convolutional network to obtain the convolved feature sequence.

[0030] In the first layer of the TCN, a convolution operation with kernel size k is used, denoted as f(·). For time-series parameter data of length T, X = {x1, x2, ..., x...} T}, where x i Given the data at time step i, the result of the convolution operation can be represented as:

[0031] C1=f(X)={c1,c2,…,c T-k+1}

[0032] where C1 is the convolved feature sequence, c i denotes the feature obtained at time step i.

[0033] S22: Max-pooling the convolved feature sequence C1 through a pooling layer of the temporal convolutional network to obtain a pooled feature sequence.

[0034] Max-pooling the convolved feature sequence C1 to obtain a pooled feature sequence P1:

[0035] P1={p1,p2,…,p T-k+1}

[0036] where P1 denotes the pooled feature sequence, p i denotes the pooling result at time step i.

[0037] S23: Concatenating a plurality of pooled feature sequences to obtain a statistical feature of the time series parameter data.

[0038] The convolution and pooling operations are repeated for multiple layers, and each layer extracts new features through convolution and pooling operations. Assuming that there are L layers of convolution and pooling operations, and the same size of convolution kernel and the same pooling operation are used in each layer. Then, after the lth layer, the pooled feature sequence P l is obtained.

[0039] The pooled feature sequences obtained by each layer are concatenated to form the final statistical feature F. If the number of convolution kernels of each layer is n, then the length of the final statistical feature F is n x L.

[0040] F={f1,f2,…,f n×L}

[0041] Through this process, the TCN converts the input time series parameter data into a higher-dimensional output feature sequence.

[0042] In this embodiment, the TCN parameters are set as input_size=10, output_size=32, num_channels=

[32] *3, kernel_size=3, and dropout=0.2. The TCN converts the input 10-dimensional time series data into a higher-dimensional 32-dimensional output feature sequence with strong correlation, as shown in Figures 2-3 .

[0043] The time series parameter data and the corresponding statistical features are visualized and compared, a time series graph is drawn, the mode and trend of the original data are displayed, and a graph of the statistical features is drawn at the same time, so that whether similar mode and trend are captured can be observed.

[0044] S3: inputting the statistical features into a trained radial basis function neural network model to obtain the geological landslide displacement of the mining area; the radial basis function neural network model is trained with the statistical features of the historical time series parameter data of the mining area as input and the historical cumulative landslide displacement of the mining area as output.

[0045] The specific steps of establishing the radial basis function neural network model include: first, determining the input size and output size of the RBFNN model, the input size is the dimension of the features extracted by the TCN, and the output size is usually the dimension of the target to be predicted; the center points of the radial basis function are determined in advance, the number of center points will be adjusted according to the feature dimension of the data set and the model complexity; the center points of the radial basis function are initialized by using a random initialization method; the calculation and output layer of the radial basis function are designed, the distance between the input sample and the center point of the radial basis function is calculated, and then the output of the radial basis function is calculated through the radial basis function.

[0046] In this embodiment, the RBFNN parameters are set as num_centers=20, hidden_size=100, num_epochs=2600, and batch_size=512.

[0047] Regarding the RBFNN, the specific operation idea is: using RBF as the "base" of the hidden unit to constitute the hidden layer space, so that the input vector can be directly mapped to the hidden space without the need of passing through the weight connection.

[0048] 1. Radial basis function (Radial Basis Function, RBF for short) is a kind of scalar function symmetrical along the radial direction. It is usually defined as a monotonic function of the Euclidean distance between any point x in space and a certain center x c , which can be denoted as k(||x-x c ||), and its role is often local, that is, when x is far away from x c , the function value is very small.

[0049] When the center point of RBF is determined, the mapping relationship is also determined. The mapping from the hidden layer space to the output space is linear, that is, the output of the network is the linear weighted sum of the hidden unit outputs, and the weight here is the adjustable parameter of the network. Among them, the role of the hidden layer is to map the vector from low dimension p to high dimension h, so that the low-dimensional linearly inseparable case can become linearly separable in high dimension, which is the main idea of the kernel function. In this way, the mapping from input to output of the network is nonlinear, while the network output is linear with respect to the adjustable parameters. The weights of the network can be directly solved by linear equations, thereby greatly speeding up the learning speed and avoiding local minimum problems.

