Geological disaster prediction method using autoregression and recurrent neural network model

Abstract

The invention discloses a geological disaster prediction method using an autoregression and recurrent neural network model. The method comprises the following steps: S1, white noise detection; carrying out white noise detection on the collected data; S2, carrying out stability detection; s3, carrying out ARMA model fitting; s4, ARMA model prediction is carried out; step S5, RNN model prediction; inputting the detected data into an RNN model for model training to obtain a trained RNN model; and step S6, ending. According to the method, the ARMA model can capture the trend and periodicity in the time sequence data, and the RNN neural network can process the long-term dependency relationship in the data, so that the dynamic change of the geological disaster can be better understood. By combining the two methods, the time and space information of the data can be utilized more comprehensively, and the accuracy and reliability of geological disaster occurrence are improved. According to the method, the occurrence condition of the geological disaster can be predicted more effectively.

Description

Technical Field

[0001] The present invention belongs to the technical field of geological disaster prediction, and specifically is a geological disaster prediction method using autoregressive and recurrent neural network models. Background Art

[0002] Geological disaster prediction refers to the identification, evaluation and prediction of possible geological disasters (such as earthquakes, debris flows, landslides, collapses, etc.) through scientific methods and technical means to reduce the impact of disasters on humans and the environment. Effective geological disaster prediction usually includes the following aspects: collecting historical disaster records, meteorological data, soil and geological characteristics and other information, and identifying potential risk areas by analyzing these data. Use geological radar, remote sensing technology, GPS, earthquake monitoring stations and other means to monitor the geological environment in real time to obtain the latest geological change information. Establish a mathematical model for the occurrence of geological disasters, such as through statistical models, machine learning and other methods to predict the probability and impact of disasters. Build a geological disaster early warning system to issue alerts to the public in a timely manner before a disaster occurs.

[0003] Predicting geohazards is a key issue for improving public safety and developing effective disaster prevention measures. This problem is challenging due to the sparsity of geohazards in space and time and the spatial heterogeneity of the environment (e.g., mountainous areas versus plains).

[0004] In the prior art, for example, the technical solution described in the patent with publication number: CN113393037: a method and system for predicting the trend of regional geological disasters, which can obtain the geological parameter information of the current region, filter it in the preset reference terrain library, obtain the reference area information, and then train it according to the historical geological disaster information in the geological parameter information of the current region and the historical geological disaster information in the reference area information to obtain the model for predicting the trend of geological disasters. Finally, the geological parameter information of the current region is input into the model for predicting the trend of geological disasters, the geological disaster trend information of the current region is obtained, and it is compared with the preset threshold information to generate geological disaster warning information.

[0005] In the existing technologies, most of the geological hazard prediction studies conducted by researchers in the field only apply classical prediction models on limited data without properly addressing the above challenges, resulting in unsatisfactory performance. The existing technologies either ignore temporal information or only use data from a small and homogeneous study area (specific geological environment) without properly handling spatial heterogeneity and temporal autocorrelation at the same time. Summary of the invention

[0006] The purpose of the present invention is to provide a geological disaster prediction method using autoregressive and recurrent neural network models to solve the problem that the existing technical solutions proposed in the background technology either ignore time information or fail to correctly handle the relationship between spatial heterogeneity and temporal autocorrelation at the same time, resulting in low algorithm prediction accuracy.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for predicting geological disasters using an autoregressive and recurrent neural network model comprises the following steps:

[0009] Step S1, white noise detection; perform white noise detection on the collected data, and discard the data if the noise of the data is too large; if the collected data passes the white noise detection, perform a stationarity detection on the data that passes the white noise detection;

[0010] Step S2, stationarity test; if the data does not meet the stationarity test, the data is stabilized until it passes the stationarity test; the data that passes the stationarity test is fitted with an ARMA model;

[0011] Step S3, ARMA model fitting;

[0012] Step S4, ARMA model prediction: input the geological disaster situation of the next month predicted by the ARMA model fitted in the previous step into the RNN prediction model with trained parameters for prediction;

[0013] Step S5, RNN model prediction: input the detected data into the RNN model for model training to obtain a trained RNN model;

[0014] Step S6, end.

