CN108875841A - A kind of pumped storage unit vibration trend forecasting method - Google Patents
A kind of pumped storage unit vibration trend forecasting method Download PDFInfo
- Publication number
- CN108875841A CN108875841A CN201810692295.1A CN201810692295A CN108875841A CN 108875841 A CN108875841 A CN 108875841A CN 201810692295 A CN201810692295 A CN 201810692295A CN 108875841 A CN108875841 A CN 108875841A
- Authority
- CN
- China
- Prior art keywords
- signal
- trend
- vibration
- unit
- value
- 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.)
- Withdrawn
Links
Classifications
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a kind of pumped storage unit vibration trend forecasting methods,It first obtains the history and real time data of unit nonstationary vibration online,User terminal is passed data to again,Time-Frequency Analysis is carried out to vibration signal followed by experience wavelet decomposition,Then Energy-Entropy and singular value comprehensive characteristics are extracted,Then analysis is associated with unit operating condition according to after certain rule progress sliding-model control to signal characteristic data acquisition system,Frequent-item is carried out using Apriori algorithm,Parse the temporal correlation of data characteristics and unit failure,Unit safety operation region is marked off by association analysis result,Finally construct time series models,Using time series trend prediction technique,Predict the development trend in its following finite time,And then operating states of the units trend is predicted and assessed,Technical support is provided to implement set state maintenance.The present invention can not only Accurate Prediction trend, also have the advantages that evaluation index relatively comprehensively and assessment more conveniently.
Description
Technical field
The present invention relates to a kind of pumped storage unit vibration trend forecasting methods.
Background technique
The pump turbine group operating condition of hydroenergy storage station is complicated, start and stop are frequent, and vibration position is more and distributed more widely general,
Influence factor is more (including hydraulic factors, mechanical factor and electromagnetic factors).Now widely used trend forecasting method is to pass through
On-line monitoring system obtains the historical data tendency chart of unit vibration state, possible operating status after subjective analysis unit,
Or regression fit is realized using least square method to characteristic parameter, but vibration signal is typically non-linear, non-stationary, because
This traditional mathematical statistics method is difficult to realize accurate trend prediction to its state.Due to analytic hierarchy process (AHP) weight selection not
When, tend not to the requirement for meeting judgment matrix approach, be unable to get reflection the comprehensive state of pumped storage unit evaluation index.Cause
This, there is trend prediction inaccuracy and the incomplete problem of evaluation index for existing pumped storage unit vibration trend forecasting method.
Summary of the invention
The object of the present invention is to provide a kind of pumped storage unit vibration trend forecasting methods.The present invention can not only be accurate
Anticipation trend also has the advantages that evaluation index is more comprehensive.
Technical solution of the present invention：A kind of pumped storage unit vibration trend forecasting method, includes the following steps：
A1, online data obtain, online from the monitoring of pump-up power station runout and monitoring system by using 485 communications protocol
Obtain the history and real time data of unit nonstationary vibration；
B1, data transmitting, it is non-flat to be converted into unit for the history of the unit nonstationary vibration obtained online and real time data
Steady vibration signal, and successively user is passed to by network isolating device, WEB server, hardware firewall and power plant's local area network
Terminal；
C1, signal analysis, user terminal are non-flat to unit using experience wavelet decomposition for unit Non-stationary vibration signal
Steady vibration signal carries out Time-Frequency Analysis；
D1, feature extraction, in the base for carrying out Time-Frequency Analysis to unit Non-stationary vibration signal using experience wavelet decomposition
On plinth, the comprehensive characteristics of description unit vibration state are extracted from unit Non-stationary vibration signal, obtain Energy-Entropy and singular value；
