CN107480341A - A kind of dam safety comprehensive method based on deep learning - Google Patents
A kind of dam safety comprehensive method based on deep learning Download PDFInfo
- Publication number
- CN107480341A CN107480341A CN201710598587.4A CN201710598587A CN107480341A CN 107480341 A CN107480341 A CN 107480341A CN 201710598587 A CN201710598587 A CN 201710598587A CN 107480341 A CN107480341 A CN 107480341A
- Authority
- CN
- China
- Prior art keywords
- measuring point
- dam
- monitoring
- sample
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Architecture (AREA)
- Civil Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Structural Engineering (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a kind of dam safety comprehensive method based on deep learning, and safety comprehensive judge is carried out using multi-level Monitoring System for Dam Safety.Comprise the following steps：1) dam safety monitoring measuring point is classified, and according to dam engineering characteristic, monitoring system is abstracted into tree；2) measuring point data pretreatment is monitored, missing values are filled up to measuring point data, reject obvious exceptional value；3) measuring point threshold calculations and classification, measuring point Selection Model is calculated, determines measuring point threshold value and measuring point is classified；4) structure convolutional network is trained and judged, dam and monitoring system evaluation result at different levels are obtained by professional monitoring personnel's inspection, as training set and test set, training set and test set training convolutional neural networks are then based on, safety comprehensive judge finally is carried out to dam.
Description
Technical field
It is specifically a kind of automatic to dam the present invention relates to a kind of dam safety comprehensive method based on deep learning
Change the method that Monitoring Data carries out safety comprehensive judge, belong to dam safety monitoring field..
Background technology
Deep learning is a kind of based on the method that data are carried out with representative learning in machine learning.Observation (such as a width
Image) it can be represented using various ways, such as vector of each pixel intensity value, or be more abstractively expressed as a series of
Side, the region etc. of given shape.And some specific method for expressing are used to be easier from example learning task (for example, face
Identification or human facial expression recognition).Deep learning is extracted efficient with the feature learning and layered characteristic of non-supervisory formula or Semi-supervised
Algorithm obtains feature by hand to substitute.
Dam safety monitoring is to hydraulic structure agent structure, foundation, two by Instrument observation and walkaround inspection
Bank slope, related facility and surrounding environment measurements made and observation；Monitoring not only includes fixing building measuring point by one
Determine the Instrument observation of frequency progress, also include the inspection directly perceived periodically or non-periodically to building appearance and internal a wide range of object
Look into and detected with instrument.By observation instrument and equipment, reflect dam and the change of basement rock condition and environment to greatly with timely obtain
The work such as the observation and data processing of the various data of dam effect.The purpose is to analyze estimation dam safe coefficient, so as to and
When take measures, try to ensure that dam safety run.
Traditional dam safety monitoring method is by manually according to speciality of hydroelectric power engineering construction knowledge, with reference to waterwork expertise
And model of structural mechanics is established, and with dam safety monitoring data come correction model and adjusting parameter, corresponding different monitoring class
Type need to establish corresponding mechanical model.Thus when tackling magnanimity Monitoring Data, dam is carried out with traditional structure mechanical model
Safety comprehensive, which is judged, can face that model complexity increase, model number be excessive, model parameter is difficult to adjust and can not ensure precision
Risk, and human cost can be sharply increased.
The present invention provides a kind of dam safety comprehensive method based on deep learning, and deep learning network model is based on
Dam safety monitoring data, representative learning is carried out to dam safety monitoring mass data, extracts data characteristics.What is passed through is a large amount of
Historical data learnt, deep learning network model can reflect the real-time characteristic of dam safety monitoring, save manpower into
This while, ensures the accuracy and real-time of dam safety comprehensive.
The content of the invention
Goal of the invention：More for Monitoring System for Dam Safety monitoring point, data volume is numerous and jumbled, and artificial treatment and judge difficulty are high
And efficiency it is low the problem of, the invention provides a kind of dam safety comprehensive method based on deep learning, magnanimity is monitored
Data carry out feature learning, and efficiently accurate calculate handles and drawn evaluation result.
Technical scheme：In terms of a kind of dam safety comprehensive method based on deep learning, including following four：
(1) dam safety monitoring measuring point is classified
(2) measuring point data pretreatment is monitored
(3) measuring point threshold calculations and classification
(4) build convolutional network training and judge
(1) classification of dam safety monitoring measuring point uses as follows：
The classification of dam safety monitoring measuring point is broadly divided into three flows, is measuring point division, measuring point numbering, measuring point number respectively
According to storage.Monitoring system is abstracted according to dam engineering structure first, then to monitoring measuring point numbering, finally from automation
Measuring point data is gathered in monitoring system and is stored.
1. measuring point divides：According to dam engineering characteristic, monitoring system is abstracted into tree.Root node is dam, is pressed
Main parts, monitoring project, instrument type are down divided into step by step according to monitoring system feature, and the bottom is monitoring measuring point.
2. measuring point is numbered：Numbered first for each measuring point, numbering there should be uniqueness, and can reflect basic belonging to measuring point
Position, monitoring project, instrument type.Then main parts, monitoring project, instrument type are numbered, and according to step 1.
In measuring point division, the index to its child node is established per first nodes, i.e. dam needs to index all main parts, often
Individual main parts index its corresponding monitoring project, and each monitoring project then indexes its corresponding all appts type, instrument class
All monitoring measuring points of all subordinaties of type.Be finally dam, main parts, monitoring project, instrument type establish database respectively
Table, and all monitoring measuring point number informations of dam are stored in databases according to index structure.
3. measuring point data stores：According to step 2. middle foundation measuring point number, monitoring is collected from automatic monitoring system
Measuring point data, and store into corresponding database table.
(2) the as follows of measuring point data pretreatment use is monitored：
Monitoring measuring point data pretreatment is broadly divided into two flows：Fill up missing values and reject obvious exceptional value.Due to outer
Boundary's factor such as error of instrument failure, data transfer, sensor error can cause the data collected in automatic monitoring system
Missing values and obvious exceptional value be present.Therefore it is that have must to measuring point initial data fill up missing values and reject obvious exceptional value
Want.
1. fill up missing values：Missing data in short-term in monitoring measuring point time series is filled up, first with fitting
Method calculates missing values, then is handled with third index flatness.For long-time deletion sequence, the data after filling up can be present very
Big error, it is regarded as one of instrument exception, factor as judge.
Missing data time point D is determined first, and it is N time serieses to choose length forward, selects mean square error threshold value T；So
It is N time series application approximating method to length afterwards, and calculates mean square error E；If E<T, then when being calculated according to approximating method
Between point D measuring point numerical value, otherwise by N+1 and repeated application approximating method；The measuring point numerical value that last basis calculates, it is with length
N time series data, does Three-exponential Smoothing, and obtained time point D numerical value is the missing values to be filled up.
2. the method being combined using quartile and calculus of finite differences is rejected and deviates considerably from overall point and Outliers.
Quartile method calculates：Q1：It is divided into the quartering after data are arranged by ascending order, data is arranged according to ascending order mode,
And the quartering is carried out, first quartile is Q1；Q3：Third quartile；IQR=Q3-Q1.Each data is entered
Row detection, if it, which is located at section [min-1.5*IQR, max+1.5*IQR], is then considered as normal value, otherwise as abnormal data at
Reason.Quartile method, which can only be rejected, deviates considerably from overall point.
