A geological disaster prediction and early warning method

Abstract

The present application relates to the technical field of geological disaster early warning, in particular to a geological disaster prediction and early warning method, which uses a machine learning algorithm to process geological disaster trend data, and contains two geological disaster prediction models, one of which predicts trend data, and the other predicts periodic data, the geological disaster trend prediction model is tested using a test set after training and verification, the accuracy of the prediction is obtained, and a geological disaster factor correlation algorithm is used to quantify factors affecting geological disasters, and the input variables are selected according to the correlation, thereby improving the accuracy of prediction and early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geological disaster early warning, and particularly relates to a geological disaster prediction and early warning method. BACKGROUND

[0002] A reliable geological disaster prediction and early warning method is a reasonable way to reduce the risk of geological disasters. If such a method that can accurately predict and warn geological disasters can be successfully implemented, it will have great social value and important economic benefits.

[0003] Geological disaster prediction and early warning models can be divided into two categories: physical models and data-driven models. Compared with physical models, data-driven models are more popular because they are simple in process, accurate in prediction, and lower in cost.

[0004] Existing geological disaster early warning models are all based on real-time warning functions, and give warning information only when disasters occur, leaving decision makers with insufficient time to face geological disasters. Therefore, there is an urgent need for a geological disaster prediction and early warning method that can predict future geological disasters and give decision makers enough time to face geological disasters, providing strong support for disaster prevention and mitigation. SUMMARY

[0005] The purpose of the present application is to provide a geological disaster prediction and early warning method, which aims to dynamically predict geological disasters through artificial intelligence algorithms, and solves the technical problem of low prediction and early warning accuracy of traditional static geological disaster prediction and early warning methods.

[0006] To achieve the above-mentioned purpose, the present application provides a geological disaster prediction and early warning method, comprising the following steps:

[0007] Obtaining disaster and environmental data;

[0008] Performing outlier replacement processing on the disaster and environmental data;

[0009] Performing standard normalization processing on the processed data;

[0010] Establishing a geological disaster prediction and early warning model;

[0011] Respectively selecting actual trend data and actual period data of geological disasters to train the geological disaster prediction and early warning model, and obtaining prediction trend data and prediction period data;

[0012] Combining the prediction trend data and the prediction period data to obtain prediction cumulative data;

[0013] Performing inverse standard normalization processing on the prediction cumulative data to obtain final geological disaster prediction data.

[0014] The disaster and environmental data include rainfall, reservoir water level, soil type, slope gradient, slope orientation, air humidity, and soil moisture.

[0015] In the process of replacing outliers, the data to be replaced is the average of the previous and next data. If it is the first or last data, it is deleted directly.

[0016] The standard normalization process is specifically handled according to the following formula:

[0017] y i =(x i -x min ) / (x max -x min ), x i For the i-th original data, x min Let x be the minimum value in the data. max The maximum value in the x data.

[0018] The geological disaster prediction and early warning model includes calculations in two directions: forward propagation and backward propagation. Forward propagation calculates f each time. t i t o t h t C t , The numerical value is obtained by calculating and adjusting the weights and bias W by reducing the sum of error terms during each backpropagation. f W o W i W c b f b o b i b c The value of h is determined, and the sum of the error terms is minimized. Training continues for a certain number of iterations until the sum of the error terms is h. t With y t The summation of the differences, where f t Represents the forget gate at time t, i t Represents the input gate and o at time t. t Represents the output gate and h at time t. t Represents the output data at time t, y t Represents the t-th actual data, C t Represents the cell state at time t, W f The weight of the forget gate, W o The weights W of the output gate are represented. i The weights W of the input gate are represented. c Indicates the candidate weights, b f Indicates the bias of the forget gate, bi Indicates the input gate bias, b o Indicates the output gate bias, b c Indicates alternative bias, This represents the unit candidate status at time t.

[0019] Specifically, the process of obtaining actual trend data and actual periodic data of geological disasters involves using a geological disaster data decomposition algorithm to obtain actual trend data of geological disasters, and subtracting the actual trend data from the actual cumulative data to obtain the actual periodic data.

[0020] The inverse standard normalization process is specifically handled according to the following formula:

[0021] y i =x i (x max -x min )+x min x min x is the minimum value in the original x data. max The maximum value in the original x data, y i For the final prediction data, x i This is the i-th data point calculated in the model.

