Landslide disaster dynamic monitoring method based on time sequence and hybrid model

Abstract

The invention provides a landslide disaster dynamic monitoring method based on a time sequence and a hybrid model, and belongs to the technical field of intelligent monitoring, and the method comprises the steps: collecting multi-source heterogeneous data of historical landslide, carrying out the preprocessing of the multi-source heterogeneous data, obtaining first data, carrying out the type division of the first data, and carrying out the time sequence coding of each type of data, obtaining single samples under the same time sequence to obtain a sample set, and training the neural network model according to the sample set to obtain an initial model; performing model verification on the initial model, and optimizing to obtain a time sequence and a hybrid model in combination with the constructed mapping relationship between the landslide deformation and the monitoring dimension under the corresponding type; and monitoring and acquiring current heterogeneous data of the target monitoring point, inputting the current heterogeneous data into the time sequence and the hybrid model, predicting to obtain a current disaster result of the target monitoring point, and outputting a prompt. The accuracy of the model is ensured through model verification and mapping relation construction, and a convenient basis is provided for landslide disaster prediction.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and in particular to a landslide disaster dynamic monitoring method based on time series and hybrid model. Background Art

[0002] Landslide refers to a natural phenomenon in which soil or rock on a slope slides downward along a certain weak surface or weak zone, either as a whole or in a dispersed manner, under the action of gravity, due to the influence of factors such as river scouring, groundwater activity, earthquakes and artificial slope cutting.

[0003] In the analysis of disaster-causing factors, the research on landslides has clearly identified the key roles of rainfall, reservoir water level changes and topographic and geological conditions. The application of monitoring technology presents a trend of coexistence of multiple means and the integration of multiple sources. The prediction methods include time series analysis and other methods. Stability evaluation and prevention use limit equilibrium analysis and other methods and implement corresponding engineering and non-engineering measures to ensure the safety of the reservoir area. However, there are some shortcomings in the current research. On the one hand, the prediction method is affected by the coupling of multiple factors, which affects the accuracy of the prediction.

[0004] Therefore, the present invention proposes a landslide disaster dynamic monitoring method based on time series and hybrid model. Summary of the invention

[0005] The present invention provides a method for dynamic monitoring of landslide disasters based on time series and hybrid models, which is used to ensure the reliability of samples by collecting multi-source heterogeneous data and performing data preprocessing, and facilitates the training of the model through time series coding. Subsequently, the accuracy of the model is further ensured by model verification and construction of mapping relationships, providing a convenient basis for landslide disaster prediction.

[0006] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, comprising:

[0007] Step 1: Collect multi-source heterogeneous data of historical landslides, and pre-process the multi-source heterogeneous data to obtain first data, wherein the multi-source heterogeneous data involves multiple monitoring dimensions;

[0008] Step 2: Classify the first data into types, and obtain a single sample in the same time series by time series encoding each type of data, and then obtain a sample set, and then train the neural network model according to the sample set to obtain an initial model, wherein the single sample includes monitoring values ​​under different monitoring dimensions and horizontal displacement of the landslide;

[0009] Step 3: verifying the initial model, and optimizing the time series and hybrid model by combining the constructed mapping relationship between the corresponding type of landslide deformation and the monitoring dimension;

[0010] Step 4: Monitor and obtain the current heterogeneous data of the target monitoring point, and input it into the time series and hybrid model, predict the current disaster results of the target monitoring point and output a reminder.

[0011] Preferably, preprocessing the multi-source heterogeneous data to obtain the first data includes:

[0012] Classify the first data into types according to the monitoring dimension, and draw a smooth curve for each first value in each type of data in chronological order;

[0013] Determine the first absolute slope of the straight line formed by two adjacent points in the smooth curve respectively, and adjust the first absolute slope in combination with the area of ​​the curve formed by the two adjacent points to obtain a second absolute slope;

[0014] ;

[0015] Wherein, D2 is the corresponding second absolute slope; D1 is the corresponding first absolute slope; Represents the value of the highest point in the curve segment formed by the i-th point and the i+1-th point; Indicates the value of the point corresponding to the highest point in the straight line formed by the i-th point and the i+1-th point; Represents the value of the lowest point in the curve segment formed by the i-th point and the i+1-th point; Indicates the value of the point corresponding to the lowest point in the straight line formed by the i-th point and the i+1-th point; represents the closed area of ​​the curve segment formed by the i-th point and the i+1-th point; represents the closed area of ​​the straight line formed by the i-th point and the i+1-th point; n represents the total number of first values ​​that exist, and is consistent with the number of points for drawing the smooth curve;

[0016] Compare the first absolute slope and the second absolute slope of the same two adjacent points with the preset slope respectively. If both are less than or equal to the preset slope, perform a first adjustment on the latter point of the corresponding two adjacent points according to max(D1, D2), where max represents the maximum value symbol;

[0017] If both are greater than or equal to the preset slope, a second adjustment is performed on the next point of the two adjacent points according to min (D1, D2), where min represents the minimum value symbol;

[0018] If one of D1 and D2 has a slope greater than the preset slope and the other less than the preset slope, then a final slope is randomly selected according to rand((D1, D2), 1) to perform a third adjustment on the next point of the two adjacent points, where rand represents a random symbol;

[0019] The first data is obtained based on all the adjustment results.

