Slope early warning method and system based on optical fiber sensor

Abstract

The invention discloses a side slope early warning method and system based on an optical fiber sensor, and relates to the technical field of side slope monitoring, and the method comprises the steps: collecting side slope deformation monitoring data and environment monitoring data, and carrying out the preprocessing of the collected deformation monitoring data; a causal association rule between the environment monitoring data and slope deformation is constructed, and the deformation space distribution range of the whole slope is predicted based on a BP neural network; and taking the slope deformation monitoring sequence as a sample, training an RNN neural network, predicting the development trend of slope deformation, carrying out early warning analysis on the overall stability of the slope, and generating an early warning result through an alarm device. According to the method, the causal association rule is constructed, the environmental monitoring data and the deformation data are combined, the space prediction precision is improved, the RNN neural network is used for training the time sequence data, and the development trend of slope deformation can be effectively predicted; and in combination with the spatial distribution data, a future deformation trend analysis and early warning result is generated, so that the timeliness and accuracy of early warning can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of slope monitoring, and in particular to a slope early warning method and system based on optical fiber sensors. Background Art

[0002] Slope disasters are one of the major natural disasters that threaten the safety of people's lives and property and engineering construction. With the rapid development of the economy and society, engineering construction activities are becoming more and more frequent, and the impact of human engineering activities on slope stability is becoming increasingly greater. In addition, with the frequent occurrence of extreme weather events, the frequency and intensity of slope disasters are increasing. In order to effectively prevent and reduce the losses caused by slope disasters, it is urgent to carry out slope stability assessment and disaster early warning research.

[0003] Traditional slope stability assessment and early warning mainly rely on geological surveys, on-site investigations and regular inspections. These methods have problems such as insufficient monitoring data and low accuracy of early warning analysis, and are difficult to meet the current needs of slope disaster prevention and control. In recent years, with the rapid development of emerging technologies such as the Internet of Things, big data, and artificial intelligence, new means and methods have been provided for slope disaster monitoring and early warning. In particular, new monitoring technologies represented by fiber optic sensing technology have the advantages of distributed, long-distance, high-precision, and anti-interference, and show broad application prospects in slope disaster monitoring and early warning. Existing slope early warning methods mainly rely on single deformation monitoring data, resulting in low accuracy of slope early warning analysis. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is: the existing slope monitoring and early warning methods have insufficient monitoring data, low early warning analysis accuracy, and cannot fully reflect the complexity of slope deformation, as well as how to improve the accuracy and reliability of slope early warning analysis.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a slope early warning method based on optical fiber sensors, comprising collecting slope deformation monitoring data and environmental monitoring data, and preprocessing the collected deformation monitoring data; constructing causal association rules between environmental monitoring data and slope deformation, and predicting the deformation spatial distribution range of the entire slope based on a BP neural network; using the slope deformation monitoring sequence as a sample, training an RNN neural network, predicting the development trend of slope deformation, performing early warning analysis on the overall stability of the slope, and generating early warning results through an alarm device.

[0007] As a preferred solution of the slope early warning method based on optical fiber sensors described in the present invention, the preprocessing includes collecting slope deformation monitoring sequences using optical fiber sensors arranged on the slope. and environmental monitoring data; the slope deformation monitoring sequence includes the strain and displacement of each measuring point, and the environmental monitoring data includes rainfall and groundwater level; the collected deformation monitoring data is preprocessed, and the preprocessing includes filtering and missing value interpolation to obtain the preprocessed slope deformation monitoring sequence ; Filter the collected deformation monitoring data, including filtering the collected slope deformation monitoring sequence , establish slope deformation monitoring sequence Statistical model , statistical model It is expressed as: , in, Indicates at time The water level data collected, Indicates at time The collected temperature data, Indicates time The rainfall data collected For time, Indicates the collection time. is the constant term representing the slope deformation, represents the influence coefficient of temperature on slope deformation, , , , represents the polynomial order, represents the influence coefficient of rainfall on slope deformation, , , , , represents the corresponding coefficient, For time The polynomial transformation value of and slope deformation monitoring sequence , calculate the residual sequence , expressed as: , For the residual sequence Perform the gross error analysis based on the Rytter criterion, eliminate the gross errors, and obtain a new deformation monitoring sequence ;use Update Original Repeat the gross error analysis of the residual sequence until no data are eliminated by two consecutive L'Eyette criterion gross error analyses, and obtain the sequence after gross error elimination ; Using the sequence after gross error elimination , establish a Kalman filter model including state equations and observation equations; use the established Kalman filter model to predict the prior estimate of the state at time k based on the optimal estimate of the state at time k-1; use the observation value at time k to update the posterior estimate of the state at time k through Kalman filtering; repeatedly update the posterior estimate of the state at time k to obtain the optimal estimate sequence of slope deformation state at each time and the deformation monitoring sequence after random error filtering .

[0008] As a preferred solution of the slope early warning method based on optical fiber sensor described in the present invention, the preprocessing also includes interpolating missing values ​​of the collected deformation monitoring data, and filtering the deformation monitoring sequence. Extract the subsequence of measurement points containing missing values ​​as the reference sequence for spatial interpolation, and extract the subsequence of measurement points containing missing values ​​from the deformation monitoring sequence after filtering. Extract the subsequence of measurement points that do not contain missing values ​​as the comparison sequence of spatial interpolation, calculate the grey correlation between the reference sequence of spatial interpolation and the comparison sequence, obtain the comparison sequence whose correlation with the reference sequence is greater than the threshold, establish a spatial grey correlation model using the obtained comparison sequence and the reference sequence, and calculate the missing values ​​in the reference sequence using the established spatial grey correlation model as the missing values ​​of spatial interpolation; from the deformation monitoring sequence after filtering Extract the subsequence of measurement points containing missing values ​​as the reference sequence for time interpolation, perform a stationarity test on the reference sequence for time interpolation, set the model order range according to the autocorrelation and partial correlation function graphs, and use the Akaike information criterion AIC to obtain the optimal order from the model order range and , according to the optimal order and , establish an autoregressive moving average model for the reference series of time interpolation , using the least squares method to solve the autoregressive moving average model The parameters of the autoregressive moving average model are set using the parameters , calculate the missing values ​​of the reference sequence of time interpolation as the missing values ​​of time interpolation; according to the missing values ​​of spatial interpolation and time interpolation, use the weighted average algorithm to get the final missing values, and use the final missing values ​​to filter the deformation monitoring sequence Perform missing value interpolation to obtain the preprocessed slope deformation monitoring sequence .

[0009] As a preferred solution of the slope early warning method based on optical fiber sensor described in the present invention, the causal association rule includes the following: , construct the causal association rules between environmental monitoring data and slope deformation, take the collected environmental monitoring data as input, use K-means clustering algorithm to divide the environmental monitoring data of each measuring point into spatial regions, and obtain the partition results of environmental monitoring measuring points. Take the partition results of environmental monitoring measuring points as input, use projection clustering algorithm to extract features of environmental monitoring data in each partition, construct projection function and projection index function, use genetic algorithm to solve the optimal projection direction, obtain the projection clustering feature value of the measuring points in each partition, and use the projection clustering feature value of the measuring points in each partition and the corresponding slope deformation monitoring sequence As input, the FP-Growth association rule algorithm is used to construct the causal association rules between environmental monitoring data and slope deformation.

[0010] As a preferred solution of the slope early warning method based on optical fiber sensor described in the present invention, the prediction of the deformation spatial distribution range of the entire slope includes collecting environmental monitoring data and preprocessing slope deformation monitoring sequence. Time synchronization is performed to synchronize the environmental monitoring data and slope deformation monitoring sequence after time synchronization. As input, with the deformation of each measuring point as output, construct the training samples of BP neural network, use the constructed training samples, adopt the error back propagation algorithm to train the BP neural network, establish the nonlinear mapping relationship among environmental monitoring data, measuring point coordinates and slope deformation, use the trained BP neural network, with the measuring point coordinates at any position and the environmental monitoring data at the corresponding time as input, calculate the deformation of the corresponding measuring point coordinates, calculate the deformation of each measuring point coordinate in the slope, and obtain the deformation prediction value of each measuring point coordinate. According to the deformation prediction value of each measuring point coordinate, adopt the spatial interpolation method to generate the deformation spatial distribution range of the entire slope.

[0011] As a preferred solution of the slope early warning method based on optical fiber sensor described in the present invention, the prediction of the development trend of slope deformation includes: The input and output sample pairs of the RNN neural network are constructed by sliding time window. The input samples are the deformation monitoring values ​​at the current moment and the previous moment, and the output samples are the deformation monitoring prediction values ​​at the next moment. The constructed input and output sample pairs are used to train the RNN neural network using the error back propagation algorithm to obtain the deformation time series of each measuring point within the spatial distribution range of the slope deformation. According to the extracted deformation time series, the trained RNN neural network is used to recursively predict the deformation of each position in the future period of time. According to the deformation of each position, the development trend of the slope deformation in the future period of time is predicted.

[0012] As a preferred solution of the slope early warning method based on optical fiber sensor described in the present invention, the early warning analysis includes calculating the average deformation of each measuring point according to the deformation prediction value of each measuring point within the spatial distribution range of the slope deformation, as the rigid body displacement of the entire slope, taking the slope volume before deformation as a reference, using the Simpson method to calculate the slope volume change after deformation as the deformation of the entire slope, and evaluating the stability of the entire slope based on the predicted deformation of each measuring point in a future period of time and the deformation of the entire slope, comparing the obtained causal association rules with the evaluated stability of the entire slope, correcting the evaluation result of the stability of the entire slope based on the comparison result, and generating an early warning analysis result through an alarm device.

[0013] Another object of the present invention is to provide a slope warning system based on optical fiber sensors, which can predict the deformation spatial distribution range of the entire slope by constructing causal association rules between environmental monitoring data and slope deformation, thereby solving the problem of low reliability of current slope monitoring and warning technologies.

