CN106570563A - Deformation prediction method and apparatus based on Kalman filtering and BP neural network - Google Patents
Deformation prediction method and apparatus based on Kalman filtering and BP neural network Download PDFInfo
- Publication number
- CN106570563A CN106570563A CN201510671512.5A CN201510671512A CN106570563A CN 106570563 A CN106570563 A CN 106570563A CN 201510671512 A CN201510671512 A CN 201510671512A CN 106570563 A CN106570563 A CN 106570563A
- Authority
- CN
- China
- Prior art keywords
- neural network
- deformation
- kalman filtering
- model
- study
- 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.)
- Pending
Links
Classifications
-
- 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
Abstract
The invention provides a deformation prediction method and apparatus based on Kalman filtering and a BP neural network, wherein the deformation prediction method based on the Kalman filtering and the BP neural network comprises: obtaining the deformation monitoring data of a monitored object in a project as a training sample; establishing a BP neural network topology model according to the training sample; learning the BP neural network topology model by using a Kalman filtering algorithm according to preset training parameters to adjust the weights of the neurons in the BP neural network topology model; and performing deformation prediction according to the BP neural network topology model with adjusted weights. The deformation prediction method based on Kalman filtering and a BP neural network can shorten the learning time of the BP neural network and improve the establishment efficiency of deformation prediction model in a deformation prediction process.
Description
Technical field
The present invention relates to dynamical system Deformation Prediction field, more particularly to it is a kind of based on Kalman filtering and anti-
To the Deformation Prediction method and apparatus for propagating (Back Propagation, abbreviation BP) neutral net.
Background technology
Modern project building during construction and operation, due to by the natural activity (example of human development
Such as groundwater abstraction, oil recovery, mining) earth's crust that causes and earth's surface deformation, or engineering works base
Plinth designs unreasonable, basement process not science, or the impact of the factor such as natural disaster, easily produces change
Shape, if deformation just influences whether normally using for engineering works beyond the limit of regulation.So,
The highly important impact for engineering construction has of deformation monitoring and Deformation Prediction.Generally, with deformation monitoring
Based on data, prediction model of deformation is set up by various mathematical statistics methods, so as to pre- according to the deformation
Surveying model carries out Deformation Prediction analysis.
In the prior art, BP neural network algorithm has obtained more and more extensive in Deformation Prediction analysis
Using.BP neural network is a kind of Multi-layered Feedforward Networks trained by Back Propagation Algorithm, and BP is neural
The topological structure of network includes input layer, hidden layer and output layer, and per layer includes multiple neurons, and BP is refreshing
The learning process of Jing networks is divided into two stages:1st stage was forward-propagating process, i.e. according to input
Layer information calculates successively the output of each neuron of hidden layer, output layer;2nd stage was back-propagation process,
That is, if the reality output of output layer is more than predetermined threshold value with the difference of desired output, output is reversely adjusted
Layer, the weights of each neuron of hidden layer.The two processes are used repeatedly so that error tapers into,
When error reaches desirable requirement, the learning process of BP neural network terminates, it is possible to use should
BP neural network model carries out Deformation Prediction analysis as prediction model of deformation.
But, because the learning rate of BP neural network is fixed, therefore, BP neural network algorithm
Convergence rate it is slower, need longer learning time, for some challenges, BP neural network is calculated
The learning time that method needs will be very long, have a strong impact on building for prediction model of deformation during Deformation Prediction
Render meritorious service rate.
The content of the invention
The invention provides a kind of based on Kalman filtering and the Deformation Prediction method and dress of BP neural network
Put, the learning time of BP neural network can be reduced, improve Deformation Prediction mould during Deformation Prediction
Type sets up efficiency.
The present invention provide based on Kalman filtering and the Deformation Prediction method of BP neural network, including:
The deformation measurement data of monitoring object in engineering is obtained as training sample;
BP neural network topological model is set up according to the training sample;
The BP neural network topological model is carried out according to default training parameter using Kalman filtering algorithm
Study, adjust the weights of each neuron in the BP neural network topological model;
Deformation Prediction is carried out according to the BP neural network topological model after adjustment weights.
The present invention provide based on Kalman filtering and the Deformation Prediction device of BP neural network, including:
Acquisition module, for obtaining engineering in monitoring object deformation measurement data as training sample;
Model building module, for setting up BP neural network topological model according to the training sample;
Study module, it is refreshing for carrying out the BP according to default training parameter using Kalman filtering algorithm
The study of Jing network topology models, adjusts the weights of each neuron in the BP neural network topological model;
Prediction module, for being deformed according to the BP neural network topological model after adjustment weights
Prediction.
