A loess slope deformation monitoring data processing method based on MEMS inertial sensor data and GNSS data

Abstract

The application provides a loess slope deformation monitoring data processing method based on MEMS inertial sensor data and GNSS data, and relates to the technical field of slope deformation monitoring. The specific steps comprise the following steps: calculating absolute displacement and relative displacement, generating cumulative displacement, and collecting rainfall characteristic parameters; iteratively optimizing clustering and clustering center points, confirming the number of each clustering sample, merging or splitting the clustering when the number does not meet the threshold, until the preset threshold is met, determining the final clustering and clustering center points; determining the belonging clustering of the current time period characteristic vector, and screening out the same kind of historical time period; predicting the next time period cumulative displacement, and performing instability risk grade evaluation. The application adopts the clustering iteration method to continuously update the clustering and clustering center points, dynamically adjusts the clustering structure, effectively avoids the over-subdivision or over-merging of clustering, and accurately predicts the loess slope instability risk grade by relying on the dynamic clustering optimization mechanism.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of slope deformation monitoring, in particular to a loess slope deformation monitoring data processing method based on MEMS inertial sensor data and GNSS data. BACKGROUND

[0002] As a special geological structure, the stability of loess slope is easily affected by external environment (such as rainfall, earthquake, etc.), especially in the rainfall environment, the rainfall infiltration will cause the increase of loess water content and the decrease of soil shear strength, thereby causing landslide, collapse and other geological disasters. At present, the mainstream slope deformation monitoring technology includes GNSS and MEMS inertial sensor. By using the long-term displacement trend information provided by GNSS and the high-frequency dynamic deformation characteristics provided by MEMS, the deformation characteristics of the slope can be more comprehensively reflected.

[0003] However, in the clustering analysis of rainfall characteristic parameters, the existing technology usually adopts fixed clustering number or experience setting clustering number, lacking adaptive adjustment mechanism for dynamic change of sample distribution. Such rigid clustering division method is easy to cause uneven distribution of data in the cluster, and the phenomenon of over-subdivision or over-merging of the cluster, affecting the subsequent feature extraction and model training effect.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a loess slope deformation monitoring data processing method based on MEMS inertial sensor data and GNSS data, to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0007] A loess slope deformation monitoring data processing method based on MEMS inertial sensor data and GNSS data, the specific steps comprising:

[0008] S1. Using GNSS receiver and three-axis accelerometer based on MEMS technology to collect three-dimensional absolute coordinates and three-axis acceleration data of monitoring points in the current time period and the previous n historical time periods, calculate absolute displacement and relative displacement, and generate cumulative displacement, and collect rainfall characteristic parameters of each time period;

[0009] S2. Combine the rainfall characteristic parameters of each historical time period into a feature vector sample, take all the feature vector samples as a training set, set an initial cluster number and a cluster feature vector sample quantity threshold, iteratively optimize the cluster and cluster center point, confirm the quantity of each cluster sample, merge or split the cluster when the quantity threshold is not met, until the preset threshold is met, determine the final cluster and cluster center point;

[0010] S3. Calculate the distance between the feature vector of the current time period and the final cluster center point to determine its cluster;

[0011] S4. According to the cluster to which the current time period belongs, filter out the historical time periods with the same cluster, named as the same type of historical time period, splice the rainfall characteristic parameters and cumulative displacement of each same type of historical time period and its previous m-1 time periods in time order to form a time series feature vector;

[0012] S5. Take the time series feature vector of each same type of historical time period and its previous m-1 time periods as input, and the cumulative displacement of the next time period of the same type of historical time period as label output, train the cumulative displacement prediction model;

[0013] S6. Input the time series feature vector of the current time period and its previous m-1 time periods into the model, predict the cumulative displacement of the next time period, and calculate the deformation rate of the next time period, combine the cumulative displacement to predict the loess slope instability risk grade.

[0014] Further, the rainfall characteristic parameters include rainfall intensity, rainfall amount and rainfall duration.

[0015] Further, the current time period and the previous historical time period are divided as follows:

[0016] The current time period and the previous historical time period are divided as follows: The length of each time period is The set of continuous time periods is denoted as , is the index of the time period, , is the index of the current time period, and the interval range of the time period is , wherein is the starting time of the time period, which is also the ending time of the time period, is the ending time of the time period, which is also the starting time of the time period;

[0017] Based on the three-dimensional absolute coordinates, the absolute displacement of each time period is calculated, and the formula is as follows:

[0018]

[0019] Wherein, is the absolute displacement of the first time period, , , are the coordinates of the monitoring point on the first time period end moment axis, axis, axis, are the coordinates of the monitoring point on the first time period start moment axis, axis, axis;

[0020] The three-axis acceleration data is twice integrated to calculate the relative displacement of each time period, and the formula is as follows:

[0021]

[0022] Wherein, is the instantaneous velocity at any time in the first time period, is the velocity at the start moment in the first time period, is the instantaneous acceleration at any time in the first time period, is the time variable from the start moment to the end moment of the first time period, is the time variable from the start moment to the time of the first time period, is the relative displacement of the first time period;

[0023] The absolute displacement and the relative displacement are weighted to obtain the cumulative displacement of each time period, and the formula is as follows:

[0024]

[0025] Wherein, is the cumulative displacement of the first time period, is the weight coefficient of the absolute displacement, The weighting coefficients for relative displacements are... Based on this, let .