[0015] According to the above technical solution, white noise means that in a time series, if there is no correlation between the data points and the mean and variance remain unchanged, the sequence is called a white noise sequence. The time series white noise test is a method to determine whether a given sequence is white noise. It determines whether there is correlation in the sequence by performing a statistical test on the sequence. This solution uses the Box-Pierce statistical test method to detect white noise. The test determines whether the sequence is white noise by calculating the statistic of the autocorrelation coefficient.

[0016] In step S1, the Box-Pierce statistical test method is used to detect white noise, as shown in the following formula:

[0017]

[0018] Where n is the number of observation periods in the series, m is the specified maximum lag order, and p is the estimated value of the autocorrelation coefficient of lag k.

[0019] According to the above technical solution, the autocorrelation coefficient is calculated as follows: first, the average value μ of the time series data is calculated, then the total variance α of the time series is calculated, and finally the formula for calculating the autocorrelation coefficient is as follows:

[0020]

[0021] In the formula, N represents the length of the time series, k represents the k-order autocorrelation coefficient, and x i represents the i-th time series data, μ represents the mean of the time series, and α represents the total variance of the time series.

[0022] According to the above technical solution, in step S1, the stationarity detection uses the ADF statistical test method to detect stationarity; specifically: determine whether the sequence has a unit root: in an autoregressive process, as shown in the following formula:

[0023] y t =by t-1 +a+ε t

[0024] If the lag term coefficient b is 1, it is called a unit root, and the series is not stationary; if the series is stationary, b is not 1, and there is no unit root.

[0025] According to the above technical solution, in step S3, the ARMA model fitting is specifically as follows:

[0026]

[0027] Where Y t is time series data, arrive is the parameter of the AR model, which is used to describe the relationship between the current value and the value at the past p time points; θ1 to θ q is the parameter of the MA model, which is used to describe the relationship between the current value and the error at the past q time points; ∈ t is the error term at time point t; c is a constant term.

[0028] According to the above technical solution, in step S4, the RNN model prediction is specifically:

[0029] Input parameters: In RNN, the output h at each time step t t Not only depends on the current input x t , but also depends on the hidden state h of the previous time step t-1 ; This dependency is expressed by the following formula:

[0030] ht =f(W·h t-1 +U·x t +b)

[0031] Among them, f is the activation function, usually tanh or ReLU function; W is the weight matrix used for the transformation of the previous hidden state; U is another weight matrix used for the transformation of the current input; b is the bias term; x in the formula t is the geological disaster situation of the next month predicted by the ARMA model, where h t-1 This is the actual geological disaster situation that occurred in the previous month.

[0032] According to the above technical solution, the final output y t Obtained by the following formula:

[0033] y t =g(V·h t +c)

[0034] Where g is expressed as a softmax or linear function according to the nature of the task, softmax is used for classification tasks, and linear function is used for regression tasks; V is the weight matrix of the output layer; c is the bias term of the output layer.

[0035] According to the above technical solution, before making a geological disaster prediction, it is necessary to process the data and add features.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] In the present invention, the ARMA model can capture the trend and periodicity in the time series data, while the RNN neural network can process the long-term dependencies in the data, so as to better understand the dynamic changes of geological disasters. By combining these two methods, the temporal and spatial information of the data can be more comprehensively utilized to improve the accuracy and reliability of the occurrence of geological disasters. By mining the inherent dynamic information about the geological structure and the dynamic time dependency of the occurrence of disasters in the historical geological disaster data, the method can more effectively predict the occurrence of geological disasters and provide more accurate prediction and decision support for disaster management and public safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of the algorithm prediction of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Embodiment 1

[0041] like Figure 1 As shown, a geological disaster prediction method using an autoregressive and recurrent neural network model includes the following steps:

[0042] Step S1, white noise detection; perform white noise detection on the collected data, and discard the data if the noise of the data is too large; if the collected data passes the white noise detection, perform a stationarity detection on the data that passes the white noise detection;

[0043] Step S2, stationarity test; if the data does not meet the stationarity test, the data is stabilized until it passes the stationarity test; the data that passes the stationarity test is fitted with an ARMA model;

[0044] Step S3, ARMA model fitting;

[0045] Step S4, ARMA model prediction: input the geological disaster situation of the next month predicted by the ARMA model fitted in the previous step into the RNN prediction model with trained parameters for prediction;

[0046] Step S5, RNN model prediction: input the detected data into the RNN model for model training to obtain a trained RNN model;

[0047] Step S6, end.