The invention provides a kind of based on Kalman filtering and the Deformation Prediction method and dress of BP neural network
Put, wherein, included based on the Deformation Prediction method of Kalman filtering and BP neural network:Obtain engineering
The deformation measurement data of middle monitoring object sets up BP neural network as training sample according to training sample
Topological model, BP neural network Top Modules are carried out using Kalman filtering algorithm according to default training parameter
The study of type, adjusts the weights of each neuron in BP neural network topological model, after adjustment weights
BP neural network topological model carry out Deformation Prediction.The Deformation Prediction method that the present invention is provided, passes through
Using Kalman filtering algorithm as BP neural network training algorithm, by the weighed value adjusting of BP neural network
Process is converted into the optimal estimation problem of weights, shortens the learning time of BP neural network, improves
Prediction model of deformation sets up efficiency during Deformation Prediction.
Description of the drawings
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to reality
Apply the accompanying drawing to be used needed for example or description of the prior art to be briefly described, it should be apparent that, under
Accompanying drawing in the description of face is some embodiments of the present invention, for those of ordinary skill in the art,
On the premise of not paying creative labor, can be with according to these other accompanying drawings of accompanying drawings acquisition.
Fig. 1 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention one is provided is pre-
The flow chart of survey method;
Fig. 2 is the BP neural network topological model based on tunnel deformation that the embodiment of the present invention one is provided
Structural representation;
Fig. 3 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention two is provided is pre-
The flow chart of survey method;
Fig. 4 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention three is provided is pre-
The flow chart of survey method;
Fig. 5 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention one is provided is pre-
Survey the structural representation of device;
Fig. 6 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention two is provided is pre-
Survey the structural representation of device.
Specific embodiment
To make purpose, technical scheme and the advantage of the embodiment of the present invention clearer, below in conjunction with this
Accompanying drawing in bright embodiment, is clearly and completely described to the technical scheme in the embodiment of the present invention,
Obviously, described embodiment is a part of embodiment of the invention, rather than the embodiment of whole.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained under the premise of creative work is not made
The every other embodiment for obtaining, belongs to the scope of protection of the invention.
Fig. 1 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention one is provided is pre-
The flow chart of survey method.As shown in figure 1, the present embodiment offer is neural based on Kalman filtering and BP
The Deformation Prediction method of network, can include:
The deformation measurement data of monitoring object is used as training sample in step 101, acquisition engineering.
Wherein, monitoring object can be any dynamical system, and the present embodiment is not specially limited, for example:
Monitoring object can be tunnel, dam, roadbed, deep basal pit, the sinking land in coalmining areas etc..Monitoring object is not
Together, its corresponding deformation measurement data is also different, for example, for tunnel deformation, affects tunnel to enclose
The factor of rock deformation is usually:Air pressure, tunnel stress, ambient temperature and time in tunnel, and tunnel
The result of road surrouding rock deformation is mainly reflected in three-D displacement deflection.
In this step, deformation measurement data specifically include affect monitoring object deformation each factor data with
And the deformation result data of monitoring object deformation, each factor data and deformation result data are by prison
Survey the given data for monitoring acquisition on the spot of object, and using each factor data and deformation result data as
Training sample.
Step 103, BP neural network topological model is set up according to training sample.
Wherein, the topological model of BP neural network is set up, specially:Determine BP neural network Top Modules
Input layer, the neuron number of each hidden layer and output layer and input-output mappings relation in type, prison
Survey object different, deformation measurement data is different, then the structure of BP neural network topological model is also different.
For example, by taking tunnel deformation as an example, Fig. 2 is the offer of the embodiment of the present invention one based on tunnel deformation
The structural representation of BP neural network topological model, as shown in Fig. 2 each of tunnel wall rock deformation will be affected
Input layer of the factor as BP neural network topological model, i.e. input layer has 4 neurons, input
Vector is X=[x1, x2, x3, x4], wherein, it is tunnel for air pressure, x3 in tunnel that x1 is ambient temperature, x2
Stress, x4 are the monitoring time;Using the result of tunnel wall rock deformation as BP neural network topological model
Output layer, i.e. output layer has 3 neurons, output vector is Z=[△ x, △ y, △ h], wherein, △ x
Add up deflection for X-direction for the accumulative deflection of X-direction, △ h for the accumulative deflection of X-direction, △ y,
Hidden layer has 5 neurons.
Step 105, using Kalman filtering algorithm according to default training parameter carry out BP neural network topology
The study of model, adjusts the weights of each neuron in BP neural network topological model.
In this step, it is Kalman filtering algorithm is refreshing as the training algorithm adjustment BP of BP neural network
The weights of each neuron in Jing network topology models, the weighed value adjusting process of BP neural network is converted into
The optimal estimation problem of weights, specifically, by the weights of each neuron in BP neural network topological model
As state vector, using the output of BP neural network as observation vector, estimated using observation vector
State vector, so as to adjust the weights of BP neural network.