E1, vibration association and quantitative analysis set empirical modal point is utilized to the data acquisition system of Energy-Entropy and singular value
Solution (EEMD) or improved experience wavelet decomposition method (EWT) decompose vibration signal, obtain a series of unifrequency component,
It carries out being associated analysis with unit operating condition after sliding-model control again, Apriori algorithm is utilized to carry out frequent-item, solution
The temporal correlation for analysing unit vibration data characteristics and unit failure, marks off unit safety operation area by association analysis result
Domain；
F1, building time serial model, multiple linear regression model and arma modeling, it is pre- using time series trend
Survey method predicts the development trend in the following finite time, and then operating states of the units trend is predicted and assessed, and is real
It applies set state maintenance and technical support is provided.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the time series trend prediction technique in step f1,
Include the following steps：
A2, the time series for extracting unit state parameter；
B2, using experience wavelet decomposition by the Time Series of set state parameter be different decomposition domain subsequence,
Obtain the fluctuation item and nonlinear trend term of stationarity；
C2, AR prediction model is used for the fluctuation item of stationarity, signal A is obtained, for nonlinear trend term using most
Small two multiply the prediction model of support vector machines, obtain signal B；
D2, signal A and signal B are reconstructed to signal, realize set state trend prediction.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, in step b1, data transmitting will can also be obtained online
The history and real time data of the unit nonstationary vibration taken are converted into unit Non-stationary vibration signal, successively fill by Network Isolation
It sets, WEB server, hardware firewall and Ali's cloud pass in APP on mobile terminal.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the prediction model of the least square method supporting vector machine
It is expressed as
Wherein X is input vector, and n is constant, and k is some value in 1~n, and m is space dimensionality, and τ is that delay time is normal
Number.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the AR model is expressed as
In formula：For auto-regressive parameter, i=1 ... p, p are order；atFor white noise, residual error is indicated.
Pass through AIC criterion in a kind of pumped storage unit vibration trend forecasting method above-mentioned, when the AR model modeling to determine
The order of model and the parameter Estimation for carrying out model.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the method for reconstruction signal in step d2, including it is following
Step：
A3, multilevel wavelet decomposition is carried out to the time series of set state parameter, obtain description trend term and fluctuates item
Each layer sub-sequence：
X=ak+d1+d2+…+dk
In formula, akFor trend term, diTo fluctuate item, i=1,2 ..., k are the wavelet decomposition number of plies；
B3, for trend term akLSSVM prediction model is established, model training and new value prediction is carried out, result is carried out
Exact evaluation；
C3, for each fluctuation item di, i=1,2 ..., k carry out AR modeling respectively, carry out precision evaluation to result；
D3, superposition calculation is carried out after obtaining the prediction result of trend term and each fluctuation item, obtains original oscillating sequence
Forecasting sequenceThe error criterion of measured value and predicted value is calculated, precision evaluation is carried out；
E3, the precision of the prediction result of original oscillating sequence is evaluated, analyzes the deviation of prediction result and actual value
Degree uses following common error criterion to the evaluation of prediction effect.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the common error criterion includes
Average relative error：With
Root-mean-square error：
Wherein k is some value in 1~N,For average value.
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the signal analysis is shown as, for a signal f
(t), EWT is broken down into the mode function f of N+1 single-frequency ingredientk(t)。
Wherein, fk(t) AM/FM amplitude modulation/frequency modulation signal (AM-FM) can be defined as：
fk(t)=Fk(t)cos(φk(t))
Wherein, k is some value in 0~N, FkFor frequency-domain function, φkAngularly to measure.