[0014] As a preferred solution of the slope early warning system based on optical fiber sensors described in the present invention, it includes: a data processing module, a prediction module, and an early warning module; the data processing module is used to collect slope deformation monitoring data and environmental monitoring data, and pre-process the collected deformation monitoring data; the prediction module is used to construct a causal association rule between environmental monitoring data and slope deformation, and predict the deformation spatial distribution range of the entire slope; the early warning module is used to train the RNN neural network with the slope deformation monitoring sequence as a sample, predict the development trend of slope deformation, perform early warning analysis on the overall stability of the slope, and generate early warning results through an alarm device.

[0015] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a slope early warning method based on an optical fiber sensor.

[0016] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a slope early warning method based on an optical fiber sensor.

[0017] Beneficial effects of the present invention: The fiber optic sensor-based slope early warning method provided by the present invention effectively improves data quality and integrity through data preprocessing, removes noise through filtering, and improves data consistency and integrity through missing value interpolation. In particular, through the application of methods such as Kalman filtering and grey correlation model, the noise and missing values ​​of monitoring data can be accurately processed to ensure the accuracy of subsequent analysis; through the construction of causal association rules, combined with environmental monitoring data and deformation data, the relationship between environmental factors (such as rainfall, temperature) and slope deformation is revealed; based on the training of BP neural network, the deformation amount of slope measuring points can be accurately calculated and the spatial distribution range of deformation can be predicted, thereby improving the accuracy of spatial prediction; using RNN neural network to train time series data can effectively predict the development trend of slope deformation; combined with spatial distribution data, future deformation trend analysis and early warning results are generated, which can improve the timeliness and accuracy of early warning and prevent the occurrence of potential disasters. The present invention achieves better results in terms of integrity, reliability and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0019] Figure 1 An overall flow chart of a slope warning method based on optical fiber sensors provided in the first embodiment of the present invention.

[0020] Figure 2 A diagram of Kalman filtering results of a slope warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0021] Figure 3 A flow chart of a monitoring feature projection clustering algorithm based on genetic algorithm optimization for a slope warning method based on an optical fiber sensor provided in the second embodiment of the present invention.

[0022] Figure 4 A flow chart of slope association rule mining based on FP-Growth algorithm for a slope early warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0023] Figure 5 A schematic diagram of the geographical location of a slope in a slope early warning method based on optical fiber sensors provided in a second embodiment of the present invention.

[0024] Figure 6A comparison diagram of measurement point partitioning results of a slope early warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0025] Figure 7 A curve diagram of the original measured values ​​of the seepage pressure measuring point 43# borehole of a slope early warning method based on an optical fiber sensor provided in the second embodiment of the present invention.

[0026] Figure 8 A BP neural network algorithm flow chart of a slope early warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0027] Fig. 9 A rainfall process line diagram of a slope early warning method based on an optical fiber sensor provided in the second embodiment of the present invention.

[0028] Fig.10 A process line diagram of the upstream reservoir water level of a slope early warning method based on an optical fiber sensor provided in the second embodiment of the present invention.

[0029] Fig.11 A diagram of measured values ​​and RNN predicted values ​​of a slope warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0030] Fig.12 A process line diagram of target slope volume change of a slope early warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0031] Fig.13 A vertical deformation process line diagram of the tpn1 measuring point of a slope early warning method based on an optical fiber sensor provided in the second embodiment of the present invention.

[0032] Fig.14 A process line diagram of measuring point displacement and slope volume change of a slope early warning method based on optical fiber sensors provided in the second embodiment of the present invention.

[0033] Fig.15 An overall flow chart of a slope warning system based on optical fiber sensors provided in the third embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0035] Example 1, reference Figure 1, as an embodiment of the present invention, provides a slope early warning method based on optical fiber sensor, comprising: S1: Collect slope deformation monitoring data and environmental monitoring data, and pre-process the collected deformation monitoring data.

[0036] Furthermore, the preprocessing includes collecting slope deformation monitoring sequences using fiber optic sensors deployed on the slope and environmental monitoring data; the slope deformation monitoring sequence includes the strain and displacement of each measuring point, and the environmental monitoring data includes rainfall and groundwater level; the collected deformation monitoring data is preprocessed, and the preprocessing includes filtering and missing value interpolation to obtain the preprocessed slope deformation monitoring sequence ; Filter the collected deformation monitoring data, including filtering the collected slope deformation monitoring sequence , establish slope deformation monitoring sequence Statistical model , statistical model It is expressed as: , in, Indicates at time The water level data collected, Indicates at time The collected temperature data, Indicates time The rainfall data collected For time, Indicates the collection time. is the constant term representing the slope deformation, represents the influence coefficient of temperature on slope deformation, , , , represents the polynomial order, represents the influence coefficient of rainfall on slope deformation, , , , , represents the corresponding coefficient, For time The polynomial transformation value of and slope deformation monitoring sequence , calculate the residual sequence , expressed as: , For the residual sequence Perform the gross error analysis based on the Rytter criterion, eliminate the gross errors, and obtain a new deformation monitoring sequence ;use Update Original Repeat the gross error analysis of the residual sequence until no data are eliminated by two consecutive L'Eyette criterion gross error analyses, and obtain the sequence after gross error elimination ; Using the sequence after gross error elimination , establish a Kalman filter model including state equations and observation equations; use the established Kalman filter model to predict the prior estimate of the state at time k based on the optimal estimate of the state at time k-1; use the observation value at time k to update the posterior estimate of the state at time k through Kalman filtering; repeatedly update the posterior estimate of the state at time k to obtain the optimal estimate sequence of slope deformation state at each time and the deformation monitoring sequence after random error filtering .

[0037] It should be noted that the preprocessing also includes interpolation of missing values ​​in the collected deformation monitoring data, and the deformation monitoring sequence after filtering. Extract the subsequence of measurement points containing missing values ​​as the reference sequence for spatial interpolation, and extract the subsequence of measurement points containing missing values ​​from the deformation monitoring sequence after filtering. Extract the subsequence of measurement points that do not contain missing values ​​as the comparison sequence of spatial interpolation, calculate the grey correlation between the reference sequence of spatial interpolation and the comparison sequence, obtain the comparison sequence whose correlation with the reference sequence is greater than the threshold, establish a spatial grey correlation model using the obtained comparison sequence and the reference sequence, and calculate the missing values ​​in the reference sequence using the established spatial grey correlation model as the missing values ​​of spatial interpolation; from the deformation monitoring sequence after filtering Extract the subsequence of measurement points containing missing values ​​as the reference sequence for time interpolation, perform a stationarity test on the reference sequence for time interpolation, set the model order range according to the autocorrelation and partial correlation function graphs, and use the Akaike information criterion AIC to obtain the optimal order from the model order range and , according to the optimal order and , establish an autoregressive moving average model for the reference series of time interpolation , using the least squares method to solve the autoregressive moving average model The parameters of the autoregressive moving average model are set using the parameters , calculate the missing values ​​of the reference sequence of time interpolation as the missing values ​​of time interpolation; according to the missing values ​​of spatial interpolation and time interpolation, use the weighted average algorithm to get the final missing values, and use the final missing values ​​to filter the deformation monitoring sequence Perform missing value interpolation to obtain the preprocessed slope deformation monitoring sequence .

[0038] It should also be noted that the slope deformation monitoring sequence refers to a series of data reflecting the deformation state of the slope that are continuously collected by monitoring equipment (such as optical fiber sensors) deployed on the slope. It is a time series representation of the slope deformation monitoring data, which records the deformation of the slope at different times. The slope deformation monitoring sequence usually contains a timestamp, which records the collection time of each monitoring data and reflects the dynamic change process of the slope deformation; the measuring point number, which identifies the monitoring points at different locations and is used to describe the spatial distribution characteristics of the slope deformation; the deformation amount, which records the deformation value of each measuring point at different times, such as displacement, strain, etc., and quantifies the degree of slope deformation; the general representation of the slope deformation monitoring sequence is: , in, Indicates Monitoring time, Indicates the measurement point number. Indicates the deformation at the corresponding time and measuring point.

[0039] It should also be noted that the Laiyite Criterion is a commonly used method for identifying and eliminating gross errors, which is used to detect and delete outliers from observed data. The Laiyite Criterion is based on the principles of probability theory and mathematical statistics, assuming that the observed data obeys a normal distribution. By calculating the degree of deviation between the observed value and the average value, outliers exceeding a certain threshold are identified. A gross error refers to an outlier in the observed data that obviously deviates from the normal range. Gross errors are usually caused by factors such as failure of the observation instrument, human error, and environmental interference, which will have an adverse effect on data analysis and modeling, resulting in deviation and distortion of the results. In our invention, a statistical model of the slope deformation monitoring sequence is established, the residual sequence is calculated, and the residual sequence is analyzed and eliminated using the Laiyite Criterion to obtain a deformation monitoring sequence after gross error elimination.

[0040] It should also be noted that the Grey Relational Grade is an indicator used in the grey system theory to describe the degree of correlation between different sequences. By comparing the geometric shape and change trend of the sequence, the mathematical expression of the Grey Relational Grade is: , in, Reference sequence Compare with sequence The grey correlation degree of is the sequence length, For the The grey correlation coefficient of two sequences at a certain moment.