[0026] Furthermore, an initial number of clusters and a threshold for the number of feature vector samples in each cluster are set. The clusters and cluster centroids are iteratively optimized to confirm the number of samples in each cluster. If the number of samples in a cluster does not meet the threshold, the clusters are merged or split. This process continues until the preset threshold is met. The logic for determining the final clusters and cluster centroids is as follows:

[0027] 1) Set the initial number of clusters to... The lower limit for the number of feature vector samples in each cluster is set to . The upper limit is ;

[0028] Initialize cluster centroids to obtain an initial set of cluster centroids. , The k-th initial cluster center in the initial set of cluster centers. The index for clustering is denoted by a vector containing three feature values: rainfall intensity, rainfall amount, and rainfall duration.

[0029] 2) Perform the first clustering iteration, calculate the distance between the feature vector samples of the historical time period and each initial cluster center, and assign each feature vector sample to the nearest cluster center to form a set of clusters to be processed. , For set The k-th cluster in, For set The number of clusters in;

[0030] Next, the number of feature vector samples in each cluster of the cluster set to be processed is counted;

[0031] If the feature vector samples within a cluster are smaller than Set a distance threshold, calculate the distance between this cluster and the center points of other clusters in the same set, and filter out clusters whose distance to the center point is less than the distance threshold and whose number of internal feature vector samples is not less than [a certain value]. If a cluster is selected as a candidate cluster, it is merged into the candidate cluster with the smallest distance from its centroid. If no candidate cluster exists, it is merged into the cluster with the smallest distance from its centroid in the same set.

[0032] If the feature vector samples within a cluster are greater than Then, the feature vector samples within the cluster are clustered using the K-value clustering method to split the cluster into two clusters;

[0033] The above merging and splitting process is repeated until the number of feature vector samples in each cluster is not less than and not greater than to obtain the final cluster set , calculate the distribution entropy value thereof, and calculate the mean of each cluster in the set , update the cluster center point, wherein is the number of clusters in the set .

[0034] 3) Based on the updated cluster center point, the clustering iteration is performed again to obtain the final cluster set after the second iteration and the distribution entropy value thereof, and calculate the mean of each cluster in the set , update the cluster center point, wherein is the number of clusters in the set .

[0035] 4) Calculate the absolute difference of the distribution entropy value between the final cluster set in the current iteration and the final cluster set in the last iteration, compare the absolute difference with the convergence threshold:

[0036] If the absolute difference is not less than the convergence threshold, continue the iteration; if the absolute difference is less than the convergence threshold, stop the clustering iteration, and obtain the final cluster set and the final cluster center point set , wherein is the number of clusters in the set .

[0037] Further, the distance between the feature vector of the current time period and each cluster center point is calculated, and the logic of its cluster is as follows:

[0038] The feature vector of the current time period , wherein respectively correspond to the rainfall intensity, rainfall amount, and rainfall duration of the current time period;

[0039]

[0040] , wherein is the distance between the feature vector of the current time period and the final th cluster center point, is the th feature value of the feature vector of the current time period, is the th feature value of the final th cluster center point, is the index of the feature value, , respectively correspond to the rainfall intensity, rainfall amount, and rainfall duration;

[0041] Compare the distances of the feature vectors of the current time period to the cluster centers, and classify them into the cluster with the smallest distance.

[0042] Furthermore, the deformation rate for the next time period is calculated using the following formula:

[0043]

[0044] in, The deformation rate for the next time period. This represents the cumulative displacement for the next time period. This represents the cumulative displacement over the current time period.

[0045] The logic for predicting the instability risk level of loess slopes by combining the deformation rate and cumulative displacement in the next time period is as follows:

[0046] If the cumulative displacement in the next time period and deformation rate If so, the risk of instability of the loess slope is determined to be low;

[0047] If the cumulative displacement in the next time period and deformation rate If so, the risk of instability of the loess slope is determined to be moderate;

[0048] If the cumulative displacement in the next time period or deformation rate If so, the loess slope is deemed to have a high risk of instability;

[0049] in, , These are the first and second thresholds for the cumulative displacement in the next time period, respectively. , , These are the first and second thresholds for the deformation rate in the next time period, respectively. .