[0022] This invention provides a method for predicting and warning of geological disasters. It uses machine learning algorithms to process geological disaster trend data and includes two geological disaster prediction models. One model predicts trend data, while the other model's algorithm part predicts periodic data. After training and validation, the geological disaster trend prediction model is tested using a test set to obtain the prediction accuracy. In addition, a geological disaster factor correlation algorithm is used to quantify the factors affecting geological disasters, and highly correlated factors are selected as input variables, thereby improving the accuracy of prediction and warning. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a geological disaster prediction and early warning method according to the present invention.

[0025] Figure 2 This is a structural diagram of the geological disaster trend prediction output model of the present invention.

[0026] Figure 3This is a specific embodiment of the present invention, showing 108 months of disaster body deformation and displacement data, rainfall data, and reservoir water level data.

[0027] Figure 4 This is a comparison diagram of the relationship between the actual cumulative displacement, actual trend displacement, and actual periodic position of the disaster body in a specific embodiment of the present invention.

[0028] Figure 5 This is a comparison chart of the predicted trend and the actual trend of the geological disaster trend prediction model according to a specific embodiment of the present invention.

[0029] Figure 6 This is a comparison chart of actual and predicted rainfall and reservoir water level when used as input variables in a specific embodiment of the present invention.

[0030] Figure 7 This is a comparison chart of the actual disaster data and the model prediction data in a specific embodiment of the present invention. Detailed Implementation

[0031] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0032] This invention provides a method for predicting and warning of geological disasters, comprising the following steps:

[0033] S1: Acquire disaster and environmental data;

[0034] S2: Perform outlier replacement processing on disaster and environmental data;

[0035] S3: Perform standard normalization on the processed data;

[0036] S4: Establish a geological disaster prediction and early warning model;

[0037] S5: Select actual trend data and actual periodic data of geological disasters respectively to train the geological disaster prediction and early warning model, and obtain prediction trend data and prediction periodic data;

[0038] S6: Combine forecast trend data and forecast cycle data to obtain forecast cumulative data;

[0039] S7: Perform inverse standardization on the predicted cumulative data to obtain the final geological disaster prediction data.

[0040] Specifically, the following will provide further explanation in conjunction with the specific implementation steps:

[0041] 1. Use high-precision GPS receivers, geological disaster monitoring sensors, remote sensing satellites, or manual measurements to monitor geological disaster deformation and obtain data on other factors that may affect geological disasters through various means, such as rainfall, reservoir water level, soil type, slope gradient, slope orientation, air humidity, soil moisture, etc.

[0042] 2. Perform outlier replacement processing on the obtained geological hazard data. The replacement data is the average of the previous and next data. If it is the first or last data, delete it directly.

[0043] 3. After replacing outliers, all data are normalized to make the data more stable.

[0044] 4. Standard normalization process: y i =(x i -x min ) / (x max -x min ), x i For the i-th original data, x min Let x be the minimum value in the data. max The maximum value in the x data.

[0045] 5. A geological disaster prediction and early warning model was established, and a geological disaster data decomposition algorithm, a geological disaster factor correlation algorithm, and a geological disaster trend prediction model were proposed. The flowchart of the entire algorithm is shown below. Figure 1 As shown.

[0046] 6. Actual cumulative disaster data = actual disaster trend data + actual disaster cycle data. The actual trend data of geological disasters is obtained by using a geological disaster data decomposition algorithm. The actual cycle data is obtained by subtracting the actual trend data from the actual cumulative data.

[0047] 7. The internal parameters of the correlation algorithm for geological disaster factors are: T is the actual trend data, D is the actual cumulative data; α is a constant term between 0 and 1, reflecting the weighted decrease rate, and t represents the t-th time.

[0048] T t =αD t-1 +(1-α)T t-1

[0049] 8. After obtaining the trend data at time t, subtract the trend data at time t from the original cumulative data at time t to obtain the periodic data at time t.

[0050] 9. By using two geological disaster trend prediction models, the predicted trend data and predicted period data were calculated respectively, thus solving the problems of gradient explosion and gradient vanishing.

