CN-121998907-A - Method for extracting mass three-dimensional deformation monitoring data of dam and high slope based on expert knowledge

CN121998907ACN 121998907 ACN121998907 ACN 121998907ACN-121998907-A

Abstract

The embodiment of the invention discloses an extraction method of mass three-dimensional deformation monitoring data of a dam and a high slope based on expert knowledge, which comprises the steps of sampling three-dimensional deformation data of the dam and the high Bianba according to a certain time interval, dividing refined time periods with different characteristics, respectively carrying out noise reduction, measuring deformation changes of different time sections by different indexes, calculating a polynomial weighted comprehensive abnormal score, judging a deformation mode, refining subcategories for the abnormal deformation mode and the stable deformation mode, and simplifying the selection of a representative value of a daily deformation according to the subcategories. The method is characterized in that the method comprises the steps of determining the deformation mode of the vehicle, and carrying out the customized noise reduction aiming at the characteristics of different deformation stages through the segmented noise reduction driven by expert rules, and introducing the isolated forest anomaly detection and comprehensive anomaly scoring method to greatly improve the automation level and the accuracy of the deformation mode determination and ensure the wide applicability of the vehicle under different climates, terrains and hydrologic environments.

Inventors

ZHANG DONGLIN
LIAN YANYING
TANG HAN
ZHANG LING
JING ZHOU
ZHANG ZONGYING
LI LONGLONG
Zhang Shenghuang
Lin Xingcheng
YE LEI
LONG YI
SU BOWEN
SUN LUYI
LU YUXIN

Assignees

华电福新周宁抽水蓄能有限公司
南京师范大学

Dates

Publication Date: 20260508
Application Date: 20251223

Claims (9)