Because Kalman filtering algorithm is the recursive process of constantly prediction, an amendment, it using observation to
The state vector for measuring to estimate to be continually changing with the time, by merging each observational equation state equation is obtained,
And, Kalman filtering algorithm is when estimating state vector, it is not necessary to store the substantial amounts of conception of history
Data are surveyed, when new observation data are obtained, new parametric filtering value can be at any time calculated, is easy in real time
Ground processes observed result, so, compared with existing BP neural network training algorithm, by Kalman
Filtering algorithm can shorten the learning time of BP neural network as the training algorithm of BP neural network,
Target error value can be just converged to through less frequency of training, so that it is determined that the power of BP neural network
Value, and then Deformation Prediction analysis can be carried out using the BP neural network as prediction model of deformation, improve
Prediction model of deformation sets up efficiency during Deformation Prediction.And, by by Kalman filtering algorithm
As the training algorithm of BP neural network, the local that traditional BP neural network easily occurs can also be solved
Minimum problem.
Optionally, the state equation of Kalman filtering algorithm is Wk+1=FkWk+ωk, Kalman filtering algorithm
Observational equation be yk+1=Hk+1(Wk+1,uk+1)+vk+1。
Wherein, WkFor the weights of each neuron in BP neural network topological model, FkSquare is shifted for state
Battle array, ωkFor process noise matrix;
Wherein, yk+1For the output vector of BP neural network topological model, Hk+1For observing matrix, uk+1For
The input vector of BP neural network topological model, vk+1For measurement noise matrix.
Optionally, default training parameter can include:Maximum frequency of training, global error desired value and
Practise speed.Wherein, the concrete numerical value for presetting training parameter is configured as needed.Preferably, it is maximum
Frequency of training is 10000, and global error desired value is 1 × 10-3, learning rate is 0.02.
Step 107, according to adjustment weights after BP neural network topological model carry out Deformation Prediction.
Since it is determined the weights of BP neural network, and then can be pre- using the BP neural network as deformation
Surveying model carries out Deformation Prediction analysis.
Present embodiments provide it is a kind of based on Kalman filtering and the Deformation Prediction method of BP neural network,
Including:The deformation measurement data of monitoring object in engineering is obtained as training sample, is built according to training sample
Vertical BP neural network topological model, carries out BP refreshing using Kalman filtering algorithm according to default training parameter
The study of Jing network topology models, adjusts the weights of each neuron in BP neural network topological model, root
Deformation Prediction is carried out according to the BP neural network topological model after adjustment weights.The deformation that the present embodiment is provided
Forecasting Methodology, by using Kalman filtering algorithm as BP neural network training algorithm, BP is neural
The weighed value adjusting process at networking is converted into the optimal estimation problem of weights, shortens of BP neural network
The habit time, improve prediction model of deformation during Deformation Prediction sets up efficiency.
Fig. 3 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention two is provided is pre-
The flow chart of survey method, the present embodiment is on the basis of embodiment one, there is provided based on Kalman filtering and
The another embodiment of the Deformation Prediction method of BP neural network.As shown in figure 3, the present embodiment is provided
Deformation Prediction method, can include:
The deformation measurement data of monitoring object is used as training sample in step 201, acquisition engineering.
Step 203, BP neural network topological model is set up according to training sample.
Step 205, training sample is divided into steady deformation tendency data acquisition system, gradual change deformation tendency data
Set, mutation deformation tendency data acquisition system and high frequency superposition long period deformation tendency data acquisition system.
In this step, training sample is divided into into 4 data acquisition systems according to the difference of deformation tendency, is deformed
Trend is divided into:" leveling style " deformation tendency, " gradation type " deformation tendency, " saltant type " deformation become
Gesture and " high frequency superposition long period " deformation tendency.
Wherein, " leveling style " deformation tendency is usually monitoring object in operation mid-term, in this time period
Interior, larger deformation has been completed, and convergence state has been reached substantially.
" gradation type " deformation tendency is usually monitoring object and is in construction time or operation initial stage, at this moment
Between in section, monitoring object not yet reaches poised state due to the impact of construction factor, engineering mechanics performance,
Its deformation rule is usually expressed as:Initial deformation speed is very fast, and middle and late stage progressively slows down and be finally reached phase
To steady statue.
" saltant type " deformation tendency is usually monitoring object is affected by accidentalia, such as earthquake, flood
Water etc., its deformation rule is usually expressed as:Initial stage is basicly stable, and the mid-term time period occurs suddenly big
Breaking type deforms, and subsequently progressively tends towards stability again.
" high frequency superposition long period " deformation tendency probability of occurrence is less, comes across dam, high-rise and super more
In high-rise engineering works, its deformation rule is usually expressed as:Ambient parameter is (for example:Water level, wind-force,
Temperature etc.) regular vibration is presented.
Step 207, using Kalman filtering algorithm according to default training parameter carry out BP neural network topology
The study of model, adjusts the weights of each neuron in BP neural network topological model.
In this step, the instruction of BP neural network is carried out respectively according to the data acquisition system of different distortion trend
Practice, adjust the weights of each neuron in BP neural network topological model, generate be applied to different changes respectively
The prediction model of deformation of shape trend.
