A water conservancy project monitoring method based on multi-sensor technology

Through multi-sensor technology and multi-task learning methods, the hybrid neural network model is constructed, and the problem of data limitation of a single sensor is solved, and the comprehensive monitoring and prediction of dam deformation is realized, the accuracy and reliability of prediction is improved, and engineering safety is enhanced. sex.

CN118349950BActive Publication Date: 2024-11-05SHANDONG LONGXINDA CONSULTING & SUPERVISION CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410555757.0
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-07
Publication Date
2024-11-05
Estimated Expiration
2044-05-07

AI Technical Summary

Technical Problem

The existing dam deformation prediction method relies on a single sensor data. It is affected by environmental factors and equipment failures. The accuracy and reliability are limited, and the comprehensive impact of multiple factors cannot be fully considered, resulting in the inability to effectively guarantee the engineering safety.

Method used

Using multi-sensor technology, by selecting and arranging multiple types of sensors (such as temperature sensors, water pressure sensors, horizontal displacement sensors, vertical displacement sensors and tilt sensors), combining wireless communication for real-time data acquisition and transmission, and using multi-task learning methods to build Hybrid neural network model (CLGM) performs dam deformation prediction and improves prediction accuracy through data processing and feature selection.

Benefits of technology

Comprehensive monitoring and prediction of dam deformation is achieved, the accuracy, real-time and reliability of prediction are improved, and the engineering safety is enhanced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure SMS_3
    Figure SMS_3
  • Figure SMS_4
    Figure SMS_4
  • Figure SMS_5
    Figure SMS_5
Patent Text Reader

Abstract

The present invention belongs to the technical field of water conservancy projects, and in particular, relates to a water conservancy project monitoring method based on multi-sensor technology. The present invention provides a water conservancy project monitoring method based on multi-sensor technology, which realizes comprehensive monitoring of water conservancy projects by fusing different types of sensor data. Feature extraction is performed on multi-sensor data, and multi-task learning is used to simultaneously monitor and predict multiple related tasks in water conservancy projects. By sharing the parameters and features of the model, the efficiency, accuracy and reliability of the water conservancy project monitoring system are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of water conservancy projects, and in particular relates to a water conservancy project monitoring method based on multi-sensor technology. Background Art

[0002] Dams are one of the important infrastructures in water conservancy projects, and their safety is of great significance to the safety of life and property of people in the surrounding areas. my country is the country with the largest number of reservoir dams in the world, but most of them were built in the 1950s-1970s. Under the action of long-term hydrostatic pressure, temperature and other loads, the performance of dam materials continues to deteriorate, and the dam is deformed. The deformation of the dam structure may threaten its stability, so it is of great significance to predict the deformation of the dam in a timely and accurate manner.

[0003] Traditional dam deformation prediction methods mainly rely on single sensor data, such as displacement sensors or strain sensors. Although this method can provide a certain degree of information, its accuracy and reliability are limited because single sensor data is affected by environmental factors, equipment failures, etc. The deformation of the dam is affected by many factors, including geological conditions, hydrological conditions, climate change, etc. However, existing methods often only consider some factors, lack comprehensive consideration of the combined impact of multiple factors, and cannot accurately predict various deformations that may occur in the dam, resulting in the inability to effectively ensure the safety of the project. Therefore, how to conduct comprehensive monitoring and prediction of dam deformation and improve the accuracy, real-time and reliability of dam deformation prediction is an urgent problem to be solved. Summary of the invention

[0004] In view of the problems existing in the current dam deformation prediction, the present invention provides a water conservancy project monitoring method based on multi-sensor technology, which realizes comprehensive monitoring of the water conservancy project by fusing different types of sensor data.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: the present invention provides a water conservancy project monitoring method based on multi-sensor technology, comprising the following steps:

[0006] a. Selection and arrangement of sensors: Select and reasonably arrange various types of sensors at key locations or facilities of water conservancy projects;

[0007] b. Data collection and transmission: Real-time collection and transmission of sensor data through wireless communication;

[0008] c. Data processing: processing the data collected by different types of sensors;

[0009] d. Establishment of prediction model: constructing the dam deformation prediction model CLGM based on multi-sensor data fusion of hybrid neural network based on multi-task learning method;

[0010] e. Emergency response and control: when abnormal situations or potential risks are discovered, emergency measures or control measures should be taken in a timely manner.

[0011] Preferably, in step a, the various types of sensors arranged mainly include: temperature sensors, water pressure sensors, horizontal displacement sensors, vertical displacement sensors and tilt sensors.

[0012] Preferably, in step c, according to the principle of the concrete dam deformation statistical model, the horizontal displacement, vertical displacement and inclination of the dam are mainly affected by three factors: water pressure factor, temperature factor and time factor. The water pressure factor is H 1 , H 2 , H 3 , H 4 , temperature factor selection The aging factors are θ, lnθ, 1-e -θ , t is the number of days from the starting day to the observation day, θ = t / 100.