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] This invention combines rainfall characteristic parameters from different time periods into feature vector samples and uses all feature vector samples as a training set. It sets an initial number of clusters and a threshold for the number of feature vector samples in each cluster, and uses an iterative optimization method to continuously update the cluster centers. During this process, the number of samples in each cluster is counted in real time. An adjustment mechanism is used to merge clusters with too few samples and split clusters with too many samples, which can dynamically adjust the cluster structure and effectively avoid over-subdivision or over-merging of clusters.

[0052] The clustering center points are updated through clustering iteration and combined with sample reassignment, so that the clustering center points gradually approach the real distribution characteristics of the samples, and finally the reasonable and accurate clustering and the center points are determined. The final clustering and clustering center points obtained in this way provide an accurate benchmark for the clustering determination of the feature vector of the current time period, so that the clustering division is more accurate, and the accurate clustering division ensures that the real clustering same and representative historical time period can be screened out subsequently, ensuring the consistency and representativeness of the time sequence feature vector input;

[0053] Relying on the dynamic clustering optimization mechanism, the time sequence feature vector used by the cumulative displacement prediction model trained is more representative, which significantly improves the prediction accuracy of the next time period cumulative displacement and deformation rate. Combined with the calculation of the cumulative displacement, the loess slope instability risk level can be accurately predicted. BRIEF DESCRIPTION OF DRAWINGS

[0054] Fig. 1 The figure is a schematic diagram of the overall method of the present application;

[0055] Fig. 2 The figure is a fitting diagram of the relationship between the absolute displacement and the cumulative displacement of the present application;

[0056] Fig. 3 The figure is a fitting diagram of the relationship between the relative displacement and the cumulative displacement of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below with specific examples.

[0058] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent the relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0059] Example 1:

[0060] Please refer to Figs. 1-3 The present application provides a technical solution:

[0061] A loess slope deformation monitoring data processing method based on MEMS inertial sensor data and GNSS data, the specific steps include:

[0062] S1. Use GNSS receiver and three-axis accelerometer based on MEMS technology to collect three-dimensional absolute coordinates and three-axis acceleration data of monitoring points in current time period and previous n historical time periods, the data collected by GNSS receiver and three-axis accelerometer based on MEMS technology need to be time-synchronized, and the three-dimensional absolute coordinates and three-axis acceleration data are divided into current time period and previous n historical time periods according to time stamp, calculate absolute displacement and relative displacement, and generate cumulative displacement amount, and collect rainfall characteristic parameters of each time period, including rainfall intensity, rainfall amount and rainfall duration.

[0063] On the basis of the above embodiment, the rainfall intensity refers to the amount of rainfall per unit time, and the rainfall intensity reflects the rate of rainfall. Higher rainfall intensity may increase the instantaneous surface runoff, thereby triggering slope slip or deformation.

[0064] The rainfall amount refers to the cumulative rainfall amount in a time period, and the rainfall amount reflects the total water input in the time period, which is an important parameter for measuring the saturation degree and water pressure of the slope soil.

[0065] The rainfall duration refers to the length of continuous rainfall, and the rainfall duration affects the water saturation degree of the soil. Long-term rainfall may cause the strength of the soil to decrease, thereby inducing slope deformation or instability.

[0066] The above-mentioned rainfall characteristic parameters can be monitored by a rain gauge. The rain gauge can collect rainwater in real time, calculate the rainfall intensity according to the rainfall time interval, record the cumulative rainfall amount in a time period as the rainfall amount, and calculate the time difference between the start and end of rainfall when the rain gauge continuously records rainfall, which is the rainfall duration.

[0067] Among them, the rainfall intensity, rainfall amount and rainfall duration need to be standardized, and the subsequent combination of the rainfall characteristic parameters of each time period into a feature vector sample is also standardized data.

[0068] On the basis of the above embodiment, the current time period and the previous n historical time periods are divided as follows:

[0069] The current time period and the previous n historical time periods are divided as follows: The current time period and the previous n historical time periods are divided as follows: The set of n continuous time periods is denoted as , ​​​​Index of time period, , Index of current time period, interval range of the time period is , , , , , , ,

[0070] Based on three-dimensional absolute coordinates, the absolute displacement of each time period is calculated, and the formula is as follows:

[0071]

[0072] , , , , , , , , , , , , , , ,

[0073] Secondly integrate the three-axis acceleration data to calculate the relative displacement of each time period, and the formula is as follows:

[0074]

[0075] , , , , , , , , , , , , , , , the relative displacement of the time period;

[0076] On the basis of the above examples, in order to express the relationship between the cumulative displacement amount of each time period and the absolute displacement and the relative displacement, Table 1 is set.

[0077] Table 1. Change of cumulative displacement amount with absolute displacement and relative displacement

[0078]

[0079] According to Table 1, when the absolute displacement increases from 1.0 to 13.7, the cumulative displacement amount increases from 0.85 to 12.43, and the two are positively correlated, and the growth is relatively smooth, indicating that as the absolute displacement increases, the cumulative displacement amount also increases synchronously.