Specifically, according to steady deformation tendency data acquisition system, using Kalman filtering algorithm according to default instruction
Practicing parameter carries out the study of BP neural network, obtains steady prediction model of deformation;Or, according to gradual change
Deformation tendency data acquisition system, BP nerve net is carried out using Kalman filtering algorithm according to default training parameter
The study of network, obtains gradual change prediction model of deformation;Or, according to mutation deformation tendency data acquisition system, adopt
The study of BP neural network is carried out according to default training parameter with Kalman filtering algorithm, mutation is obtained and is become
Shape forecast model;Or, long period deformation tendency data acquisition system is superimposed according to high frequency, filtered using Kalman
Ripple algorithm carries out the study of BP neural network according to default training parameter, obtains high frequency superposition long period and becomes
Shape forecast model.
Step 209, according to adjustment weights after BP neural network topological model carry out Deformation Prediction.
Present embodiments provide it is a kind of based on Kalman filtering and the Deformation Prediction method of BP neural network,
Including:The deformation measurement data of monitoring object in engineering is obtained as training sample, is built according to training sample
Vertical BP neural network topological model, is divided into training sample steady deformation tendency data acquisition system, gradual change and becomes
Shape trend data set, mutation deformation tendency data acquisition system and high frequency superposition long period deformation tendency data set
Close, of BP neural network topological model is carried out according to default training parameter using Kalman filtering algorithm
Practise, adjust the weights of each neuron in BP neural network topological model, it is refreshing according to the BP after adjustment weights
Jing network topology models carry out Deformation Prediction.The Deformation Prediction method that the present embodiment is provided, by training
Sequence is divided into different data acquisition systems according to different deformation tendencies, for each deformation tendency data set
Close by using Kalman filtering algorithm as BP neural network training algorithm, by the power of BP neural network
Value adjustment process is converted into the optimal estimation problem of weights, shortens the learning time of BP neural network,
Improve prediction model of deformation during Deformation Prediction sets up efficiency.
Fig. 4 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention three is provided is pre-
The flow chart of survey method, the present embodiment is on the basis of embodiment one and embodiment two, there is provided based on card
Another embodiment of the Deformation Prediction method of Kalman Filtering and BP neural network.As shown in figure 4,
The Deformation Prediction method that the present embodiment is provided, can include:
The deformation measurement data of monitoring object is used as training sample in step 301, acquisition engineering.
In this step, deformation measurement data not only includes training sample, also including test sample.Wherein,
Training sample is used for the study of BP neural network, and test sample is used to evaluate the BP neural network as change
Shape forecast model carries out the effect of Deformation Prediction analysis.
Deformation measurement data is divided into into training sample and test sample can various implementations, this enforcement
Example is not any limitation as, for example:Deformation measurement data is certain 1 year deformation measurement data in monitoring point, then may be used
With sequentially in time, using the deformation measurement data in the monitoring point January~November as training sample, use
In the study of BP neural network, and using the deformation measurement data in the monitoring point December as test sample,
The effect of Deformation Prediction analysis is carried out as prediction model of deformation for evaluating the BP neural network.
Step 303, BP neural network topological model is set up according to training sample.
Step 305, using Kalman filtering algorithm according to default training parameter carry out BP neural network topology
The study of model, adjusts the weights of each neuron in BP neural network topological model.
Step 307, according to adjustment weights after BP neural network topological model carry out Deformation Prediction.
Step 309, BP neural network topological model is tested according to test sample.
Present embodiments provide it is a kind of based on Kalman filtering and the Deformation Prediction method of BP neural network,
Including:The deformation measurement data of monitoring object in engineering is obtained as training sample, is built according to training sample
Vertical BP neural network topological model, carries out BP refreshing using Kalman filtering algorithm according to default training parameter
The study of Jing network topology models, adjusts the weights of each neuron in BP neural network topological model, root
Deformation Prediction is carried out according to the BP neural network topological model after adjustment weights, it is refreshing to BP according to test sample
Jing network topology models are tested.The Deformation Prediction method that the present embodiment is provided, by the way that Kalman is filtered
Ripple algorithm is converted into the weighed value adjusting process of BP neural network as the training algorithm of BP neural network
The optimal estimation problem of weights, shortens the learning time of BP neural network, improves Deformation Prediction mistake
Prediction model of deformation sets up efficiency in journey.
Fig. 5 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention one is provided is pre-
Survey the structural representation of device, the change based on Kalman filtering and BP neural network that the present embodiment is provided
Shape prediction meanss, for perform embodiment illustrated in fig. 1 offer based on Kalman filtering and BP nerve net
The Deformation Prediction method of network.As shown in figure 5, the present embodiment offer is refreshing based on Kalman filtering and BP
The Deformation Prediction device of Jing networks, can include:
Acquisition module 11, for obtaining engineering in monitoring object deformation measurement data as training sample.
Model building module 13, for setting up BP neural network topological model according to training sample.