[0020] Preferably, performing a first adjustment on the latter point of the corresponding two adjacent points according to max(D1, D2) includes:

[0021]

[0022] in, represents the value after the first adjustment of the latter point; Indicates the value before adjustment of the next point; Represents the variance of D1-D2 based on all two adjacent points; Indicates the positive or negative slope of the corresponding i-th point and i+1-th point.

[0023] Preferably, performing a second adjustment on the latter point of the corresponding two adjacent points according to min (D1, D2) includes:

[0024]

[0025] in, Indicates the value after the second adjustment of the latter point.

[0026] Preferably, a final slope is randomly selected according to rand((D1, D2), 1) to perform a third adjustment on the next point of the corresponding two adjacent points, including:

[0027]

[0028]

[0029] in, represents the value after the third adjustment of the latter point; represents the adjustment function for two adjacent points; rand((D1,D2),1) represents a function that randomly selects a final slope from D1 and D2 corresponding to two adjacent points; Indicates the preset slope.

[0030] Preferably, the time series and the mixed model are optimized, including:

[0031] Extracting first multi-source data and second source data at two adjacent moments from the sample set, wherein the first multi-source data corresponds to a first historical moment, the second multi-source data corresponds to a second historical moment, and the first historical moment is before the second historical moment;

[0032] Inputting the first multi-source data into the initial model to obtain a verification type and a first verification coefficient under different monitoring dimensions;

[0033] According to the historical displacement and mapping relationship of the corresponding disaster type, the first disaster coefficient of each monitoring dimension and the corresponding disaster type in the same single sample is determined respectively;

[0034] Obtain all first disaster coefficients of different monitoring dimensions involved in the same disaster type, and construct the first coefficient vector of each monitoring dimension under the same disaster type;

[0035] Extracting a first verification coefficient and a verification type of the first multi-source data corresponding to the first coefficient vector to obtain a verification coefficient vector and a verification type vector, and combining the first coefficient vector to obtain a first matrix;

[0036] Based on the first matrix, a difference factor based on each monitoring dimension in the same disaster type actually determined in history is obtained;

[0037] Relying on the first historical difference between the second multi-source data and the first multi-source data, the type difference between the historical actual disaster type under the second multi-source data and the verification type of the first multi-source data, and combining the difference factor of each monitoring dimension based on the corresponding historical actual disaster type, constructing the optimization vector corresponding to the first historical moment;

[0038] The initial model is optimized based on all optimization vectors to obtain a time series and a mixed model.

[0039] Preferably, according to the historical displacement and mapping relationship of the corresponding disaster type, the first disaster coefficient of each monitoring dimension and the corresponding disaster type in the same single sample is determined respectively, including:

[0040] Obtain the historical displacements of different disaster types from the time series-type-displacement comparison table;

[0041] Determine N1 first samples that are time-series associated with the single sample, sort the monitoring values ​​of each monitoring dimension in the N1+1 samples based on the time sequence, and determine whether there is a difference between the two sides of the corresponding single sample;

[0042] If there is a difference between the two sides, then according to and The size relationship and and The size relationship of is used to determine the important coefficient of the corresponding single sample under the corresponding monitoring dimension, where: Indicates the left difference corresponding to a single sample; Represents the right-hand difference corresponding to a single sample; Indicates the difference threshold under the corresponding monitoring dimension; Indicates the monitoring value of a single sample under the corresponding monitoring dimension; , They respectively represent the left monitoring value and the right monitoring value based on the corresponding single sample in the corresponding monitoring dimension; , They represent constants respectively;

[0043] If there is only one side difference, then according to Determine the important coefficient of the corresponding single sample under the corresponding monitoring dimension, where: Represents the average value of the monitoring values ​​corresponding to N1+1 samples under the corresponding monitoring dimension; Indicates the maximum value of the monitoring values ​​corresponding to N1+1 samples; Represents the variance of the monitoring value corresponding to N1+1 samples, where , the corresponding importance coefficient is judged to be 0; otherwise, the corresponding importance coefficient is judged to be ;

[0044] Based on the mapping relationship of the corresponding type and all historical displacements under the corresponding type, the setting weight of each monitoring dimension in the corresponding disaster type is determined, and combined with the important coefficient, the first disaster coefficient of the corresponding monitoring dimension in the corresponding disaster type is determined;

[0045]

[0046] Among them, Ui represents the first disaster coefficient of the i-th monitoring dimension in the corresponding disaster type; Indicates the historical disaster value of the corresponding disaster type; represents the set weight of the ith monitoring dimension in the corresponding disaster type; represents the important coefficient of the i-th monitoring dimension in the corresponding disaster type; It represents the sum of the set weights of all monitoring dimensions under the corresponding disaster type, and its value is 1; It represents the sum of the important coefficients of all monitoring dimensions under the corresponding disaster type; m represents the number of monitoring dimensions.

[0047] Preferably, obtaining the difference factors based on each monitoring dimension in the same disaster type actually determined in history based on the first matrix includes:

[0048] If all verification types involved in the first matrix are consistent with the corresponding disaster types, then the absolute value of the difference between the first disaster coefficient and the first verification coefficient in each column vector is calculated, and the coefficient variance is obtained and combined with the corresponding disaster type to obtain the difference factor;

[0049] If there are types among all the verification types involved in the first matrix that are inconsistent with the corresponding disaster types, the number of occurrences of each inconsistent type is extracted respectively, and the absolute value of the difference between all the first disaster coefficients and the first verification coefficients under the inconsistent type is combined to obtain the current factor of each inconsistent type and thus form a difference factor.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] By collecting multi-source heterogeneous data and performing data preprocessing, the reliability of the samples is guaranteed, and through time series coding, the training of the model is facilitated. Subsequently, through model verification and the construction of mapping relationships, the accuracy of the model is further guaranteed, providing a convenient basis for landslide disaster prediction.