[0080] The relative displacement increases from 0.5 to 7.2, and also corresponds to the increase of the cumulative displacement amount, and the two are positively correlated.

[0081] According to Figs. 2-3 , the cumulative displacement amount increases linearly with the increase of the absolute displacement, and the fitting line of the absolute displacement and the cumulative displacement amount can well reflect the linear relationship between the two, indicating that the absolute displacement and the cumulative displacement amount are significantly positively correlated;

[0082] The cumulative displacement amount also increases linearly with the increase of the relative displacement, and the fitting line of the relative displacement and the cumulative displacement amount also reflects the linear relationship between the two, indicating that the relative displacement and the cumulative displacement amount also have obvious positive correlation.

[0083] On the basis of the above examples, the absolute displacement and the relative displacement are calculated by weighting, and the cumulative displacement amount of each time period is obtained, and the formula is as follows:

[0084]

[0085] , wherein, is the cumulative displacement amount of the i-th time period, and the cumulative displacement amount is used to combine the two index parameters of absolute displacement and relative displacement to evaluate the instability risk degree of the loess slope, and the greater the cumulative displacement amount, the higher the instability risk; On this basis, it needs to be explained that:

[0086] The absolute displacement refers to the overall position change amount of the soil body obtained by GNSS measurement, which reflects the overall trend deformation of the slope. When the absolute displacement increases, it indicates that the overall deformation of the loess slope intensifies, the stability of the soil body decreases, and the instability risk increases.

[0087]

[0088] ​The relative displacement is based on the local displacement calculated by the MEMS sensor, reflects the subtle deformation and movement inside the soil body, and when the relative displacement monitored by the MEMS increases, it shows that the internal structure of the soil body is active, indicating a potential risk of sliding, and the risk of instability increases;

[0089] Therefore, the absolute displacement and the relative displacement are positively correlated with the cumulative displacement.

[0090] In addition, the overall movement (absolute displacement) of the slope is a macroscopic deformation performance, which is an important indicator of the degradation of slope stability, and the micro movement (relative displacement) inside the slope reflects the stress adjustment, crack expansion or micro sliding of the local structure, which is often a precursor of instability. The absolute displacement and the relative displacement reflect the deformation behavior of the soil body under different scales and different physical mechanisms, respectively, and the two constitute the superposition of the overall deformation of the slope. In a small range of soil deformation, the deformation response is approximately linear, and the displacement can be regarded as a linear combination of different physical contributions.

[0091] Based on the above, the function relationship between the cumulative displacement and the absolute displacement and the relative displacement is expressed by the function form.

[0092] In the formula, is the weight coefficient of the absolute displacement, is the weight coefficient of the relative displacement.

[0093] The key to slope stability lies in whether the soil body has overall displacement or sliding. The absolute displacement, as the position change of the soil body under the ground reference system, directly reflects the macroscopic situation of the slope deformation, and is the most core indicator in risk assessment, so the weight is large.

[0094] During the integral calculation of speed and displacement, the MEMS accelerometer has problems such as cumulative error and zero drift, which leads to lower accuracy and stability of the relative displacement than the absolute displacement. The relative displacement reflects the local micro movement, which is easily affected by vibration, temperature change, electromagnetic interference, etc. The signal has more noise components, which may introduce misjudgment. In order to prevent the relative displacement noise from having too great an impact on the cumulative displacement, the weight coefficient is usually set to reduce its contribution to the final result.

[0095] Based on the above, . .

[0096] As an embodiment, the value range of is 0.5-1, and the value range of is 0-0.5. The specific value is set by the technical personnel according to the actual situation, which is not limited here.

[0097] S2. Combine the rainfall characteristic parameters of each historical time period into a feature vector sample, take all the feature vector samples as a training set, set an initial cluster number and a threshold of the number of feature vector samples of each cluster, iteratively optimize the cluster and cluster center point, confirm the number of samples of each cluster, merge or split the cluster when the number of samples does not meet the threshold, until the preset threshold is met, and finally determine the cluster and cluster center point;

[0098] On the basis of the above embodiment, the initial cluster number and the threshold of the number of feature vector samples of each cluster are set, the cluster and the cluster center point are iteratively optimized, the number of samples of each cluster is confirmed, the cluster is merged or split when the number of samples does not meet the threshold, until the preset threshold is met, and the final cluster and cluster center point are determined. The logic is as follows:

[0099] 1) Set the initial cluster number as , set the lower limit of the number of feature vector samples of each cluster as , and the upper limit as ;

[0100] Initialize the cluster center point to obtain an initial cluster center point set , is the kth initial cluster center point in the initial cluster center point set, is the index of the cluster, and each cluster center point is represented by a vector containing three feature values, i.e., rainfall intensity, rainfall amount, and rainfall duration;