Study module 15, it is neural for carrying out BP according to default training parameter using Kalman filtering algorithm
The study of network topology model, adjusts the weights of each neuron in BP neural network topological model.
Prediction module 17, it is pre- for carrying out deforming according to the BP neural network topological model after adjustment weights
Survey.
Optionally, default training parameter can include:Maximum frequency of training, global error desired value and
Practise speed.Wherein, the concrete numerical value for presetting training parameter is configured as needed.Preferably, it is maximum
Frequency of training is 10000, and global error desired value is 1 × 10-3, learning rate is 0.02.
Optionally, the state equation of Kalman filtering algorithm is Wk+1=FkWk+ωk, Kalman filtering algorithm
Observational equation be yk+1=Hk+1(Wk+1,uk+1)+vk+1。
Wherein, WkFor the weights of each neuron in BP neural network topological model, FkSquare is shifted for state
Battle array, ωkFor process noise matrix;
Wherein, yk+1For the output vector of BP neural network topological model, Hk+1For observing matrix, uk+1For
The input vector of BP neural network topological model, vk+1For measurement noise matrix.
Optionally, deformation measurement data also includes test sample.Study module 15 is additionally operable to, according to test
Sample is tested BP neural network topological model.
Present embodiments provide it is a kind of based on Kalman filtering and the Deformation Prediction device of BP neural network,
Including:Acquisition module, model building module, study module and prediction module, wherein, acquisition module is used
In the deformation measurement data of monitoring object in engineering is obtained as training sample, model building module is used for root
BP neural network topological model is set up according to training sample, study module is used to adopt Kalman filtering algorithm
The study of BP neural network topological model, adjustment BP neural network topology are carried out according to default training parameter
The weights of each neuron in model, prediction module is used for according to the BP neural network topology after adjustment weights
Model carries out Deformation Prediction.The Deformation Prediction device that the present embodiment is provided, by by Kalman filtering algorithm
As the training algorithm of BP neural network, the weighed value adjusting process of BP neural network is converted into into weights
Optimal estimation problem, shortens the learning time of BP neural network, improves and become during Deformation Prediction
Shape forecast model sets up efficiency.
Fig. 6 is that the deformation based on Kalman filtering and BP neural network that the embodiment of the present invention two is provided is pre-
The structural representation of device is surveyed, the present embodiment is on the basis of embodiment one, there is provided filter based on Kalman
The another embodiment of the Deformation Prediction device of ripple and BP neural network, for performing Fig. 1~Fig. 4 institutes
Show embodiment provide based on Kalman filtering and the Deformation Prediction method of BP neural network.Such as Fig. 6 institutes
Show, the present embodiment provide based on Kalman filtering and the Deformation Prediction device of BP neural network, can be with
Including:Acquisition module 11, model building module 13, study module 15, prediction module 17 and data point
Solution module 19.
Wherein, data decomposing module 19 is used for, by training sample be divided into steady deformation tendency data acquisition system,
Gradual change deformation tendency data acquisition system, mutation deformation tendency data acquisition system and high frequency superposition long period deformation tendency
Data acquisition system.
Study module 15 is additionally operable to, according to steady deformation tendency data acquisition system, using Kalman filtering algorithm
The study of BP neural network is carried out according to default training parameter, steady prediction model of deformation is obtained;Or,
According to gradual change deformation tendency data acquisition system, BP is carried out according to default training parameter using Kalman filtering algorithm
The study of neutral net, obtains gradual change prediction model of deformation;Or, according to mutation deformation tendency data set
Close, the study of BP neural network is carried out according to default training parameter using Kalman filtering algorithm, obtain
Mutation prediction model of deformation;Or, long period deformation tendency data acquisition system is superimposed according to high frequency, using card
Kalman Filtering algorithm carries out the study of BP neural network according to default training parameter, obtains high frequency superposition length
Cyclomorphosis forecast model.
Finally it should be noted that:Various embodiments above is only illustrating technical scheme rather than right
Its restriction;Although being described in detail to the present invention with reference to foregoing embodiments, this area it is common
Technical staff should be understood:It still can modify to the technical scheme described in foregoing embodiments,
Either equivalent is carried out to which part or all technical characteristic;And these modifications or replacement, and
The scope of the essence disengaging various embodiments of the present invention technical scheme of appropriate technical solution is not made.
Claims (9)
1. a kind of based on Kalman filtering and the Deformation Prediction method of BP neural network, it is characterised in that
Including:
The deformation measurement data of monitoring object in engineering is obtained as training sample;
BP neural network topological model is set up according to the training sample;
The BP neural network topological model is carried out according to default training parameter using Kalman filtering algorithm
Study, adjust the weights of each neuron in the BP neural network topological model;
Deformation Prediction is carried out according to the BP neural network topological model after adjustment weights.