[0051] 10. Use the geological hazard factor correlation algorithm to calculate the correlation between each environmental factor and the actual periodic data. Select factors whose geological hazard factor correlation algorithm values ​​exceed a certain set value as input variables for the prediction periodic data model. Assume there are two variables A = {a} i {i = 1, 2, ..., n} and B = {b} i Let i = 1, 2, ..., n, where n is the number of variables. Then the algorithm for the correlation degree of geological disaster factors is as follows:

[0052]

[0053] Where p(a) is the marginal probability density of A; p(b) is the marginal probability density of B; and p(a,b) is the joint probability density of A and B. Let D = {(a,b)} i ,b i Let the set of elements D, i = 1, 2, ..., n, consist of a finite set of ordered pairs. Let G divide the range of A into x segments and the range of B into y segments. G represents an x×y grid. MI(A,B) is calculated in each grid partition. There are multiple grids with the same x×y dimension, and the MI value of G is obtained based on the maximum value of MI(A,B) among different grids. The maximum MI equation for D under partition G is defined as:

[0054] MI * (D,x,y)=max MI(D|G)

[0055] Where D|G represents partitioning the data D using G. Although the geological hazard factor correlation algorithm relies on MI to represent grid quality, it is not simply an estimation of MI. The maximum normalized MI values ​​obtained under different partitions form a feature matrix, defined as MI(D). x，y The feature matrix can be defined as:

[0056]

[0057] The equation for the correlation degree algorithm of geological disaster factors is shown.

[0058]

[0059] Where B(n) represents the upper bound of the grid partition x×y.

[0060] 11. Geological Disaster Trend Prediction Algorithm: Geological disaster trend prediction is a typical neural network model with an input layer, hidden layers, and an output layer. The input layer connects to the hidden layers, and the hidden layers connect to the output layer. The hidden layer contains n nodes, which are also interconnected, and each connection has its own weight. The input data is divided into a training set, a validation set, and a test set. The training set is input from the input layer to train the geological disaster trend prediction model and improve its performance. After training, the validation set is used for cross-validation to optimize the model. The output layer outputs the prediction results, which are compared with the test set to obtain the model's evaluation accuracy.

[0061] The internal parameters of the geological disaster trend prediction model are: x t This represents the t-th input data (environmental data), f t Represents the forget gate at time t, i t Represents the input gate and o at time t. t Represents the output gate and h at time t. t Represents the output data at time t, y t Represents the t-th actual data (displacement data), C t Represents the cell state at time t, W f The weight of the forget gate, W o The weights W of the output gate are represented. i The weights W of the input gate are represented. c Indicates the candidate weights, b f Indicates the bias of the forget gate, b i Indicates the input gate bias, b o Indicates the output gate bias, b c σ represents the alternative bias, σ represents the sigmoid function, and tanh represents the tanh function. This represents the unit candidate status at time t. This represents the model learning rate. [] indicates that two vectors are concatenated, · indicates the product of matrices, and h... ft Indicates the forward forward time data output status, h bt H indicates the output status of reverse time data. t This is the output state of the geological hazard trend prediction model. The structure diagram of the geological hazard trend prediction output model is shown below. Figure 2 As shown.

[0062] f t =σ(w f ·[h t-1 ,x t ]+b f )

[0063] i t =σ(w i ·[ht-1 ,x t ]+b i )

[0064] o t =σ(w o ·[h t-1 ,x t ]+b o )

[0065]

[0066] h t =o t ·tanh(C t )

[0067]

[0068]

[0069] Geological hazard trend prediction involves two propagation directions: forward propagation and backward propagation. Forward propagation calculates f each time. t i t o t h t C t , The numerical value is obtained by calculating and adjusting the weights and bias W by reducing the sum of error terms during each backpropagation. f W o W i W c b f b o b i b c The value of h is determined, and the sum of the error terms is minimized. Training continues for a certain number of iterations until the sum of the error terms is h. t With y t The sum of the differences is calculated. As the number of training iterations increases, the sum of the error terms decreases, indicating that the model's prediction accuracy will increase and the model's parameters will better reflect the current geological disaster situation.

[0070] After training, the geological disaster trend prediction model was validated using validation set data, and adjustments were made to include... All parameters, including those listed above.

[0071] After validation, the geological disaster trend prediction model was tested using a test set to obtain the accuracy of the prediction.

[0072] Warning levels are issued based on pre-set thresholds.

[0073] 12. Forecast cumulative data = Forecast trend data + Forecast period data.

[0074] 13. The predicted data are subjected to inverse standard normalization to obtain the final geological disaster prediction data of the geological disaster prediction and early warning model.

[0075] 14. Inverse standard normalization process: y i =x i (x max -x min )+x min x min x is the minimum value in the original x data, xmax is the maximum value in the original x data, and y is the maximum value in the original x data. i For the final prediction data, x i This is the i-th data point calculated in the model.

[0076] Furthermore, the present invention also proposes a specific embodiment, please refer to [link / reference]. Figures 3 to 7 :

[0077] 1. Using data from the ZG118 monitoring point of a geological disaster in China, covering 108 months from January 2004 to December 2012, including disaster body deformation and displacement data, rainfall data, and reservoir water level data, such as... Figure 3 As shown.