The decomposition method of non-stationary signal is intended to that the f of AM-FM ingredient can be decomposited from original signalk(t)。EWT
The AM-FM ingredient of signal is extracted by adaptive segmentation Fourier spectrum wavelet structure filter.Assuming that by Fourier branch
It holds [0, π] and is divided into N number of continuous part.Every section with ∧nIt indicates：
Λn=[ωn-1, ωn], n=1,2 ..., N
Wherein, ωnFor each section boundary, ωnThere are many mode of selection, general to take two consecutive roots in Fourier spectrum
Terminal between big value point, and ω0=0, ωn=π.With ωnCentered on to define width be Tn=2 τnChangeover portion.According to
The make of Meyer small echo is to each ∧nConstruct bandpass filter, experience wavelet functionWith experience scale letter
NumberIt is defined as：
Wherein,
Assuming that Fourier's variation is denoted as F [], inverse Fourier transform is denoted as F-1[·]；Then according to traditional wavelet
Solution mode defines experience wavelet transformation, and the detail coefficients of experience wavelet transformation are by experience wavelet function Ψn(t) with signal f
(t) inner product acquires：
Experience wavelet approximation coefficients are by signal f (t) and scaling function Ψ1(t) inner product acquires：
Wherein,WithIt is the Fourier transformation of experience wavelet function and scaling function respectively；
WithIt is the conjugate complex number of experience wavelet function and scaling function respectively.
Original signal can rule of thumb wavelet function and scaling function reconstruct：
Wherein, * indicates convolution；∧ indicates Fourier transformation.
It is hereby achieved that the mode f that experience small echo is decomposedk(t)：
In a kind of pumped storage unit vibration trend forecasting method above-mentioned, the feature extraction is shown as, it is assumed that original signal
A series of mode, c are obtained after experience wavelet transformations(t), s=1,2 ..., K, K are the number of plies after decomposing, E1,E2...,
EKIt is the corresponding energy value of every layer of mode.The calculating process of energy feature is as follows：
The energy feature of every layer of mode is calculated first：
Then the total energy value of all mode is calculated：
Finally calculate Energy-Entropy feature；
The singular value features of all mode composition matrixes will be also extracted after signal experience wavelet transformation.Singular value features are pair
Matrix carries out the series of features vector obtained after singular value decomposition.According to matrix theory principle, singular value is to reflect a square
One value of battle array build-in attribute, and the value has stability.Meanwhile singular value has scale invariability and rotational invariance.
Therefore singular value features are that a very reliable evaluation index is used to distinguish different failures
For original signal after experience wavelet transformation, every layer of energy has first layer successively to successively decrease to last one layer, when
When preceding Q layers of mode almost can include most energy informations of original signal, Q layers of mode are initial to construct before just taking
Eigenmatrix A：
A=[c1,c2..., cQ]T
Assuming that the size of real number matrix A is PQ, while meeting the following conditions：
A=U ∧ VT
Wherein U and V is respectively the orthogonal matrix that size is what PP of QQ, and ∧ is the fault signature matrix that a size is QP,
And all component σi(i=1,2 ..., Q) (Q≤P) is arranged from low to high according to the mode number of plies.These ffault matrix
Element is some singular values of matrix A.Therefore by vector [σ1, σ2..., σQ] as another fault signature.
Compared with prior art, present invention improves over existing pumped storage unit vibration trend forecasting methods, anti-by establishing
The trend prediction model of film projector group vibrational state key characterization parameter from multiple orientation optimization prediction algorithms, and constructs when being based on
Between combined sequence prediction model；The minutia that signal is extracted using wavelet transformation theory, by unit vibration state parameter point
Solution is the fluctuation item of nonlinear trend term and stationarity, is utilized respectively least square method supporting vector machine (LSSVM) theory and oneself
It returns (AR) model and carries out trend prediction, finally realize that the trend of unit vibration state parameter is pre- using addition rule reconstruction signal
It surveys, the trend prediction of pumped storage unit vibration is relatively accurate and evaluation index is more comprehensive.In addition, the present invention is by above-mentioned model use in unit
In state on_line monitoring system, and under the original frame of pump-up power station Unit State Monitor System, by extension equipment and
Software interface (mainly including WEB server, lateral isolation device, hardware firewall and the network switch), transfers data to
On mobile terminal equipped with TN8000APP, realization is monitored set state by mobile terminal, analyzes, and pushes away in real time
The functions such as early warning and alarming and fault message are sent, set state is monitored by mobile terminal TN8000APP, and can be to unit
Vibration trend is analyzed automatically, and pump-up power station operation maintenance personnel is facilitated to assess set state.Therefore, the present invention can not only
Enough Accurate Prediction trend also has the advantages that evaluation index is relatively comprehensive and assesses more conveniently.