[0013] Preferably, in the step c, due to the influence of factors such as the original air temperature, water temperature, different deformation sampling intervals, equipment failure, and data transmission failure, the data collected by the sensor has data missing. The missing data is filled with linear interpolation at a time interval of 1 day. At the same time, in order to avoid excessive differences between data features caused by different sensor data, the maximum and minimum normalization methods are used to map the numerical ranges of different features to the same scale. The normalization formula is:

[0014]

[0015] Among them, x ik is the current sensor data, x imax is the maximum value of the sensor measurement data, x min The minimum value of the sensor measurement data.

[0016] Preferably, in step d, when predicting the deformation of the dam, the deformation of the dam is comprehensively evaluated from the horizontal displacement, vertical displacement and inclination angle of the dam, and the prediction results output by the CLGM model are the horizontal displacement, vertical displacement and inclination angle of the dam.

[0017] Preferably, the specific operation method of step d is:

[0018] d1. Based on the feature selection method combining the maximum information coefficient MIC and the correlation feature selection CFS, the influencing factors of dam deformation are determined and used as the input of the CLGM model. The calculation steps for feature variable selection are as follows:

[0019] d11. Calculate the MIC value between each influencing factor and horizontal displacement, vertical displacement and tilt:

[0020]

[0021] Where X is the influencing factor matrix, Y is the horizontal displacement, vertical displacement and tilt matrix, g(x) and h(y) are the marginal probability density functions of X and Y respectively, f(x,y) is the joint probability density function of X and Y, D is the m×n grid after the discretization of the sample space of X and Y, and B is the maximum upper limit of the number of m×n grids;

[0022] d12, D={X1,X2,…,X k} combination for encoding,

[0023] And calculate the evaluation function M of each feature subset erits :

[0024]

[0025] Among them, k is the number of feature variables in the selected feature subset D; r is the maximum information coefficient MIC, is the average MIC of k features and Y, is the average MIC among k features;

[0026] d13, select M erits The largest characteristic combination is the final dam deformation influencing factor;

[0027] d2. Perform singular spectrum analysis (SVD) on the factors affecting dam deformation to remove the influence of high noise in the data;

[0028] d3. Based on the denoised data, a CLGM model is constructed. The CLGM model consists of three one-dimensional convolutional layers, one one-dimensional pooling layer, one LSTM layer, one GRU layer, three fully connected layers and a Softmax classifier.

[0029] d4. Based on the prediction results of the horizontal displacement, vertical displacement and inclination of the dam output by the CLGM model, the Dempster fusion rule is used to comprehensively judge the deformation of the dam. The fusion rule is as follows:

[0030]

[0031] Among them, f(y) is the Dempster fusion rule function; ζ is the mass function; y1, y2, y3 are the predicted results of the dam horizontal displacement, vertical displacement and inclination output by the CLGM model; K is the normalization constant.

[0032] d5. Select the test result supported by the most evidence as the test result of dam deformation.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are:

[0034] 1. The present invention provides a water conservancy project monitoring method based on multi-sensor technology, which realizes comprehensive monitoring of water conservancy projects by fusing different types of sensor data. Feature extraction is performed on multi-sensor data, and multi-task learning is used to simultaneously monitor and predict multiple related tasks in water conservancy projects. By sharing the parameters and features of the model, the efficiency, accuracy and reliability of the water conservancy project monitoring system are improved. DETAILED DESCRIPTION

[0035] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in Example 1. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0037] Embodiment 1: This embodiment provides a water conservancy project monitoring method based on multi-sensor technology

[0038] In this embodiment, concrete arch dams are taken as the research object. Considering that most existing dam deformation prediction studies rely mainly on single sensor data, although this method can provide some information, its accuracy and reliability are limited because single sensor data is affected by environmental factors, equipment failures, etc. This embodiment uses multiple sensors to detect the deformation of the dam to make up for the defects of a single sensor, so that the collected data is more accurate and complete, and the efficiency, accuracy and reliability of the water conservancy project monitoring system are improved.

[0039] The specific operation method is to first arrange horizontal displacement sensors, vertical position sensors, tilt sensors, water pressure sensors and temperature sensors in the arch dam section, construction transverse joints, surface holes, middle holes, top arches, arch crown beams, 1 / 4 top arch axis position beams and 3 / 4 top arch axis position beams of the dam. The embodiment of the present invention collects monitoring data through wireless sensor network technology without wiring. According to the principle of the statistical model of concrete dam deformation, the deformation of the dam is mainly affected by the upstream and downstream water levels, ambient temperature changes and time effects. In this embodiment, the water pressure factor is the water depth H in front of the dam on the observation day. 1 The first, second, third and fourth powers are denoted as H 1 , H 2 , H3 , H 4 ; Temperature factor selection The aging factors are θ, lnθ, 1-e -θ , where t is the number of days from the start date to the observation date, θ = t / 100.