[0048] In the present invention, the ARMA model can capture the trend and periodicity in the time series data, while the RNN neural network can process the long-term dependencies in the data, so as to better understand the dynamic changes of geological disasters. By combining these two methods, the temporal and spatial information of the data can be more comprehensively utilized to improve the accuracy and reliability of the occurrence of geological disasters. By mining the inherent dynamic information about the geological structure and the dynamic time dependency of the occurrence of disasters in the historical geological disaster data, the method can more effectively predict the occurrence of geological disasters and provide more accurate prediction and decision support for disaster management and public safety.

[0049] Embodiment 2

[0050] This embodiment is a further refinement of the first embodiment.

[0051] White noise refers to a time series in which there is no correlation between the data points and the mean and variance remain unchanged. The time series white noise test is a method to determine whether a given sequence is white noise. It determines whether there is correlation in the sequence by performing a statistical test on the sequence. This solution uses the Box-Pierce statistical test method to detect white noise. The test determines whether the sequence is white noise by calculating the statistic of the autocorrelation coefficient.

[0052] In step S1, the Box-Pierce statistical test method is used to detect white noise, as shown in the following formula:

[0053]

[0054] Where n is the number of observation periods in the series, m is the specified maximum lag order, and p is the estimated value of the autocorrelation coefficient of lag k.

[0055] The calculation of the autocorrelation coefficient is as follows: first calculate the mean value μ of the time series data, then calculate the total variance α of the time series, and finally calculate the formula for the autocorrelation coefficient as follows:

[0056]

[0057] In the formula, N represents the length of the time series, k represents the k-order autocorrelation coefficient, and x i represents the i-th time series data, μ represents the mean of the time series, and α represents the total variance of the time series.

[0058] A stationary series means that its mean, variance and covariance do not change over time. This solution uses the Augmented Dickey-Fuller test (ADF) statistical test method to detect stationarity. The ADF test is to determine whether the series has a unit root.

[0059] In step S1, the stationarity test uses the ADF statistical test method to detect stationarity; specifically: determine whether the sequence has a unit root: in an autoregressive process, as shown in the following formula:

[0060] y t =by t-1 +a+ε t

[0061] If the lag term coefficient b is 1, it is called a unit root, and the series is not stationary; if the series is stationary, b is not 1, and there is no unit root.

[0062] In step S3, the ARMA model fitting is specifically as follows:

[0063]

[0064] Where Y t is time series data, arrive is the parameter of the AR model, which is used to describe the relationship between the current value and the value at the past p time points; θ1 to θ q is the parameter of the MA model, which is used to describe the relationship between the current value and the error at the past q time points; ∈ t is the error term at time point t; c is a constant term.

[0065] In step S4, the RNN model prediction is specifically:

[0066] Input parameters: In RNN, the output h at each time step t t Not only depends on the current input x t , but also depends on the hidden state h of the previous time step t-1 ; This dependency is expressed by the following formula:

[0067] h t =f(W·h t-1 +U·x t +b)

[0068] Among them, f is the activation function, usually tanh or ReLU function; W is the weight matrix used for the transformation of the previous hidden state; U is another weight matrix used for the transformation of the current input; b is the bias term; x in the formula t is the geological disaster situation of the next month predicted by the ARMA model, where h t-1 This is the actual geological disaster situation that occurred in the previous month.

[0069] The final output y t Obtained by the following formula:

[0070] y t =g(V·h t +c)

[0071] Where g is expressed as a softmax or linear function according to the nature of the task, softmax is used for classification tasks, and linear function is used for regression tasks; V is the weight matrix of the output layer; c is the bias term of the output layer.