Calculus of finite differences calculates the absolute value of adjacent 2 points of difference | y2-y1|,|y3-y2|,…,|yn-yn-1|, obtain this group of number
Median median, if | yi+1-yi|≤8*median, then regarding yi+1To be normal, if | yi+1-yi|>8*median, then judge
Before it | yi-yi-1| it is whether abnormal and below | yi+2-yi+1| whether also to be abnormal, if the two conditions meet simultaneously,
yi+1It is abnormal data.Otherwise, it may be possible to the data of continuous more days relatively before presence relatively large deviation, such case is not abnormal.
Because median has stability in data sequence, therefore Outliers can be avoided to influence to differentiate by median.
(3) measuring point threshold calculations and classification use as follows：
Measuring point threshold calculations and classification are broadly divided into five flows：Determine that sample size, Selection Model calculate, calculate sample
Mean square error, division measuring point threshold value, measuring point classification.
1. determine sample size：The sample data imported after step (2) pretreatment.Home window length is chosen,
Divide training sample and test sample.
2. Selection Model calculates：To candidate family (exponential model, quadratic polynomial model, cubic function interpolation model, three
Secondary Spline-Interpolation Model and Lagrange's interpolation model) it is fitted respectively, the equation of corresponding fitting is obtained, uses fit equation
The radial displacement of the previous day of forecast date needed for calculating obtains verifying error, verification error is as to model as verification
Verification, can be with the reasonability of analysis model in itself and the accuracy of prediction.
3. calculate sample mean square deviation：Judgment criteria of the sample mean square error (RMSE) as forecast model quality, it is calculated
Formula is：Wherein n is the quantity of sample, and i represents i-th of sample, yiFor the estimate of sample,For estimation
The average value of value, n-1 are the free degree of sample mean square error when calculating simple regression analysis.
Then sample size (i.e. step-length adds 1) is subjected to aforesaid operations every time.If increase sample size sample mean square error subtracts
It is small, then should to continue to look for optimal models, if continuous 7 times all no renewal sample mean square errors, and update sample after this
The probability of this mean square error is less than 0.003 (if increase by 1 regards Bernoulli Jacob's reality as to sample to solve local optimum every time
Test, and think that it is separate, because optimal each time with being likely to be breached, and think that it reaches most in each experiment
It is excellent or not reach optimal probability be 1/2, then it can not all improve sample mean square error by continuous 7 times experiments, that
The probability for improving sample square mean error amount after 7 times again is less than 0.00391 (i.e. 2-8) be small probability event), due to one
During secondary random experiment, the probability that this small probability event occurs is near 0, so having reason to find local optimum.
4. divide measuring point threshold value：The fitting result of each model is counted, models fitting is used as using sample square mean error amount
Measurement index, compare the size of the sample mean square error in above two model to determine final dynamic optimal model, use
The result of calculation of optimal models is predicted value, according to measuring point measured value and predictor calculation relative error, as threshold value.
5. measuring point is classified：Classified calculating methods of the monitoring point P in date D is as follows：
1) real-time of monitoring is considered, 15 days data of monitoring point to date D calculate sample as sample before taking date D
The average mean and standard deviation std of this threshold value；
2) measuring point P is set in date D monitor value as x, defines tetra- kinds of bases of Perfect, Regular, Examine, Problem
This state, wherein Perfect represent the perfection of dam running status, and Regular represents dam normal operation, and Examine represents big
Dam, which is run, to be needed to check, Problem represents dam operation and problem be present, it is necessary to check.State state calculation formula are as follows：
State=Perfect, mean-std<x<mean+std
State=Regular, mean-2*std<x<mean+2*std
State=Examine, mean-3*std<x<mean+3*std
State=Problem, else
Judge according to Perfect, Regular, Examine, Problem order, finally give the classification of measuring point.
3) measuring point classification is quantified, including occurs the situation of long-term sequence missing (Missing) in pretreatment, will
Missing is considered as the superposition of Examine and Problem states, and calculation is：
Perfect=[1,0,0,0]
Regular=[0,1,0,0]
Examine=[0,0,1,0]
Problem=[0,0,0,1]
Missing=[0,0,1,1]
(4) build convolutional network training and judge the as follows of use：
The training of structure convolutional network and judge are broadly divided into four flows：Prepare data set, structure network model, training and
Assessment models, dam safety comprehensive.Actual conditions and engineering experience are run according to dam, check to obtain greatly by monitoring personnel
Dam and monitoring system evaluation result at different levels, as training set and test set；It is then based on training set and test set training volume
Product neutral net, finally carries out safety comprehensive judge to dam.
1. prepare data set：Actual conditions and engineering experience are run according to dam, by monitoring personnel check to obtain dam with
And the evaluation result that monitoring system is at different levels, as training set and test set.
2. build network model：The dam tree feature divided according to step (1), bottom monitoring measure-point amount are far super
Main parts, monitoring project, the interstitial content of instrument type.In bottom measuring point level, data volume is very big, and convolutional network can pass through
Shared convolution kernel, efficient process high dimensional data, and without manual selected characteristic, train weight, it is good to produce tagsort effect.
In main parts, monitoring project, instrument type and dam level, data volume can be reduced drastically, using convolutional network not now
Feature extraction precision can be increased, and training time cost and space cost can be increased.Therefore multilayer Softmax sorter networks are used instead
To be trained and judge, training cost is greatly decreased while nicety of grading is ensured.
1) convolutional network：Including input layer, two layers of convolution and pond layer, full articulamentum, Dropout layers and output layer.
Parameter initialization：In order to create convolutional network model, it is necessary to create substantial amounts of weight and bias term, weight is initial
A small amount of noise is added during change to break symmetry and avoid 0 gradient.ReLU (amendment linear unit) is all used per layer network,
Therefore the problem of with a less integer to initialize bias term to avoid neuron node output perseverance from being 0.
Convolution and pond：Convolution calculates from higher-dimension degrees of data and extracts feature, then with pond to different features
Aggregate statistics are carried out, these summary statistics features have much lower dimension, and effectively avoid over-fitting.Convolution and pondization operation
Initial data is mapped to hidden layer feature space.
Full articulamentum：Full articulamentum uses bulk redundancy parameter, and to the characteristic weighing of extraction, fusion preceding networks study is arrived
Depth characteristic, map that to sample labeling space.The output of last layer network is multiplied by weight matrix, plus biasing, so
After be used for ReLU (amendment linear unit).
Dropout layers：In order to reduce over-fitting, Dropout is added before output layer.The output of setting neuron exists
Constant probability is kept in Dropout, and enables Dropout in the training process, Dropout is closed in test process.
Output layer：Softmax is added, obtains classification results.
2) multilayer Softmax networks：It is made up of input layer, hidden layer, output layer
Input layer：For a collection of measuring point to be trained, seclected time sequence, if the measure-point amount to be classified is n, represents and divide
The vector length of class be 4, therefore by measuring point data expand into length be 4n one-dimensional vector, in the training process each batch with
Machine selects the measuring point data at multiple time points as input.
Hidden layer：Activation primitive is sigmoid, updates hidden layer weight and offset parameter, extraction using back propagation
Data characteristics.
Output layer：Different classifications allocation probability, and output category result are given with Softmax.
According to dam tree, different network models is selected for different node levels, and by each layer of network
Connected by data sharing, final structure one has multi-level tree network group model.