[0040] Affected by factors such as the original air temperature, water temperature, and deformation sampling intervals, as well as equipment failures and data transmission failures, the data collected by the sensor is missing. The missing data is filled with linear interpolation at a time interval of 1 day. At the same time, in order to avoid excessive differences between data features caused by different sensor data, the maximum and minimum normalization methods are used to map the numerical ranges of different features to the same scale. The normalization formula is:

[0041]

[0042] Among them, x ik is the current sensor data, x imax is the maximum value of the sensor measurement data, x min The minimum value of the sensor measurement data.

[0043] The quality of the data model is directly related to the selection of input feature variables. There is a large correlation between the variables that affect the accuracy of the dam deformation prediction model. If all related variables are used as input variables, redundant information will be introduced, which will reduce the quality of the model. At the same time, dam deformation usually has strong volatility, and the data directly obtained through sensors often has strong nonlinearity. Traditional methods are usually unable to obtain satisfactory prediction results. In this embodiment, before establishing the prediction model, the sensor data is first selected based on the maximum information coefficient MIC and the correlation feature selection CFS combined feature variable selection method to determine the dam deformation influencing factors, and then the obtained influencing factors are subjected to singular value decomposition SVD to reduce the volatility of the original sequence and provide technical support for subsequent deformation prediction. The specific operation steps are as follows:

[0044] First, calculate the MIC value between each influencing factor and horizontal displacement, vertical displacement and tilt:

[0045]

[0046] Where X is the influencing factor matrix, Y is the horizontal displacement, vertical displacement and tilt matrix, g(x) and h(y) are the marginal probability density functions of X and Y respectively, f(x,y) is the joint probability density function of X and Y, D is the m×n grid after the discretization of the sample space of X and Y, and B is the maximum upper limit of the number of m×n grids;

[0047] Then, all possible feature subsets D = {X1, X2, ..., X k} combination for encoding, and calculate the evaluation function M of each feature subset erits :

[0048]

[0049] Among them, k is the number of feature variables in the selected feature subset D; r is the maximum information coefficient MIC, is the average MIC of k features and Y, is the average MIC among k features, and selects M erits The largest characteristic combination is the final dam deformation influencing factor;

[0050] Singular value decomposition (SVD) is a method that uses the principle of orthogonality to extract data periodic terms, trend terms, and seasonal terms. For the dam deformation influencing factor sequence, in order to obtain an ideal trajectory matrix, it is necessary to select a suitable window length L. The selection of L is generally related to the periodicity of the actual sequence. Combined with the periodic characteristics of dam deformation, in this embodiment, L is taken as one year, that is, L = 365. Then, according to the SVD theory, the trajectory matrix is ​​constructed, and the trajectory matrix is ​​decomposed and divided into non-intersecting component groups according to different singular value trajectory matrices, and then reconstructed according to the orthogonal functions and principal components of each component group. Finally, the original dam deformation influencing factor sequence is decomposed into several stable subsequences, which can effectively analyze the dam deformation influencing factor sequence without losing any deformation information.

[0051] Based on the denoised data, a CLGM model is constructed. The CLGM model consists of three one-dimensional convolutional layers, one one-dimensional pooling layer, one LSTM layer, one GRU layer, three fully connected layers and a Softmax classifier. The CNN convolutional neural network captures the local correlation between data through convolution and pooling layers, and describes the spatial information of the data with a small number of parameters. In this embodiment, the dam deformation influencing factor sequence is a set of one-dimensional data, so a one-dimensional convolutional layer and a pooling layer are used to capture the internal features of the input data, and feature extraction is performed on the historical deformation data of the dam. Since the LSTM and GRU networks can process time series data well, the LSTM network is used to make a preliminary prediction of the dam deformation, and then the GRU network is used to compensate for the prediction error, and the dam deformation prediction value and the error compensation value are reconstructed.

[0052] Multi-task learning helps improve the generalization performance of all tasks by utilizing useful information contained in multiple related tasks. Traditional machine learning methods usually only focus on the performance of the model on a specific task, pay too much attention to a single task, and ignore the useful information hidden in related tasks that can help optimize the task. In this embodiment, the horizontal displacement, vertical displacement and inclination of the dam jointly determine the deformation of the dam, and there is a correlation between the various parameters. Therefore, in this embodiment, multi-task learning is introduced into the prediction of dam deformation, and the horizontal displacement, vertical displacement and inclination prediction information of the dam are simultaneously output through three fully connected layer Softmax classifiers.