Detailed description of the invention
Fig. 1 is pumped storage unit vibration trend prediction model of the invention；
Fig. 2 is Apriori algorithm flow chart；
Fig. 3 is combination forecasting schematic diagram.
Specific embodiment
The present invention is further illustrated with reference to the accompanying drawings and examples, but be not intended as to the present invention limit according to
According to.
Embodiment.A kind of pumped storage unit vibration trend forecasting method, as shown in Figure 1, the prediction technique mainly includes six
Step, i.e.,：Online data acquisition, data transmitting, signal analysis, feature extraction, the association of vibration and quantitative analysis, trend are pre-
It surveys.
First step online data obtains first, is designed and is configured by relevant interface, from the monitoring of certain pump-up power station runout, prison
The systems such as control obtain the history and real time data of unit nonstationary vibration online, for signal analysis, feature extraction and state evaluation
And fault diagnosis provides information resources.
The transmitting of second step data, it is non-to be converted into unit for the history of the unit nonstationary vibration obtained online and real time data
Stationary vibration signal, and successively use is passed to by network isolating device, WEB server, hardware firewall and power plant's local area network
Family terminal；Or the history of the unit nonstationary vibration obtained online and real time data are converted into unit nonstationary vibration letter
Number, successively passed in the APP on mobile terminal by network isolating device, WEB server, hardware firewall and Ali's cloud.
Third step signal analysis, to Mr. Yu's pumped storage unit Non-stationary vibration signal, using experience wavelet decomposition (EWT) to vibration
Dynamic signal carries out Time-Frequency Analysis.
The mode function f of N+1 single-frequency ingredient is broken down into for a signal f (t), EWTk(t)。
Wherein, fk(t) AM/FM amplitude modulation/frequency modulation signal (AM-FM) can be defined as：
fk(t)=Fk(t)cos(φk(t))
Wherein, k is some value in 0~N, FkFor frequency-domain function, φkAngularly to measure.
The decomposition method of non-stationary signal is intended to that the f of AM-FM ingredient can be decomposited from original signalk(t)。EWT
The AM-FM ingredient of signal is extracted by adaptive segmentation Fourier spectrum wavelet structure filter.Assuming that by Fourier branch
It holds [0, π] and is divided into N number of continuous part.Every section with ∧nIt indicates：
∧n=[ωn-1, ωn], n=1,2 ..., N
Wherein, ωnFor each section boundary, ωnThere are many mode of selection, general to take two consecutive roots in Fourier spectrum
Terminal between big value point, and ω0=0, ωn=π.With ωnCentered on to define width be Tn=2 τnChangeover portion.According to
The make of Meyer small echo is to each ∧nConstruct bandpass filter, experience wavelet functionWith experience scale letter
NumberIt is defined as：
Wherein,
Assuming that Fourier's variation is denoted as F [], inverse Fourier transform is denoted as F-1[·]；Then according to traditional wavelet
Solution mode defines experience wavelet transformation, and the detail coefficients of experience wavelet transformation are by experience wavelet function Ψn(t) with signal f
(t) inner product acquires：
Experience wavelet approximation coefficients are by signal f (t) and scaling function Ψ1(t) inner product acquires：
Wherein,WithIt is the Fourier transformation of experience wavelet function and scaling function respectively；
WithIt is the conjugate complex number of experience wavelet function and scaling function respectively.
Original signal can rule of thumb wavelet function and scaling function reconstruct：
Wherein, * indicates convolution；∧ indicates Fourier transformation.