3. training and assessment models：Cross entropy is the inefficiencies that the prediction for weighing network model is used to describe truth.
In the present invention convolutional network and Softmax networks that build, cross entropy is not only used for weighing a pair of single predictions and true
Real value, but a summation for training the cross entropy of all measuring points in batch, for the prediction table of whole training lot data point
Now than single data point performance can preferably descriptive model performance.Cross entropy is defined as follows, wherein y be we predict it is general
Rate is distributed, and y ' is actual distribution, and i refers to training i-th of batch in batch.
Hy′(y)=- ∑iy′ilog(yi)
After defining loss function, learning rate is set.Softmax Web vector graphics gradient descent algorithm is with certain study
Rate finely tunes parameter, constantly reduces cost.Convolutional network does gradient steepest decline with more complicated ADAM optimizers, and adds
Extra parameter controls Dropout ratios.
4. dam safety comprehensive：By deep learning and parameter adjustment, each hierarchical network is trained, finally gives one
Multi-level tree network group model, store the structure and parameter of whole network group.All measuring points of dam are judged step by step,
And realize dam general safety Comprehensive Evaluation.
Beneficial effect：Compared with prior art, the dam safety comprehensive side provided by the invention based on deep learning
Method to dam safety monitoring measuring point by being classified, monitoring measuring point data pretreatment, measuring point threshold calculations and classification, structure convolution net
Dam safety comprehensive is realized in network training and judge, is improved dam safety comprehensive efficiency and precision, is carried for dam monitoring
For reliable basis.
Brief description of the drawings
Fig. 1 is the overall framework figure of the dam safety comprehensive method based on deep learning of present example；
Fig. 2 is the measuring point partition structure figure of present example.
Embodiment
With reference to specific embodiment, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limitation the scope of the present invention, after the present invention has been read, various equivalences of the those skilled in the art to the present invention
The modification of form falls within the application appended claims limited range.
Fig. 1 be the dam safety comprehensive method provided by the invention based on deep learning overall framework figure, its work
It is described below to make process：
1. dam safety monitoring measuring point is classified.Monitoring system is abstracted according to dam engineering structure, marks off monitoring
Measuring point, then to monitoring measuring point numbering, measuring point data is finally gathered from automatic monitoring system and is stored.Measuring point division is such as attached
Shown in Fig. 2, root node is dam, and main parts, monitoring project, instrument class are down divided into step by step according to monitoring system feature
Type, the bottom are monitoring measuring point.
Then unique number is carried out step by step to monitoring measuring point, main parts, monitoring project, instrument type, and according to measuring point
Division, the index to its child node is established per first nodes, i.e. dam needs to index all main parts, each basic portion
Position index is to its corresponding monitoring project, and each monitoring project then indexes its corresponding all appts type, under instrument type is all
All monitoring measuring points of category.Be finally dam, main parts, monitoring project, instrument type establish database table respectively, and according to
All monitoring measuring point number informations of dam are stored in databases by index structure.
Numbered according to measuring points at different levels, monitoring measuring point data is collected from automatic monitoring system, and stored to corresponding number
According in the table of storehouse.
2. monitor measuring point data pretreatment.Including filling up missing values and rejecting obvious exceptional value.
Fill up missing values：Missing data in short-term in monitoring measuring point time series is filled up, first with fitting side
Method calculates missing values, then is handled with third index flatness.Missing data time point D is determined first, when selection length is N forward
Between sequence, select mean square error threshold value T；Then it is N time series application approximating method to length, and calculates mean square error E；
If E<T, then the measuring point numerical value at time point D is calculated according to approximating method, otherwise by N+1 and repeated application approximating method；Last root
According to the measuring point numerical value calculated, with the time series data that length is N, Three-exponential Smoothing is done, obtained time point D numerical value
The missing values as to be filled up.
The method being combined using quartile and calculus of finite differences is rejected and deviates considerably from overall point and Outliers.Quartile
Number method calculates：Q1：It is divided into the quartering after data are arranged by ascending order, data is arranged according to ascending order mode, and carries out the quartering, the
One quartile is Q1；Q3：Third quartile；IQR=Q3-Q1.Each data is detected, if its position
In section, [min-1.5*IQR, max+1.5*IQR] is then considered as normal value, otherwise as dealing of abnormal data.Quartile method is only
It can reject and deviate considerably from overall point.
Calculus of finite differences calculates the absolute value of adjacent 2 points of difference | y2-y1|,|y3-y2|,…,|yn-yn-1|, obtain this group of number
Median median, if | yi+1-yi|≤8*median, then regarding yi+1To be normal, if | yi+1-yi|>8*median, then judge
Before it | yi-yi-1| it is whether abnormal and below | yi+2-yi+1| whether also to be abnormal, if the two conditions meet simultaneously,
yi+1It is abnormal data.Otherwise, it may be possible to the data of continuous more days relatively before presence relatively large deviation, such case is not abnormal.
Because median has stability in data sequence, therefore Outliers can be avoided to influence to differentiate by median.
3. measuring point threshold calculations and classification.Including determine sample size, Selection Model calculate, calculate sample mean square error,
Divide measuring point threshold value, measuring point classification.
Determine sample size：The sample data imported after pretreatment.Choose home window length, division training sample
Sheet and test sample.
Selection Model calculates：To candidate family (exponential model, quadratic polynomial model, cubic function interpolation model, three times
Spline-Interpolation Model and Lagrange's interpolation model) it is fitted respectively, the equation of corresponding fitting is obtained, with fit equation meter
The radial displacement of the previous day of forecast date needed for calculation obtains verifying error, verification error is as the school to model as verification
Test.
Calculate sample mean square deviation：Judgment criteria of the sample mean square error (RMSE) as forecast model quality, it calculates public
Formula is：Wherein n be sample quantity, yiFor the estimate of sample,For the average value of estimate, n-1 is
The free degree of sample mean square error when calculating simple regression analysis.
Then sample size (i.e. step-length adds 1) is subjected to aforesaid operations every time.If increase sample size sample mean square error subtracts
It is small, then should to continue to look for optimal models, if continuous 7 times all no renewal sample mean square errors, then it is assumed that find part most
The figure of merit.
Divide measuring point threshold value：The fitting result of each model is counted, the weighing apparatus of models fitting is used as using sample square mean error amount
Figureofmerit, compare the size of the sample mean square error in above two model to determine final dynamic optimal model, using most
The result of calculation of excellent model is predicted value, according to measuring point measured value and predictor calculation relative error, as threshold value.
Measuring point is classified：Classified calculating methods of the monitoring point P in date D is as follows：
1) real-time of monitoring is considered, 15 days data of monitoring point to date D calculate sample as sample before taking date D
The average mean and standard deviation std of this threshold value；
2) measuring point P is set in date D monitor value as x, defines tetra- kinds of bases of Perfect, Regular, Examine, Problem
This state, wherein Perfect represent the perfection of dam running status, and Regular represents dam normal operation, and Examine represents big
Dam, which is run, to be needed to check, Problem represents dam operation and problem be present, it is necessary to check.State state calculation formula are as follows：
State=Perfect, mean-std<x<mean+std
State=Regular, mean-2*std<x<mean+2*std
State=Examine, mean-3*std<x<mean+3*std
State=Problem, else
Judge according to Perfect, Regular, Examine, Problem order, finally give the classification of measuring point.