[0041] It should also be noted that the specific steps of extracting the subsequences of measuring points and other subsequences of measuring points include traversing the deformation monitoring sequence after filtering, and for each subsequence of measuring points, determining whether it contains missing values; if the subsequence of measuring points contains missing values, using it as a reference sequence for spatial interpolation; if the subsequence of measuring points does not contain missing values, using it as a comparison sequence for spatial interpolation; repeating until all subsequences of measuring points are traversed to obtain a reference sequence set and a comparison sequence set; after extracting the reference sequence and the comparison sequence, the comparison sequence with the strongest correlation with the reference sequence is screened out by calculating the grey correlation degree, and used to establish a spatial grey correlation model; extracting other subsequences of measuring points that do not contain missing values ​​from the deformation monitoring sequence after filtering as a comparison sequence for spatial interpolation; data organization includes grouping the deformation monitoring sequence after filtering according to measuring points, and the data of each measuring point constitutes a subsequence; assuming that there are measuring points, the deformation monitoring sequence can be expressed as: , in, Indicates The subsequence of each measuring point contains the deformation monitoring value of the measuring point at different times. The missing value judgment includes the subsequence of each measuring point. , to determine whether it contains missing values, you can do so in the following ways: set a missing value marker, such as a null value (None) or a special character (such as -99999), and mark the missing monitoring value as the special value during the data collection and storage phase; use the program to traverse the subsequence For each element of , determine whether it is equal to the missing value mark. If there is a missing value mark, it is considered that the subsequence contains missing values. Subsequence extraction includes extracting the measurement point subsequence that does not contain missing values ​​according to the result of missing value judgment as a comparison sequence for spatial interpolation. The extraction result can be expressed as: , in, Indicates a subsequence of test points that does not contain missing values, is the number of measurement points without missing values, ; Data storage includes storing the extracted comparison sequence according to a certain data structure to facilitate subsequent grey relational calculation and spatial interpolation analysis. Commonly used data structures include arrays, lists, data frames, etc.

[0042] It should also be noted that outliers in monitoring data refer to data that are obviously inconsistent with the laws of normal monitoring values. Based on the statistical characteristic values ​​of the data, the Leyte criterion can be used to identify and eliminate outliers. In general, the random error of the measurement data has the normal distribution characteristics of zero expectation and fixed variance. According to the principle of small probability events, the absolute value of the random error exceeds The situation is a very unlikely event; therefore, is the threshold value to judge abnormal values; if the monitoring value Then remove it, then retain; and Respectively represent the mean and mean square error of the monitoring data, expressed as: , , Taking slope deformation monitoring data as an example, it may be affected by environmental factors such as rainfall and temperature. The slopes of reservoir banks or river banks may also be affected by reservoir water levels or river water levels, and sometimes there is a certain trend of change.

[0043] It should also be noted that the Kalman filter method is mainly based on the measured value at the current moment and the optimal estimated value at the previous moment, and according to the recursive formula, the estimated value at the next moment is obtained; the slope deformation monitoring data is a time series collected at a certain time interval, and the Kalman filter model of the discretized system can be used for error analysis; the Kalman filter model is expressed by two parts: the state equation and the observation equation, and its discrete form is: , , in, and They are Moment and The filtered value of the state vector at time instant, , is the control vector (input vector), and denote the dynamic noise vector and the observation noise vector respectively, is the state vector coefficient matrix, is the dynamic noise coefficient matrix, , is the control vector coefficient matrix, is the observation vector, is the observation matrix; when the system does not consider deterministic input, the state equation and observation equation are: , , make is the dynamic noise covariance matrix, is the observation noise covariance matrix, then The state estimate at time is expressed as: , in, and Respectively Moment and The prior state estimate at the time, that is, the value of the previous time time( The best estimate prediction at time time( The covariance matrix estimate is expressed as: , The state update equation is: , , , in, express The posterior state estimate at time t is The best estimate of the time; and Respectively The prior estimated covariance and the posterior estimated covariance of the moment; is the filter gain matrix; is a unit vector.

[0044] It should also be noted that the absolute correlation has uniqueness and symmetry; the uniqueness is reflected in that when the sample size changes, the correlation coefficient remains unchanged; the symmetry means that when two data sequences are used as reference sequences for calculation, the correlation coefficient and correlation degree obtained are the same, so the present invention uses the absolute correlation for analysis; for discrete data, the absolute correlation reflects the closeness of the slopes of the two data sequence curves at each corresponding point. If the difference in the slopes of the two curves at each corresponding point is small, it indicates that the correlation between the two is large; otherwise, the correlation is small; for the reference sequence after initialization and compare sequences , expressed as: , , in, System behavior sequence In the The value of the moment, is the length of the time series, For the Comparison sequence In the The value of the moment, To compare the number of sequences, the difference quotient slope is calculated as: , , in, is the system behavior sequence at adjacent moments and The slope of the difference between For the The comparison sequence at adjacent times and The difference between the slopes is used to calculate the correlation coefficient. : , Calculate the correlation : , Grey correlation interpolation, determine the reference sequence and several comparison sequences, and perform initialization on them respectively; calculate the correlation between the reference sequence and each comparison sequence respectively ; According to the calculated correlation, arrange them in ascending order, , select the comparison sequence with the greatest correlation with the reference sequence , and its correlation is , the initial value image of the comparison sequence and the initial value image of the reference sequence are: , , in, Indicates Normalized comparison sequence of moments The value of To indicate the Normalized system behavior sequence at each moment The value of The data in the comparison sequence The correlation coefficient of each data is approximately taken as the overall correlation, expressed as: , Depend on , we can find the first The initial value of the value , then the first Value .

[0045] It should also be noted that the slope deformation time series model is based on the deformation monitoring data series. The mathematical model established by the above method and parameter estimation can be used to analyze the long-term development trend of monitoring data and can also be used to interpolate missing data. is a time series that has been tested for stationarity and conforms to normality and zero mean. The value of is not only related to the value of the first n items of the sequence, but also affected by the noise of the first m steps of the sequence. The ARMA conventional model is as follows: , in, is the autoregressive parameter, is the sliding average parameter, is a white noise sequence with a mean of 0 and a variance of The normal distribution of Model; the autocorrelation coefficient and partial autocorrelation coefficient of this model have tailing characteristics, that is, they decay at an exponential rate; when , the model becomes: , This model is called The order autoregressive model is denoted as The autocorrelation coefficient ACF of this model has obvious tailing, that is, it does not drop to 0 after a certain step, but decays at an exponential rate, and the partial autocorrelation coefficient PACF has The order of truncation, that is, After the step, it is 0; when , the model becomes: , This model is called The moving average model is denoted as ; The autocorrelation coefficient ACF of this model has The order of truncation, that is, The partial autocorrelation coefficient has a tailing property, that is, it does not drop to 0 after a certain step, but decays at an exponential rate. When the autoregressive model (AR) is used to interpolate monitoring data, the parameters of the AR model can be estimated by the least squares method. When determining the model order, first determine the ACF function graph and the PACF function graph. , Then the AIC criterion (Akaike Information Criterion) is used to determine the best , Take the value with the smallest AIC value , As the best fitting model order; the slope deformation monitoring data series has temporal and spatial correlation, so establishing a combined model that reflects the temporal and spatial correlation of deformation measurements is helpful to improve the interpolation accuracy of missing data; the interpolation formula of the combined interpolation method is expressed as: , in, is the grey relational interpolation value of the measuring point, Impute values ​​for time series models, , is the weight coefficient, which can be determined by the equal weight method and the deviation weight method; the present invention adopts the equal weight method, preferably taking .

[0046] S2: Construct causal association rules between environmental monitoring data and slope deformation, and predict the spatial distribution range of deformation of the entire slope.

[0047] Furthermore, the causal association rules include the collected environmental monitoring data and slope deformation monitoring sequence , construct the causal association rules between environmental monitoring data and slope deformation, take the collected environmental monitoring data as input, use K-means clustering algorithm to divide the environmental monitoring data of each measuring point into spatial regions, and obtain the partition results of environmental monitoring measuring points. Take the partition results of environmental monitoring measuring points as input, use projection clustering algorithm to extract features of environmental monitoring data in each partition, construct projection function and projection index function, use genetic algorithm to solve the optimal projection direction, obtain the projection clustering feature value of the measuring points in each partition, and use the projection clustering feature value of the measuring points in each partition and the corresponding slope deformation monitoring sequence As input, the FP-Growth association rule algorithm is used to construct the causal association rules between environmental monitoring data and slope deformation.

[0048] It should be noted that the prediction of the deformation spatial distribution range of the entire slope includes the collected environmental monitoring data and the pre-processed slope deformation monitoring sequence. Time synchronization is performed to synchronize the environmental monitoring data and slope deformation monitoring sequence after time synchronization. As input, with the deformation of each measuring point as output, construct the training samples of BP neural network, use the constructed training samples, adopt the error back propagation algorithm to train the BP neural network, establish the nonlinear mapping relationship among environmental monitoring data, measuring point coordinates and slope deformation, use the trained BP neural network, with the measuring point coordinates at any position and the environmental monitoring data at the corresponding time as input, calculate the deformation of the corresponding measuring point coordinates, calculate the deformation of each measuring point coordinate in the slope, and obtain the deformation prediction value of each measuring point coordinate. According to the deformation prediction value of each measuring point coordinate, adopt the spatial interpolation method to generate the deformation spatial distribution range of the entire slope.

[0049] It should also be noted that the steps of applying the K-means algorithm to monitoring data mining are as follows: initializing the cluster center, In the sample, randomly determine Cluster Center ; Preliminarily divide the data set, and obtain each data object and Cluster Center Distance , and divide each data object into the closest cluster, expressed as: , in, is a data object in the data set T, where and Both dimensional array, the cluster center is regenerated as: , in, It is Cluster centers, , yes The number of data objects in the corresponding cluster; algorithm convergence judgment, using the square error criterion to measure the number of data objects in the cluster Similarity: , in, yes Corresponding to the data object in the class cluster, yes The average value of all data objects in the corresponding cluster, yes The sum of square errors of all data objects in the corresponding cluster; the cluster center is updated repeatedly until the cluster center no longer changes, the loop ends and the clustering result is output.

