CN-121808289-B - High slope deformation prediction system and method based on frequency domain analysis and deep learning

CN121808289BCN 121808289 BCN121808289 BCN 121808289BCN-121808289-B

Abstract

The invention provides a high slope deformation prediction system and method based on frequency domain analysis and deep learning, which relate to the technical field of slope deformation prediction and comprise a data acquisition module, a time sequence processing module, a frequency domain analysis module, a deformation prediction module and a deformation distribution module; the method comprises the steps of arranging monitoring points at equal intervals by a data acquisition module, dividing a basic time window, calculating deformation difference values of the monitoring points relative to a reference monitoring point, splicing multi-source time sequence data by a time sequence processing module to generate a plurality of cross overlapped sliding windows, outputting an effective frequency band component and a residual error component by a frequency domain analysis module, outputting deformation prediction values of large deformation categories by a deformation prediction module by using a deep learning model, distributing deformation total amount by a deformation distribution module, and selecting a maximum value as an integral deformation prediction result. The method solves the problems of single data dimension, prediction lag and the like in the prior art, and realizes dual and accurate output of single-point deformation predicted values of all monitoring points and high slope integral deformation predicted values.

Inventors

ZHAO JIE
WANG QIANG
GAO YAN

Assignees

大连大学

Dates

Publication Date: 20260508
Application Date: 20260309

Claims (8)