2. method according to claim 1, it is characterised in that the employing Kalman filtering algorithm
Carry out before the study of the BP neural network topological model, also including according to default training parameter:
By the training sample be divided into steady deformation tendency data acquisition system, gradual change deformation tendency data acquisition system,
Mutation deformation tendency data acquisition system and high frequency superposition long period deformation tendency data acquisition system;
The employing Kalman filtering algorithm carries out the BP neural network topology according to default training parameter
The study of model, adjusts the weights of each neuron in the BP neural network topological model, specifically includes:
According to the steady deformation tendency data acquisition system, using Kalman filtering algorithm according to the default instruction
Practicing parameter carries out the study of the BP neural network, obtains steady prediction model of deformation;Or,
According to the gradual change deformation tendency data acquisition system, using Kalman filtering algorithm according to the default instruction
Practicing parameter carries out the study of the BP neural network, obtains gradual change prediction model of deformation;Or,
According to the mutation deformation tendency data acquisition system, using Kalman filtering algorithm according to the default instruction
Practicing parameter carries out the study of the BP neural network, obtains mutation prediction model of deformation;Or,
According to the high frequency be superimposed long period deformation tendency data acquisition system, using Kalman filtering algorithm according to
The default training parameter carries out the study of the BP neural network, obtains high frequency superposition long period deformation
Forecast model.
3. method according to claim 1, it is characterised in that the deformation measurement data also includes
Test sample, in the adjustment BP neural network topological model after the weights of each neuron, also
Including:
The BP neural network topological model is tested according to the test sample.
4. method according to claim 1, it is characterised in that the default training parameter includes:
Maximum frequency of training, global error desired value and learning rate.
5. method according to claim 4, it is characterised in that the maximum frequency of training is 10000,
The global error desired value is 1 × 10-3, the learning rate is 0.02.
6. according to the arbitrary described method of claim 1 to 5, it is characterised in that the Kalman filtering
The state equation of algorithm is:Wk+1=FkWk+ωk;The observational equation of the Kalman filtering algorithm is:
yk+1=Hk+1(Wk+1,uk+1)+vk+1;
Wherein, WkFor the weights of each neuron in the BP neural network topological model, FkTurn for state
Move matrix, ωkFor process noise matrix;
Wherein, yk+1For the output vector of the BP neural network topological model, Hk+1For observing matrix, uk+1
For the input vector of the BP neural network topological model, vk+1For measurement noise matrix.
7. a kind of based on Kalman filtering and the Deformation Prediction device of BP neural network, it is characterised in that
Including:
Acquisition module, for obtaining engineering in monitoring object deformation measurement data as training sample;
Model building module, for setting up BP neural network topological model according to the training sample;
Study module, it is refreshing for carrying out the BP according to default training parameter using Kalman filtering algorithm
The study of Jing network topology models, adjusts the weights of each neuron in the BP neural network topological model;
Prediction module, for being deformed according to the BP neural network topological model after adjustment weights
Prediction.
8. device according to claim 7, it is characterised in that also include:Data decomposing module;
The data decomposing module is used for, by the training sample be divided into steady deformation tendency data acquisition system,
Gradual change deformation tendency data acquisition system, mutation deformation tendency data acquisition system and high frequency superposition long period deformation tendency
Data acquisition system;
The study module is additionally operable to, and according to the steady deformation tendency data acquisition system, is filtered using Kalman
Ripple algorithm carries out the study of the BP neural network according to the default training parameter, obtains steady deformation
Forecast model;Or,
According to the gradual change deformation tendency data acquisition system, using Kalman filtering algorithm according to the default instruction
Practicing parameter carries out the study of the BP neural network, obtains gradual change prediction model of deformation;Or,
According to the mutation deformation tendency data acquisition system, using Kalman filtering algorithm according to the default instruction
Practicing parameter carries out the study of the BP neural network, obtains mutation prediction model of deformation;Or,
According to the high frequency be superimposed long period deformation tendency data acquisition system, using Kalman filtering algorithm according to
The default training parameter carries out the study of the BP neural network, obtains high frequency superposition long period deformation
Forecast model.