[0052] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0053] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 It is a flow chart of a method for dynamic monitoring of landslide disasters based on time series and hybrid model in an embodiment of the present invention;

[0056] Figure 2 It is a comparison diagram of the actual and predicted results of the present invention. DETAILED DESCRIPTION

[0057] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0058] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, such as Figure 1 As shown, including:

[0059] Step 1: Collect multi-source heterogeneous data of historical landslides, and pre-process the multi-source heterogeneous data to obtain first data, wherein the multi-source heterogeneous data involves multiple monitoring dimensions;

[0060] Step 2: Classify the first data into types, and obtain a single sample in the same time series by time series encoding each type of data, and then obtain a sample set, and then train the neural network model according to the sample set to obtain an initial model, wherein the single sample includes monitoring values ​​under different monitoring dimensions and horizontal displacement of the landslide;

[0061] Step 3: verifying the initial model, and optimizing the time series and hybrid model by combining the constructed mapping relationship between the corresponding type of landslide deformation and the monitoring dimension;

[0062] Step 4: Monitor and obtain the current heterogeneous data of the target monitoring point, and input it into the time series and hybrid model, predict the current disaster results of the target monitoring point and output a reminder.

[0063] In this embodiment, the neural network model includes: an input layer, an embedding layer, an encoder layer, a bidirectional long short-term memory network, a fully connected layer, and an output layer. Specifically:

[0064] Input layer: The input dimension is set to 4, which will be explained in detail in the parameter setting.

[0065] Embedding layer: The model first contains an embedding layer, whose function is to transform the input features. Here, the input feature dimension is 4. After being processed by the linear layer, the output feature dimension becomes 64 and has a bias. The role of this embedding layer is to map the original input data so that it can better adapt to the processing requirements of the subsequent model layers.

[0066] Encoder layer: It contains several important components:

[0067] Multi-head Self-Attention: The input feature dimension is processed from 64 through its internal output projection layer, and the output is still 64 dimensions. It is used to capture the correlation information between different positions of the input sequence, so that the model can pay attention to different parts of the data and their relationships.

[0068] Linear transformation layer (Linear1 and Linear2): Linear layer 1 expands the input 64-dimensional features to 2048 dimensions, and then undergoes a random dropout operation (randomly discarding some neurons with a probability of 0.1 to prevent overfitting). Linear layer 2 then changes the dimension from 2048 dimensions back to 64 dimensions. This transformation is used to extract and integrate data features.

[0069] Normalization layer (Norm1 and Norm2): There are two layer normalization operations, and the eps in their parameters is set to 1e-05. Their function is to normalize the input data of each layer, which helps to stabilize the gradient during model training, accelerate convergence and improve model performance.

[0070] Additional Dropout layers (Dropout1 and Dropout2): These two Dropout layers also randomly drop neurons with probability 0.1 to further help prevent overfitting.

[0071] Bidirectional Long Short-Term Memory Network (BiLSTM):

[0072] This part is a bidirectional LSTM, which receives the 64-dimensional feature input from the previous processing. The LSTM output in each direction is 32-dimensional, but because it is bidirectional, the final output dimension is 32×2=64. The bidirectional structure used in this layer means that it can consider the forward and reverse information of the sequence data at the same time, and more comprehensively capture the temporal dependencies in the sequence. At the same time, the Dropout operation is set with a probability of 0.1 to avoid overfitting of the model.

[0073] Fully connected layer:

[0074] Then there is a fully connected layer, which receives the features from the bidirectional LSTM output. Since BiLSTM is bidirectional, the output dimension should be 512 dimensions after merging and other operations (the corresponding dimensions are integrated according to the characteristics of the bidirectional LSTM), and then it is converted into a 1-dimensional output through this linear layer. This layer also has a bias to make appropriate adjustments to the output results.

[0075] Output layer:

[0076] This is the last layer of the model, which produces a 1-dimensional output of the prediction model's processing of the 4-dimensional input data.

[0077] Set its core parameters and training parameters, then introduce the sample set for model training, and the parameter setting table is as follows:

[0078]

[0079] In the parameter setting, since the input data includes four features: horizontal displacement, elevation, rainfall, and reservoir water level, the input dimension is set to 4. This model aims to obtain disaster conditions, so the output dimension is 1. The length of the input sequence is adjusted to 8 according to the scale of the data set. In order to improve the generalization ability of the model and reduce overfitting, the random deactivation technique is used, and the deactivation rate is usually set to 0.1~0.2.

[0080] In the optimization of training parameters, after repeated trials and adjustments, the most suitable training configuration for the current model and data set was found. The number of iterations was set to 25, which not only ensured that the model had enough iterations to learn, but also avoided the problem of overfitting. Since the Adam optimization algorithm was used, the learning rate was set to 0.001. The number of training samples per batch was set to 32, which can strike a balance between memory usage and computational efficiency and help enhance the generalization performance of the model. The proportion of the validation set was set to 20%, which helps to reasonably distribute training and validation data, so as to more accurately evaluate the effect of the model.