3) measuring point classification is quantified, including occurs the situation of long-term sequence missing (Missing) in pretreatment, will
Missing is considered as the superposition of Examine and Problem states, and calculation is：
Perfect=[1,0,0,0]
Regular=[0,1,0,0]
Examine=[0,0,1,0]
Problem=[0,0,0,1]
Missing=[0,0,1,1]
The vector that the corresponding length of each state is 4, state classification is represented with vector.
4. build convolutional network training and judge.Including prepare data set, structure network model, training and assessment models,
Dam safety comprehensive.Actual conditions and engineering experience are run according to dam, are checked to obtain dam and prison by monitoring personnel
Examining system evaluation result at different levels, as training set and test set；It is then based on training set and test set training convolutional nerve net
Network, safety comprehensive judge finally is carried out to dam.
Prepare data set：Actual conditions and engineering experience are run according to dam, by monitoring personnel check to obtain dam and
Monitoring system evaluation result at different levels, as training set and test set.
Build network model：In bottom measuring point level, using multilayer convolutional network.In main parts, monitoring project, instrument
Type and dam level, it is trained and judges using multilayer Softmax sorter networks.
1) convolutional network：Including input layer, two layers of convolution and pond layer, full articulamentum, Dropout layers and output layer.
Parameter initialization：Weight initialization when add a small amount of noise to break symmetry and avoid 0 gradient, with compared with
The problem of small integer is to initialize bias term to avoid neuron node output perseverance from being 0.
Convolution and pond：Convolution calculates from higher-dimension degrees of data and extracts feature, then with pond to different features
Carry out aggregate statistics.Initial data is mapped to hidden layer feature space by convolution and pondization operation.
Full articulamentum：Full articulamentum uses bulk redundancy parameter, and to the characteristic weighing of extraction, fusion preceding networks study is arrived
Depth characteristic, map that to sample labeling space.The output of last layer network is multiplied by weight matrix, plus biasing, so
After be used for ReLU (amendment linear unit).
Dropout layers：In order to reduce over-fitting, Dropout is added before output layer.The output of setting neuron exists
Constant probability is kept in Dropout, and enables Dropout in the training process, Dropout is closed in test process.
Output layer：Softmax is added, obtains classification results.
2) multilayer Softmax networks：It is made up of input layer, hidden layer, output layer
Input layer：For a collection of measuring point to be trained, seclected time sequence, if the measure-point amount to be classified is n, represents and divide
The vector length of class be 4, therefore by measuring point data expand into length be 4n one-dimensional vector, in the training process each batch with
Machine selects the measuring point data at multiple time points as input.
Hidden layer：Activation primitive is sigmoid, using back propagation undated parameter, extracts data characteristics.
Output layer：Different classifications allocation probability, and output category result are given with Softmax.
According to dam tree, different network models is selected for different node levels, and by each layer of network
Connected by data sharing, final structure one has multi-level tree network group model.
Training and assessment models：In the convolutional network and Softmax networks that the present invention is built, cross entropy is a training
The summation of the cross entropy of all measuring points in batch.Cross entropy is defined as follows, and wherein y is the probability distribution that we predict, y ' is real
The distribution on border.
Hy′(y)=-Σiy′ilog(yi)
Learning rate is set, Softmax Web vector graphics gradient descent algorithm is finely tuned parameter with certain learning rate, constantly reduced
Cost.Convolutional network does gradient steepest decline with more complicated ADAM optimizers, and adds extra parameter to control
Dropout ratios.
Dam safety comprehensive：By deep learning and parameter adjustment, each hierarchical network is trained, is finally given more than one
The tree network group model of level, store the structure and parameter of whole network group.All measuring points of dam are judged step by step, and
Realize dam general safety Comprehensive Evaluation.
It can be seen from above example, for Traditional Man computational structural mechanics model to carry out dam safety comprehensive
Model complexity it is high, computational efficiency is low, judges the problem of precision is low, and human cost is high, method of the invention, is pacified by dam
Full monitoring measuring point classification, monitor measuring point data pretreatment, measuring point threshold calculations and classification, the training of structure convolutional network and judge,
Dam safety comprehensive efficiency and precision are improved, reliable basis are provided for dam monitoring.
Claims (6)
1. a kind of dam safety comprehensive method based on deep learning, the safety comprehensive for dam monitoring system is judged,
It is characterised in that it includes four aspects：Dam safety monitoring measuring point point, monitoring measuring point data pretreatment, measuring point threshold calculations and
Classification, the training of structure convolutional network and judge；
1) dam safety monitoring measuring point is classified, and monitoring system is abstracted according to dam engineering structure, then to monitoring measuring point
Numbering, measuring point data is finally gathered from automatic monitoring system and is stored；
2) measuring point data pretreatment is monitored, missing values is filled up and rejects obvious exceptional value；
3) measuring point threshold calculations and classification, to pretreated data, it is first determined sample size, then Selection Model calculating,
And sample mean square error is calculated, choose optimal models and divide threshold value, finally measuring point is classified；
4) build convolutional network training and judge, actual conditions and engineering experience are run according to dam, checked by monitoring personnel
The evaluation result at different levels to dam and monitoring system, as training set and test set；It is then based on training set and test is assembled for training
Practice convolutional neural networks, safety comprehensive judge finally is carried out to dam.
2. the dam safety comprehensive method based on deep learning as claimed in claim 1, it is characterised in that dam safety
Monitoring measuring point classification is broadly divided into three flows, is measuring point division, measuring point numbering, measuring point data storage respectively.Basis first
Dam engineering structure is abstracted to monitoring system, and then monitoring measuring point numbering, the step 1) are comprised the steps of：
1.1) measuring point divides：According to dam engineering characteristic, monitoring system is abstracted into tree；Root node is dam, according to
Monitoring system feature is down divided into main parts, monitoring project, instrument type step by step, and the bottom is monitoring measuring point.
1.2) measuring point is numbered：Numbered first for each measuring point, numbering there should be uniqueness, and can reflect the basic portion belonging to measuring point
Position, monitoring project, instrument type.Then main parts, monitoring project, instrument type are numbered, and according to step 1.1)
In measuring point division, the index to its child node is established per first nodes, i.e. dam needs to index all main parts, often
Individual main parts index its corresponding monitoring project, and each monitoring project then indexes its corresponding all appts type, instrument class
All monitoring measuring points of all subordinaties of type；Be finally dam, main parts, monitoring project, instrument type establish database respectively
Table, and all monitoring measuring point number informations of dam are stored in databases according to index structure；
1.3) measuring point data stores：Numbered according to the measuring point established in step 1.2), monitoring is collected from automatic monitoring system
Measuring point data.
3. the dam safety comprehensive method based on deep learning as claimed in claim 1, it is characterised in that the step
3) comprise the steps of：
3.1) sample size is determined：The sample data imported after pretreatment, choose home window length, division training sample
Sheet and test sample.；
3.2) Selection Model calculates：Candidate family is fitted respectively, the equation of corresponding fitting is obtained, with fit equation meter
The radial displacement of the previous day of forecast date needed for calculation obtains verifying error, verification error is as the school to model as verification
Test；
3.3) sample mean square deviation is calculated, its calculation formula is：Wherein n is the quantity of sample, and i is represented i-th
Sample, yiFor the estimate of sample,For the average value of estimate, n-1 is sample mean square error when calculating simple regression analysis
The free degree；Then every time sample size (i.e. step-length adds 1) 3.2) operate；If increase sample size sample mean square error
Reduce, then should continue to look for optimal models, if continuous 7 times all no renewal sample mean square errors, then it is assumed that find part
Optimal value.