[0101] 2) Perform the first iteration of clustering, calculate the distance between the feature vector samples of the historical time period and each initial cluster center point using the Euclidean distance formula, assign each feature vector sample to the nearest cluster center point, and form a set of clusters to be processed , is the kth cluster in the set , and is the number of clusters in the set ;

[0102] Then, count the number of feature vector samples in each cluster in the set of clusters to be processed;

[0103] If the number of feature vector samples in a cluster is less than , set a distance threshold, the distance threshold is the mean or median of the Euclidean distances between all cluster centers, calculate the distance between the cluster and other cluster center points in the same set, and select the clusters with a distance less than the distance threshold and a number of internal feature vector samples not less than as candidate clusters, merge the cluster into the candidate cluster with the smallest distance to the center point, and if there is no candidate cluster, merge the cluster into the cluster with the smallest distance to the center point in the same set;

[0104] If the number of feature vector samples in a cluster is greater than All feature vector samples in the cluster are extracted to form a data set to be split, and a K value clustering algorithm with a cluster number of 2 is used to cluster the data set, and the samples are divided into two mutually exclusive subsets according to the eigenvalues of the samples, and the two sub-clusters divided out are used to replace the original cluster;

[0105] The above merging and splitting process is repeated until the number of feature vector samples in each cluster is not less than and not greater than to obtain the final cluster set , calculate the distribution entropy value, and calculate the mean of each cluster in the set , update the cluster center point, wherein is the number of clusters in the set ;

[0106] 3) Based on the updated cluster center point, the clustering iteration is performed again to obtain the final cluster set after the second iteration and its distribution entropy value, and the mean of each cluster in the set is calculated, and the cluster center point is updated, wherein is the number of clusters in the set ;

[0107] 4) Calculate the absolute difference value of the distribution entropy value of the final cluster set in this iteration and the last iteration, and the formula is as follows:

[0108]

[0109] Wherein, is the absolute difference value of the distribution entropy value of the th clustering iteration and the th clustering iteration, is the distribution entropy value of the th clustering iteration, is the distribution entropy value of the th clustering iteration, is the number of feature vector samples of the th cluster in the th clustering iteration, is the number of clusters in the th clustering iteration, is the index of the clustering iteration, , is the total number of clustering iterations;

[0110] Compare the absolute difference value with the convergence threshold value:

[0111] If the absolute difference is not less than the convergence threshold, iteration is continued; if the absolute difference is less than the convergence threshold, clustering iteration is stopped, and a final clustering set is obtained , and a final clustering center point set , wherein, is the number of clusters in the set .

[0112] On the basis of the above, it should be noted that:

[0113] By setting the upper and lower threshold values of the number of clustering samples, and dynamically merging the clusters with too few samples or splitting the clusters with too many samples during the clustering iteration process, the automatic adjustment of the number of clusters is realized, which effectively avoids the over-subdivision or over-merging of clusters, and makes the sample size of each cluster evenly distributed, providing accurate clustering and center points for the subsequent clustering of the current time period and screening out the same historical time period as the clustering. Each iteration calculates the mean of the cluster as a new center point, which helps to quickly converge to a more reasonable cluster structure and overcome the accidental and local optimal problems caused by the initial selection of the center point in the traditional single K-means.

[0114] The convergence threshold is a small positive number set in advance, which is used to determine whether the absolute difference of the distribution entropy values of the current iteration and the previous iteration is less than the threshold, so as to determine whether the clustering iteration is terminated. By using the distribution entropy values of the clustering results and their difference between adjacent two iterations as the termination condition, the stability of the clustering structure and the change of the clustering distribution can be effectively quantified, and the premature stop or meaningless repeated iteration in the clustering iteration process can be avoided. The convergence threshold is preferably set in the range of , so as to balance the clustering accuracy and computational efficiency.

[0115] On the basis of the above embodiment, the distance between the feature vector of the current time period and each clustering center point is calculated, and the logic of its clustering is as follows:

[0116] The feature vector of the current time period , wherein, respectively correspond to the rainfall intensity, rainfall amount, and rainfall duration of the current time period;

[0117]

[0118] , wherein, is the distance between the feature vector of the current time period and the final th clustering center point, is the th feature value of the feature vector of the current time period, is the th feature value of the final th clustering center point, an index of an eigenvalue, respectively corresponding to rainfall intensity, rainfall amount, and rainfall duration;

[0119] On the basis of the above embodiment, the feature vector of the current time period is compared to the distance of each cluster center point, and it is classified into the cluster with the smallest distance.