[0050] The activation function of the RBFNN model can be represented as:

[0051]

[0052] where x p is the pth input sample, n is the number of output samples or the number of classifications, c i is the kernel function center, and σ is the radial basis width of the function, which determines how fast the radial basis function decreases. RBF like Gaussian function has good local characteristics, which responds significantly only in a neighborhood near the center point, while the function value decreases monotonically with the increase of the distance from the center point and gradually approaches 0.

[0053] The structure of the radial basis neural network can obtain the output of the network as:

[0054]

[0055] Of course, the least squares loss function is represented as:

[0056]

[0057] where L represents the value of the loss function, n represents the number of samples, which is the total number of samples in the data set, m represents the number of features, j represents a specific sample index, y j represents the actual value of the jth sample, represents the predicted value of the jth sample, and θ j represents the parameters corresponding to the jth sample.

[0058] 2. Self-organizing center learning method:

[0059] Step 1: Unsupervised learning process, solving the center and variance of the hidden layer basis function; Step 2: Supervised learning process, solving the weight between the hidden layer and the output layer.

[0060] First, select h centers for k-means clustering. For the radial basis of the Gaussian kernel function, the variance is solved by the formula:

[0061]

[0062] c max is the maximum distance between the selected center points.

[0063] The connection weights between the neurons of the hidden layer and the output layer can be calculated directly by the least square method, i.e. the partial derivative of the loss function with respect to w is set to 0, which can be simplified to the following calculation formula:

[0064]

[0065] 3. Complete interpolation method:

[0066] The reason why RBF can fit any function can be understood from the perspective of interpolation method. To fit a curve, the expression function of the curve can be obtained by interpolation method, such as polynomial interpolation, Lagrange interpolation, etc. RBF interpolation is a combination of a series of accurate interpolation methods; that is, the surface must pass through each measured sampling value.

[0067] RBF interpolation selects an approximation function or mapping function of a norm, such as:

[0068]

[0069] where the approximation function represents the weighted sum of each RBF, and each RBF corresponds to a different center and the weight is represented by an approximation coefficient . Since the approximation function is a linear weighted average of each RBF, the weight coefficient can be obtained by solving the linear equation set by the least square method.

[0070] The nonlinear function is RBF, and its parameters depend on the Euclidean norm of n-dimensional space, such as:

[0071]

[0072] For RBF interpolation, its characteristics are: in the input data set, the points close to the center point have the greatest contribution to the mapping function.

[0073] The complete interpolation method requires all interpolation points to pass through the surface. Since RBF interpolation uses each x, it is a complete interpolation form. The problem is that when the sample contains noise, the neural network will fit an incorrect surface, thereby reducing the generalization ability. In addition, if the data of sample x is much larger than the nonlinear function The solution becomes unstable, that is, the solution is over-determined equations. Therefore, it is necessary to introduce a regularization method, that is, a regularization term is usually added.

[0074]

[0075] In the inference process of the RBFNN model, RBF is used as the "base" of the hidden unit to form the hidden layer space, and the input vector is directly mapped to the hidden space (without passing through the weight mapping). When the center point of the RBF is determined, the mapping relationship is also determined (the center point is obtained through K-MEANS clustering). The mapping from the hidden layer space to the output space is linear (the weight value is obtained through the least square method). The relationship between the factors that make the linearly non-separable into linearly separable is a complex nonlinear relationship. The RBFNN model can better approximate these complex relationships through its nonlinear radial basis function activation unit, thereby improving the expression ability and prediction accuracy of the model.

[0076] Further, the training process of the radial basis function neural network model is:

[0077] (1) Construct a data set; the data set is the historical time series parameter data of the mining area, and the label is the historical cumulative landslide displacement (mm); according to the ratio of 8:1:1, the data set is randomly divided into a training set, a validation set and a test set.

[0078] The data used in this embodiment is part of a public data set. The data set records the monitoring of the Twin Towers landslide in Zenna, near Northern California, every hour; the number is about 23600.

[0079] (2) Extract the first column and the last column of the data set as the timestamp and label; use the TCN network to extract statistical features to obtain a set of high-dimensional feature representations.

[0080] (3) Take the extracted statistical features as input and the historical cumulative landslide displacement as label, train the radial basis function neural network model, and evaluate the trained model: the evaluation indicators include root mean square error (RMSE), mean absolute error (MAE), and determination coefficient (R 2 ), to evaluate the prediction performance and generalization ability of the model. During network training, the validation set is tested every 30 rounds. Through the loss and evaluation indicators in the validation set, as shown in Table 1, the best performing model is finally saved. Figures 4-5 The loss of the training process and the loss of the validation process are shown in Table 1.