3.4) measuring point threshold value is divided：The fitting result of each model is counted, the weighing apparatus of models fitting is used as using sample square mean error amount
Figureofmerit, compare the size of the sample mean square error in above two model to determine final dynamic optimal model, using most
The result of calculation of excellent model is predicted value, according to measuring point measured value and predictor calculation relative error, as threshold value.
3.5) measuring point is classified, according to classification of the running status of dam to measuring point.
4. the dam safety comprehensive method based on deep learning as claimed in claim 3, it is characterised in that monitoring point P
It is as follows in date D classified calculating method：
1. in view of the real-time of monitoring, 15 days data of monitoring point to date D calculate sample threshold as sample before taking date D
The average mean and standard deviation std of value；
2. setting measuring point P in date D monitor value as x, tetra- kinds of basic shapes of Perfect, Regular, Examine, Problem are defined
State, wherein Perfect represent the perfection of dam running status, and Regular represents dam normal operation, and Examine represents dam fortune
Capable to need to check, Problem represents dam operation and problem be present, it is necessary to check；State state calculation formula are as follows：
State=Perfect, mean-std<x<mean+std
State=Regular, mean-2*std<x<mean+2*std
State=Examine, mean-3*std<x<mean+3*std
State=Problem, else
Judge according to Perfect, Regular, Examine, Problem order, finally give the classification of measuring point；
3. quantifying to measuring point classification, including occur the situation of long-term sequence missing (Missing) in pretreatment, will
Missing is considered as the superposition of Examine and Problem states, and calculation is：
Perfect=[1,0,0,0]
Regular=[0,1,0,0]
Examine=[0,0,1,0]
Problem=[0,0,0,1]
Missing=[0,0,1,1].
5. the dam safety comprehensive method based on deep learning as claimed in claim 1, it is characterised in that the step
4) comprise the steps of：
4.1) data set is prepared：Actual conditions and engineering experience are run according to dam, by monitoring personnel check to obtain dam and
Monitoring system evaluation result at different levels, as training set and test set；
4.2) network model is built：In bottom measuring point level, using multilayer convolutional network；In main parts, monitoring project, instrument
Type and dam level, it is trained and judges using multilayer Softmax sorter networks；
4.3) training and assessment models：In the convolutional network and Softmax networks built, cross entropy is institute in a training batch
There is the summation of the cross entropy of measuring point；Cross entropy is defined as follows, and wherein y is the probability distribution that we predict, y ' is actual distribution；
Hy′(y)=- Σiy′ilog(yi)
Set learning rate, Softmax Web vector graphics gradient descent algorithm finely tunes parameter with the learning rate that sets, constantly reduce into
This；Convolutional network does gradient steepest decline with ADAM optimizers, and adds extra parameter to control Dropout ratios.
4.4) dam safety comprehensive：By deep learning and parameter adjustment, each hierarchical network is trained, is finally given more than one
The tree network group model of level, store the structure and parameter of whole network group；All measuring points of dam are judged step by step, and
Realize dam general safety Comprehensive Evaluation.
6. the dam safety comprehensive method based on deep learning as claimed in claim 5, it is characterised in that the step
4) comprise the steps of：
1. convolutional network：Including input layer, two layers of convolution and pond layer, full articulamentum, Dropout layers and output layer；
Parameter initialization：Weight adds a small amount of noise to break symmetry and avoid 0 gradient in initialization, and use is less
The problem of integer is to initialize bias term to avoid neuron node output perseverance from being 0；
Convolution and pond：Convolution is calculated from higher-dimension degrees of data and extracts feature, and then different features is carried out with pond
Aggregate statistics；Initial data is mapped to hidden layer feature space by convolution and pondization operation.
Full articulamentum：Full articulamentum uses bulk redundancy parameter, to the characteristic weighing of extraction, merges the depth that preceding networks learn
Feature is spent, maps that to sample labeling space；The output of last layer network is multiplied by weight matrix, it is then right plus biasing
It uses ReLU (amendment linear unit)；
Dropout layers：In order to reduce over-fitting, Dropout is added before output layer；The output of setting neuron exists
Constant probability is kept in Dropout, and enables Dropout in the training process, Dropout is closed in test process；
Output layer：Softmax is added, obtains classification results.
2. multilayer Softmax networks：It is made up of input layer, hidden layer, output layer
Input layer：For a collection of measuring point to be trained, seclected time sequence, if the measure-point amount to be classified is n, classification is represented
Vector length is 4, therefore measuring point data is expanded into the one-dimensional vector that length is 4n, and each batch is selected at random in the training process
The measuring point data at multiple time points is selected as input；
Hidden layer：Activation primitive is sigmoid, using back propagation undated parameter, extracts data characteristics；
Output layer：Different classifications allocation probability, and output category result are given with Softmax；
According to dam tree, different network models is selected for different node levels, and each layer of network is passed through
Data sharing connects, and final structure one has multi-level tree network group model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710598587.4A CN107480341B (en) | 2017-07-21 | 2017-07-21 | A kind of dam safety comprehensive method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710598587.4A CN107480341B (en) | 2017-07-21 | 2017-07-21 | A kind of dam safety comprehensive method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107480341A true CN107480341A (en) | 2017-12-15 |
| CN107480341B CN107480341B (en) | 2018-10-23 |
Family
ID=60595335
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710598587.4A Active CN107480341B (en) | 2017-07-21 | 2017-07-21 | A kind of dam safety comprehensive method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107480341B (en) |
Cited By (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108549341A (en) * | 2018-04-19 | 2018-09-18 | 广州裕申电子科技有限公司 | Workshop Production approaches to IM, system and device based on Internet of Things |
| CN108563739A (en) * | 2018-04-11 | 2018-09-21 | 平安科技（深圳）有限公司 | Weather data acquisition methods and device, computer installation and readable storage medium storing program for executing |