1. The method for extracting mass three-dimensional deformation monitoring data of the dam and the high slope based on expert knowledge is characterized by comprising the following steps of: Sampling three-dimensional deformation data of the dam and the height Bianba according to a certain time interval to obtain a three-dimensional deformation data sequence X # ....A.) dividing refined time periods with different characteristics according to priori knowledge and combining geographical environments of the dam, and respectively carrying out noise reduction on three-dimensional deformation data of the time periods; According to the three-dimensional deformation data of the segments after filtering and denoising, aiming at priori knowledge of different time periods, different indexes are adopted to measure deformation changes of different segments, a mutant type abnormal point in the three-dimensional deformation data is identified by combining an isolated forest method, the indexes and the abnormal scores are respectively combined with weight values, a polynomial weighted comprehensive abnormal score is calculated, and the deformation mode is judged by utilizing the comprehensive abnormal score, so that the daily deformation mode process of a dam and a high slope is judged; and refining the subcategories for the abnormal deformation mode and the stable deformation mode, and simplifying and selecting the representative value of the daily deformation according to the subcategories.
2. The extraction method according to claim 1, characterized in that: sampling three-dimensional deformation data of a dam and a height Bianba according to a certain time interval, and dividing the refinement time periods with different characteristics according to priori knowledge and the geographic environment of the dam, wherein the refinement time periods comprise the following specific steps: According to the prior knowledge of three-dimensional deformation monitoring of the dam and the side slope, the data of one day is divided into three typical stages according to the trend and the periodical change of the deformation, namely a night stable stage X_stable, a water level lifting stage X_ wlevel and a temperature sensitive stage X_ tsense, wherein the night stable stage X_stable is a time period with relatively stable temperature and water level, small overall fluctuation and no obvious change trend, the water level lifting stage X_ wlevel is a time period with obvious trend change when the water level of the dam gradually rises or falls under the physical action of tides and the like, and the temperature sensitive stage X_ tsense is a time period with severe daily temperature change and obvious periodicity.
3. The extraction method according to claim 2, characterized in that: the step of respectively denoising the three-dimensional deformation data of the time period comprises the following steps: For three-dimensional deformation data of each night stabilization period, moving average filtering is adopted, wherein in the moving average filtering, the window size is odd The window unit is the deformation monitoring and collecting interval Respectively supplementing two ends of sequence data of three-dimensional deformation data Points, for each three-dimensional deformation data Calculate its front and back co-ordinates Arithmetic mean of points and output smoothed As shown in formula (1): Formula (1); three-dimensional deformation data for each water level elevation period Filtering with Savitzky-Golay filter, and taking the size of filter window The order of the polynomial fit is scaled to cover the typical time scale of the water level change Constructing a matrix Each row corresponds to the position of a point in the filter window, the first Line 1 Is arranged as Calculating a filter coefficient vector Wherein Symmetrically filling m points at two ends of the three-dimensional deformation data sequence, and carrying out three-dimensional deformation data Linear combination is carried out to obtain As shown in formula (2): formula (2) For three-dimensional deformation data X_ tsense of each temperature sensitive period, a frequency domain window is defined by adopting Fourier low-pass filtering Is provided with At the position of The rest is set to zero, low-frequency filtering is reserved, and three-dimensional deformation data in the segment is processed Performing discrete Fourier transform to obtain Structure of Removing unnecessary frequency components, performing inverse discrete Fourier transform, and taking the real part as the filtering result 。
4. The extraction method according to claim 3, characterized in that: According to the three-dimensional deformation data of the segments after filtering and denoising, different indexes are adopted to measure deformation changes of different segments according to priori knowledge of different time periods, and the method specifically comprises the following steps: For sampling intervals Observation data after filtering and noise reduction Wherein Constructing a time vector Perform decentering , ; For three-dimensional deformation data after denoising at night stabilization period By means of fluctuation standard deviation As an index to reflect the fluctuation strength in a stable state, the data average value in the night stable period is Standard deviation of fluctuation Expressed as: Formula (3) For three-dimensional deformation data after denoising in water level lifting period Using linear trend slope And goodness of fit Reflecting trend intensity and trend significance, linear trend slope Expressed as: , Formula (4) Goodness of fit Represented as Formula (5) Wherein the sum of the squares is Regression sum of squares The predicted value is The intercept is expressed as ; For three-dimensional deformation data after denoising in temperature sensitive stage water level lifting stage Performing discrete Fourier transform using the de-centered receipt; ; N represents the data quantity in the temperature sensitive period, which is calculated by dividing the time of the temperature sensitive period by the sampling time interval, wherein the sampling frequency Calculating corresponding period of temperature sensitive period Frequency point of (2) Pass frequency point Calculating cycle amplitude And spectral significance Reflecting the amplitude of the main period and the degree of prominence of the period relative to other frequencies, Amplitude of cycle Expressed as: , Spectral significance Expressed as: , ; the method for identifying the mutant abnormal points in the three-dimensional deformation data by combining the isolated forest method specifically comprises the following steps: For the deformation sequence of the filtered noise reduction treatment Given the number of trees in a forest Sub-sample size for each tree Maximum tree depth Referring to the average height of the binary search tree, a normalization constant is defined: , ; Wherein the method comprises the steps of For Euler constant, for each Calculating anomaly score Wherein Indicating that Is easy to be isolated, has high possibility of abnormality, Is difficult to be isolated and tends to be normal, Using average anomaly scores To represent the anomaly index of the overall data.
5. The extraction method according to claim 4, wherein: the method comprises the steps of calculating polynomial weighted comprehensive abnormal scores by combining the indexes and the abnormal scores respectively, and judging a deformation mode by using the comprehensive abnormal scores, wherein the method specifically comprises the following steps: Calculating the comprehensive abnormal score of the daily deformation data by using a formula (6), Formula (6) Wherein, the To normalize the score, weight Is defined empirically by an expert and sets a threshold for the composite anomaly score when When the threshold value is more than or equal to the threshold value, the deformation is in an abnormal deformation mode, when And when the deformation is smaller than the threshold value, the deformation belongs to a stable deformation mode.
6. The extraction method according to claim 5, characterized in that: The sub-categories of the abnormal deformation mode and the stable deformation mode are refined, wherein the abnormal deformation mode is divided into a trend abnormal type, a period abnormal type and a mutation abnormal type, and the stable deformation mode is divided into an absolute stable type, a trend stable type and a period stable type.
7. The extraction method according to claim 6, characterized in that: the representative value of the daily deformation is simplified and selected according to the subcategory, which is specifically as follows: For the trend anomaly, an extremum method of the trend direction is adopted, if the trend is positive, namely, the trend is upward, and the representative value takes the maximum value of all days: if the trend is negative, namely the descending trend, the representative value takes the minimum value of all days: Wherein Representing a deformation monitoring data set of a whole day after segmented noise reduction; for the period anomaly type, a maximum period amplitude method is adopted, and a period peak-valley range maximum point pair (maximum amplitude) is determined according to Fourier analysis: Or selecting the maximum absolute value of the amplitude, wherein The peak value representing the most typical period is, A valley representing the most typical period; For the mutation anomaly, an isolated forest anomaly score extremum method is adopted, and the absolute value maximum point with the highest anomaly score is selected from anomaly data points detected by the isolated forest; For absolute stability, a steady median method is adopted, after abnormal points are detected in isolated forests are removed, Wherein Representing a normal data set remaining after filtering out the outlier data points; for trend stationary type, a trend fitting robust method is adopted, all-day data are linearly fitted first, and the midpoint (i.e. the middle moment) on a trend fitting line is calculated: Wherein For the middle time of the day, And Fitting the slope and intercept of the line for trend; For cyclostationary, a typical cycle amplitude center method is used to determine the main cycle amplitude using fourier analysis: I.e. the middle point of the peak-to-valley of the most typical period of the day, where The peak value representing the most typical period is, The valley representing the most typical period.
8. The extraction method according to any one of claims 1 to 7, characterized in that: The three-dimensional deformation data are millimeter wave radar data, laser radar (LiDAR) scanning data, continuous GNSS monitoring data, three-dimensional photogrammetry reconstruction data, ground synthetic aperture radar (GB-SAR) monitoring data or three-dimensional deformation data acquired by an intelligent sensor network (IoT device).
9. An electronic device, comprising: the method for extracting mass three-dimensional deformation monitoring data of dams and highways based on expert knowledge according to any one of claims 1-8 comprises a processor and a memory, wherein the processor and the memory are connected through a bus, the memory is suitable for storing instructions or programs executable by the processor, and the processor executes the instructions stored by the memory.