[0120] S4. According to the cluster to which the current time period belongs, the same cluster of historical time periods is screened out, named as the same cluster historical time period, and the rainfall feature parameters and cumulative displacement of each same cluster historical time period and the previous m-1 time periods are spliced in time sequence to form a time sequence feature vector;

[0121] S5. The time sequence feature vector of each same cluster historical time period and the previous m-1 time periods is taken as input, and the cumulative displacement of the next time period of the same cluster historical time period is taken as label output, and a cumulative displacement prediction model is trained;

[0122] Relying on the dynamic clustering optimization mechanism, the time sequence feature vector used by the trained cumulative displacement prediction model is more representative.

[0123] On the basis of the above embodiment, the cumulative displacement prediction model is constituted by a deep learning network based on a multilayer perceptron, the deep neural network of the multilayer perceptron includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer, the first hidden layer, the second hidden layer, and the third hidden layer each have at least two neurons, and each uses ReLU (Rectified Linear Unit) as an activation function;

[0124] The collected time sequence feature vectors of the same cluster historical time period and the previous m-1 time periods are divided into a training set, a validation set, and a test set, and a common division ratio is 7:2:1, wherein the training set is used for learning model parameters; the validation set is used for adjusting hyperparameters in the training process to prevent overfitting; and the test set is used for evaluating the generalization ability of the model after training is completed.

[0125] In the cumulative displacement prediction model, the input features of the deep learning network of the multilayer perceptron include the time sequence feature vectors of each same cluster historical time period and the previous m-1 time periods;

[0126] The structure of the deep learning network of the multilayer perceptron is as follows:

[0127] Input layer: used for receiving the time sequence feature vectors of each same cluster historical time period and the previous m-1 time periods;

[0128] First hidden layer: having 128 neurons, using ReLU as an activation function;

[0129] ​The second hidden layer has 64 neurons, and also uses the ReLU activation function;

[0130] The third hidden layer has 32 neurons, and uses the ReLU activation function;

[0131] The output layer has 1 neuron, which is used to output the accumulated displacement of the next time period after the same type of historical time period.

[0132] The process of training the accumulated displacement prediction model is as follows:

[0133] The time series feature vector of each same type of historical time period and the previous m-1 time periods is input, and the accumulated displacement of the next time period after the same type of historical time period is output as a label, and the accumulated displacement prediction model is trained, and the mean square error is used as the loss function. When the mean square error is in the range of , the training of the accumulated displacement prediction model is stopped.

[0134] S6. The time series feature vector of the current time period and the previous m-1 time periods is input into the model, the accumulated displacement of the next time period is predicted, and the deformation rate of the next time period is calculated, and the accumulated displacement is combined to predict the loess slope instability risk grade.

[0135] On the basis of the above embodiment, the deformation rate of the next time period is calculated, and the formula is as follows:

[0136]

[0137] wherein, is the deformation rate of the next time period, is the accumulated displacement of the next time period, is the accumulated displacement of the current time period;

[0138] On this basis, it should be noted that:

[0139] The greater the accumulated displacement, the higher the energy accumulation and deformation degree inside the slope, and the lower the strength and stability of the soil body, and the development of the slip surface is accelerated, and eventually the critical condition of landslide may be reached;

[0140] The higher the deformation rate, the faster the release speed of the shear stress of the slope, and the deformation process gradually loses control, and the system approaches the instability boundary from the stable state.

[0141] When both are small, i.e. the accumulated displacement is small, and the slip surface has not been penetrated; the deformation rate is small, and the system is still in a relatively stable state.

[0142] When both are large, i.e. the accumulated displacement is large, and the slip surface has developed significantly, and the critical condition of landslide is close; the deformation rate is large, and the deformation of the slope is accelerated, and the system approaches the instability critical state.

[0143] The logic for predicting the loess slope instability risk level in combination with the deformation rate and the cumulative displacement amount of the next time period is as follows:

[0144] If the cumulative displacement amount of the next time period is less than the first threshold value of the cumulative displacement amount, and the deformation rate of the next time period is less than the first threshold value of the deformation rate, it is determined that the loess slope instability risk is low. If the cumulative displacement amount of the next time period is greater than the first threshold value of the cumulative displacement amount, and the deformation rate of the next time period is less than the first threshold value of the deformation rate, it is determined that the loess slope instability risk is medium.

[0145] If the cumulative displacement amount of the next time period is less than the first threshold value of the cumulative displacement amount, and the deformation rate of the next time period is greater than the first threshold value of the deformation rate, it is determined that the loess slope instability risk is high.

[0146] wherein the first threshold value and the second threshold value of the cumulative displacement amount of the next time period are respectively a first adjustment coefficient and a standard deviation of the cumulative displacement amount, and the first threshold value and the second threshold value of the deformation rate of the next time period are respectively a third adjustment coefficient and a standard deviation of the deformation rate.

[0147] wherein the first threshold value and the second threshold value of the cumulative displacement amount of the next time period are respectively a first adjustment coefficient and a standard deviation of the cumulative displacement amount, and the first threshold value and the second threshold value of the deformation rate of the next time period are respectively a third adjustment coefficient and a standard deviation of the deformation rate.