[0081] The training dataset is used for the training process by using the provided model architecture and parameter configuration. The purpose of training is to enable the model to recognize and understand the patterns and features in the data so that it can make more accurate predictions. Figure 2 As shown in the figure, the horizontal displacement prediction results of GNSS02 monitoring point are displayed, where the red line represents the actually observed horizontal displacement data, and the blue line represents the predicted value generated by the model.

[0082] In this embodiment, historical landslides refer to areas where there is a downward sliding movement along a certain weak surface or weak zone, either as a whole or in a dispersed manner, and various data related to the historical landslides are known, including: surface displacement monitoring, rainfall monitoring, reservoir water level change monitoring, etc., and surface displacement monitoring uses GNSS real-time monitoring to obtain the surface displacement changes of the landslide to grasp the deformation dynamics of the landslide; rainfall monitoring uses rain gauges to obtain real-time information; reservoir water level change monitoring uses data collection to provide basic data for landslide deformation analysis and early warning.

[0083] In this embodiment, according to the basic characteristics and deformation features of the landslide body and in combination with the existing survey data, it is determined that the main monitoring contents of the landslide are surface displacement monitoring, auxiliary, rainfall monitoring and reservoir water level change monitoring.

[0084] In this embodiment, multi-source heterogeneous data refers to data in five dimensions, among which there are four input dimensions: horizontal displacement, elevation, rainfall and reservoir water level, and the output dimension is the disaster situation. It should be noted that the disaster situation is the horizontal displacement of the landslide.

[0085] That is: single sample = {horizontal displacement, elevation, rainfall, reservoir water level in the corresponding time series, horizontal displacement of landslide in the corresponding time series}.

[0086] In this embodiment, preprocessing refers to adjusting the result values ​​involved in multi-source heterogeneous data. Therefore, there may be certain measurement errors in the monitoring process. Therefore, the values ​​are fine-tuned based on continuous time series to ensure the reliability of the data and avoid reducing the accuracy of the model due to errors.

[0087] In this embodiment, the type classification is to divide the first data according to five dimensions, and the data under each dimension is a corresponding type of data.

[0088] In this embodiment, constructing a mapping relationship between corresponding types of landslide deformation and influencing factors refers to determining the impact of data under the corresponding dimension on the landslide. The greater the impact, the greater the weight corresponding to the dimension, and the influencing factor refers to the impact of the corresponding dimension.

[0089] In this embodiment, time series coding refers to setting the collection time for different data to facilitate the subsequent construction of samples.

[0090] In this embodiment, the current heterogeneous data of the target monitoring point refers to the data under four input dimensions obtained by measurement.

[0091] In this embodiment, the current disaster is related to the predicted horizontal displacement result of the target monitoring point. If the horizontal displacement result exceeds the set displacement, a reminder is required. Otherwise, no reminder is required, and the reminder refers to reminding the relevant monitoring personnel. If the predicted result is that the horizontal displacement under the current disaster is 10cm, but the set displacement is 8cm, then an early warning is required.

[0092] The beneficial effects of the above technical solution are: by collecting multi-source heterogeneous data and performing data preprocessing, the reliability of the sample is guaranteed, and through time series coding, the training of the model is facilitated, and the subsequent verification of the model and the construction of mapping relationships are used to further ensure the accuracy of the model, providing a convenient basis for landslide disaster prediction.

[0093] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which preprocesses the multi-source heterogeneous data to obtain first data, including:

[0094] Classify the multi-source heterogeneous data into types according to the monitoring dimension, and draw a smooth curve for each first value in each type of data in chronological order;

[0095] Determine the first absolute slope of the straight line formed by two adjacent points in the smooth curve respectively, and adjust the first absolute slope in combination with the area of ​​the curve formed by the two adjacent points to obtain a second absolute slope;

[0096]

[0097] Wherein, D2 is the corresponding second absolute slope; D1 is the corresponding first absolute slope; Represents the value of the highest point in the curve segment formed by the i-th point and the i+1-th point; Indicates the value of the point corresponding to the highest point in the straight line formed by the i-th point and the i+1-th point; Represents the value of the lowest point in the curve segment formed by the i-th point and the i+1-th point; Indicates the value of the point corresponding to the lowest point in the straight line formed by the i-th point and the i+1-th point; represents the closed area of ​​the curve segment formed by the i-th point and the i+1-th point; represents the closed area of ​​the straight line formed by the i-th point and the i+1-th point; n represents the total number of first values ​​that exist, and is consistent with the number of points for drawing the smooth curve;

[0098] Compare the first absolute slope and the second absolute slope of the same two adjacent points with the preset slope respectively. If both are less than or equal to the preset slope, perform a first adjustment on the latter point of the corresponding two adjacent points according to max(D1, D2), where max represents the maximum value symbol;

[0099] If both are greater than or equal to the preset slope, a second adjustment is performed on the next point of the two adjacent points according to min (D1, D2), where min represents the minimum value symbol;

[0100] If one of D1 and D2 has a slope greater than the preset slope and the other less than the preset slope, then a final slope is randomly selected according to rand((D1, D2), 1) to perform a third adjustment on the next point of the two adjacent points, where rand represents a random symbol;

[0101] The first data is obtained based on all the adjustment results.

[0102] In this embodiment, the adjusted data in each dimension is counted to obtain the first data.