| CN108764601A (en) * | 2018-04-03 | 2018-11-06 | 哈尔滨工业大学 | A kind of monitoring structural health conditions abnormal data diagnostic method based on computer vision and depth learning technology |
| CN108805192A (en) * | 2018-06-01 | 2018-11-13 | 河海大学 | Analysis on monitoring data method based on hierarchical network architecture |
| CN108984932A (en) * | 2018-07-25 | 2018-12-11 | 中国地震局地球物理研究所 | Building structure characteristic rapid extracting method and corresponding loss appraisal calculation method |
| CN109145257A (en) * | 2018-09-07 | 2019-01-04 | 国网湖南省电力有限公司 | A kind of dam safety monitoring analysis method based on the public abnormal time of multi-measuring point |
| CN109163997A (en) * | 2018-09-18 | 2019-01-08 | 天津大学 | A kind of rock surface strength detection method based on sonograph deep learning |
| CN109255555A (en) * | 2018-10-16 | 2019-01-22 | 中国电力科学研究院有限公司 | Electric power big data life period of equipment estimation method based on historical operational information |
| CN109272039A (en) * | 2018-09-19 | 2019-01-25 | 北京航空航天大学 | A kind of dam periphery method for monitoring abnormality and device based on unmanned plane |
| CN109284851A (en) * | 2018-06-11 | 2019-01-29 | 西安交通大学 | A kind of user power utilization behavior classification method suitable for Demand Side Response |
| CN109409406A (en) * | 2018-09-19 | 2019-03-01 | 北京航空航天大学 | A kind of the space base detection method and device of dam body abnormal conditions |
| CN109543746A (en) * | 2018-11-20 | 2019-03-29 | 河海大学 | A kind of sensor network Events Fusion and decision-making technique based on node reliability |
| CN109612423A (en) * | 2018-12-19 | 2019-04-12 | 沈阳天眼智云信息科技有限公司 | Tunnel collapsing prediction technique and system based on displacement sensor |
| CN110188399A (en) * | 2019-05-07 | 2019-08-30 | 河海大学 | A kind of dam safety monitoring list measuring point evaluation method based on more correlated series |
| CN110260914A (en) * | 2019-05-06 | 2019-09-20 | 河海大学 | A kind of project security monitoring system realm division methods based on measuring point space-time characteristic |
| CN110532694A (en) * | 2019-08-29 | 2019-12-03 | 四川大学 | A kind of arch dam safety evaluation method based on geomechanics model comprehensive method test |
| CN110595535A (en) * | 2019-08-19 | 2019-12-20 | 湖南强智科技发展有限公司 | Monitoring method, device and storage medium |
| CN110728397A (en) * | 2019-09-26 | 2020-01-24 | 河海大学 | Dam single-measuring-point threshold calculation method based on automatic model matching |
| CN110765679A (en) * | 2019-09-30 | 2020-02-07 | 国电南京自动化股份有限公司 | Dam monitoring web display method based on finite element model and SVM regression algorithm |
| CN111027680A (en) * | 2019-12-06 | 2020-04-17 | 北京瑞莱智慧科技有限公司 | Monitoring quantity uncertainty prediction method and system based on variational self-encoder |
| CN111090634A (en) * | 2019-11-06 | 2020-05-01 | 长江勘测规划设计研究有限责任公司 | Intelligent safety monitoring data compilation analysis system based on cloud service |
| CN111105097A (en) * | 2019-12-25 | 2020-05-05 | 河海大学常州校区 | Dam deformation prediction system and method based on convolutional neural network |
| CN111121711A (en) * | 2019-12-24 | 2020-05-08 | 河南理工大学 | Method and device for acquiring movement deformation of missing monitoring point in mining influence area |
| CN111177218A (en) * | 2019-12-25 | 2020-05-19 | 深圳市东深电子股份有限公司 | Dam safety analysis method based on big data analysis |
| CN111860202A (en) * | 2020-06-28 | 2020-10-30 | 中铁大桥科学研究院有限公司 | Beam yard pedestal state identification method and system combining image identification and intelligent equipment |
| CN112381309A (en) * | 2020-11-23 | 2021-02-19 | 珠江水利委员会珠江水利科学研究院 | Reservoir dam safety monitoring and early warning method, device and system and storage medium |
| CN112434750A (en) * | 2020-12-04 | 2021-03-02 | 中国电建集团华东勘测设计研究院有限公司 | Method for identifying dam monitoring data development pattern based on convolutional neural network |
| CN112487356A (en) * | 2020-11-30 | 2021-03-12 | 北京航空航天大学 | Structural health monitoring data enhancement method |
| CN112580682A (en) * | 2020-11-16 | 2021-03-30 | 河海大学 | Dam monitoring system missing data reconstruction method based on multitask Gaussian process |
| CN112598287A (en) * | 2020-12-25 | 2021-04-02 | 武汉烽火信息集成技术有限公司 | Rail transit health assessment method and system |
| CN112766511A (en) * | 2019-11-01 | 2021-05-07 | 伊姆西Ip控股有限责任公司 | Method, apparatus and computer program product for model adaptation |
| CN112949900A (en) * | 2021-01-18 | 2021-06-11 | 水利部交通运输部国家能源局南京水利科学研究院 | Reservoir dam safety information intelligent perception fusion early warning method and terminal equipment |
| CN112966856A (en) * | 2021-02-10 | 2021-06-15 | 四川水利职业技术学院 | Mountain torrent risk prediction method and prediction system |
| CN112990372A (en) * | 2021-04-27 | 2021-06-18 | 北京瑞莱智慧科技有限公司 | Data processing method, model training device and electronic equipment |
| CN113033702A (en) * | 2021-04-21 | 2021-06-25 | 上海勘察设计研究院(集团)有限公司 | Structure safety early warning method based on space-time data |
| CN113298426A (en) * | 2021-06-17 | 2021-08-24 | 华能澜沧江水电股份有限公司 | Knowledge graph driven dam safety evaluation weight dynamic drafting method and system |
| CN113344470A (en) * | 2021-08-02 | 2021-09-03 | 山东炎黄工业设计有限公司 | Intelligent power supply system management method based on block chain |
| CN113554849A (en) * | 2021-08-09 | 2021-10-26 | 广东智云工程科技有限公司 | Air-ground monitoring slope system and instability risk assessment method thereof |
| CN113591668A (en) * | 2021-07-26 | 2021-11-02 | 南京大学 | Wide-area unknown dam automatic detection method using deep learning and spatial analysis |
| CN113807570A (en) * | 2021-08-12 | 2021-12-17 | 水利部南京水利水文自动化研究所 | Reservoir dam risk level evaluation method and system based on XGboost |
| CN114611833A (en) * | 2022-03-30 | 2022-06-10 | 中国华能集团有限公司江西分公司 | Dam body deep learning model construction method based on dual-drive combination |
| CN114881074A (en) * | 2022-04-24 | 2022-08-09 | 同济大学 | Engineering structure deformation prediction method based on clustering-deep learning |
| CN116628473A (en) * | 2023-05-17 | 2023-08-22 | 国网上海市电力公司 | Power equipment state trend prediction method based on multi-factor neural network algorithm |
| CN117078627A (en) * | 2023-08-18 | 2023-11-17 | 中国水利水电科学研究院 | Method and system for monitoring and safely evaluating defects of dam body of silt dam |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070162241A1 (en) * | 2005-12-01 | 2007-07-12 | Siemens Corporate Research, Inc. | Robust Sensor Correlation Analysis For Machine Condition Monitoring |
| CN103410516A (en) * | 2013-05-08 | 2013-11-27 | 浙江广川工程咨询有限公司 | Surrounding rock deformation early-warning construction method for small-diameter tunnel full-section excavation |
| CN104281130A (en) * | 2014-09-22 | 2015-01-14 | 国家电网公司 | Hydroelectric equipment monitoring and fault diagnosis system based on big data technology |
-
2017