9. device according to claim 7, it is characterised in that the deformation measurement data also includes
Test sample;
The study module is additionally operable to, according to the test sample to the BP neural network topological model
Tested.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510671512.5A CN106570563A (en) | 2015-10-13 | 2015-10-13 | Deformation prediction method and apparatus based on Kalman filtering and BP neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510671512.5A CN106570563A (en) | 2015-10-13 | 2015-10-13 | Deformation prediction method and apparatus based on Kalman filtering and BP neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106570563A true CN106570563A (en) | 2017-04-19 |
Family
ID=58508700
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510671512.5A Pending CN106570563A (en) | 2015-10-13 | 2015-10-13 | Deformation prediction method and apparatus based on Kalman filtering and BP neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106570563A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109062248A (en) * | 2018-08-02 | 2018-12-21 | 西北工业大学 | Space non-cooperative target appearance rail integration method for parameter estimation based on self-organizing network |
CN109061506A (en) * | 2018-08-29 | 2018-12-21 | 河海大学常州校区 | Lithium-ion-power cell SOC estimation method based on Neural Network Optimization EKF |
CN109145463A (en) * | 2018-08-27 | 2019-01-04 | 北京住总集团有限责任公司 | A kind of deformation analysis method for during tunnel excavation |
CN109238358A (en) * | 2018-09-12 | 2019-01-18 | 国网福建省电力有限公司 | A kind of distribution type electric energy batch meter multi-sensor data collection and anti-interference processing method |
CN109300276A (en) * | 2018-07-27 | 2019-02-01 | 昆明理工大学 | A kind of car inside abnormity early warning method based on Fusion |
CN109828211A (en) * | 2018-12-25 | 2019-05-31 | 宁波飞拓电器有限公司 | A kind of emergency light battery SOC estimation method based on neural network adaptive-filtering |
CN109977464A (en) * | 2019-02-18 | 2019-07-05 | 江苏科技大学 | A kind of prediction technique of the piston machining deflection based on BP neural network |
CN110242310A (en) * | 2019-06-14 | 2019-09-17 | 西安电子科技大学 | Shield axis method for correcting error based on deep neural network in conjunction with association analysis |
CN110457795A (en) * | 2019-07-26 | 2019-11-15 | 东南大学 | High-rise charming appearance and behaviour displacement state appraisal procedure neural network based |
CN110796234A (en) * | 2019-10-21 | 2020-02-14 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
CN111479982A (en) * | 2017-11-15 | 2020-07-31 | 吉奥奎斯特系统公司 | In-situ operating system with filter |
CN111813858A (en) * | 2020-07-10 | 2020-10-23 | 电子科技大学 | Distributed neural network hybrid synchronous training method based on self-organizing grouping of computing nodes |
CN112307414A (en) * | 2020-10-29 | 2021-02-02 | 上海勘察设计研究院(集团)有限公司 | Tunnel scanning point cloud resolving method based on nonlinear Kalman filtering |
CN115688251A (en) * | 2022-12-19 | 2023-02-03 | 山东大学 | Earthquake multi-occurrence-zone tunnel risk decision method and system based on deep learning |
CN117272872A (en) * | 2023-11-21 | 2023-12-22 | 四川大学 | Panel rock-fill dam deformation monitoring method based on component separation |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101344389A (en) * | 2008-08-20 | 2009-01-14 | 中国建筑第八工程局有限公司 | Method for estimating tunnel surrounding rock displacement by neural network |
US20110153529A1 (en) * | 2009-12-23 | 2011-06-23 | Bracy Anne W | Method and apparatus to efficiently generate a processor architecture model |
CN103116804A (en) * | 2013-01-31 | 2013-05-22 | 苏州科技学院 | Fuzzy neural network model and intelligent prediction method for deep excavation deformation |
CN103810524A (en) * | 2014-03-08 | 2014-05-21 | 辽宁工程技术大学 | Method for predicting ground subsidence in underground metro construction process |
-
2015
- 2015-10-13 CN CN201510671512.5A patent/CN106570563A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101344389A (en) * | 2008-08-20 | 2009-01-14 | 中国建筑第八工程局有限公司 | Method for estimating tunnel surrounding rock displacement by neural network |
US20110153529A1 (en) * | 2009-12-23 | 2011-06-23 | Bracy Anne W | Method and apparatus to efficiently generate a processor architecture model |
CN103116804A (en) * | 2013-01-31 | 2013-05-22 | 苏州科技学院 | Fuzzy neural network model and intelligent prediction method for deep excavation deformation |
CN103810524A (en) * | 2014-03-08 | 2014-05-21 | 辽宁工程技术大学 | Method for predicting ground subsidence in underground metro construction process |
Non-Patent Citations (2)
Title |
---|
刘建国: "基于卡尔曼滤波的BP神经网络模型在桥梁形变中的应用", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
唐蕾: "神经网络与卡尔曼滤波算法的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111479982A (en) * | 2017-11-15 | 2020-07-31 | 吉奥奎斯特系统公司 | In-situ operating system with filter |
CN109300276A (en) * | 2018-07-27 | 2019-02-01 | 昆明理工大学 | A kind of car inside abnormity early warning method based on Fusion |
CN109062248A (en) * | 2018-08-02 | 2018-12-21 | 西北工业大学 | Space non-cooperative target appearance rail integration method for parameter estimation based on self-organizing network |
CN109145463A (en) * | 2018-08-27 | 2019-01-04 | 北京住总集团有限责任公司 | A kind of deformation analysis method for during tunnel excavation |
CN109061506A (en) * | 2018-08-29 | 2018-12-21 | 河海大学常州校区 | Lithium-ion-power cell SOC estimation method based on Neural Network Optimization EKF |