[0081] Table 1 Evaluation indicators of the best performing model

[0082] Loss RMSE MAE [R 2 ]]> 0.0788 0.2807 0.1278 0.8507

[0083] The calculation formula of the root mean square error RMSE is:

[0084]

[0085] The calculation formula of the mean absolute error MAE is:

[0086]

[0087] The calculation formula of the determination coefficient R 2 is:

[0088]

[0089] Wherein, n is the sample quantity, y k is the observed actual sample value, is the average value of the actual sample value, is the model prediction value.

[0090] In the application process, the trained RBFNN model is used for landslide displacement prediction, and the output actual label and prediction value comparison chart is output, and the normal range and abnormal range are automatically divided according to the prediction value. Specifically, the mean and standard deviation of the prediction value are calculated, a normal range is defined according to the mean and standard deviation, for example, it is considered to be normal within the range of mean plus or minus twice the standard deviation, and the prediction value exceeding the normal range is marked as an abnormal value, and the model prediction is completed, as shown in Figure 6 .

[0091] This embodiment can realize accurate prediction and analysis of the geological landslide situation in the mining area by combining the geological landslide related monitoring data and the geological characteristics of the mining area, using the radial basis function neural network model for modeling and training. This method not only can make full use of multi-source monitoring data information, but also can adapt to complex conditions under different geological environments, and improve the accuracy and reliability of the geological landslide situation analysis.

[0092] Embodiment two

[0093] A computer device, comprising: a memory, a processor to store a computer program on the memory and executable on the processor, and the processor executes the computer program to realize the radial basis function neural network based landslide displacement prediction method in the mining area in embodiment one.

[0094] Embodiment three

[0095] A computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to realize the radial basis function neural network based landslide displacement prediction method in the mining area in embodiment one.

[0096] Embodiment four

[0097] A computer program product comprises a computer program which, when executed by a processor, implements the method for predicting landslide displacement of a mine area based on a radial basis function neural network in the embodiment one.

[0098] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments of the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments of the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments of the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, etc., without being limited thereto.

[0099] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0100] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application range will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.

Claims

1.A method for predicting geological landslide displacement in a mining area based on a radial basis function neural network, characterized in that, The method comprises: obtaining time series parameter data of a mining area; the time series parameter data comprises upper deep displacement, upper shallow displacement, middle deep displacement, middle shallow displacement, lower deep displacement, lower shallow displacement, vertical length, total horizontal stress, effective horizontal stress and hourly rainfall at each time point; statistical features of the time series parameter data are extracted by using a time convolution network; the statistical features comprise mean, variance, maximum and minimum; statistical features of the time series parameter data are extracted by using a time convolution network, specifically comprising: convolution operation is performed on the time series parameter data by a convolution layer of the time convolution network to obtain a feature sequence after convolution; maximum pooling is performed on the feature sequence after convolution by a pooling layer of the time convolution network to obtain a feature sequence after pooling; a plurality of feature sequences after pooling are spliced to obtain statistical features of the time series parameter data; the statistical features are input into a trained radial basis function neural network model to obtain geological landslide displacement of the mining area; the radial basis function neural network model is trained by taking statistical features of historical time series parameter data of the mining area as input and taking historical cumulative landslide displacement of the mining area as output; the radial basis function neural network model adopts a least square loss function; an expression of the loss function is: in, This represents the value of the loss function. The number of samples indicates the total number of samples in the dataset. Indicates the number of features, Indicates a specific sample index. Indicates the first The actual value of each sample Indicates the first The predicted value for each sample, Indicates the first The parameters corresponding to each sample. 2.The mine geological landslide displacement prediction method based on the radial basis function neural network according to claim 1, wherein, an expression of an activation function of the radial basis function neural network model is: wherein, is an activation function of the radial basis function neural network model, is the p-th input sample, is the i-th kernel function center, and σ is a radial basis width of the kernel function. 3.The mine geological landslide displacement prediction method based on the radial basis function neural network according to claim 1, wherein, The method further comprises: evaluating prediction performance and generalization ability of the trained radial basis function neural network model by evaluation indexes; the evaluation indexes comprise root mean square error, mean absolute error and determination coefficient. 4.The mine geological landslide displacement prediction method based on the radial basis function neural network according to claim 3, characterized in that, a calculation formula of the root mean square error RMSE is: a calculation formula of the mean absolute error MAE is: The decision coefficient R 2 The calculation formula is: where n is the number of samples, is the observed actual sample value, is the mean of the actual sample values, is the model predicted value.

Images