[0050] Considering the characteristics of K-means algorithm for partitioning clusters, which is simple and easy to implement and has good effects, the K-means algorithm is used to partition the slope measurement points. The key lies in flexibly constructing the distance measurement index of the measurement points and comprehensively considering the similarity between different measurement points. Among them, the spatial distance index can be used to measure the distance between different measurement points, and the attribute distance index can be used to measure the attribute difference between different measurement points. At the same time, the deformation of the slope is closely related to its stability. Deformation, as an external direct reflection of the dynamic evolution of the internal stability state of the slope, can capture physical information closely related to the stability of the slope. Therefore, the slope displacement measurement value is used here to calculate the attribute distance, and the horizontal and vertical spatial coordinates of the measurement points are used to calculate the spatial distance. The calculation formulas of different distance indicators are as follows: , , in, is the attribute distance index of the measuring point, is the spatial distance index of the measuring points, , , Represents the horizontal and vertical spatial coordinates of the measuring point. Characterization Measurement point The displacement measurement at a moment, Characterization Measurement point The displacement measurement at the moment, where , ; Comprehensively consider the attribute distance and spatial distance, and weightedly determine the comprehensive distance index of the measuring point: , in, is the comprehensive distance index of the measuring point. and are the weights of the attribute distance and spatial distance indicators, respectively, satisfying , usually taken On the basis of constructing the comprehensive distance index of the measuring points, the spatial area division of the displacement monitoring points is realized based on the principle of K-means algorithm.

[0051] Based on the measurement point partitioning results of the K-means clustering algorithm, the projection clustering algorithm is used to further extract data features and compress the data level; the projection clustering algorithm includes dimensionless data processing, and the number of measurement points is The time series length is Monitoring data set Perform dimensionless processing to eliminate the dimension differences between input data as much as possible, and convert the processed input data into It indicates that, , , Refers to The first values; in order to fully extract the characteristics of the monitoring data set, the data set is described by taking the annual mean and annual extreme values ​​as two statistical features to characterize the numerical size, extreme value distribution and measured value changes of the data set; therefore, the monitoring data set Convert to mean set and extreme value sets Then the dimensionless processing is performed to facilitate subsequent calculations.

[0052] It should also be noted that in order to realize the projection of high-dimensional data in linear space, it is necessary to construct a projection function, where Dimensional data The projection direction is , is the projection cluster sequence value in the projection direction, and there exists ; Construct projection index. The projection index function helps to select the optimal projection direction and is also the key to mapping high-dimensional data to low-dimensional space; define the projection index function for: , , , in, The projection point The standard deviation of The projection point The local density of The projection point The average value of is the distance between samples, ; is a unit step function, when When , its value is 0, otherwise its value is 1; is the window radius of the local data space density.

[0053] Optimize projection index function, projection direction Determines the projection index function value The size of , so the selection of the optimal projection direction can be transformed into the problem of solving the maximum value of the projection index function, expressed as: , , , in, is the projection vector The comprehensive cluster analysis of the components Substitute to get the projection feature result , .

[0054] The key to projection clustering is the selection of the optimal projection direction; the genetic algorithm is introduced to optimize the calculation process of projection clustering; the genetic algorithm needs to determine the objective function, and the algorithm objective function writing: , The fitness function is an evaluation criterion for the quality of individuals in a population. It requires the algorithm structure to be simple and the calculation result to be non-negative in order to reduce the complexity of the algorithm as much as possible. Here, the fitness function is set according to the objective function. , thus calculating the fitness value of each individual in the population; in addition, rewrite the constraint function of the genetic algorithm as follows: .

[0055] S3: Using the slope deformation monitoring sequence as a sample, train the RNN neural network, predict the development trend of slope deformation, conduct early warning analysis on the overall stability of the slope, and generate early warning results through the alarm device.

[0056] Furthermore, the prediction of slope deformation development trend includes the pre-processed slope deformation monitoring sequence The input and output sample pairs of the RNN neural network are constructed by sliding time window. The input samples are the deformation monitoring values ​​at the current moment and the previous moment, and the output samples are the deformation monitoring prediction values ​​at the next moment. The constructed input and output sample pairs are used to train the RNN neural network using the error back propagation algorithm to obtain the deformation time series of each measuring point within the spatial distribution range of the slope deformation. According to the extracted deformation time series, the trained RNN neural network is used to recursively predict the deformation of each position in the future period of time. According to the deformation of each position, the development trend of the slope deformation in the future period of time is predicted.

[0057] It should be noted that the early warning analysis includes calculating the average deformation of each measuring point based on the deformation prediction value of each measuring point within the spatial distribution range of the slope deformation, as the rigid body displacement of the entire slope, taking the slope volume before deformation as a benchmark, and using the Simpson method to calculate the slope volume change after deformation as the deformation of the entire slope, and evaluating the stability of the entire slope based on the predicted deformation of each measuring point in a future period of time and the deformation of the entire slope, comparing the causal association rules obtained with the evaluated stability of the entire slope, and correcting the evaluation result of the stability of the entire slope based on the comparison result, and generating an early warning analysis result through an alarm device.

[0058] Example 2, reference Figure 2-Figure 14, which is an embodiment of the present invention, provides a slope early warning method based on optical fiber sensors. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0059] Seven horizontal displacement observation points are set up on a slope for periodic repetitive observations with a frequency of once a day. The 20-period monitoring data of point A are processed by Kalman filtering. To verify the denoising effect of the Kalman filtering method, noise with a mean of 0 and a variance of 6.3 is added to the 4th, 8th, 12th, and 15th period data at random. The data after adding noise are shown in Table 1.

[0060] Table 1 Noise data table , There are many factors that affect slope deformation. Here, the state parameters only consider horizontal displacement and deformation rate, so the observation system is one-dimensional and the dynamic system of the deformed body is two-dimensional. The Kalman filter model is established, and the initial state vector is taken as follows: , , the state transfer matrix , control matrix , dynamic noise and observation noise Take respectively , , Matlab programming is used to calculate and filter the 20-period data, and the filtered values ​​are shown in Table 2. The true values, observed values ​​and filtered values ​​are plotted as Figure 2 shown.

[0061] Table 2 Kalman filter value table , Depend on Figure 2 It can be seen that the Kalman filter method can effectively filter out the noise added in advance, and the filtering result is very close to the actual measured value; the example shows that the Kalman filter method can be used to process random errors in slope monitoring data.

[0062] At present, the commonly used correlation calculation methods include Deng's correlation and absolute correlation. Assume that the system behavior sequence and correlation sequence after initialization are: , ; Deng's correlation calculation, let is the difference sequence between the system behavior sequence and the initial value image of the correlation sequence, , ,make They represent the maximum difference between the two levels and the minimum difference between the two levels, that is, the maximum and minimum values ​​of the absolute values ​​of the difference between the reference sequence and the comparison sequence at each moment: , , the reference data sequence is The Deng's correlation coefficient of the comparison data series at each corresponding time is calculated by the following formula: , where is the resolution coefficient. In order to avoid the maximum value being too large and causing distortion of the Dunn correlation coefficient, Its value range is (0, 1), usually ; Indicates Comparison sequence Initial value image and reference sequence at time The absolute value of the difference between the initial value images at time instant.

[0063] It is known that there are 9 measuring points arranged on a certain slope. The missing data interpolation analysis is carried out by taking the 1# measuring point as an example. The 35 settlement deformation data of the 1# measuring point are analyzed. Assuming that the last 5 data are missing, the first 30 data are used to model the interpolation of the last 5 data. The grey correlation interpolation model is established. The monitoring data series of 2# and 3# measuring points on the same slope surface (35 monitoring data) are selected as the comparison data series, and their absolute correlation is calculated. The absolute correlation calculation results of the measured value series of 2#, 3# and 1# measuring points are shown in Table 3. It can be seen that the correlation between 2# and 3# measuring points and 1# measuring point is good. The measured value series of 2# measuring point with a larger correlation is interpolated. The initial value image and the original measured value of the reference sequence are estimated in turn, and the estimated value obtained each time is involved in the calculation of the next measured value estimation. The interpolated measured values ​​of 31-35 periods of the 1# measuring point are listed in Table 4.

[0064] Table 3 Absolute correlation analysis table , Table 4 Grey correlation interpolation results of 1# measuring point , A time series interpolation AR model is established. In the established time series model, the ADF test method is first used to test the data stationarity. The calculation result is H=1. The deformation monitoring data sequence of the measuring point is judged to be a stationary time series by the time series diagram and H=1. The autocorrelation coefficient is tailing and the partial autocorrelation coefficient is 20th order truncated. Therefore, the model is selected as AR(m). The AIC values ​​when m is 0 to 20 are calculated respectively. The result is that the AIC value is the smallest when m=7. Therefore, the AR(7) model is adopted. The model is used to predict the monitoring data of the next five periods. The results are listed in Table 5.

[0065] Table 5 Comparison of prediction results , It can be seen from Table 5 that the prediction value of the method combining grey relational analysis with time series analysis is closer to the true value than the result of single grey relational prediction or time series prediction. The combination of the two reduces the error of interpolation by a single method and makes the interpolation effect better. Therefore, when interpolating missing values ​​in actual engineering, if the data accuracy requirement is high, it is appropriate to use a combination of different interpolation methods for comprehensive analysis to reduce the error of a single interpolation method and improve the prediction accuracy.

[0066] The safety monitoring system of hydraulic rock-soil structures such as high slopes on reservoir banks and earth-rock dams has the attributes of multiple measuring points and multiple projects. Its safety monitoring information is a typical spatiotemporal data with spatial, temporal and attribute characteristics. In view of the multidimensional spatiotemporal characteristics of monitoring information, clustering analysis algorithm and association rule algorithm are introduced. Taking slope monitoring as an example, spatiotemporal data mining is carried out for massive monitoring information of multiple measuring points and multiple projects, and effective safety judgment information is mined from a large amount of seemingly irregular original monitoring information. In order to maximize the spatiotemporal characteristics of monitoring data, partition clustering and projection clustering algorithms are used to carry out spatiotemporal data mining, and the monitoring data are divided into measuring point areas and features are extracted respectively to achieve the purpose of simplifying monitoring data.