[0103] In this embodiment, multi-source heterogeneous data under each monitoring dimension can be divided into types to obtain data under different dimensions. The horizontal axis of the smooth curve drawn is time, and the vertical axis is the value monitored under the corresponding dimension, and the values ​​of all monitored values ​​are greater than or equal to 0.

[0104] In this embodiment, for example, there are point 1, point 2, and point 3 arranged in time sequence in the smooth curve. At this time, point 1 and point 2, and point 2 and point 3 are adjacent points respectively.

[0105] In this embodiment, the first absolute slope=|(value of the next point−value of the previous point) / time interval|.

[0106] In this embodiment, for example, the curve area for point 1 and point 2 is the area of ​​a closed area formed by vertically cutting the smooth curve with point 1 and point 2 as reference points in the corresponding smooth curve and the time axis.

[0107] In this embodiment, the purpose of adjusting the slope is to reduce the area difference between the area enclosed by the curve and the area enclosed by the straight line, so as to minimize the inaccuracy of subsequent analysis of the disaster situation due to measurement errors.

[0108] In this embodiment, the value of the preset slope is generally 0.1.

[0109] In this embodiment, the reliability of the corresponding value adjustment is further guaranteed by comparing the two absolute slopes with the preset slope, thereby reducing the data error.

[0110] The beneficial effect of the above technical solution is: a smooth curve is drawn based on the data in each dimension, and the reliability of the data is determined by comparing the slope before and after adjustment between two points with the preset slope, providing an accurate basis for subsequent training models.

[0111] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which performs a first adjustment on the latter point of two corresponding adjacent points according to max (D1, D2), including:

[0112]

[0113] in, represents the value after the first adjustment of the latter point; Indicates the value before adjustment of the next point; Represents the variance of D1-D2 based on all two adjacent points; Indicates the positive or negative slope of the corresponding i-th point and i+1-th point.

[0114] In this embodiment, the positive or negative condition of the slope refers to whether the slope is greater than 0, less than 0, or equal to 0 before the absolute value is added to the slope.

[0115] The beneficial effect of the above technical solution is that the first adjustment is made to the relevant value based on the positive or negative of the slope.

[0116] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which performs a second adjustment on the latter point of two corresponding adjacent points according to min (D1, D2), including:

[0117]

[0118] in, Indicates the value after the second adjustment of the latter point.

[0119] The beneficial effect of the above technical solution is that the second adjustment is made to the relevant value based on the positive or negative of the slope.

[0120] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which randomly selects a final slope according to rand((D1, D2), 1) to perform a third adjustment on the latter point of the corresponding two adjacent points, including:

[0121]

[0122]

[0123] in, represents the value after the third adjustment of the next point; D3 represents the adjustment function for the corresponding two adjacent points; rand((D1, D2), 1) represents a function that randomly selects a final slope from D1 and D2 corresponding to the two adjacent points; Indicates the preset slope.

[0124] The beneficial effect of the above technical solution is: the third adjustment is made to the relevant values ​​based on the random function.

[0125] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which optimizes the time series and hybrid model, including:

[0126] Extracting first multi-source data and second source data at two adjacent moments from the sample set, wherein the first multi-source data corresponds to a first historical moment, the second multi-source data corresponds to a second historical moment, and the first historical moment is before the second historical moment;

[0127] Inputting the first multi-source data into the initial model to obtain a verification type and a first verification coefficient under different monitoring dimensions;

[0128] According to the historical displacement and mapping relationship of the corresponding disaster type, the first disaster coefficient of each monitoring dimension and the corresponding disaster type in the same single sample is determined respectively;

[0129] Obtain all first disaster coefficients of different monitoring dimensions involved in the same disaster type, and construct the first coefficient vector of each monitoring dimension under the same disaster type;

[0130] Extracting a first verification coefficient and a verification type of the first multi-source data corresponding to the first coefficient vector to obtain a verification coefficient vector and a verification type vector, and combining the first coefficient vector to obtain a first matrix;

[0131] Based on the first matrix, a difference factor based on each monitoring dimension in the same disaster type actually determined in history is obtained;

[0132] Relying on the first historical difference between the second multi-source data and the first multi-source data, the type difference between the historical actual disaster type under the second multi-source data and the verification type of the first multi-source data, and combining the difference factor of each monitoring dimension based on the corresponding historical actual disaster type, constructing the optimization vector corresponding to the first historical moment;

[0133] The initial model is optimized based on all optimization vectors to obtain a time series and a mixed model.

[0134] In this embodiment, the first disaster coefficient is known, that is, the monitoring values ​​under different dimensions can be obtained by matching them from the dimension-monitoring-disaster comparison table, wherein the dimension-monitoring-disaster comparison table includes different monitoring values ​​under different input dimensions and the disaster coefficients corresponding to the values, and the disaster coefficient has a value range of 0 to 1. Generally, the closer the monitoring value under the corresponding dimension is to the pre-set warning value, the larger the corresponding disaster coefficient is. That is, before determining the horizontal displacement, it is necessary to perform a disaster analysis on the monitoring value under each monitoring dimension, and then comprehensively obtain the horizontal displacement to ensure the accuracy of subsequent model predictions. Among them, in the case of known measurement data and horizontal displacement under four dimensions, the corresponding disaster type already exists and is evaluated in advance by experts. Therefore, during the operation of the model, there will be relevant verification coefficients and verification types under different dimensions, but the final output is the horizontal displacement.