[0148] wherein the first threshold value of the cumulative displacement amount of the next time period is used to divide the low risk and the medium risk, and the second threshold value of the cumulative displacement amount of the next time period is used to divide the medium risk and the high risk.

[0149] The first threshold value of the cumulative displacement amount is usually a product of the first adjustment coefficient and the standard deviation of the cumulative displacement amount, plus the mean value of the cumulative displacement amount, and the second threshold value of the cumulative displacement amount is usually a product of the second adjustment coefficient and the standard deviation of the cumulative displacement amount, plus the mean value of the cumulative displacement amount. wherein the mean value and the standard deviation of the cumulative displacement amount are obtained through the cumulative displacement amount of the historical time period, the second adjustment coefficient is greater than the first adjustment coefficient, the second adjustment coefficient is set between 2.0 and 3.0, and the first adjustment coefficient is set between 1.0 and 2.0.

[0150] The adjustment coefficient multiplied by the standard deviation plus the mean value forms the threshold value, which represents the risk dividing line given based on the data distribution in statistics, wherein the second adjustment coefficient is greater than the first adjustment coefficient, which means that the second threshold value is higher than the first threshold value, thereby forming a strict division of the low, medium and high risk levels. Similarly, the fourth adjustment coefficient is greater than the third adjustment coefficient, which is used to set the risk level boundary of the deformation rate.

[0151]

[0152] ​​​​​​​​​​​​​The first threshold of the deformation rate is usually a product of the third adjustment coefficient and the standard deviation of the deformation rate, plus the mean value of the deformation rate, and the second threshold of the deformation rate is usually a product of the fourth adjustment coefficient and the standard deviation of the deformation rate, plus the mean value of the deformation rate. The first threshold of the deformation rate is usually a product of the third adjustment coefficient and the standard deviation of the deformation rate, plus the mean value of the deformation rate, and the second threshold of the deformation rate is usually a product of the fourth adjustment coefficient and the standard deviation of the deformation rate, plus the mean value of the deformation rate.

[0153] The mean value and the standard deviation of the deformation rate are obtained through the deformation rate of a historical time period, the fourth adjustment coefficient is greater than the third adjustment coefficient, the fourth adjustment coefficient is set between 2.0 and 3.0, and the third adjustment coefficient is set between 1.0 and 2.0.

[0154] The above formulas are all dimensionless numerical calculations, the formulas are obtained by collecting a large amount of data to simulate the latest real situation, and the preset parameters in the formulas are set by a person skilled in the art according to actual conditions.

[0155] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0156] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0157] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered within the protection scope of the present application.

Claims

1. A data processing method for loess slope deformation monitoring based on MEMS inertial sensor data and GNSS data, characterized in that, The specific steps include: S1. Collect three-dimensional absolute coordinates and three-axis acceleration data of the monitoring points in the current time period and the previous n historical time periods using a GNSS receiver and a three-axis accelerometer based on MEMS technology, calculate absolute displacement and relative displacement, and generate cumulative displacement, while collecting rainfall characteristic parameters in each time period; S2. Combine the rainfall characteristic parameters of each historical time period into a feature vector sample, use all the feature vector samples as a training set, set an initial cluster number and a threshold for the number of feature vector samples in each cluster, iteratively optimize the clustering and cluster center points, and confirm the number of samples in each cluster. If the number of samples does not meet the threshold, merge or split the clusters until the preset threshold is met, and determine the final clusters and cluster center points; S3. Calculate the distance between the feature vector of the current time period and the final cluster center points to determine its cluster; S4. According to the cluster to which the current time period belongs, filter out the historical time periods with the same cluster, which are named as similar historical time periods. Combine the rainfall characteristic parameters and cumulative displacement of each similar historical time period and its previous m-1 time periods in chronological order to form a time series feature vector; S5. Take the time series feature vector of each similar historical time period and its previous m-1 time periods as input, and take the cumulative displacement of the next time period of the similar historical time period as label output to train the cumulative displacement prediction model; S6. Input the time series feature vector of the current time period and the previous m-1 time periods into the model to predict the cumulative displacement of the next time period, calculate the deformation rate of the next time period, and combine the cumulative displacement to predict the loess slope instability risk level.

2. The method according to claim 1, wherein, The rainfall characteristic parameters include rainfall intensity, rainfall amount, and rainfall duration.