[0072] Before making geological disaster predictions, the data needs to be processed and features added.

[0073] Embodiment 3

[0074] This embodiment provides a specific implementation method.

[0075] The present invention introduces the ARMA model, which is a model that combines the advantages of the AR model and the MA model and can handle more complex time series problems.

[0076] 1. The AR part is used to process the autoregressive part of the time series, which takes into account the impact of observations from several past periods on the current value.

[0077] 2. The MA part is used to process the moving average part of the time series, which takes into account the impact of past forecast errors on current values.

[0078] Combining these two parts, the ARMA model can capture the trend changes of data and handle data with temporary, sudden changes or large noise. Therefore, the ARMA model performs well in many time series forecasting problems.

[0079] Mathematical expressions of AR and MA models:

[0080] AIZ.:

[0081] MA:Y t =μ+ε t +θ1ε t-1 +θ2ε t-2 +…+θ q ε t-q (2)

[0082] Formally, the formula of the ARMA model can be expressed as:

[0083]

[0084] In this formula (3), Y t is time series data, arrive is the parameter of the AR model, which is used to describe the relationship between the current value and the value at the past p time points; θ1 to θ q is the parameter of the MA model, which is used to describe the relationship between the current value and the error at the past q time points; ∈ t is the error term at time point t; c is a constant term.

[0085] A recurrent neural network (RNN) is a neural network used to process sequence data. Its design allows the network to pass information in a time series. The core principle of RNN lies in its cyclic structure, which enables the network to update its internal state at each time step, thereby capturing the dynamic characteristics of the sequence.

[0086] In RNN, the output h at each time step t is tNot only depends on the current input x t , but also depends on the hidden state h of the previous time step t-1 This dependency can be expressed by the following formula:

[0087] h t =f(W·h t-1 +U·x t+b ) (4)

[0088] y t =g(V·h t +c) (5)

[0089] Among them, f is the activation function, usually tanh or ReLU function; W is the weight matrix used for the transformation of the previous hidden state; U is another weight matrix used for the transformation of the current input; b is the bias term; x in the formula t is the geological disaster situation of the next month predicted by the ARMA model, where h t-1 is the actual geological disaster situation that occurred in the previous month; g is usually a softmax or linear function, depending on the nature of the task; V is the weight matrix of the output layer; c is the bias term of the output layer.

[0090] Combining ARMA and RNN to realize geological disaster prediction:

[0091] 1. Data processing and feature addition

[0092] The current geological disaster statistics are all unilateral data statistics of a single disaster event, and the data features are shown in Table 1. From the data features, it can be seen that the characteristics of each geological disaster are for a single event. When predicting the number of geological disasters in months, there are no generalized features. Therefore, we select the common features of each event, such as weather and geological conditions, from the features, and summarize and integrate them according to the month.

[0093] In addition, since the statistical data is a unilateral statistics for a single event, when the disaster involves multiple locations, multiple records with the same time and location will be formed. In this case, the data needs to be deduplicated according to the ID number, as shown in Table 1.

[0094] Table 1

[0095]

[0096] 2. Combine ARMA and RNN algorithm flow chart

[0097] First, the data sequence is tested for white noise. If the data noise is too large, the data is discarded. The data that has passed the white noise test is tested for stationarity. If the data does not meet the stationarity test, the data is stabilized until it meets the stationarity test.

[0098] Next, the tested data is input into the ARMA model for model fitting. In this algorithm, the occurrence of geological disasters in the previous six months is used to predict the occurrence of geological disasters in the next month. Specifically, y in formula 3 t-1 ~y t-p The actual geological disaster occurrence data is collected, and the data features are generated based on Table 1. Here, the parameter p of Formula 3 is set to 6, that is, the geological disaster occurrence in the past 6 months is used to predict the geological disaster occurrence in the next month. By substituting the actual data, the fitted ARMA formula can be obtained.