[0067] Now combined with the slope spatiotemporal monitoring data, such as Figure 3 As shown in the figure, the steps of the projection clustering method optimized by the genetic algorithm are introduced as follows: set the algorithm parameters, set the operating parameters of the genetic algorithm according to the input monitoring data, for example, determine the population size by the number of measuring points, determine the variable constraints and value range by the measurement range, and set other algorithm parameters such as crossover probability, mutation probability and number of iterations; form an initial population, use randomly generated binary codes as genotypes to describe the monitoring sequence of each measuring point data, form an initial population of monitoring data containing different genotypes, calculate the fitness value of each population individual respectively, and sort the calculation results in sequence; perform proportional selection operation, perform proportional selection operation based on the idea of ​​spinning the roulette wheel, extract some excellent individuals in the initial population of monitoring data and copy them to the new generation of individuals.

[0068] Execute the crossover operation according to the crossover probability, arbitrarily select the gene segments in the middle of two individuals for exchange, and generate a new generation of individuals for monitoring data in the replacement and recombination process; execute the discrete mutation operation according to the mutation probability, and further train to improve the local search ability of the algorithm by exchanging the positions of the gene codes of the two chromosomes; output the projection clustering result, and when the number of iterations pre-set by the genetic algorithm is met, the algorithm is judged to be terminated, otherwise the operation is repeated. In this process, the maximum value of the objective function is solved, and the optimal projection direction vector of the monitoring data is output, so as to obtain the projection clustering feature values ​​of the regional measurement points and the single measurement points.

[0069] like Figure 4As shown in the figure, based on the FP-Growth algorithm, causal association rule mining is used to explore the causality and correlation between environmental quantities and effect quantities, and causal association rule mining is performed on slope monitoring data; according to the experience of experts in the field, rainfall and reservoir water are one of the causes of slope instability, which mainly cause slope deformation through four effects: softening, saturated water loading, static water pressure, and dynamic water pressure; therefore, the antecedent of the association rule is set based on the monitoring environmental quantity data such as rainfall measurement, reservoir water level measurement, and temperature measurement. At the same time, the monitoring effect data such as displacement measurement value and seepage measurement value are used as the latter item of the association rule. The basis for setting up; spatial association rule mining, exploring the potential connection between the measured data of different spatial positions of the measuring points for the slope monitoring data, and conducting spatial association rule mining; the monitoring data of the slope effect quantity is composed of the measured values ​​of multiple measuring points, each of which has a different spatial position. Introducing the consideration of spatial position in the process of association rule mining is more helpful to gain a deeper understanding of the overall structure and internal laws of the slope; setting the rule antecedent and contingent based on the monitoring effect quantity data, the antecedent Characterizes the change of the measured value of a certain monitoring project. Characterizes the changes in measured values ​​at another measuring point of the same monitoring project at the same time.

[0070] Attribute conversion of measured data. Before the association rule mining begins, the degree of data change and data discreteness need to be interpreted into concise and easy-to-understand qualitative language based on the existing monitoring level. The reason is that each slope monitoring data is an accurate quantitative value and cannot reflect any qualitative information. At the same time, the measured value sequence expressed in the form of continuous values ​​is not convenient to directly participate in the calculation. Therefore, it is necessary to realize the attribute conversion of measured data. The specific method is to divide the data value domain into several interval combinations without destroying the data distribution, and give different intervals clear physical meanings respectively. Assuming that there is a fixed feature of the slope, the n measuring points containing this feature constitute the item set set , , then the itemset set In all item sets The contribution ratio is: , based on the idea of ​​equidistant binning, combined with the extreme value of contribution ratio and , the contribution ratio Divide into three categories with equal intervals, obtain the three-level division points in turn, and form an interval arithmetic progression: , , according to the three-level division points, the item set collection intervals with different contribution ratios are generated, including: the item set collection interval with low contribution ratio is , the interval of the item set with the contribution ratio is , the interval of the itemset collection with high contribution ratio is ; By judging the interval attributes of the slope measurement data, the attribute conversion of the measurement data can be achieved.

[0071] Constructing a frequent pattern tree, after transforming the attributes of the measured data, constructing a priority tree for slope spatiotemporal data mining on the basis of retaining the item set information, using a bottom-up approach to comprehensively scan the database and screen out and The association rules under the constraints of two thresholds, namely, generate strong association rules; generate slope association rules, and finally, evaluate and interpret the generated rules one by one in comparison with the actual operation of the project, interpret the essence of the physical phenomena contained in the screened rules, and thus mine effective information reflecting the operation characteristics of the project in the slope spatiotemporal monitoring data.

[0072] like Figure 5 As shown in the figure, the high slope of a certain reservoir bank is 700m long along the river, with a relative slope height of 500m~700m; the average slope is 25~45° above the height of 1400m, and 22~25° below the elevation of 1400m, and there are multiple gentle slope sections; the slope strikes about S60°E, is 700m long along the river, the rock formation dips nearly EW / S∠30°~35°, and the slope is a monocline slope with a binary structure; due to the different stability of the slope sections along the river, the slope below the elevation of 1400m is divided into zones I, II, and III from upstream to downstream according to the different geological structures and instability failure modes of the slope during the preliminary design. Since 1985, after multiple periods of preliminary investigation, dam construction, reservoir water storage and project operation, the slope has accumulated a large amount of monitoring data including surface deformation, deep deformation and environmental quantities. The existing monitoring information is summarized in Table 6.

[0073] Table 6 Monitoring information of the example slope , The measurement point area division based on the partitioning clustering algorithm is based on a total of 24 surface displacement measurement points in Area I, Area II and Area III, including displacement monitoring data in three directions: downstream, downstream and vertical. The K-means clustering algorithm is used to divide the measurement point area. Before the partitioning clustering, the spatial distance and attribute distance of the measurement points are first calculated to obtain the comprehensive distance index of the measurement points. The X-coordinate, Y-coordinate and elevation are used to calculate the spatial distance between each measurement point. At the same time, the three-dimensional displacement measurement values ​​of downstream, downstream and vertical are used to calculate the attribute distance of each measurement point. After the attribute distance and spatial distance are standardized, the comprehensive distance index of the measurement point is obtained by weighted calculation. The initial number of clusters is set to 3. Without specifying the cluster center, the square Euclidean distance between different measurement points is calculated to measure the similarity between the measurement points, and a cyclic iteration method is used to achieve the optimal effect of the measurement point area division.

[0074] like Figure 6As shown in the figure, comparing the measurement point partitioning results based on the partitioning clustering algorithm with the original measurement point partitioning results based on the preliminary data, among the 24 displacement measurement points, except for measurement points D03, intersection 5 and intersection 23, the partitioning results of the remaining 21 measurement points are consistent with the measurement point partitioning results based on the preliminary data, and the partitioning accuracy rate is 87.5%, indicating that the measurement point partitioning results based on the K-means algorithm are more authentic and reliable; for the convenience of calculation, the following is based on the measurement point partitioning results provided by the preliminary data, and the projection clustering algorithm optimized by the genetic algorithm is used to further compress the data level and extract data features.

[0075] Data feature extraction based on projection clustering algorithm, on the basis of measuring point partitioning, the projection clustering algorithm improved by genetic algorithm is used to further extract data features of slope monitoring information; firstly, the annual mean and annual extreme values ​​of the three-way displacement monitoring data are obtained for the measured values ​​of 24 surface displacement measuring points from 1997 to 2019, and the data features of the slope monitoring data set are preliminarily extracted to form 6 displacement measurement matrices with n=23 and m=24 , ; On the basis of standardizing the measured data, the operating parameters of the genetic algorithm are set, including: number of iterations k=50, window radius coefficient Alpha=0.1, variable lower bound LB=-1, variable upper bound UB=1; according to the principle of projection clustering algorithm, the corresponding objective function, fitness function and constraint function are further set.

[0076] Taking the annual average displacement of the slope area as an example, after iterative calculation of the genetic algorithm, the optimal projection direction Best_a=(-0.188, 0.324, -0.258, 0.066, -0.206, -0.222, 0.140, 0.275, -0.105, -0.114, 0.165, 0.167, 0.400, 0.501, -0.025, 0.275, 0.229, 0.008, -0.039, -0.170, -0.129, 0.040, 0.185) is obtained; thus, the projection clustering result of the annual average displacement of the slope area is obtained: =(0.463, 0.468, 0.359, 0.237, 0.000, 0.479, 0.394, 0.352, 0.330, 0.332, 0.172, 0.274, 0.172, 0.052, 0.040, 0.893, 0.301, 0.322, 0.371, 0.201, 0.525, 0.037, 0.530, 1.000), The values ​​in each column correspond to the projection clustering characteristic values ​​of each surface displacement measuring point; similarly, the optimal projection directions of the annual mean and annual extreme values ​​of displacement measuring points in other directions and the corresponding projection clustering characteristic results of the measuring points can be obtained.

[0077] Observe the projection clustering results of displacement measurement points based on genetic algorithm in the figure above. Taking the projection clustering eigenvalue as the judgment index, the measurement points that need to be focused on can be screened out. Taking the projection clustering calculation results of the displacement measurement along the river as an example, it can be seen that the annual mean and annual extreme value projection clustering eigenvalues ​​are the largest at measurement point 11, and the smallest at measurement point 7, which correspond to the maximum and minimum values ​​in the measurement sequence along the river, respectively. The process lines of the two measurement points extend in the positive and negative directions respectively, and the measurement points whose projection clustering eigenvalues ​​are close to the median level, such as measurement point 10, have a small fluctuation range. And there is no obvious increasing or decreasing trend, and the operating status of the measuring points is relatively safe; according to the above rules, the measuring points that need to be focused on in the other two directions are selected; among the measuring points along the slope, the projection clustering eigenvalues ​​of most measuring points are less than 0.6, and the measured value changes are relatively stable, while the measuring points II03 and Jiao 6 in Zone III, whose projection eigenvalues ​​are close to 1, show an obvious increasing trend, and the measured value variation in 20 years reaches 600mm; among the vertical measuring points, except for the measuring point Jiao 11 in Zone II, the eigenvalues ​​of the other measuring points are also less than 0.6, the measured value change law is similar and the development trend is stable.