[0078] 2. Normalize all data.

[0079] 3. The raw data is correlated with geological hazard factors, with α set to 0.25, to calculate the trend and periodic data. The relationship between the actual cumulative displacement, actual trend displacement, and actual periodic location of the hazard body is compared as follows: Figure 4 As shown.

[0080] 4. The correlation degree algorithm for geological disaster factors was used to calculate the correlation degree between geological disaster data and reservoir water level and precipitation. The results were all greater than 0.3. Therefore, reservoir water level and precipitation were selected as input variables for the prediction period data.

[0081] 5. The acquired geological disaster trend data, geological disaster cycle data, rainfall and reservoir water level data are divided into training set, validation set and test set according to the ratio of 8:4, 12, 12.

[0082] 6. Train one geological hazard trend prediction model using the geological hazard trend data training set, and train another geological hazard trend prediction model using the geological hazard cycle data, rainfall data, and reservoir water level data training set.

[0083] 7. W f W o W i W c bf b o b i b c Set the learning rate to the preset value.

[0084] 8. In each iteration, calculate f. t i t o t h t C t , H t h bt h ft The value of y is used to calculate y. t with h t The differences are calculated and summed to obtain the sum of the error terms. The sum of the error terms is then applied to W. f W o W i W c b f b o b i b c Taking the derivative yields the derivative W of the sum of the error terms with respect to each weight. f W o W i W c b f b o b i b c Subtracting the product of the derivative and the learning rate, we get the new W. f W o W i W c b f b o b i b c .

[0085] 9. Iterate until the set number of iterations or the sum of the error terms is less than the set value, at which point the model training is complete.

[0086] 10. Train the model using validation set data.

[0087] 11. Use the final W of the model f W o W i W c b f b o b i b c The learning rate is kept constant, and it is increased from a very small value each time until it approaches 1. The sum of the error terms is recorded each time, and the learning rate with the smallest sum of error terms is used as the final learning rate.

[0088] 12. Using geological hazard trend data as input variables, the geological hazard trend prediction model predicts future geological hazard trends. See details. Figure 5 .

[0089] 13. Using rainfall and reservoir water levels as input variables, the geological disaster trend prediction model predicts future geological disaster cycle data, such as... Figure 6 As shown.

[0090] 14. Compare the model prediction results with the actual geological disaster results, and use (the sum of the error between the prediction results and the actual results of the test set / the actual data of the test set) as the accuracy to evaluate the model.

[0091] 15. The prediction results are inversely normalized.

[0092] 16. Forecast cumulative data = forecast trend data + forecast period data; geological disaster data forecast completed.

[0093] A comparison of actual disaster data and model prediction data can be found in [link / reference]. Figure 7 .

[0094] 17. Issue the corresponding warning level 2 based on the pre-set warning threshold.

[0095] Warning level 1 2 3 4 5 Model prediction value Warning threshold Less than 2300 2310-2330 2330-2350 2350-2400 Above 2400 2328

[0096] In summary, the present invention has the following beneficial effects:

[0097] 1. The statistical characteristics and development patterns of time series data were taken into account, which improved the accuracy of geological disaster prediction and early warning;

[0098] 2. The impact of environmental factors on the occurrence and development of geological disasters has been considered, which has improved the accuracy of geological disaster prediction and early warning;

[0099] 3. Geological disaster prediction and early warning can be carried out with less actual geological disaster data and less data on various influencing factors, reducing the difficulty of prediction;

[0100] 4. Compared with traditional static geological disaster prediction and early warning algorithms, the geological disaster prediction and early warning model proposed in this paper is a dynamic prediction algorithm, which is more adaptable to real-world needs;

[0101] 5. Solve the gradient explosion and gradient vanishing problems that occur in neural network algorithms;

[0102] 6. Divide the data into training set, validation set, and test set to make the geological disaster prediction and early warning model more accurate;

[0103] 7. For the first time, machine learning algorithms were used to process geological disaster trend data;

[0104] 8. A geological hazard factor correlation algorithm was used to quantify the factors affecting geological hazards, and highly correlated factors were selected as input variables to improve the accuracy of prediction and early warning.