[0135] In this embodiment, the first multi-source data involves monitoring values ​​in four dimensions, the first disaster coefficient of each monitoring value, the corresponding disaster type and the horizontal displacement. Therefore, by inputting the data into the model for verification, the following can be obtained: the first verification coefficient, verification type and verification displacement in four dimensions.

[0136] In this embodiment, the difference factor refers to the coefficient and type of the difference.

[0137] In this embodiment, the first historical difference={differences in monitoring values ​​under four monitoring dimensions, differences in disaster coefficients corresponding to each monitoring value, differences in disaster types, and differences in horizontal displacements}.

[0138] In this embodiment, the type difference is the difference between the actual historical disaster type at the second historical moment and the verified type at the first historical moment.

[0139] In this embodiment, the optimization vector={first historical difference, type difference, difference factor}.

[0140] In this embodiment, the timing can be 1s, that is, it is determined by a short time, and the result obtained by inputting the information of the known moment is used as the prediction result of the next moment. Then, the optimization vector can be constructed through two comparisons to ensure the effective output of the displacement corresponding to the subsequent monitoring of the four dimensions of the current monitoring point.

[0141] In this embodiment, the first coefficient vector={the first disaster coefficient of each time series under the same monitoring dimension under the corresponding disaster type}.

[0142] In this embodiment, the verification coefficient vector={the first verification coefficient of each time series under the same monitoring dimension under the corresponding disaster type}.

[0143] In this embodiment, the verification type vector={verification type at each time sequence}.

[0144] In this embodiment, .

[0145] The beneficial effect of the above technical solution is: through two consecutive moments, the previous moment is used to verify the next moment, and specifically, a matrix is ​​constructed through a vector composed of disaster coefficients, verification coefficients, and disaster types to obtain difference factors, and an optimization vector is constructed by combining historical differences and type differences of multi-source data to achieve reliable optimization of the model and ensure model accuracy.

[0146] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which determines the first disaster coefficient of each monitoring dimension and the corresponding disaster type in the same single sample according to the historical displacement and mapping relationship of the corresponding disaster type, including:

[0147] Obtain the historical displacements of different disaster types from the time series-type-displacement comparison table;

[0148] Determine N1 first samples that are time-series associated with the single sample, sort the monitoring values ​​of each monitoring dimension in the N1+1 samples based on the time sequence, and determine whether there is a difference between the two sides of the corresponding single sample;

[0149] If there is a difference between the two sides, then according to and The size relationship and and The size relationship of is used to determine the important coefficient of the corresponding single sample under the corresponding monitoring dimension, where: Indicates the left difference corresponding to a single sample; Represents the right-hand difference corresponding to a single sample; Indicates the difference threshold under the corresponding monitoring dimension; Indicates the monitoring value of a single sample under the corresponding monitoring dimension; , They respectively represent the left monitoring value and the right monitoring value based on the corresponding single sample in the corresponding monitoring dimension; , They represent constants respectively;

[0150] If there is only one side difference, then according to Determine the important coefficient of the corresponding single sample under the corresponding monitoring dimension, where: Represents the average value of the monitoring values ​​corresponding to N1+1 samples under the corresponding monitoring dimension; Indicates the maximum value of the monitoring values ​​corresponding to N1+1 samples; Represents the variance of the monitoring value corresponding to N1+1 samples, where , the corresponding importance coefficient is judged to be 0; otherwise, the corresponding importance coefficient is judged to be ;

[0151] Based on the mapping relationship of the corresponding type and all historical displacements under the corresponding type, the setting weight of each monitoring dimension in the corresponding disaster type is determined, and combined with the important coefficient, the first disaster coefficient of the corresponding monitoring dimension in the corresponding disaster type is determined;

[0152]

[0153] Among them, Ui represents the first disaster coefficient of the i-th monitoring dimension in the corresponding disaster type; Indicates the historical disaster value of the corresponding disaster type; represents the set weight of the ith monitoring dimension in the corresponding disaster type; represents the important coefficient of the i-th monitoring dimension in the corresponding disaster type; It represents the sum of the set weights of all monitoring dimensions under the corresponding disaster type, and its value is 1; It represents the sum of the important coefficients of all monitoring dimensions under the corresponding disaster type; m represents the number of monitoring dimensions.

[0154] In this embodiment, weight is set = , where Yc represents the influence degree of the mapping relationship under the corresponding type; sum1 represents the sum of all influence degrees under the corresponding type; represents the average value of all historical displacements of the corresponding type, Indicates the maximum value among all historical positions of the corresponding type; Indicates all types of The sum of , and each type of The calculation results are different.

[0155] In this embodiment, the time series-type-displacement comparison table includes different historical actual disaster types and the corresponding historical displacements, which are all stored for easy direct retrieval and use, and the historical displacement under each type is unique because the time series is unique. Among them, the disaster type can be a step-type wading landslide type, a push-type landslide, etc., and the uniqueness of each type is obtained by analysis after the landslide occurs.

[0156] In this embodiment, N1 is greater than or equal to 10.

[0157] In this embodiment, the time series association means that the continuous time series sequence only needs to include a corresponding single sample.

[0158] In this embodiment, = the monitoring value of the corresponding single sample in the corresponding dimension - the monitoring value of the left sample of the single sample in the corresponding dimension, =The monitoring value of the corresponding single sample in the corresponding dimension - the monitoring value of the sample on the right side of the single sample in the corresponding dimension.