[0078] The key to the association analysis of slope monitoring data based on FP-Growth algorithm is to set the before and after items of monitoring data, and mainly select the monitoring data of environmental quantity and effect quantity as the basis for constructing association rules. Based on the antecedent and the latter, the monitoring information association mining and knowledge discovery are carried out from the two perspectives of causal association and spatial association respectively; among them, the environmental quantity data comes from the monitoring data of the reservoir water level, and the effect quantity data comes from the displacement monitoring data of the displacement measuring points along the slope, along the river, and vertically, as well as the groundwater level monitoring data of the borehole measuring points; before the association analysis begins, the attribute conversion processing is first performed on each measured data, and the corresponding mathematical statistics are used to characterize the time series distribution and the intensity of the change of the measured data. Then, the arithmetic progression is used to obtain the three-level division points in turn, and each part of the data is divided into equidistant intervals and the names are converted; taking the attribute conversion of the reservoir water level data as an example, assuming that the reservoir water level data , respectively obtain the three-level division points of the interval , , , ,when When , named "normal water level", when When it is called “high water level”.

[0079] After the attribute conversion of the measured data, the Priority tree of slope spatiotemporal data mining is constructed on the basis of retaining the item set information, and the database is fully scanned in a bottom-up manner to screen out strong association rules that conform to the actual project; the causal association rules between the reservoir water level time series distribution and the borehole water level are mined, and the causal association rules between the reservoir water level time series distribution and the borehole water level time series distribution are calculated. , first set the previous item To monitor the environmental quantity, i.e. the temporal distribution of the reservoir water level, the latter term is set is the monitoring effect quantity, that is, the time series distribution of the water level of boreholes 43#, 44#, and 45#; after attribute conversion, the rule antecedent is in the form of The item set of the rule is a set of itemsets, and the rule posterior is of the form On this basis, causal association rules are developed of excavation work.

[0080] In the first association rule mining process using the FP-Growth algorithm, a series of association rules were obtained based on the measured data of the 43# borehole. However, the calculated association rules representing the temporal distribution of reservoir water level and the temporal distribution of borehole water level had a low degree of correlation, especially the association rules of the form "low groundwater level The support and confidence of the association rule of "low borehole water level" are 17.13% and 17.39% respectively, which are far below the requirements of strong association rules. Figure 7 ) found that after the monitoring data of the 43# measuring hole increased sharply in 2003, the measured values ​​were not updated. It was judged that the instrument might be faulty or damaged, resulting in the inconsistency between the association rule mining results and the actual engineering cognition; the abnormal measured values ​​of the 43# measuring hole after 2003 were removed, and the FP-Growth algorithm was used for the second association rule mining. The calculation results of the selected strong association rules are shown in Table 7.

[0081] Table 7 Association rule calculation table , Based on the monitoring data of boreholes 43#, 44#, and 45#, the results of association rule mining were obtained. After removing the abnormal measurement values, the association rule results recalculated based on the monitoring data of borehole 43# showed that the time series distribution of the water level in borehole 43# was strongly correlated with the time series distribution of the reservoir water level, and the support and confidence were significantly improved compared with those before the abnormal measurement values ​​were removed. It can be seen that association analysis is helpful to identify abnormal phenomena at monitoring points; from the perspective of confidence, it can be seen that the confidence of the extracted association rules exceeds 95%, and the maximum confidence reaches 100%. The reliability and credibility of the rules are high, and it can be obtained that the time series distribution of the water level in borehole 43# is strongly correlated with the time series distribution of the reservoir water level. The temporal distribution of groundwater level elevation is highly correlated with the temporal distribution of reservoir water level. From the perspective of support, it can be seen that the temporal distribution of water level at different elevations in borehole 43# is relatively uniform, while the support of borehole 45 reaches 91.29%, and the water level elevation is basically located in the high water level interval. Engineering experience shows that the temporal distribution of groundwater level and the temporal distribution of reservoir water level have a causal effect, and the above-mentioned association rule mining results based on FP-Growth show that the temporal distribution of groundwater level elevation also has a significant correlation effect with the temporal distribution of reservoir water level, indicating that the groundwater level of the slope is affected by the rise and fall of the reservoir water level and the rock permeability is good.

[0082] Mining causal association rules between reservoir water level and displacement measurement, calculating the association rules between the intensity of reservoir water level changes and the intensity of displacement measurement changes ; Set the previous item The latter term is set to the degree of change in the measured value of the reservoir water level is the degree of change in the displacement values ​​of the representative measuring points (intersection 8, intersection 9, intersection 10) in three directions (along the slope, along the river, and vertical) of the slope zones I, II, and III; at this time, the first term of the rule is in the form of The item set of the rule is a set of itemsets, and the rule posterior is of the form On this basis, the FP-Growth algorithm is used to develop causal association rules The data mining work is carried out; the causal association rules with higher credibility are screened out as shown in Table 8.

[0083] Table 8 Association rule calculation table , Analyzing the above calculation results, the following conclusions are drawn: By comparing different monitoring projects, it can be seen that the correlation between the three-way displacement measurement and the reservoir water level is significantly lower than the correlation between the groundwater level and the reservoir water level. The reason is that the displacement measurement is affected by the comprehensive influence of factors such as temperature, water level, and rainfall, while the groundwater level elevation is mainly affected by the reservoir water level; by comparing different distribution areas, among the representative measuring points in slope zones I, II, and III, the displacement measurement changes of the intersection 8 measuring points in slope zone II are more sensitive to the changes in the reservoir water level. The corresponding confidence levels in the three directions are 93.33%, 93.33%, and 80.00%, respectively, which indicates that the credibility of the association rules is high. The correlation is strong; comparing different displacement directions, the change of displacement measurement value along the river is most correlated with the change of reservoir water level measurement value, and the rule confidences corresponding to the three measuring points are 93.33%, 83.33% and 80.00% respectively. Therefore, it can be concluded that the change of displacement measurement value along the river is most affected by reservoir water level; from the support level, it can also be seen that the degree of change of displacement measurement values ​​in the three directions is small, which is consistent with the stability analysis conclusion that the current slope deformation is relatively stable but has a slow growth trend; it can be seen that causal association rule mining can extract and mine the correlation and causality between environmental quantities and effect quantities, and can provide a new perspective for engineering safety monitoring theory.

[0084] Mining spatial association rules of multi-point monitoring information, calculating the association rules between the development laws of multiple displacement measurement points , set the previous item Set the rule after the rule for the development of the measured value of a measuring point is the development law of the measured value of another measuring point. Here, the three-way monitoring data of 12 displacement measuring points, including intersection 10, intersection 8, intersection 12, intersection 13, intersection 16, intersection 5, intersection 6, intersection 7, intersection 9, intersection 11, intersection 14, and intersection 15, from 1998 to 2019 are selected for association rule calculation; at this time, the rule antecedent is in the form of The item set of the rule is a set of itemsets, and the rule posterior is of the form On this basis, the FP-Growth algorithm is used to develop association rules The data mining work is carried out; the spatial association rules with higher credibility are shown in Table 9.

[0085] Table 9 Association rule calculation table , Based on the spatial association rules calculated based on the measured data in three directions, it can be seen that: in the measured data in three directions, the degree of synchronous change of the measured points along the river is generally low, while the measured points along the slope and vertical directions show a significant trend of synchronous change; the synchronous degree of the measured point changes of the three measuring points Jiao 10, Jiao 5 and Jiao 13 in the three directions is very high; among them, the synchronous support of the three-directional changes of the measuring points Jiao 10 and Jiao 5 are 100%, 77.27% and 100% respectively, the synchronous support of the three-directional changes of the measuring points Jiao 13 and Jiao 5 are 100%, 90.91% and 100% respectively, and the synchronous support of the three-directional changes of the measuring points Jiao 10 and Jiao 13 are 100%, 68.18% and 100% respectively; on the basis of calculating the spatial association rules of different measuring points, the measured value change process lines of the three measuring points Jiao 10, Jiao 5 and Jiao 13 in different directions are compared to verify the output rules.

[0086] The BP neural network consists of three parts: input layer, hidden layer, and output layer. Studies have shown that the three-layer BP neural network structure can approximate any nonlinear continuous function with arbitrary accuracy. Therefore, in practical problem solving, a three-layer structure is usually used; its calculation process is implemented by two steps: sample forward propagation and error reverse propagation. When the output result does not meet the expected value, the error reverse propagation is performed to continuously adjust the network weights and thresholds until the output meets the set error result; the BP neural network algorithm flow chart is as follows Figure 8 shown.

[0087] Recurrent Neural Networks (RNN) is a neural network with a special structure. Its characteristic is that different neural units are activated at one time. The hidden layer node at the current moment serves as the feedback item of the output content of the hidden layer node at the previous moment. At the same time, it processes its own input as the output layer and the input of the hidden layer node at the previous time point, and then passes it to the hidden layer node of the next time node after processing the information at that moment. RNN is always a process of continuous backward recursion in the time dimension. The results of the previous moment have a direct impact on the training of the next moment, but because the weights at each time step are the same, different units will share the same set of weights at the same moment.