[0105] The above description discloses only one preferred embodiment of the present invention, and should not be construed as limiting the scope of the present invention. Those skilled in the art will understand that all or part of the processes of the above embodiments can be implemented, and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for predicting and warning of geological disasters, characterized in that, Includes the following steps: Obtain disaster and environmental data; Outlier replacement processing is performed on disaster and environmental data; The processed data is then normalized. Establish geological disaster prediction and early warning models; The geological disaster prediction and early warning model includes calculations in two directions: forward propagation and backward propagation. Forward propagation calculates f each time. t i t o t h t C t , The numerical value is obtained by calculating and adjusting the weights and bias W by reducing the sum of error terms during each backpropagation. f W o W i W c b f b o b i b c The value of h is determined, and the sum of the error terms is minimized. Training continues for a certain number of iterations until the sum of the error terms is h. t With y t The summation of the differences, where f t Represents the forget gate at time t, i t Represents the input gate and o at time t. t Represents the output gate and h at time t. t Represents the output data at time t, y t Represents the t-th actual data, C t Represents the cell state at time t, W f The weight of the forget gate, W o The weights W of the output gate are represented. i The weights W of the input gate are represented. c Indicates the candidate weights, b f Indicates the bias of the forget gate, b i Indicates the input gate bias, b o Indicates the output gate bias, b c Indicates alternative bias, This represents the unit's alternative status at time t; Where [] indicates that two vectors are connected, H t This is the output state of the geological disaster trend prediction model; The geological disaster prediction and early warning model was trained by selecting actual trend data and actual periodic data of geological disasters respectively, and prediction trend data and prediction periodic data were obtained. The process of obtaining actual trend data and actual periodic data of geological disasters is as follows: the actual trend data of geological disasters is obtained by using a geological disaster data decomposition algorithm, and the actual periodic data is obtained by subtracting the actual trend data from the actual cumulative data. The internal parameters of the geological hazard factor correlation algorithm are: T is the actual trend data, D is the actual cumulative data; α is a constant term between 0 and 1, reflecting the weighted decrease rate, and t represents the t-th time. T t =αD t-1 +(1-α)T t-1 After obtaining the trend data at time t, subtract the trend data at time t from the original cumulative data at time t to obtain the periodic data at time t. The correlation algorithm for geological hazard factors is used to calculate the correlation between each environmental factor and the actual periodic data. Factors whose correlation algorithm values ​​exceed a certain set value are selected as input variables for the prediction periodic data model. Assume there are two variables A = {a...} i {i = 1, 2, ..., n} and B = {b} i Let i = 1, 2, ..., n, where n is the number of variables. Then the algorithm for the correlation degree of geological disaster factors is as follows: Where p(a) is the marginal probability density of A; p(b) is the marginal probability density of B; and p(a,b) is the joint probability density of A and B. Let D = {(a,b)} i ,b i ), i = 1, 2, ..., n} consists of a finite set of ordered pairs. G is defined to divide the range of A into x segments and the range of B into y segments. G represents an x×y grid. MI(A,B) is calculated in each grid partition. There are multiple grids with the same x×y dimension. The MI value of G is obtained based on the maximum value of MI(A,B) among different grids. The maximum MI equation for partitioning G under D is defined as: MI * (D,x,y)=max MI(D|G) Where D|G means partitioning data D using G; The maximum normalized MI values ​​obtained under different partitions form a feature matrix, defined as MI(D). x,y The feature matrix can be defined as: The equation for the correlation degree algorithm of geological disaster factors is shown below: Where B(n) represents the upper bound of the grid partition x×y; Accumulated forecast data is obtained by combining forecast trend data and forecast cycle data; The predicted cumulative data is subjected to inverse standardization to obtain the final geological disaster prediction data.

2. The geological disaster prediction and early warning method as described in claim 1, characterized in that, The disaster and environmental data include rainfall, reservoir water level, soil type, slope gradient, slope orientation, air humidity, and soil moisture.

3. The geological disaster prediction and early warning method as described in claim 1, characterized in that, During the outlier replacement process, the replacement data is the average of the previous and next data. If it is the first or last data, it is deleted directly.

4. The geological disaster prediction and early warning method as described in claim 1, characterized in that, The standard normalization process is specifically handled according to the following formula: y i =(x i -x min ) / (x max -x min ), x i For the i-th original data, x min Let x be the minimum value in the data. max The maximum value in the x data.

5. The geological disaster prediction and early warning method as described in claim 1, characterized in that, The process of inverse standard normalization is specifically handled according to the following formula: y i =x i (x max -x min )+x min x min x is the minimum value in the original x data. max The maximum value in the original x data, y i For the final prediction data, x i This is the i-th data point calculated in the model.

Images