Short-time-space-cycle geological disaster prediction method based on weather forecast data

Abstract

A short time-space cycle geological disaster prediction method based on weather forecast data is characterized by comprising the following steps: step 1, a data acquisition module acquires multi-source weather data a with a time sequence attribute, and performs data fusion processing on the multi-source weather data a to obtain actually measured data b; the data acquisition module transmits measured data b to the data preprocessing module; step 2, a data preprocessing module performs preprocessing operation on the actually measured data b, performs data format conversion on the preprocessed data, converts the preprocessed data into three-dimensional input data c, and transmits the three-dimensional input data c to a prediction module; and step 3, the prediction module performs geological disaster prediction on the three-dimensional input data c and outputs a prediction result y. The method has the advantages that various meteorological data and interaction among the meteorological data are comprehensively considered, the geological disaster prediction model is further improved, the situations of false report and missing report are reduced, and the accuracy and practicability of a prediction result are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of meteorological data fusion and disaster prediction, and in particular to a method for predicting geological disasters in a short time and space period based on meteorological forecast data. Background Art

[0002] Geological disasters are geological events caused by natural environmental factors or different human production activities, which pose a certain degree of harm to human life and property safety. Geological conditions and climate characteristics are often the main factors that cause ground deformation, debris flow, landslide and other geological disasters. A considerable number of geological disasters are related to large-scale and diverse human production activities. The specific impact is that the recurrence intensity and frequency of geological disasters are getting higher and higher, and human wealth is also damaged. Therefore, it is of great practical significance to discover the correlation between the multiple factors that lead to geological disasters and then establish an effective geological disaster early warning model.

[0003] The current early warning model has improved the accuracy of early warning to a certain extent. However, since there are many factors that lead to slope instability, and they are often the result of the interweaving of multiple influencing factors, this increases the complexity of prediction. The current landslide early warning model mainly focuses on a single parameter, such as rainfall, for setting thresholds and making judgments, but ignores the potential correlation between multiple causes. Therefore, the actual early warning results may still have certain errors.

[0004] Disadvantages of existing technologies: The model parameters considered in current disaster warning models are relatively simple, and the warning results are not accurate enough. Summary of the invention

[0005] The present invention provides a method for predicting geological disasters in a short time and space period based on meteorological forecast data, which can realize the automatic prediction and deduction of geological disasters and effectively improve the accuracy of the prediction results.

[0006] To achieve the above object, the present invention provides a method for predicting geological disasters in a short time and space period based on meteorological forecast data, the key of which is to include the following steps:

[0007] Step 1: Data collection: The data collection module collects multi-source meteorological data a with time series attributes, and performs data fusion processing on the multi-source meteorological data a to obtain measured data b. The data collection module transmits the measured data b to the data preprocessing module;

[0008] Step 2: Data preprocessing: The data preprocessing module performs preprocessing operations on the measured data b, and converts the preprocessed data into a data format, converting it into three-dimensional input data c. The data preprocessing module transmits the three-dimensional input data c to the prediction module;

[0009] Step 3: Geological disaster prediction: The prediction module predicts the geological disaster for the three-dimensional input data c and outputs the prediction result y.

[0010] Through the above design, the data acquisition module is used to collect multi-source meteorological data, and geological disaster prediction is carried out in combination with various geological disaster influencing factors, which effectively improves the accuracy of geological disaster prediction results; the data preprocessing module is used to perform preprocessing operations and data format conversion on the input data of the geological disaster prediction model, which improves the quality and reliability of the input data and enhances the fit between the input data format and the geological disaster prediction model.

[0011] The data input dimension of the three-dimensional input data c can be expressed as (data_len, seq_len, input_size), where data_len is the number of sample sets; seq_len is the length of the input feature, which represents the historical observation time, i.e. L; input_size is the number of features of the input network, and its size corresponds to the number of types of input features. The current input features include meteorological temperature data, air humidity data, dew point temperature data, atmospheric pressure data and meteorological wind speed data, so input_size is 5.

[0012] Preferably, in step 1, the multi-source meteorological data a includes meteorological temperature data, air humidity data, dew point temperature data, atmospheric pressure data and meteorological wind speed data.