[0088] The deformation volume change is used to reflect the overall deformation behavior of the slope. The BP neural network and recursive neural network modeling principles are used to study the estimation method of the slope deformation volume, and the overall deformation behavior prediction model of the slope is established, which provides a new perspective for accurately evaluating the development of the slope stability state. The overall deformation behavior prediction model of the slope takes the volume change of the slope determined by multiple measuring points as the analysis target. To determine the volume of the deformation body, it is necessary to master the spatial coordinates of the measuring points and the deformation of the measuring points at each moment, that is, the deformation field of the slope. However, the actual engineering measuring points are arranged less and relatively scattered, and some measuring point data may become invalid and have no use value due to objective factors. There are fewer measuring points that can be used for modeling. At this time, several interpolation points should be set according to the actual situation to obtain the coordinates and deformation of the interpolation points. The correlation between multiple measuring points is usually difficult to describe with a display function. For this reason, we invented the use of BP neural network to express it. Considering that slope deformation is easily affected by rainfall and temperature changes, and waterside slopes such as reservoir banks and river banks are also affected by water levels, then, Any measurement point in time The deformation vector Available expressions are: , where: , , , They are Water level, temperature, rainfall and time-dependent factors at each moment; for Coordinates of the measuring points; is the total number of observation points; using the neural network modeling principle, the water level at time t and the previous period ,temperature , rainfall , the corresponding time-effect factor , measuring point coordinates As neural network input; measurement point The measured deformation vector As output, the relationship between the measuring point coordinates and the measuring point displacement is established using the BP neural network model, which can realize the calculation of slope deformation.

[0089] After determining the deformation influencing factors, the input and output of the model are determined accordingly, and the number of input layer nodes can be determined. And the number of nodes in the output layer The number of nodes in the hidden layer can be determined by preliminarily determining the value range, and then substituting it into the trial calculation to finally determine it: , , , where , , are the number of nodes in the hidden layer, input layer, and output layer respectively; is a constant between 1 and 10.

[0090] Depend on The training samples and verification samples formed by the measured data of each measuring point can be used to establish a neural network model to obtain the distribution of slope deformation at any time; when the monitoring data of each measuring point is incomplete or asynchronous, the missing data spatiotemporal interpolation method can be used to fill it in order to form a complete neural network input and output data sample.

[0091] The established neural network model can estimate the displacement value of any point on the target slope during deformation observation. Taking the observation allowable error as the critical value, the deformation range of the target slope can be obtained, including the plane range and elevation range. , and , , , the slope body is the target slope body; then the overall deformation of the target slope body includes rigid body displacement and deformation displacement; the rigid body displacement of the slope can be obtained by averaging the displacements of each measuring point, such as , , The rigid body displacement in the direction can be expressed as: , The deformation displacement of the target slope is Target slope volume at the moment As the reference benchmark, calculate the volume of the target slope at time t , - The slope deformation volume ; Any time The target slope volume can be estimated by Simpson's method. The deformation volume of the target slope body at the moment: Obviously, the rigid body displacement and deformation displacement reflecting the overall deformation are different when the slope investigation range, i.e. the target slope body, is different. Thus, the deformation behavior change of the slope from point to local to overall can be analyzed.

[0092] The slope deformation volume RNN prediction model can obtain the rigid body displacement of the slope in three directions based on the measured deformation data of each measuring point. and the time series of slope deformation volume change , T is the total number of deformation observations at each measuring point; A statistical regression model can be established; considering the time-varying nature of slope deformation, the present invention adopts the RNN model to predict the change of slope deformation volume, and evaluates the overall stability of the slope by the change trend of the deformation volume; the prediction method of slope rigid body displacement is similar to the prediction method of measuring point displacement; according to the characteristics of slope deformation, different early measured values ​​and early influencing factors can be used as input, such as The deformation of the moment is taken as output, , , …, ( The displacement of the slope measuring point and the slope deformation volume at the previous impact time (days or hours) and the corresponding impact factors are used as input; then the measured time series of slope deformation is By sliding the window, the RNN training sample set can be obtained; the recursive neural network (RNN) model established by the training samples can predict the changes in the displacement of the slope measuring points and the deformation volume of the slope.

[0093] The high slope of a hydropower station is located on the downstream side of the dam shoulder on the right bank of the power station hub, and is about 300 meters long along the river. Since the excavation and construction of the dam road was completed in 1997, the slope has collapsed and slid on a large scale, and local soil collapse and sliding will occur every rainy season. Due to the excavation and blasting of the dam road, the local rock mass on the slope of the road has loosened and the joints have opened, making the local rock mass unstable. Some rock masses have creeped, and affected by the cutting of the diabase vein, the lower surface of the diabase vein is in the shape of an inverted slope in many places. The slope is large in range and high in height, and its lower part is the dam road, the power station access road, the power station indoor switch station and the management room. The unstable slope seriously threatens the safety of highway traffic, engineering construction and operation management, so it is necessary to monitor and analyze its deformation.

[0094] The high slope monitoring system consists of environmental quantity monitoring, surface displacement monitoring, deep displacement monitoring and groundwater monitoring. In 2000, the slope surface displacement monitoring system arranged 7 displacement measuring points, namely tp01-07. In 2001, 8 measuring points, namely tpn1-8, were added to strengthen the monitoring of the slope. The present invention selects 6 surface displacement measuring points, namely tpn1 and tpn4-8, whose monitoring data are relatively complete, whose values ​​are reliable and whose positions are relatively concentrated, for modeling and analysis. The spatial coordinates of the 6 measuring points are shown in Table 10. The rainfall Fig. 9 As shown in Figure 2, the water level of the upstream reservoir is Fig.10 shown.

[0095] Table 10 List of initial coordinates of slope deformation measurement points , The neural network spatiotemporal model of slope deformation is established by , and , Determine the plane range of the target slope body, that is, determine the plane range of the slope body according to the coordinates of each modeling measurement point in Table 10: , , the bottom elevation of the target slope body is taken as 200m; combined with the position distribution of 6 known measuring points, with the principle of even and reasonable distribution of measuring points as much as possible, a total of 12 interpolation points NX1~NX12 are set, the target slope body is divided into 9 blocks, and 16 grid points are formed on the slope surface, of which 4 are observation points and 12 are unknown deformation points. The unknown deformation will be calculated by the neural network spatiotemporal model established later; the target measuring point has a total of 69 displacement monitoring data from January 2011 to December 2016, and the observation frequency is 1 month / time; taking the data in January 2011 as an example, the three-dimensional coordinates of TPn1, TPn4~TPn8 are used as input, and the displacement value of the measuring point in January is used as output to train the model. After 358 random combination trials, the output error is 0.178, which meets the target error requirement; the trained network can be used for prediction; taking itself as a test sample, the displacement value of the known measuring point is predicted by the model and compared with the actual displacement value. The results are shown in Table 11.

[0096] Table 11 Vertical deformation prediction table of BP neural network model in January 2011 , As can be seen from the table above, the model prediction value is not much different from the actual value, with an error of 0.01% to 3.53%, and the maximum error is no more than 5%, indicating that the model prediction accuracy is high, and it is more reliable to use this model for slope deformation interpolation; the displacement of the interpolation points NX1 to NX12 is calculated using the trained network, and then the volume V1 of the target slope body at each period is obtained; the volume of the deformed body at the initial moment is calculated ; Calculate the deformation volume V of the target slope.

[0097] The RNN prediction model for the displacement and deformation volume of the slope measuring point was established. A three-step-one-step model was selected, that is, the water level, temperature, rainfall and the slope volume change values ​​of the first three periods were used as input to deduce the volume change value of the fourth period, and then the volume change value of the fifth period was deduced by the environmental quantities and volume change values ​​of the 2nd, 3rd and 4th periods, and the volume change value was deduced in sequence. Since the frequency of deformation monitoring of the measuring point is 1 month / time, and the frequency of monitoring of water level, temperature and rainfall is 1 day / time, the water level in the modeling is replaced by the average water level of one month, and the rainfall is the cumulative rainfall of the 15 days before the previous observation day. Finally, an RNN neural network with 12 input layer nodes, 1 output layer node and 9 hidden layer nodes was established. Matlab programming was used to calculate and draw the curve of the predicted value and the true value of the deformation volume RNN model as shown in the figure. Fig.11 ;Depend on Fig.11 It can be seen that the predicted value obtained by the RNN recursive neural network is relatively close to the true value and can accurately reflect the actual situation. This model can be used to predict and analyze the deformation volume value of the slope with time-varying characteristics.

[0098] Establishment and comparative analysis of conventional statistical regression model of slope deformation volume. To verify the feasibility of RNN modeling, conventional statistical regression models were established for slope volume change value and deformation value δ of single measuring point tpn1 for comparison. Slope deformation is mainly affected by water level, temperature, rainfall and time. The following statistical model was used for analysis: , where: is a constant term, , , , Respectively represent water pressure component, temperature component, time component and rainfall component; water pressure component expression, , where represents the regression coefficient, To monitor the reservoir water level on the day; the temperature component expression is: , in, represents the temperature factor regression coefficient, represents the cumulative number of days from the monitoring day to the first day of modeling; the expression of the time-dependent component, , where is the regression coefficient of the time-dependent component, Cumulative number of days from the monitoring day to the first day of modeling Divide by 100; the rainfall component expression is: , in, represents the regression coefficient of rainfall factor, They represent the average rainfall on the monitoring day, the previous 4 days, the previous 15 days, and the previous 30 days respectively.