1. The utility model provides a high slope deformation prediction system based on frequency domain analysis and degree of depth study which characterized in that includes: The data acquisition module is used for arranging monitoring points on the surface of the high slope at equal intervals along the slope inclination direction, defining a monitoring interval and dividing a basic time window, taking the lowest monitoring point of the slope foot as a reference monitoring point, acquiring slope deformation and environment data of other monitoring points in each basic time window, and calculating the deformation difference value of each monitoring point relative to the reference monitoring point; the time sequence processing module is used for splicing the slope deformation, the environment data and the deformation difference values of the rest monitoring points into multi-source time sequence data, and sliding and intercepting the multi-source time sequence data according to preset sliding window parameters to generate a plurality of cross overlapped sliding windows; The frequency domain analysis module is used for clustering the multi-source time sequence data in each sliding window to obtain small, medium and large deformation categories, eliminating the small, medium deformation category data, carrying out frequency domain analysis only after carrying out data mean value operation in the large deformation category, outputting corresponding frequency band components and residual error components, and simultaneously carrying out time sequence-frequency domain feature matching screening on the frequency band components of the large deformation category to obtain effective frequency band components; The deformation prediction module is used for inputting the effective frequency band component and the residual component of the large deformation class corresponding to each sliding window into the trained deep learning model, and predicting the deformation predicted value of the large deformation class in the next basic time window; And the deformation distribution module is used for summing the slope deformation predicted values of the large deformation category under the latest sliding window to obtain a deformation total predicted value, distributing the deformation total according to the deformation difference value between each monitoring point in the large deformation category and the reference monitoring point to obtain the deformation predicted value of the next basic time window of each monitoring point, and selecting the maximum value as the high slope overall deformation predicted result.
2. The high slope deformation prediction system based on frequency domain analysis and deep learning according to claim 1, wherein the environmental data comprises rainfall, slope temperature and soil humidity, and the frequency band components comprise a low frequency component, a medium frequency component and a high frequency component.
3. The high slope deformation prediction system based on frequency domain analysis and deep learning according to claim 1, wherein the specific logic is as follows: and taking the current slope deformation monitoring moment as an end point, tracing forward for a preset time length to form a deformation monitoring interval, and uniformly dividing the deformation monitoring interval into a plurality of continuous non-overlapping basic time windows.
4. The high slope deformation prediction system based on frequency domain analysis and deep learning according to claim 1, wherein the deformation difference of each monitoring point relative to a reference monitoring point is calculated, and the specific logic is as follows: setting a reference monitoring point as The rest monitoring points are Wherein Is the first The other monitoring points are used for monitoring the position of the monitoring points, , The total number of the rest monitoring points; First, the The rest monitoring points In the first place The deformation in each basic time window is Reference monitoring point In the first place The deformation in each basic time window is Then (1) The rest monitoring points Relative reference monitoring point In the first place Deflection difference for each base time window The method comprises the following steps: Wherein, the As an index to the basic time window, , Is the total number of base time windows.
5. The high slope deformation prediction system based on frequency domain analysis and deep learning according to claim 4, wherein the sliding interception is performed according to preset sliding window parameters to generate a plurality of sliding windows which are overlapped in a crossing way, and the specific logic is as follows: the sliding window parameters comprise sliding window length and sliding step length; Wherein the sliding window length is The sliding step length is , And Are all positive integers, and ; Co-inclusion of multi-source time series data A base time window, and satisfies ; According to the sliding step length Slide from left to back in turn, cut length to Form a plurality of sliding windows which are overlapped in a crossing way and are marked as ; Wherein the method comprises the steps of Is the first A plurality of sliding windows are arranged on the inner surface of the frame, For the index of the sliding window, , For the total number of sliding windows, the following is satisfied: Wherein, the Is rounded downwards; First, the Sliding windows The intercepted basic time window index range is And the index range completely falls into the basic time window index interval And (3) inner part.
6. The high slope deformation prediction system based on frequency domain analysis and deep learning according to claim 5, wherein the effective frequency band component is obtained by performing time sequence-frequency domain feature matching screening on the large deformation class frequency band component, and the specific logic is as follows: Extracting all frequency band components corresponding to the large deformation category in each sliding window, and a deformation difference time sequence of each monitoring point of the large deformation category in each sliding window according to the basic time window index sequence; For the time sequence of the frequency band component and the deformation difference value in a sliding window, respectively calculating the variation of the frequency band component and the deformation difference value in the corresponding adjacent basic time window, wherein the variation of the frequency band component is as follows: Wherein, the For the amount of change in the frequency band components within a sliding window over the adjacent base time window, In the first frequency band The value of the individual base time window is, In the first frequency band A value of a base time window; The amount of change in the time series sequence of the deformation difference values is as follows: Wherein, the Is the variation of the time sequence of the variation difference in a sliding window in the adjacent basic time window, Is the time sequence of the deformation difference value The value of the individual base time window is, Is the time sequence of the deformation difference value A value of a base time window; Counting the signs of the change slopes of the two, and calculating the ratio of the number of slopes with consistent signs to the number of total slopes, wherein the ratio is the corresponding trend consistency coefficient, and the number of total slopes is ; And taking the average value of all the trend consistency coefficients as an adaptive threshold value, and screening out frequency band components with the trend consistency coefficients larger than the threshold value, namely the effective frequency band components.
7. The high slope deformation prediction system based on frequency domain analysis and deep learning according to claim 5, wherein the total deformation amount is distributed according to the deformation difference value between each monitoring point and the reference monitoring point in the large deformation category, so as to obtain the deformation amount prediction value of the next basic time window of each monitoring point, and the specific logic is as follows: Setting the predicted value of the total deformation amount corresponding to the large deformation category under the latest sliding window as All monitoring points in the large deformation category are at the first The sum of the deformation differences of the basic time windows is ; Calculate the first The monitoring points are at Distribution ratio of each basic time window The specific formula is as follows: Calculate the first The deformation forecast value of the next basic time window of each monitoring point comprises the following specific formula: Wherein, the Is the first Deformation prediction values of next basic time windows of the monitoring points.
8. A high slope deformation prediction method based on frequency domain analysis and deep learning, the method being performed by the high slope deformation prediction system based on frequency domain analysis and deep learning according to any one of claims 1 to 7, characterized by comprising the specific steps of: s1, arranging monitoring points on the surface of a high slope at equal intervals along the slope inclination direction, defining a monitoring interval, dividing a basic time window, taking the lowest monitoring point of a slope foot as a reference monitoring point, collecting slope deformation and environment data of other monitoring points in each basic time window, and calculating the deformation difference value of each monitoring point relative to the reference monitoring point; S2, splicing the slope deformation, the environmental data and the deformation difference values of the rest monitoring points into multi-source time sequence data, and performing sliding interception according to preset sliding window parameters to generate a plurality of sliding windows which are overlapped in a crossing mode; s3, clustering monitoring points on multi-source time sequence data in each sliding window to obtain small, medium and large deformation categories, eliminating the small, medium deformation category data, carrying out frequency domain analysis only after data average value operation in the large deformation category, outputting corresponding frequency band components and residual error components, and simultaneously carrying out time sequence-frequency domain feature matching screening on the frequency band components of the large deformation category to obtain effective frequency band components; S4, inputting the effective frequency band component and the residual component of the large deformation class corresponding to each sliding window into a trained deep learning model, and predicting the deformation predicted value of the large deformation class in the next basic time window; S5, summing the slope deformation predicted values of the large deformation category under the latest sliding window to obtain a deformation total predicted value, distributing the deformation total according to the deformation difference value of each monitoring point in the large deformation category and the reference monitoring point to obtain the deformation predicted value of the next basic time window of each monitoring point, and selecting the maximum value as a high slope overall deformation predicted result.