[0013] Preferably, in step 1, the data acquisition module includes a temperature sensor, a humidity sensor, a dew point sensor, an air pressure sensor and a wind speed sensor.

[0014] The temperature sensor is used to collect meteorological temperature data, the humidity sensor is used to collect air humidity data, the dew point sensor is used to collect dew point temperature data, the air pressure sensor is used to collect atmospheric pressure data, and the wind speed sensor is used to collect meteorological wind speed data.

[0015] Preferably, in step 1, a Kalman filtering method is used to implement data fusion processing of multi-source meteorological data a.

[0016] Kalman filtering recursively combines prior information and measured observation data to continuously update the state estimate, thereby improving the accuracy and robustness of the fusion result. The Kalman filter algorithm takes into account the dynamic evolution of the system, so that when performing data fusion on a time series, it can better adapt to system changes and make full use of past and current information.

[0017] Preferably, in step 2, the data preprocessing module performs preprocessing operations on the measured data b, including outlier processing and noise processing, and the specific steps are as follows:

[0018] Step 2.1: outlier processing: the measured data b is processed for outliers using the interquartile range method. First, a moving window of a specific size is used to determine the outlier data in the measured data b. Then, all the values ​​in the moving window are sorted, and the central value data after sorting, as well as the values ​​at the one-quarter position and the three-quarter position after sorting are selected to calculate the threshold value to obtain the threshold value.

[0019] Compare each value in the moving window with the threshold value respectively, if the comparison result of a value in the moving window is an abnormal value, use the central value data to replace the value data;

[0020] The moving window size is often determined based on a percentage of the data, such as choosing a moving window that encompasses 5% or 10% of the overall data.

[0021] Step 2.2: Noise processing: Use wavelet threshold denoising to perform denoising and smoothing on the data after outlier processing. First, select the wavelet function to perform n-layer wavelet decomposition on the input signal to obtain the wavelet coefficients at each scale. Then, based on the wavelet coefficients at each scale, select the threshold calculation method to calculate the threshold, and perform threshold processing on the decomposed wavelet coefficients of each layer. Finally, according to the wavelet function selected during decomposition, reconstruct the wavelet coefficients of each scale after threshold processing to obtain the signal after wavelet threshold denoising.

[0022] The outlier processing and noise processing can improve the quality and accuracy of input data, thereby improving the accuracy of prediction results.

[0023] Preferably, in step 3, the prediction module is a trained geological disaster prediction model, and the geological disaster prediction model is provided with a convolution input layer, a pooling sampling layer and a fully connected output layer, and the convolution input layer, the pooling sampling layer and the fully connected output layer are connected in sequence.

[0024] The convolution input layer is used to capture the spatiotemporal relationship in the three-dimensional input data c, especially the capture of geographical location information and time series information in meteorological data; the pooling sampling layer is used to convolve the convolution result data d of the convolution input layer. i,j Down sampling is performed to reduce the data dimension, retain key information, and speed up the calculation speed of the geological disaster prediction model; the fully connected output layer is used to introduce nonlinear relationships to provide a more powerful modeling capability for the geological disaster prediction model.

[0025] Preferably, the convolution input layer performs a convolution operation on the three-dimensional input data c to obtain convolution result data d i,j , and its calculation expression is as follows:

[0026]

[0027] Among them, K represents the convolution kernel, f1 represents the height of the convolution kernel, f2 represents the width of the convolution kernel, i and j are the indexes of the elements in the output matrix, indicating the position in the output matrix; m and n are the indexes of the elements in the convolution kernel, indicating the position of the convolution kernel;

[0028] The pooling sampling layer performs convolution on the convolution result d i,j Perform pooling operation to obtain pooling result data e i,j , and its calculation expression is as follows:

[0029]

[0030] Among them, s represents the stride of the pooling window;

[0031] The fully connected output layer performs pooling on the result data e i,j After the full connection operation, the full connection result data x is obtained, and then the prediction result y is output through the Relu activation function. Its activation expression is as follows:

[0032] y = ReLU(x) = max(0,x)

[0033] Among them, y represents the prediction result.

[0034] Preferably, the prediction result y is a probability value in the range of (0, 1). When the prediction result y is greater than a disaster warning threshold, it is determined that a disaster has occurred; otherwise, it is determined that no disaster has occurred.