[0053] Then, according to the prediction results of the horizontal displacement, vertical displacement and inclination of the dam output by the CLGM model, the Dempster fusion rule is used to comprehensively judge the deformation of the dam. The fusion rule is as follows:

[0054]

[0055] Among them, f(y) is the Dempster fusion rule function; ζ is the mass function; y1, y2, y3 are the predicted results of the dam horizontal displacement, vertical displacement and inclination output by the CLGM model; K is the normalization constant.

[0056] Finally, the test result supported by the most evidence was selected as the test result of dam deformation.

[0057] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A water conservancy project monitoring method based on multi-sensor technology, characterized in that: The following steps are involved: a. Selection and arrangement of sensors: Select and reasonably arrange various types of sensors at key locations or facilities of water conservancy projects; b. Data collection and transmission: Real-time collection and transmission of sensor data through wireless communication; c. Data processing: processing the data collected by different types of sensors; d. Establishment of prediction model: constructing the dam deformation prediction model CLGM based on multi-sensor data fusion of hybrid neural network based on multi-task learning method; e. Emergency response and control: when abnormal situations or potential risks are discovered, emergency measures or control measures should be taken in a timely manner; Wherein, in the above step d, when predicting the deformation of the dam, the deformation of the dam is comprehensively evaluated from the horizontal displacement, vertical displacement and inclination angle of the dam, and the prediction results output by the CLGM model are the horizontal displacement, vertical displacement and inclination angle of the dam; The specific operation method of constructing the prediction model in step d is: d1. Based on the characteristic variable selection method combining the maximum information coefficient MIC and the correlation feature selection CFS, the dam deformation influencing factors are determined and used as the input of the CLGM model. The calculation steps for characteristic variable selection are as follows: d11. Calculate the MIC value between each influencing factor and horizontal displacement, vertical displacement and tilt: Among them, X is the impact factor matrix, Y is the horizontal displacement, vertical displacement and tilt matrix, g(x) and h(y) are the marginal probability density functions of X and Y respectively, f(x,y) is the joint probability density function of X and Y, D is the m×n grid after the discretization of the sample space of X and Y, and B is the maximum upper limit of the number of m×n grids; d12, D={X1,X2,…,X k } combination for encoding, And calculate the evaluation function M of each feature subset erits : Among them, k is the number of feature variables in the selected feature subset D; r is the maximum information coefficient MIC, is the average MIC of k features and Y, is the average MIC among k features; d13, select M erits The largest characteristic combination is the final dam deformation influencing factor; d2. Perform singular spectrum analysis (SVD) on the factors affecting dam deformation to remove the influence of high noise in the data; d3. Based on the denoised data, a CLGM model is constructed. The CLGM model consists of three one-dimensional convolutional layers, one one-dimensional pooling layer, one LSTM layer, one GRU layer, three fully connected layers and a Softmax classifier. d4. Based on the prediction results of the horizontal displacement, vertical displacement and inclination of the dam output by the CLGM model, the Dempster fusion rule is used to comprehensively judge the deformation of the dam. The fusion rule is as follows: Among them, f(y) is the Dempster fusion rule function; ζ is the mass function; y1, y2, y3 are the prediction results of the dam horizontal displacement, vertical displacement and inclination output by the CLGM model; K is the normalization constant; d5. Select the test result supported by the most evidence as the test result of dam deformation.

2. A water conservancy project monitoring method based on multi-sensor technology according to claim 1, characterized in that: In the step a, the various types of sensors arranged mainly include: temperature sensors, water pressure sensors, horizontal displacement sensors, vertical displacement sensors and tilt sensors.

3. The water conservancy project monitoring method based on multi-sensor technology according to claim 1 is characterized in that: According to the principle of the concrete dam deformation statistical model, the horizontal displacement, vertical displacement and inclination of the dam are mainly affected by three factors: water pressure, temperature and aging. In step c, the water pressure factor is H 1 , H 2 , H 3 , H 4 , temperature factor selection The aging factors are θ, lnθ, 1-e -θ , t is the number of days from the starting day to the observation day, θ = t / 100.

4. The water conservancy project monitoring method based on multi-sensor technology according to claim 1 is characterized in that: In step c, due to the influence of factors such as the original air temperature, water temperature, different deformation sampling intervals, equipment failure, and data transmission failure, the data collected by the sensor has data missing. The missing data is filled by linear interpolation at a time interval of 1 day. At the same time, in order to avoid the large difference between the data features caused by different sensor data, the maximum and minimum normalization methods are used to map the numerical ranges of different features to the same scale. The normalization formula is: Among them, x ik is the current sensor data, x imax is the maximum value of the sensor measurement data, x min The minimum value of the sensor measurement data.