Description

High slope deformation prediction system and method based on frequency domain analysis and deep learning Technical Field The invention relates to the technical field of high slope deformation prediction, in particular to a high slope deformation prediction system and method based on frequency domain analysis and deep learning. Background Along with the rapid development of intelligent monitoring technology and deep learning algorithm, more and more technical schemes are applied to the field of high slope deformation prediction, and aims to solve the problems of high labor cost, low data acquisition frequency, early warning lag and other pain points in the traditional manual monitoring, improve the accuracy and timeliness of deformation prediction and promote the high slope safety management to be upgraded from passive response to active prevention. At present, a plurality of deformation prediction methods based on data processing and deep learning are disclosed in the prior art, wherein the engineering structure deformation prediction method based on clustering-deep learning provided by the publication No. CN114881074A is one of more typical technical schemes. The prior art (CN 114881074A) specifically comprises the following steps of S1, collecting deformation time sequence data of different monitoring points in each monitoring project in an actual project in real time, S2, preprocessing the collected deformation time sequence data, including abnormal point detection, data resampling and monitoring point clustering, S3, carrying out EMD (empirical mode decomposition) on the preprocessed data to obtain a plurality of IMF components and a residual component, S4, respectively inputting each IMF component and the residual component into a corresponding prediction model for training, obtaining corresponding future prediction values through the trained prediction model, and summing the future prediction values corresponding to each component to obtain the deformation prediction value of the final monitoring project. The method has the advantages of good prediction effect, capability of effectively aiming at various different monitoring deformation data in actual engineering, reference value for actual engineering application, accurate prediction, improvement of prediction hysteresis, reduction of calculated amount and the like. However, the method has the defects that the prior art time sequence processing layer does not integrate multi-source data, does not adopt sliding window interception, is difficult to capture the real-time sequence characteristics of the high slope, is easy to generate prediction lag, cannot accurately track the dynamic deformation process of the high slope, and the characteristic processing layer adopts EMD decomposition, does not perform cluster screening on monitoring points to focus on deformation key areas, so that core deformation cannot be captured. The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art. Disclosure of Invention The invention aims to provide a high slope deformation prediction system and method based on frequency domain analysis and deep learning, so as to solve the problems in the background technology. In order to achieve the above purpose, the present invention provides the following technical solutions: A high slope deformation prediction system based on frequency domain analysis and deep learning, comprising: The data acquisition module is used for arranging monitoring points on the surface of the high slope at equal intervals along the slope inclination direction, defining a monitoring interval and dividing a basic time window, taking the lowest monitoring point of the slope foot as a reference monitoring point, acquiring slope deformation and environment data of other monitoring points in each basic time window, and calculating the deformation difference value of each monitoring point relative to the reference monitoring point; the time sequence processing module is used for splicing the slope deformation, the environment data and the deformation difference values of the rest monitoring points into multi-source time sequence data, and sliding and intercepting the multi-source time sequence data according to preset sliding window parameters to generate a plurality of cross overlapped sliding windows; The frequency domain analysis module is used for clustering the multi-source time sequence data in each sliding window to obtain small, medium and large deformation categories, eliminating the small, medium deformation category data, carrying out frequency domain analysis only after carrying out data mean value operation in the large deformation category, outputting corresponding frequency band components and residual error components, and simultaneou