[0099] Next, the detected data is input into the RNN model for model training to obtain a trained RNN model. Then the geological disaster situation of the next month predicted by the ARMA model fitted in the previous step is input into the RNN prediction model with trained parameters for prediction. Specifically, x in Formula 4 t The ARMA model predicts the geological disasters in the next month. t-1 is the actual geological disaster situation that occurred in the previous month. The final model prediction result can be obtained according to Formula 4 and Formula 5.

[0100] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0101] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting geological disasters using an autoregressive and recurrent neural network model, characterized in that: The following steps are involved: Step S1, white noise detection; perform white noise detection on the collected data, and discard the data if the noise of the data is too large; if the collected data passes the white noise detection, perform a stationarity detection on the data that passes the white noise detection; Step S2, stationarity test; if the data does not meet the stationarity test, the data is stabilized until it passes the stationarity test; the data that passes the stationarity test is fitted with an ARMA model; Step S3, ARMA model fitting; Step S4, ARMA model prediction: input the geological disaster situation of the next month predicted by the ARMA model fitted in the previous step into the RNN prediction model with trained parameters for prediction; Step S5, RNN model prediction: input the detected data into the RNN model for model training to obtain a trained RNN model; Step S6, end.

2. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 1, characterized in that: White noise means that in a time series, if there is no correlation between the data points and the mean and variance remain unchanged, the sequence is called a white noise sequence. The time series white noise test is a method to determine whether a given sequence is white noise. It determines whether there is correlation in the sequence by performing a statistical test on the sequence. This solution uses the Box-Pierce statistical test method to detect white noise. The test determines whether the sequence is white noise by calculating the statistic of the autocorrelation coefficient. In step S1, the Box-Pierce statistical test method is used to detect white noise, as shown in the following formula: Where n is the number of observation periods in the series, m is the specified maximum lag order, and p is the estimated value of the autocorrelation coefficient of lag k.

3. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 2, characterized in that: The calculation of the autocorrelation coefficient is as follows: first calculate the mean value μ of the time series data, then calculate the total variance α of the time series, and finally calculate the formula for the autocorrelation coefficient as follows: In the formula, N represents the length of the time series, k represents the k-order autocorrelation coefficient, and x i represents the i-th time series data, μ represents the mean of the time series, and α represents the total variance of the time series.

4. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 3, characterized in that: In step S1, the stationarity test uses the ADF statistical test method to detect stationarity; specifically: determine whether the sequence has a unit root: in an autoregressive process, as shown in the following formula: y t =by t-1 +a+ε t If the lag term coefficient b is 1, it is called a unit root, and the series is not stationary; if the series is stationary, b is not 1, and there is no unit root.

5. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 1, characterized in that: In step S3, the ARMA model fitting is specifically as follows: Where Y t is time series data, arrive is the parameter of the AR model, which is used to describe the relationship between the current value and the value at the past p time points; θ1 to θ q is the parameter of the MA model, which is used to describe the relationship between the current value and the error at the past q time points; ∈ t is the error term at time t; c is a constant term.

6. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 1, characterized in that: In step S4, the RNN model prediction is specifically: Input parameters: In RNN, the output h at each time step t t Not only depends on the current input x t , but also depends on the hidden state h of the previous time step t-1 ; This dependency is expressed by the following formula: h t =f(W·h t-1 +U·x t +b) Among them, f is the activation function, usually tanh or ReLU function; W is the weight matrix used for the transformation of the previous hidden state; U is another weight matrix used for the transformation of the current input; b is the bias term; x in the formula t is the geological disaster situation of the next month predicted by the ARMA model, where h t-1 This is the actual geological disaster situation that occurred in the previous month.

7. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 1, characterized in that: The final output y t Obtained by the following formula: y t =g(V·h t +C) Where g is expressed as a softmax or linear function according to the nature of the task, softmax is used for classification tasks, and linear function is used for regression tasks; V is the weight matrix of the output layer; c is the bias term of the output layer.

8. The method for predicting geological disasters using an autoregressive and recurrent neural network model according to claim 1, characterized in that: Before making geological disaster predictions, the data needs to be processed and features added.

Images