[0099] Regression models were established for the slope deformation volume and the displacement value of the 1# measuring point. Fig.12 and Fig.13 The measured value and the fitted value process line are shown in Figure 2. The fitting accuracy of the slope volume change process line is relatively high, with a complex correlation coefficient of 0.82. The water pressure component and the temperature component account for a small proportion, and the time-effect component accounts for more than 80%. The slope volume shows an obvious periodic change law, fluctuating in the range of 0 to 20 m3 and gradually stabilizing. The complex correlation coefficient of the vertical deformation process line of the tpn1 measuring point is only 0.46, and the model fitting accuracy is poor. Fig.12 and Fig.13 It can be seen that the mutation error of the displacement measurement value of a single measuring point has a large interference, the change law is not obvious, and the fitting effect of the single-point model is also poor. It is difficult to predict the development trend of slope deformation using a single-point model; the slope deformation volume change process line is relatively smooth. The application of the slope deformation volume can reduce the analysis of the overall deformation of the slope by the observation error of a single measuring point, and the slope deformation volume model has a high accuracy; Fig.14The volume change process line and the displacement process lines of the six measuring points involved in the modeling were plotted. It can be seen that the overall change trend of the displacement process lines of each measuring point is consistent, which just verifies that the deformation between points has a certain correlation. Among them, the volume change process line shows a certain periodic change and tends to be stable, which is consistent with the actual change law of the slope, indicating that the deformation volume of the slope can better reflect the actual situation of the overall deformation of the slope.

[0100] Example 3, reference Fig.15 , as an embodiment of the present invention, provides a slope warning system based on optical fiber sensors, including a data processing module, a prediction module, and a warning module.

[0101] The data processing module is used to collect slope deformation monitoring data and environmental monitoring data, and to pre-process the collected deformation monitoring data; the prediction module is used to construct the causal association rules between environmental monitoring data and slope deformation, and to predict the deformation spatial distribution range of the entire slope; the early warning module is used to use the slope deformation monitoring sequence as a sample, train the RNN neural network, predict the development trend of slope deformation, conduct early warning analysis on the overall stability of the slope, and generate early warning results through the alarm device.

[0102] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0103] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0104] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk case (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0105] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logical function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0106] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A slope early warning method based on optical fiber sensor, characterized in that: include: Collect slope deformation monitoring data and environmental monitoring data, and pre-process the collected deformation monitoring data; Construct the causal association rules between environmental monitoring data and slope deformation, and predict the deformation spatial distribution range of the entire slope based on BP neural network; Taking the slope deformation monitoring sequence as a sample, the RNN neural network is trained to predict the development trend of slope deformation, conduct early warning analysis on the overall stability of the slope, and generate early warning results through the alarm device.

2. The slope early warning method based on optical fiber sensor according to claim 1, characterized in that: The preprocessing includes using optical fiber sensors deployed on the slope to collect slope deformation monitoring sequences and environmental monitoring data; The slope deformation monitoring sequence includes the strain and displacement of each measuring point, and the environmental monitoring data includes rainfall and groundwater level; The collected deformation monitoring data is preprocessed, which includes filtering and missing value interpolation to obtain the slope deformation monitoring sequence after preprocessing. ; The collected deformation monitoring data are filtered, including the collected slope deformation monitoring sequence , establish slope deformation monitoring sequence Statistical model , statistical model It is expressed as: ， in, Indicates at time The water level data collected, Indicates at time The collected temperature data, Indicates time The rainfall data collected, For time, Indicates the collection time. is the constant term representing the slope deformation, represents the influence coefficient of temperature on slope deformation, , , , represents the polynomial order, represents the influence coefficient of rainfall on slope deformation, , , , , represents the corresponding coefficient, For time The polynomial transformation value of ; According to the statistical model and slope deformation monitoring sequence , calculate the residual sequence , expressed as: ， For the residual sequence Perform the gross error analysis based on the Rytter criterion, eliminate the gross errors, and obtain a new deformation monitoring sequence ; use Update Original Repeat the gross error analysis of the residual sequence until no data are eliminated by two consecutive L'Eyette criterion gross error analyses, and obtain the sequence after gross error elimination ; Using the sequence after gross error elimination , establish a Kalman filter model including state equation and observation equation; Using the established Kalman filter model, the state prior estimate at time k is predicted based on the optimal estimate of the state at time k-1; Using the observation value at time k, the Kalman filter updates the state posterior estimate at time k; Repeatedly update the state posterior estimation at time k to obtain the optimal estimation sequence of slope deformation state at each time and the deformation monitoring sequence after random error filtering. .

3. The slope early warning method based on optical fiber sensor according to claim 2, characterized in that: The preprocessing also includes interpolating missing values ​​of the collected deformation monitoring data, and filtering the deformation monitoring sequence. Extract the subsequence of measurement points containing missing values ​​as the reference sequence for spatial interpolation, and extract the subsequence of measurement points containing missing values ​​from the deformation monitoring sequence after filtering. Extract the subsequence of measured points without missing values ​​as the comparison sequence of spatial interpolation, calculate the grey correlation degree between the reference sequence of spatial interpolation and the comparison sequence, obtain the comparison sequence whose correlation degree with the reference sequence is greater than the threshold, establish the spatial grey correlation model by using the obtained comparison sequence and the reference sequence, and calculate the missing values ​​in the reference sequence by using the established spatial grey correlation model as the missing values ​​of spatial interpolation; From the deformation monitoring sequence after filtering Extract the subsequence of measurement points containing missing values ​​as the reference sequence for time interpolation, perform a stationarity test on the reference sequence for time interpolation, set the model order range according to the autocorrelation and partial correlation function graphs, and use the Akaike information criterion AIC to obtain the optimal order from the model order range and , according to the optimal order and , establish an autoregressive moving average model for the reference series of time interpolation , using the least squares method to solve the autoregressive moving average model The parameters of the autoregressive moving average model are set using the parameters , calculate the missing values ​​of the reference sequence of time interpolation as the missing values ​​of time interpolation; According to the missing values ​​of spatial interpolation and time interpolation, the weighted average algorithm is used to obtain the final missing values, and the final missing values ​​are used to filter the deformation monitoring sequence after processing. Perform missing value interpolation to obtain the preprocessed slope deformation monitoring sequence .

4. The slope early warning method based on optical fiber sensor according to claim 3, characterized in that: The causal association rules include collecting environmental monitoring data and slope deformation monitoring sequence , construct the causal association rules between environmental monitoring data and slope deformation, take the collected environmental monitoring data as input, use K-means clustering algorithm to divide the environmental monitoring data of each measuring point into spatial regions, and obtain the partition results of environmental monitoring measuring points. Take the partition results of environmental monitoring measuring points as input, use projection clustering algorithm to extract features of environmental monitoring data in each partition, construct projection function and projection index function, use genetic algorithm to solve the optimal projection direction, obtain the projection clustering feature value of the measuring points in each partition, and use the projection clustering feature value of the measuring points in each partition and the corresponding slope deformation monitoring sequence As input, the FP-Growth association rule algorithm is used to construct the causal association rules between environmental monitoring data and slope deformation.

5. The slope early warning method based on optical fiber sensor according to claim 4, characterized in that: The prediction of the deformation spatial distribution range of the entire slope includes collecting environmental monitoring data and preprocessing the slope deformation monitoring sequence Time synchronization is performed to synchronize the environmental monitoring data and slope deformation monitoring sequence after time synchronization. As input, with the deformation of each measuring point as output, construct the training samples of BP neural network, use the constructed training samples, adopt the error back propagation algorithm to train the BP neural network, establish the nonlinear mapping relationship among environmental monitoring data, measuring point coordinates and slope deformation, use the trained BP neural network, with the measuring point coordinates at any position and the environmental monitoring data at the corresponding time as input, calculate the deformation of the corresponding measuring point coordinates, calculate the deformation of each measuring point coordinate in the slope, and obtain the deformation prediction value of each measuring point coordinate. According to the deformation prediction value of each measuring point coordinate, adopt the spatial interpolation method to generate the deformation spatial distribution range of the entire slope.

6. The slope early warning method based on optical fiber sensor according to claim 5, characterized in that: The prediction of the development trend of slope deformation includes using the pre-processed slope deformation monitoring sequence The input and output sample pairs of the RNN neural network are constructed by sliding time window. The input samples are the deformation monitoring values ​​at the current moment and the previous moment, and the output samples are the deformation monitoring prediction values ​​at the next moment. The constructed input and output sample pairs are used to train the RNN neural network using the error back propagation algorithm to obtain the deformation time series of each measuring point within the spatial distribution range of the slope deformation. According to the extracted deformation time series, the trained RNN neural network is used to recursively predict the deformation of each position in the future period of time. According to the deformation of each position, the development trend of the slope deformation in the future period of time is predicted.

7. The slope early warning method based on optical fiber sensor according to claim 6, characterized in that: The early warning analysis comprises calculating the average deformation of each measuring point according to the deformation prediction value of each measuring point within the spatial distribution range of the slope deformation, taking the average deformation of the slope as the rigid body displacement of the whole slope, taking the volume of the slope before deformation as a reference, calculating the volume change of the slope after deformation by using the Simpson method, taking the deformation of the whole slope as the deformation, evaluating the stability of the whole slope according to the predicted deformation of each measuring point in a future period of time and the deformation of the whole slope, comparing the causal association rule obtained with the evaluated stability of the whole slope, revising the evaluation result of the stability of the whole slope according to the comparison result, and generating the early warning analysis result through the alarm device.

8. A system using the slope early warning method based on optical fiber sensors as claimed in any one of claims 1 to 7, characterized in that: Including data processing module, prediction module and early warning module; The data processing module is used to collect slope deformation monitoring data and environmental monitoring data, and pre-process the collected deformation monitoring data; The prediction module is used to construct the causal association rules between environmental monitoring data and slope deformation, and predict the deformation spatial distribution range of the entire slope; The early warning module is used to train the RNN neural network using the slope deformation monitoring sequence as a sample, predict the development trend of the slope deformation, perform early warning analysis on the overall stability of the slope, and generate early warning results through the alarm device.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the slope early warning method based on optical fiber sensor according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the slope warning method based on optical fiber sensors according to any one of claims 1 to 7 are implemented.

Images