[0159] In this embodiment, The value of is 0.5. The value of is 0.4.

[0160] like ,and ,At this time, it is determined that the importance coefficient of the corresponding single sample in the corresponding monitoring dimension is 1;

[0161] like ,and ,At this time, it is determined that the importance coefficient of the corresponding single sample in the corresponding monitoring dimension is 0.5;

[0162] like ,and At this time, the importance coefficient of the corresponding single sample in the corresponding monitoring dimension is determined to be 0.5.

[0163] The beneficial effect of the above technical solution is: by obtaining samples in a continuous time series, the important coefficients of a single sample with a one-sided difference and a two-sided difference are determined, and the difference types in different dimensions are determined by combining the mapping relationship and historical displacement, which facilitates the acquisition of the optimization vector.

[0164] The present invention provides a landslide disaster dynamic monitoring method based on time series and hybrid model, which obtains the difference factor based on each monitoring dimension in the same disaster type actually determined in history based on the first matrix, including:

[0165] If all verification types involved in the first matrix are consistent with the corresponding disaster types, then the absolute value of the difference between the first disaster coefficient and the first verification coefficient in each column vector is calculated, and the coefficient variance is obtained and combined with the corresponding disaster type to obtain the difference factor;

[0166] If there are types among all the verification types involved in the first matrix that are inconsistent with the corresponding disaster types, the number of occurrences of each inconsistent type is extracted respectively, and the absolute value of the difference between all the first disaster coefficients and the first verification coefficients under the inconsistent type is combined to obtain the current factor of each inconsistent type and thus form a difference factor.

[0167] In this embodiment, if the row vectors corresponding to the verification type in the first matrix are all consistent with the disaster type, then the difference factor is: coefficient variance-disaster type.

[0168] In this embodiment, if they are different, at this time, the average value of the absolute value of the difference under the inconsistent type / the corresponding number of occurrences is calculated, and then combined with the inconsistent type to obtain the difference factor.

[0169] The beneficial effect of the above technical solution is: through type comparison analysis, the difference factors in the cases of consistent types and inconsistent types are determined, providing a basis for constructing an optimization vector.

[0170] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A dynamic monitoring method for landslide disasters based on time series and hybrid model, characterized in that: include: Step 1: Collect multi-source heterogeneous data of historical landslides, and pre-process the multi-source heterogeneous data to obtain first data, wherein the multi-source heterogeneous data involves multiple monitoring dimensions; Step 2: Classify the first data into types, and obtain a single sample in the same time series by time series encoding each type of data, and then obtain a sample set, and then train the neural network model according to the sample set to obtain an initial model, wherein the single sample includes monitoring values ​​under different monitoring dimensions and horizontal displacement of the landslide; Step 3: verifying the initial model, and optimizing the time series and hybrid model by combining the constructed mapping relationship between the corresponding type of landslide deformation and the monitoring dimension; Step 4: Monitor and obtain the current heterogeneous data of the target monitoring point, and input it into the time series and hybrid model, predict the current disaster results of the target monitoring point and output a reminder.

2. The landslide disaster dynamic monitoring method based on time series and hybrid model according to claim 1 is characterized in that: Preprocessing the multi-source heterogeneous data to obtain first data includes: Classify the first data into types according to the monitoring dimension, and draw a smooth curve for each first value in each type of data in chronological order; Determine the first absolute slope of the straight line formed by two adjacent points in the smooth curve respectively, and adjust the first absolute slope in combination with the area of ​​the curve formed by the two adjacent points to obtain a second absolute slope; ； Wherein, D2 is the corresponding second absolute slope; D1 is the corresponding first absolute slope; Represents the value of the highest point in the curve segment formed by the i-th point and the i+1-th point; Indicates the value of the point corresponding to the highest point in the straight line formed by the i-th point and the i+1-th point; Represents the value of the lowest point in the curve segment formed by the i-th point and the i+1-th point; Indicates the value of the point corresponding to the lowest point in the straight line formed by the i-th point and the i+1-th point; represents the closed area of ​​the curve segment formed by the i-th point and the i+1-th point; represents the closed area of ​​the straight line formed by the i-th point and the i+1-th point; n represents the total number of first values ​​that exist, and is consistent with the number of points for drawing the smooth curve; Compare the first absolute slope and the second absolute slope of the same two adjacent points with the preset slope respectively. If both are less than or equal to the preset slope, perform a first adjustment on the latter point of the corresponding two adjacent points according to max(D1, D2), where max represents the maximum value symbol; If both are greater than or equal to the preset slope, a second adjustment is performed on the next point of the two adjacent points according to min (D1, D2), where min represents the minimum value symbol; If one of D1 and D2 has a slope greater than the preset slope and the other less than the preset slope, then a final slope is randomly selected according to rand((D1, D2), 1) to perform a third adjustment on the next point of the two adjacent points, where rand represents a random symbol; The first data is obtained based on all the adjustment results.

3. The landslide disaster dynamic monitoring method based on time series and hybrid model according to claim 2 is characterized in that: The first adjustment is performed on the next point of the corresponding two adjacent points according to max (D1, D2), including: ； in, represents the value after the first adjustment of the latter point; Indicates the value before adjustment of the next point; Represents the variance of D1-D2 based on all two adjacent points; Indicates the positive or negative slope of the corresponding i-th point and i+1-th point.