[0035] The prediction result is a probability value, which can intuitively show whether a disaster will occur, making it convenient for relevant staff to take corresponding geological disaster prevention measures based on the geological disaster prediction results.

[0036] Preferably, in step 3, the prediction module is a trained geological disaster prediction model, and the training process of the geological disaster prediction model is as follows:

[0037] Step S1: The data acquisition module acquires multi-source meteorological data with time series attributes, performs data fusion processing on the data, and then transmits the data to the data preprocessing module for preprocessing operations and data format conversion to obtain three-dimensional input data, and divides the three-dimensional input data into training set data and verification set data;

[0038] Step S2: Use the long short-term memory network LSTM to build a geological disaster prediction model;

[0039] Step S3: using the training set data to train the geological disaster prediction model, using the Bayesian optimization method to optimize the model parameters of the geological disaster prediction model, and obtaining the optimal model parameters, wherein the model parameters include the number of neurons in the hidden layer of the model, the number of samples selected for one training, Batch-size, the learning rate, and the hyperparameter Dropout;

[0040] Step S4: Use the validation set data to evaluate the performance of the geological disaster prediction model. Input the validation set data into the trained geological disaster prediction model to obtain the prediction results. Use precision (P), recall (R), and average precision (AP) to evaluate the accuracy of the geological disaster prediction results.

[0041] The long short-term memory network LSTM is a time recurrent neural network used to solve the long-term dependency problem of general recurrent neural networks RNN. All recurrent neural networks RNN have a chain form of repeated neural network modules. In the standard recurrent neural network RNN, this repeated structural module has only a very simple structure, such as a tanh layer.

[0042] Through the above design, the training set data and the verification set data are used to perform training optimization and performance verification on the geological disaster prediction model to ensure the accuracy and reliability of the geological disaster prediction results.

[0043] Preferably, in step S4, the accuracy rate P is the ratio of the number of correctly identified geological disasters to the total number of geological disasters. The calculation formula of the accuracy rate P is as follows:

[0044]

[0045] Among them, TP is the number of correctly predicted positive samples, and FP is the number of incorrectly predicted positive samples;

[0046] The recall rate R is the ratio of the number of correctly identified geological disasters to the total number of true geological disaster values, and is used to measure the omission of the classifier on the geological disaster data set. The calculation formula of the recall rate R is as follows:

[0047]

[0048] Among them, NP is the number of samples that are incorrectly predicted as negative samples;

[0049] The average precision AP is the average precision of the same type of targets under different recall rates. The calculation formula of the average precision AP is as follows:

[0050]

[0051] The larger the values ​​of the precision (P), recall (R) and average precision (AP), the better the prediction performance of the geological disaster prediction model.

[0052] The beneficial effects of the present invention are: by comprehensively considering a variety of meteorological data and the interactions between them, the geological disaster prediction model is further improved, the false alarms and missed alarms are reduced, and the accuracy and practicality of the prediction results are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the process of the present invention;

[0054] Figure 2 Schematic diagram of the wavelet threshold denoising process in the embodiment. DETAILED DESCRIPTION

[0055] The present invention is further described in detail below in conjunction with the accompanying drawings and specific examples. The following examples or drawings are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0056] like Figure 1 As shown: A method for predicting geological disasters in a short time and space period based on meteorological forecast data, the key is that it includes the following steps:

[0057] Step 1: Data collection: The data collection module collects multi-source meteorological data a with time series attributes, and performs data fusion processing on the multi-source meteorological data a to obtain measured data b. The data collection module transmits the measured data b to the data preprocessing module;

[0058] Step 2: Data preprocessing: The data preprocessing module performs preprocessing operations on the measured data b, and converts the preprocessed data into a data format, converting it into three-dimensional input data c. The data preprocessing module transmits the three-dimensional input data c to the prediction module;

[0059] Step 3: Geological disaster prediction: The prediction module predicts the geological disaster for the three-dimensional input data c and outputs the prediction result y.

[0060] Specifically, in step 1, the multi-source meteorological data a includes meteorological temperature data, air humidity data, dew point temperature data, atmospheric pressure data and meteorological wind speed data.