Description

Method for extracting mass three-dimensional deformation monitoring data of dam and high slope based on expert knowledge Technical Field The invention relates to the field of deformation monitoring and measurement, in particular to an extraction method of mass three-dimensional deformation monitoring data of a dam and a high slope based on expert knowledge, which can be widely applied to long-term safety monitoring, dangerous case early warning and health assessment of important infrastructures such as ultra-high dams, deep-excavation slopes and the like and provides key technical support for intelligent operation and maintenance of the infrastructures. Background The dam is used as a core infrastructure for water resource management and flood control power generation, and the structural health of the dam is directly related to the safety and regional economic development of millions of downstream population. The traditional deformation monitoring mainly relies on point contact means such as total stations and leveling measures, and has the bottlenecks of low monitoring frequency, limited space coverage, incapability of working in severe weather and the like. In recent years, synthetic aperture radar interferometry technology is becoming an important tool for monitoring deformation of a dam due to the advantages of all-weather, high-precision and planar monitoring. By continuously acquiring deformation time sequence data with millimeter-level precision, the InSAR not only can capture dynamic response of the dam under the multi-factor coupling effects of temperature, water level, geology and the like, but also can provide key data support for structural safety early warning. However, facing massive and high-dimensional radar monitoring data, how to realize the span from data accumulation to knowledge discovery is still a scientific difficulty to be broken through in the engineering safety field. The method for extracting and analyzing deformation data of the dam and the high slope commonly seen in the prior art comprises the following steps: 1. short period analysis, capturing burst deformation in real time Short period analysis (hours/days) is suitable for emergency monitoring scenarios in construction period, extreme weather (e.g., storms/floods), seismic events, etc. According to the method, sudden deformation can be captured in real time through sliding window statistics (such as average value per hour) and mutation point detection (such as CUSUM algorithm), and timely data support is provided for emergency response. Sliding window statistics rapidly identifies abnormal fluctuations in data by calculating the mean and standard deviation over a short window of time. The mutation point detection algorithm (such as accumulation and control chart CUSUM) accurately detects the deformed mutation point by monitoring the accumulation change of the data. The advantage of short-period analysis is its fast response capability, enabling detection of abnormal changes in deformation in a short time, thus supporting timely decisions and actions. For example, during a storm or flood, the method can monitor deformation conditions of the dam in real time, discover potential safety hazards in time, and provide scientific basis for emergency management and decision making. 2. Mid-cycle analysis, separation of environmental factor effects The mid-cycle analysis (cycle/month level) is applicable to the case of seasonal water level changes and periodic loads (e.g. impoundment/flood discharge). Referring to fig. 1, the influence of environmental factors can be effectively separated and long-term trends of deformation can be clearly identified by extracting periodic components and STL decomposition (Seasonal-trend decomposition) through fourier transform. Fourier transforms help identify deformations caused by quaternary water level changes or other periodic loading by converting time series data to the frequency domain, extracting the main periodic component. STL decomposition then decomposes the time series into seasonal, trending and residual components, further separating the effects of environmental factors. The method has the advantages that short-term fluctuation can be filtered, periodic rules of deformation are focused, and important basis is provided for understanding the behaviors of the dam under different environmental conditions. For example, during a quaternary water level change, mid-cycle analysis may identify deformation patterns caused by the water level change, helping to assess the long term stability of the dam. Long-period analysis to evaluate full-lifecycle performance degradation Long-period analysis (years/years) is mainly aimed at slow processes such as material aging, foundation settlement and the like. Referring to fig. 2, performance degradation of the dam over the life cycle can be comprehensively assessed by linear regression fitting the long term trend and Mann-Kendall test trend significance. Li