4. The landslide disaster dynamic monitoring method based on time series and hybrid model according to claim 3 is characterized in that: The second adjustment is performed on the next point of the two corresponding adjacent points according to min (D1, D2), including: ； in, Indicates the value after the second adjustment of the latter point.

5. The method for dynamic monitoring of landslide disasters based on time series and hybrid model according to claim 3 is characterized in that: According to rand((D1,D2),1), a final slope is randomly selected to make a third adjustment to the next point of the corresponding two adjacent points, including: ； ； in, represents the value after the third adjustment of the next point; D3 represents the adjustment function for the corresponding two adjacent points; rand((D1, D2), 1) represents a function that randomly selects a final slope from D1 and D2 corresponding to the two adjacent points; Indicates the preset slope.

6. The method for dynamic monitoring of landslide disasters based on time series and hybrid model according to claim 1 is characterized in that: Optimized time series and mixed models, including: Extracting first multi-source data and second source data at two adjacent moments from the sample set, wherein the first multi-source data corresponds to a first historical moment, the second multi-source data corresponds to a second historical moment, and the first historical moment is before the second historical moment; Inputting the first multi-source data into the initial model to obtain a verification type and a first verification coefficient under different monitoring dimensions; According to the historical displacement and mapping relationship of the corresponding disaster type, the first disaster coefficient of each monitoring dimension and the corresponding disaster type in the same single sample is determined respectively; Obtain all first disaster coefficients of different monitoring dimensions involved in the same disaster type, and construct the first coefficient vector of each monitoring dimension under the same disaster type; Extracting a first verification coefficient and a verification type of the first multi-source data corresponding to the first coefficient vector to obtain a verification coefficient vector and a verification type vector, and combining the first coefficient vector to obtain a first matrix; Based on the first matrix, a difference factor based on each monitoring dimension in the same disaster type actually determined in history is obtained; Relying on the first historical difference between the second multi-source data and the first multi-source data, the type difference between the historical actual disaster type under the second multi-source data and the verification type of the first multi-source data, and combining the difference factor of each monitoring dimension based on the corresponding historical actual disaster type, constructing the optimization vector corresponding to the first historical moment; The initial model is optimized based on all optimization vectors to obtain a time series and a mixed model.

7. The method for dynamic monitoring of landslide disasters based on time series and hybrid model according to claim 6 is characterized in that: According to the historical displacement and mapping relationship of the corresponding disaster type, the first disaster coefficient of each monitoring dimension and the corresponding disaster type in the same single sample is determined respectively, including: Obtain the historical displacements of different disaster types from the time series-type-displacement comparison table; Determine N1 first samples that are time-series associated with the single sample, sort the monitoring values ​​of each monitoring dimension in the N1+1 samples based on the time sequence, and determine whether there is a difference between the two sides of the corresponding single sample; If there is a difference between the two sides, then according to and The size relationship and and The size relationship of is used to determine the important coefficient of the corresponding single sample under the corresponding monitoring dimension, where: Indicates the left difference corresponding to a single sample; Represents the right-hand difference corresponding to a single sample; Indicates the difference threshold under the corresponding monitoring dimension; Indicates the monitoring value of a single sample under the corresponding monitoring dimension; , They respectively represent the left monitoring value and the right monitoring value based on the corresponding single sample in the corresponding monitoring dimension; , They represent constants respectively; If there is only one side difference, then according to Determine the important coefficient of the corresponding single sample under the corresponding monitoring dimension, where: Represents the average value of the monitoring values ​​corresponding to N1+1 samples under the corresponding monitoring dimension; Indicates the maximum value of the monitoring values ​​corresponding to N1+1 samples; Represents the variance of the monitoring value corresponding to N1+1 samples, where , the corresponding importance coefficient is judged to be 0; otherwise, the corresponding importance coefficient is judged to be ; Based on the mapping relationship of the corresponding type and all historical displacements under the corresponding type, the setting weight of each monitoring dimension in the corresponding disaster type is determined, and combined with the important coefficient, the first disaster coefficient of the corresponding monitoring dimension in the corresponding disaster type is determined; ； Among them, Ui represents the first disaster coefficient of the i-th monitoring dimension in the corresponding disaster type; Indicates the historical disaster value of the corresponding disaster type; represents the set weight of the ith monitoring dimension in the corresponding disaster type; represents the important coefficient of the i-th monitoring dimension in the corresponding disaster type; It represents the sum of the set weights of all monitoring dimensions under the corresponding disaster type, and its value is 1; It represents the sum of the important coefficients of all monitoring dimensions under the corresponding disaster type; m represents the number of monitoring dimensions.

8. The method for dynamic monitoring of landslide disasters based on time series and hybrid model according to claim 7 is characterized in that: Based on the first matrix, the difference factors based on each monitoring dimension in the same disaster type actually determined in history are obtained, including: If all verification types involved in the first matrix are consistent with the corresponding disaster types, then the absolute value of the difference between the first disaster coefficient and the first verification coefficient in each column vector is calculated, and the coefficient variance is obtained and combined with the corresponding disaster type to obtain the difference factor; If there are types among all the verification types involved in the first matrix that are inconsistent with the corresponding disaster types, the number of occurrences of each inconsistent type is extracted respectively, and the absolute value of the difference between all the first disaster coefficients and the first verification coefficients under the inconsistent type is combined to obtain the current factor of each inconsistent type and thus form a difference factor.

Images