3. The method according to claim 2, wherein, Current time period and previous The logic for dividing historical periods is as follows: a current time period and a previous history time period are grouped into consecutive time periods, and a length of each time period is , a set of consecutive time periods is denoted as , , an index of a time period, , , an index of a current time period, an interval range of a time period is , wherein , a start time of a time period, and an end time of a time period, , an end time of a time period, and a start time of a time period; Based on three-dimensional absolute coordinates, the absolute displacement of each time period is calculated according to the following formula: in, For the first Absolute displacement over a time period , , The monitoring points are respectively at the 1st The end time of each time period axis, axis, Axis coordinates The monitoring points are respectively at the 1st The start time of each time period axis, axis, Axis coordinates; The relative displacement of each time period is calculated by twice integrating the three-axis acceleration data according to the following formula: in, For the first Any time within a time period instantaneous speed, For the first The velocity at the start time within a time period, For the first Any time within a time period Instantaneous acceleration, For the first The time variable from the start time to the end time of each time period. For the first The time period from start time to time The time variable, For the first Relative displacement over a time period; The cumulative displacement of each time period is obtained by weighted calculation of the absolute displacement and the relative displacement according to the following formula: wherein, is the cumulative displacement amount for the first time period, is the weight coefficient of the absolute displacement, is the weight coefficient of the relative displacement, and is the cumulative displacement amount for the second time period.

4. The method according to claim 3, wherein, The logic of setting the initial cluster number and the threshold for the number of feature vector samples in each cluster, iteratively optimizing the clustering and cluster center points, and confirming the number of samples in each cluster is as follows: 1) Set the initial number of clusters as , set the lower limit of the number of feature vector samples of each cluster as , and the upper limit as ; Initializing the cluster center points to obtain an initial cluster center point set , is the kth initial cluster center point in the initial cluster center point set, is the index of the cluster, and each cluster center point is represented by a vector containing three feature values, namely rainfall intensity, rainfall amount, and rainfall duration. 2) perform the first clustering iteration, calculate the distance between the feature vector samples of the historical time period and each initial cluster center point, assign each feature vector sample to the nearest cluster center point to form a set of to-be-processed clusters , is the kth cluster in the set , is the number of clusters in the set ; Then, the number of feature vector samples in each cluster in the to-be-processed cluster set is counted; If the number of feature vector samples in a cluster is less than , a distance threshold is set, the distance between the cluster center and other cluster centers in the same set is calculated, and clusters with a distance less than the distance threshold and a number of feature vector samples not less than are screened out as candidate clusters. The cluster is merged into the candidate cluster with the smallest distance to the cluster center, or into the cluster with the smallest distance to the cluster center in the same set if there is no candidate cluster. If the number of feature vector samples in a cluster is greater than the K-value clustering method is used to cluster the feature vector samples in the cluster to split the cluster into two clusters. The above merging and splitting process is repeated until the number of feature vector samples in each cluster is not less than and not greater than to obtain a final cluster set , the distribution entropy value thereof is calculated, the mean of each cluster in the set is calculated, and the cluster center point is updated, wherein is the number of clusters in the set . 3) perform clustering iteration again based on the updated cluster center points to get the final clustering set after the 2nd iteration and the distribution entropy value, and calculate the mean value of each cluster in the set , update the cluster center points, wherein, is the number of clusters in the set ​ 4) Calculate the absolute difference between the distribution entropy value of the final cluster set in the current iteration and that in the previous iteration, and compare the absolute difference with the convergence threshold: If the absolute difference is not less than the convergence threshold, iteration is continued; if the absolute difference is less than the convergence threshold, clustering iteration is stopped, and a final clustering set is obtained , and a final clustering center point set , wherein, is the number of clusters in the set .

5. The method according to claim 4, wherein, The logic of calculating the distance between the feature vector of the current time period and each cluster center point to obtain its cluster is as follows: a feature vector of the current time period wherein, corresponding to the rainfall intensity, the rainfall amount, and the rainfall duration of the current time period, respectively wherein, is the distance of the feature vector of the current time period to the final th cluster center point, is the th feature value of the feature vector of the current time period, is the th feature value of the final th cluster center point, is the index of the feature value, , respectively correspond to the rainfall intensity, the rainfall amount, and the rainfall duration. Compare the distance between the feature vector of the current time period and each cluster center point, and classify it into the cluster with the smallest distance.

6. The method according to claim 5, wherein, The formula for calculating the deformation rate of the next time period is as follows: wherein, is a deformation rate for the next time period, is a cumulative displacement amount for the next time period, is a cumulative displacement amount for the current time period; The logic for predicting the loess slope instability risk level based on the deformation rate and cumulative displacement of the next time period is as follows: If the next time period accumulates displacement amount , and the deformation rate , then it is determined that the loess slope is low in instability risk; If the next time period accumulates displacement amount , and the deformation rate , then it is determined that the loess slope instability risk is medium; If the next time period accumulates displacement amount , or deformation rate , then determine that the loess slope is high risk of instability; wherein, , are respectively a first threshold value, a second threshold value of the cumulative displacement amount of the next time period, and , , are respectively a first threshold value, a second threshold value of the deformation rate of the next time period, and .

Images