- 2017-07-21 CN CN201710598587.4A patent/CN107480341B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070162241A1 (en) * | 2005-12-01 | 2007-07-12 | Siemens Corporate Research, Inc. | Robust Sensor Correlation Analysis For Machine Condition Monitoring |
| CN103410516A (en) * | 2013-05-08 | 2013-11-27 | 浙江广川工程咨询有限公司 | Surrounding rock deformation early-warning construction method for small-diameter tunnel full-section excavation |
| CN104281130A (en) * | 2014-09-22 | 2015-01-14 | 国家电网公司 | Hydroelectric equipment monitoring and fault diagnosis system based on big data technology |
Non-Patent Citations (2)
| Title |
|---|
| 张晓春 等: "径向基函数神经网络在大坝安全监测数据处理中的应用", 《武汉大学学报(工学版)》 * |
| 赵鲲鹏 等: "ABC-RBF神经网络在平板坝裂缝开合度监测中的应用", 《水利水电技术》 * |
Cited By (65)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108764601A (en) * | 2018-04-03 | 2018-11-06 | 哈尔滨工业大学 | A kind of monitoring structural health conditions abnormal data diagnostic method based on computer vision and depth learning technology |
| CN108563739A (en) * | 2018-04-11 | 2018-09-21 | 平安科技（深圳）有限公司 | Weather data acquisition methods and device, computer installation and readable storage medium storing program for executing |
| CN108563739B (en) * | 2018-04-11 | 2021-05-04 | 平安科技（深圳）有限公司 | Weather data acquisition method and device, computer device and readable storage medium |
| WO2019196278A1 (en) * | 2018-04-11 | 2019-10-17 | 平安科技（深圳）有限公司 | Weather data acquisition method and apparatus, computer apparatus and readable storage medium |
| CN108549341A (en) * | 2018-04-19 | 2018-09-18 | 广州裕申电子科技有限公司 | Workshop Production approaches to IM, system and device based on Internet of Things |
| CN108805192A (en) * | 2018-06-01 | 2018-11-13 | 河海大学 | Analysis on monitoring data method based on hierarchical network architecture |
| CN108805192B (en) * | 2018-06-01 | 2019-04-02 | 河海大学 | Analysis on monitoring data method based on hierarchical network architecture |
| CN109284851A (en) * | 2018-06-11 | 2019-01-29 | 西安交通大学 | A kind of user power utilization behavior classification method suitable for Demand Side Response |
| CN108984932A (en) * | 2018-07-25 | 2018-12-11 | 中国地震局地球物理研究所 | Building structure characteristic rapid extracting method and corresponding loss appraisal calculation method |
| CN108984932B (en) * | 2018-07-25 | 2022-11-08 | 中国地震局地球物理研究所 | Building structure characteristic rapid extraction method and corresponding loss evaluation calculation method |
| CN109145257A (en) * | 2018-09-07 | 2019-01-04 | 国网湖南省电力有限公司 | A kind of dam safety monitoring analysis method based on the public abnormal time of multi-measuring point |
| CN109145257B (en) * | 2018-09-07 | 2022-07-05 | 国网湖南省电力有限公司 | Dam safety monitoring analysis method based on multi-measuring-point public abnormal time |
| CN109163997A (en) * | 2018-09-18 | 2019-01-08 | 天津大学 | A kind of rock surface strength detection method based on sonograph deep learning |
| CN109272039A (en) * | 2018-09-19 | 2019-01-25 | 北京航空航天大学 | A kind of dam periphery method for monitoring abnormality and device based on unmanned plane |
| CN109409406A (en) * | 2018-09-19 | 2019-03-01 | 北京航空航天大学 | A kind of the space base detection method and device of dam body abnormal conditions |
| CN109255555A (en) * | 2018-10-16 | 2019-01-22 | 中国电力科学研究院有限公司 | Electric power big data life period of equipment estimation method based on historical operational information |
| CN109255555B (en) * | 2018-10-16 | 2023-10-27 | 中国电力科学研究院有限公司 | Electric power big data equipment life cycle estimation method based on historical operation information |
| CN109543746A (en) * | 2018-11-20 | 2019-03-29 | 河海大学 | A kind of sensor network Events Fusion and decision-making technique based on node reliability |
| CN109612423A (en) * | 2018-12-19 | 2019-04-12 | 沈阳天眼智云信息科技有限公司 | Tunnel collapsing prediction technique and system based on displacement sensor |
| CN110260914A (en) * | 2019-05-06 | 2019-09-20 | 河海大学 | A kind of project security monitoring system realm division methods based on measuring point space-time characteristic |
| CN110260914B (en) * | 2019-05-06 | 2020-06-19 | 河海大学 | Engineering safety monitoring system region division method based on measuring point space-time characteristics |
| CN110188399A (en) * | 2019-05-07 | 2019-08-30 | 河海大学 | A kind of dam safety monitoring list measuring point evaluation method based on more correlated series |
| CN110188399B (en) * | 2019-05-07 | 2020-10-16 | 河海大学 | Dam safety monitoring single-measuring-point evaluation method based on multiple correlation sequences |
| CN110595535A (en) * | 2019-08-19 | 2019-12-20 | 湖南强智科技发展有限公司 | Monitoring method, device and storage medium |
| CN110532694A (en) * | 2019-08-29 | 2019-12-03 | 四川大学 | A kind of arch dam safety evaluation method based on geomechanics model comprehensive method test |
| CN110532694B (en) * | 2019-08-29 | 2023-02-28 | 四川大学 | Arch dam safety evaluation method based on geomechanical model comprehensive method test |
| CN110728397A (en) * | 2019-09-26 | 2020-01-24 | 河海大学 | Dam single-measuring-point threshold calculation method based on automatic model matching |
| CN110728397B (en) * | 2019-09-26 | 2021-03-02 | 河海大学 | Dam single-measuring-point threshold calculation method based on automatic model matching |
| CN110765679A (en) * | 2019-09-30 | 2020-02-07 | 国电南京自动化股份有限公司 | Dam monitoring web display method based on finite element model and SVM regression algorithm |
| CN110765679B (en) * | 2019-09-30 | 2023-07-07 | 国电南京自动化股份有限公司 | Dam monitoring web display method based on finite element model and SVM regression algorithm |
| CN112766511A (en) * | 2019-11-01 | 2021-05-07 | 伊姆西Ip控股有限责任公司 | Method, apparatus and computer program product for model adaptation |
| CN111090634B (en) * | 2019-11-06 | 2023-01-24 | 长江勘测规划设计研究有限责任公司 | Intelligent safety monitoring data compilation analysis system based on cloud service |
| CN111090634A (en) * | 2019-11-06 | 2020-05-01 | 长江勘测规划设计研究有限责任公司 | Intelligent safety monitoring data compilation analysis system based on cloud service |
| CN111027680A (en) * | 2019-12-06 | 2020-04-17 | 北京瑞莱智慧科技有限公司 | Monitoring quantity uncertainty prediction method and system based on variational self-encoder |
| CN111121711A (en) * | 2019-12-24 | 2020-05-08 | 河南理工大学 | Method and device for acquiring movement deformation of missing monitoring point in mining influence area |
| CN111121711B (en) * | 2019-12-24 | 2022-01-18 | 河南理工大学 | Method and device for acquiring movement deformation of missing monitoring point in mining influence area |
| CN111177218A (en) * | 2019-12-25 | 2020-05-19 | 深圳市东深电子股份有限公司 | Dam safety analysis method based on big data analysis |
| CN111105097A (en) * | 2019-12-25 | 2020-05-05 | 河海大学常州校区 | Dam deformation prediction system and method based on convolutional neural network |
| CN111177218B (en) * | 2019-12-25 | 2022-08-30 | 深圳市东深电子股份有限公司 | Dam safety analysis method based on big data analysis |
| CN111860202A (en) * | 2020-06-28 | 2020-10-30 | 中铁大桥科学研究院有限公司 | Beam yard pedestal state identification method and system combining image identification and intelligent equipment |