CN109238358A (en) * | 2018-09-12 | 2019-01-18 | 国网福建省电力有限公司 | A kind of distribution type electric energy batch meter multi-sensor data collection and anti-interference processing method |
CN109828211A (en) * | 2018-12-25 | 2019-05-31 | 宁波飞拓电器有限公司 | A kind of emergency light battery SOC estimation method based on neural network adaptive-filtering |
CN109977464B (en) * | 2019-02-18 | 2023-11-24 | 江苏科技大学 | Prediction method of piston cutting deformation based on BP neural network |
CN109977464A (en) * | 2019-02-18 | 2019-07-05 | 江苏科技大学 | A kind of prediction technique of the piston machining deflection based on BP neural network |
CN110242310B (en) * | 2019-06-14 | 2020-08-11 | 西安电子科技大学 | Shield axis deviation rectifying method based on combination of deep neural network and correlation analysis |
CN110242310A (en) * | 2019-06-14 | 2019-09-17 | 西安电子科技大学 | Shield axis method for correcting error based on deep neural network in conjunction with association analysis |
CN110457795A (en) * | 2019-07-26 | 2019-11-15 | 东南大学 | High-rise charming appearance and behaviour displacement state appraisal procedure neural network based |
CN110796234A (en) * | 2019-10-21 | 2020-02-14 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
CN110796234B (en) * | 2019-10-21 | 2022-07-12 | 苏州浪潮智能科技有限公司 | Method and device for predicting computer state |
CN111813858A (en) * | 2020-07-10 | 2020-10-23 | 电子科技大学 | Distributed neural network hybrid synchronous training method based on self-organizing grouping of computing nodes |
CN111813858B (en) * | 2020-07-10 | 2022-06-24 | 电子科技大学 | Distributed neural network hybrid synchronous training method based on self-organizing grouping of computing nodes |
CN112307414A (en) * | 2020-10-29 | 2021-02-02 | 上海勘察设计研究院(集团)有限公司 | Tunnel scanning point cloud resolving method based on nonlinear Kalman filtering |
CN112307414B (en) * | 2020-10-29 | 2022-07-29 | 上海勘察设计研究院(集团)有限公司 | Tunnel scanning point cloud resolving method based on nonlinear Kalman filtering |
CN115688251A (en) * | 2022-12-19 | 2023-02-03 | 山东大学 | Earthquake multi-occurrence-zone tunnel risk decision method and system based on deep learning |
CN117272872A (en) * | 2023-11-21 | 2023-12-22 | 四川大学 | Panel rock-fill dam deformation monitoring method based on component separation |
CN117272872B (en) * | 2023-11-21 | 2024-01-30 | 四川大学 | Panel rock-fill dam deformation monitoring method based on component separation |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106570563A (en) | Deformation prediction method and apparatus based on Kalman filtering and BP neural network | |
Hung et al. | An artificial neural network model for rainfall forecasting in Bangkok, Thailand | |
CN114997541B (en) | Urban waterlogging prediction method and early warning platform based on digital twin technology | |
Jan et al. | Neural network forecast model in deep excavation | |
EP2831804B1 (en) | System and method for automatic local grid refinement in reservoir simulation systems | |
Zhang et al. | Application of optimized grey discrete Verhulst–BP neural network model in settlement prediction of foundation pit | |
CN107748942B (en) | Radar Echo Extrapolation prediction technique and system based on velocity field sensing network | |
CN109992900A (en) | A kind of more real-time online cooperative intelligent emulation modes of mass concrete and system | |
CN105735370B (en) | Foundation settlement Deformation Prediction method based on Rayleigh waves | |
CN109559098A (en) | A kind of low influence development facility analogy method in sponge city pilot area | |
CN112113146B (en) | Synchronous self-adaptive check method for roughness coefficient and node water demand of water supply pipe network pipeline | |
CN109492756A (en) | More element conductor galloping method for early warning and relevant apparatus based on deep learning | |
JP6813865B1 (en) | Information processing method, program, information processing device and model generation method | |
CN111666621A (en) | Method for determining safe support pressure interval of excavation face of tunnel in clay stratum | |
CN105678417A (en) | Prediction method and device for tunnel face water inflow of construction tunnel | |
CN107368928A (en) | A kind of combination forecasting method and system of ancient building sedimentation | |
CN109711063B (en) | Station underground excavation footage adjusting method and device suitable for soft upper and hard lower strata | |
El-Ghandour et al. | Groundwater management using a new coupled model of flow analytical solution and particle swarm optimization | |
NO20200978A1 (en) | Optimized methodology for automatic history matching of a petroleum reservoir model with ensemble kalman filter | |
CN116976227A (en) | Storm water increasing forecasting method and system based on LSTM machine learning | |
CN103796217B (en) | A kind of estimation range partitioning method and device based on drive test data | |
CN114676486A (en) | Method for analyzing influence of river water seepage on river-crossing tunnel excavation | |
CN106934729A (en) | Building Testing and appraisal method and device | |
CN114764546A (en) | Method for predicting residual oil of fault block oil reservoir based on generation of antagonistic neural network | |
CN108549745A (en) | The wing perf ormance of low reynolds number flying condition and the prediction technique for restoring boundary |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170419 |