[0061] The data acquisition module includes a temperature sensor, a humidity sensor, a dew point sensor, an air pressure sensor and a wind speed sensor.

[0062] The Kalman filtering method is used to realize the data fusion processing of multi-source meteorological data a.

[0063] In step 2, the data preprocessing module performs preprocessing operations on the measured data b, including outlier processing and noise processing. The specific steps are as follows:

[0064] Step 2.1: outlier processing: the measured data b is processed for outliers using the interquartile range method. First, a moving window of a specific size is used to determine the outlier data in the measured data b. Then, all the values ​​in the moving window are sorted, and the central value data after sorting, as well as the values ​​at the one-quarter position and the three-quarter position after sorting are selected to calculate the threshold value to obtain the threshold value.

[0065] Compare each value in the moving window with the threshold value respectively, if the comparison result of a value in the moving window is an abnormal value, use the central value data to replace the value data;

[0066] Step 2.2: Noise processing: Use wavelet threshold denoising to perform denoising and smoothing on the data after outlier processing. First, select a wavelet function to perform n-layer wavelet decomposition on the input signal to obtain wavelet coefficients at each scale. Then, based on the wavelet coefficients at each scale, select a threshold calculation method to calculate the threshold, perform threshold processing on the decomposed wavelet coefficients at each layer, and finally, reconstruct the wavelet coefficients of each scale after threshold processing according to the wavelet function selected during decomposition to obtain the signal after wavelet threshold denoising; Figure 2 shown.

[0067] In step 3, the prediction module is a trained geological disaster prediction model, and the geological disaster prediction model is provided with a convolution input layer, a pooling sampling layer and a fully connected output layer, and the convolution input layer, the pooling sampling layer and the fully connected output layer are connected in sequence.

[0068] The convolution input layer performs a convolution operation on the three-dimensional input data c to obtain the convolution result data d i,j , and its calculation expression is as follows:

[0069]

[0070] Among them, K represents the convolution kernel, f1 represents the height of the convolution kernel, f2 represents the width of the convolution kernel, i and j are the indexes of the elements in the output matrix, indicating the position in the output matrix; m and n are the indexes of the elements in the convolution kernel, indicating the position of the convolution kernel;

[0071] The pooling sampling layer performs convolution on the convolution result d i,j Perform pooling operation to obtain pooling result data e i,j , and its calculation expression is as follows:

[0072]

[0073] Among them, s represents the stride of the pooling window;

[0074] The fully connected output layer performs pooling on the result data e i,j After the full connection operation, the full connection result data x is obtained, and then the prediction result y is output through the Relu activation function. Its activation expression is as follows:

[0075] y = ReLU(x) = max(0,x)

[0076] Among them, y represents the prediction result.

[0077] The prediction result y is a probability value in the range of (0, 1). When the prediction result y is greater than the disaster warning threshold, it is determined that a disaster has occurred; otherwise, it is determined that no disaster has occurred.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. 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 in short time and space periods based on meteorological forecast data, characterized in that: The following steps are involved: Step 1: Data collection: The data collection module collects multi-source meteorological data a with time series attributes, and performs data fusion processing on the multi-source meteorological data a to obtain measured data b. The data collection module transmits the measured data b to the data preprocessing module; Step 2: Data preprocessing: The data preprocessing module performs preprocessing operations on the measured data b, and converts the preprocessed data into a data format, converting it into three-dimensional input data c. The data preprocessing module transmits the three-dimensional input data c to the prediction module; Step 3: Geological disaster prediction: The prediction module predicts the geological disaster for the three-dimensional input data c and outputs the prediction result y.

2. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1 is characterized by: In step 1, the multi-source meteorological data a includes meteorological temperature data, air humidity data, dew point temperature data, atmospheric pressure data and meteorological wind speed data.

3. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1, characterized in that: In step 1, the data acquisition module includes a temperature sensor, a humidity sensor, a dew point sensor, an air pressure sensor and a wind speed sensor.

4. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1, characterized in that: In step 1, a Kalman filtering method is used to implement data fusion processing of multi-source meteorological data a.

5. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1, characterized in that: In step 2, the data preprocessing module performs preprocessing operations on the measured data b, including outlier processing and noise processing. The specific steps are as follows: Step 2.1: outlier processing: the measured data b is processed for outliers using the interquartile range method. First, a moving window of a specific size is used to determine the outlier data in the measured data b. Then, all the values ​​in the moving window are sorted, and the central value data after sorting, as well as the values ​​at the one-quarter position and the three-quarter position after sorting are selected to calculate the threshold value to obtain the threshold value. Compare each value in the moving window with the threshold value respectively, if the comparison result of a value in the moving window is an abnormal value, use the central value data to replace the value data; Step 2.2: Noise processing: Use wavelet threshold denoising to perform denoising and smoothing on the data after outlier processing. First, select the wavelet function to perform n-layer wavelet decomposition on the input signal to obtain the wavelet coefficients at each scale. Then, based on the wavelet coefficients at each scale, select the threshold calculation method to calculate the threshold, and perform threshold processing on the decomposed wavelet coefficients of each layer. Finally, according to the wavelet function selected during decomposition, reconstruct the wavelet coefficients of each scale after threshold processing to obtain the signal after wavelet threshold denoising.

6. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1, characterized in that: In step 3, the prediction module is a trained geological disaster prediction model, and the geological disaster prediction model is provided with a convolution input layer, a pooling sampling layer and a fully connected output layer, and the convolution input layer, the pooling sampling layer and the fully connected output layer are connected in sequence.

7. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 6 is characterized by: The convolution input layer performs a convolution operation on the three-dimensional input data c to obtain the convolution result data d i,j , and its calculation expression is as follows: Among them, K represents the convolution kernel, f1 represents the height of the convolution kernel, f2 represents the width of the convolution kernel, i and j are the indexes of the elements in the output matrix, indicating the position in the output matrix; m and n are the indexes of the elements in the convolution kernel, indicating the position of the convolution kernel; The pooling sampling layer performs the convolution result d i,j Perform pooling operation to obtain pooling result data e i,j , and its calculation expression is as follows: Among them, s represents the stride of the pooling window; The fully connected output layer performs pooling on the result data e i,j After the full connection operation, the full connection result data x is obtained, and then the prediction result y is output through the Relu activation function. Its activation expression is as follows: y = ReLU(x) = max(0,x) Among them, y represents the prediction result.

8. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1, characterized in that: The prediction result y is a probability value in the range of (0, 1). When the prediction result y is greater than the disaster warning threshold, it is determined that a disaster has occurred; otherwise, it is determined that no disaster has occurred.

9. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 1, characterized in that: In step 3, the prediction module is a trained geological disaster prediction model, and the training process of the geological disaster prediction model is as follows: Step S1: The data acquisition module acquires multi-source meteorological data with time series attributes, performs data fusion processing on the data, and then transmits the data to the data preprocessing module for preprocessing operations and data format conversion to obtain three-dimensional input data, and divides the three-dimensional input data into training set data and verification set data; Step S2: Use the long short-term memory network LSTM to build a geological disaster prediction model; Step S3: using the training set data to train the geological disaster prediction model, using the Bayesian optimization method to optimize the model parameters of the geological disaster prediction model, and obtaining the optimal model parameters, wherein the model parameters include the number of neurons in the hidden layer of the model, the number of samples selected for one training, Batch-size, the learning rate, and the hyperparameter Dropout; Step S4: Use the validation set data to evaluate the performance of the geological disaster prediction model. Input the validation set data into the trained geological disaster prediction model to obtain the prediction results. Use precision (P), recall (R), and average precision (AP) to evaluate the accuracy of the geological disaster prediction results.

10. The method for predicting geological disasters in short time and space periods based on meteorological forecast data according to claim 9, characterized in that: In step S4, the accuracy rate P is the ratio of the number of correctly identified geological disasters to the total number of geological disasters. The calculation formula of the accuracy rate P is as follows: Among them, TP is the number of correctly predicted positive samples, and FP is the number of incorrectly predicted positive samples; The recall rate R is the ratio of the number of correctly identified geological disasters to the total number of true geological disaster values, and is used to measure the omission of the classifier on the geological disaster data set. The calculation formula of the recall rate R is as follows: Among them, NP is the number of samples that are incorrectly predicted as negative samples; The average precision AP is the average precision of the same type of targets under different recall rates. The calculation formula of the average precision AP is as follows:

Images