| CN112580682A (en) * | 2020-11-16 | 2021-03-30 | 河海大学 | Dam monitoring system missing data reconstruction method based on multitask Gaussian process |
| CN112381309A (en) * | 2020-11-23 | 2021-02-19 | 珠江水利委员会珠江水利科学研究院 | Reservoir dam safety monitoring and early warning method, device and system and storage medium |
| CN112487356A (en) * | 2020-11-30 | 2021-03-12 | 北京航空航天大学 | Structural health monitoring data enhancement method |
| CN112434750A (en) * | 2020-12-04 | 2021-03-02 | 中国电建集团华东勘测设计研究院有限公司 | Method for identifying dam monitoring data development pattern based on convolutional neural network |
| CN112598287A (en) * | 2020-12-25 | 2021-04-02 | 武汉烽火信息集成技术有限公司 | Rail transit health assessment method and system |
| CN112949900A (en) * | 2021-01-18 | 2021-06-11 | 水利部交通运输部国家能源局南京水利科学研究院 | Reservoir dam safety information intelligent perception fusion early warning method and terminal equipment |
| CN112949900B (en) * | 2021-01-18 | 2022-01-11 | 水利部交通运输部国家能源局南京水利科学研究院 | Reservoir dam safety information intelligent perception fusion early warning method and terminal equipment |
| CN112966856A (en) * | 2021-02-10 | 2021-06-15 | 四川水利职业技术学院 | Mountain torrent risk prediction method and prediction system |
| CN113033702A (en) * | 2021-04-21 | 2021-06-25 | 上海勘察设计研究院(集团)有限公司 | Structure safety early warning method based on space-time data |
| CN113033702B (en) * | 2021-04-21 | 2022-12-13 | 上海勘察设计研究院(集团)有限公司 | Structure safety early warning method based on space-time data |
| CN112990372A (en) * | 2021-04-27 | 2021-06-18 | 北京瑞莱智慧科技有限公司 | Data processing method, model training device and electronic equipment |
| CN113298426A (en) * | 2021-06-17 | 2021-08-24 | 华能澜沧江水电股份有限公司 | Knowledge graph driven dam safety evaluation weight dynamic drafting method and system |
| CN113591668A (en) * | 2021-07-26 | 2021-11-02 | 南京大学 | Wide-area unknown dam automatic detection method using deep learning and spatial analysis |
| CN113591668B (en) * | 2021-07-26 | 2023-11-21 | 南京大学 | Wide area unknown dam automatic detection method using deep learning and space analysis |
| CN113344470A (en) * | 2021-08-02 | 2021-09-03 | 山东炎黄工业设计有限公司 | Intelligent power supply system management method based on block chain |
| CN113344470B (en) * | 2021-08-02 | 2021-11-16 | 山东炎黄工业设计有限公司 | Intelligent power supply system management method based on block chain |
| CN113554849A (en) * | 2021-08-09 | 2021-10-26 | 广东智云工程科技有限公司 | Air-ground monitoring slope system and instability risk assessment method thereof |
| CN113807570A (en) * | 2021-08-12 | 2021-12-17 | 水利部南京水利水文自动化研究所 | Reservoir dam risk level evaluation method and system based on XGboost |
| CN113807570B (en) * | 2021-08-12 | 2024-02-02 | 水利部南京水利水文自动化研究所 | XGBoost-based reservoir dam risk level assessment method and system |
| CN114611833A (en) * | 2022-03-30 | 2022-06-10 | 中国华能集团有限公司江西分公司 | Dam body deep learning model construction method based on dual-drive combination |
| CN114881074A (en) * | 2022-04-24 | 2022-08-09 | 同济大学 | Engineering structure deformation prediction method based on clustering-deep learning |
| CN114881074B (en) * | 2022-04-24 | 2023-08-08 | 同济大学 | Engineering structure deformation prediction method based on clustering-deep learning |
| CN116628473A (en) * | 2023-05-17 | 2023-08-22 | 国网上海市电力公司 | Power equipment state trend prediction method based on multi-factor neural network algorithm |
| CN117078627A (en) * | 2023-08-18 | 2023-11-17 | 中国水利水电科学研究院 | Method and system for monitoring and safely evaluating defects of dam body of silt dam |
| CN117078627B (en) * | 2023-08-18 | 2024-02-27 | 中国水利水电科学研究院 | Method and system for monitoring and safely evaluating defects of dam body of silt dam |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107480341B (en) | 2018-10-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107480341B (en) | A kind of dam safety comprehensive method based on deep learning | |
| Kisi et al. | Modeling discharge–sediment relationship using neural networks with artificial bee colony algorithm | |
| CN104063612B (en) | A kind of Tunnel Engineering risk profiles fuzzy evaluation method and assessment system | |
| CN108875161A (en) | Flow grade prediction technique based on convolutional neural networks deep learning | |
| CN108009674A (en) | Air PM2.5 concentration prediction methods based on CNN and LSTM fused neural networks | |
| CN104834977B (en) | Traffic alert grade prediction technique based on learning distance metric | |
| CN109242149A (en) | A kind of student performance early warning method and system excavated based on educational data | |
| CN106250667A (en) | The monitoring method of a kind of landslide transition between states of paddling and device | |
| Jalalkamali et al. | Groundwater modeling using hybrid of artificial neural network with genetic algorithm | |
| CN106408120B (en) | Local area landslide prediction device and method | |
| CN111797364B (en) | Landslide multilayer safety evaluation method based on composite cloud model | |
| CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
| CN107295537A (en) | A kind of method and system for wireless sensor network reliability of testing and assessing | |
| CN107918837A (en) | A kind of fruit or vegetable type food security risk Forecasting Methodology | |
| CN107798431A (en) | A kind of Medium-and Long-Term Runoff Forecasting method based on Modified Elman Neural Network | |
| CN110321493A (en) | A kind of abnormality detection of social networks and optimization method, system and computer equipment | |
| CN109491709B (en) | Software code controllability comprehensive evaluation method based on AHP and neural network | |
| CN106780053A (en) | A kind of overlap community discovery method and system based on node connection preference | |
| CN112434887B (en) | Water supply network risk prediction method combining network kernel density estimation and SVM | |
| CN111582634B (en) | Multi-factor safety grading method and system for underground large-space construction | |
| TWI587222B (en) | Neural networks based water level predicting system and method for reservoirs | |
| Khadr et al. | Data-driven stochastic modeling for multi-purpose reservoir simulation | |
| CN116402951A (en) | Digital twinning-based side slope safety management method and system | |
| CN113269384B (en) | Method for early warning health state of river system | |
| CN115330526A (en) | Enterprise credit scoring method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20171215 Assignee: HUANENG LANCANG RIVER HYDROPOWER Inc. Assignor: HOHAI University Contract record no.: X2021320000034 Denomination of invention: A comprehensive evaluation method for dam safety based on deep learning Granted publication date: 20181023 License type: Common License Record date: 20210525 |
|
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Qi Hai Inventor after: Mao Yingchi Inventor after: Chen Hao Inventor after: Li Zhitao Inventor after: Wang Longbao Inventor before: Mao Yingchi Inventor before: Qi Hai Inventor before: Chen Hao Inventor before: Li Zhitao Inventor before: Wang Longbao |