It is hereby achieved that the mode f that experience small echo is decomposedk(t)：
4th step feature extraction extracts Energy-Entropy and singular value etc. and describes unit vibration shape on the basis of signal analysis
The comprehensive characteristics of state.
Entropy is characterized in measurement uncertainty and probabilistic index, and when a failure occurs, trouble location can generate impact,
Therefore vibration signal relative to frequency response can also change, the distribution of energy and size will also change.First will
Signal obtains a series of mode after experience wavelet transformation, then seeks energy feature to these mode and Energy-Entropy is special
Sign.Since compared to fault-signal, working normally signal has stronger certainty and stationarity, therefore the signal of nominal situation
Property metaphor fault-signal entropy it is big.
Assuming that original signal obtains a series of mode, c after experience wavelet transformations(t), s=1,2 ..., K, K are point
The number of plies after solution, E1, E2..., EKIt is the corresponding energy value of every layer of mode.The calculating process of energy feature is as follows：
The energy feature of every layer of mode is calculated first：
Then the total energy value of all mode is calculated：
Finally calculate Energy-Entropy feature；
The singular value features of all mode composition matrixes will be also extracted after signal experience wavelet transformation.Singular value features are pair
Matrix carries out the series of features vector obtained after singular value decomposition.According to matrix theory principle, singular value is to reflect a square
One value of battle array build-in attribute, and the value has stability.Meanwhile singular value has scale invariability and rotational invariance.
Therefore singular value features are that a very reliable evaluation index is used to distinguish different failures
For original signal after experience wavelet transformation, every layer of energy has first layer successively to successively decrease to last one layer, when
When preceding Q layers of mode almost can include most energy informations of original signal, Q layers of mode are initial to construct before just taking
Eigenmatrix A：
A=[c1,c2..., cQ]T
Assuming that the size of real number matrix A is PQ, while meeting the following conditions：
A=U ∧ VT
Wherein U and V is respectively the orthogonal matrix that size is what PP of QQ, and ∧ is the fault signature matrix that a size is QP,
And all component σi(i=1,2 ..., Q) (Q≤P) is arranged from low to high according to the mode number of plies.These ffault matrix
Element is some singular values of matrix A.Therefore by vector [σ1, σ2..., σQ] as another fault signature.
The association and quantitative analysis of 5th step vibration, to certain pump-up power station unit runout signal characteristic data acquisition system according to one
It establishes rules and is associated analysis with unit operating condition after then carrying out sliding-model control, Apriori algorithm is utilized to carry out frequent episode digging
Pick parses the temporal correlation of unit vibration data feature and unit failure, marks off unit safety by association analysis result
Operation area.
Wherein Apriori algorithm is the valid data association analysis of a kind of mining data implicit information, discovery frequent item set
Method generates discovery frequent item set by the way that limitation is candidate, and flow chart is as shown in Figure 2.After excavating discretization using Apriori algorithm
Pump-storage generator runout data the step of it is as follows：Firstly, the pump-storage generator runout signal characteristic established by scanning
With running state analysis database, add up the frequency of occurrence of operating states of the units amount, and collects the minimum support for meeting setting
Status items, find out the set of frequent 1 item collection, be denoted as L1.Then, according to Apriori property, frequent 2 item collection is found out using L1
Set L2, find out L3 using L2, so go down, until frequent k item collection cannot be found again.Pass through runout signal and set state
Between correlation analysis, obtain unit safety operation region, for unit operational safety domain differentiation support is provided.
6th step trend prediction constructs time serial model, multiple linear regression model, arma modeling, when use
Between Sequence Trend prediction technique, predict the development trend in its following finite time, and then carry out to operating states of the units trend
Prediction and estimation provide technical support to implement set state maintenance.
Wherein the 6th pre- flow gauge of step unit vibration state trend is as follows：1) unit state parameter time series is extracted；2)
The Time Series of set state parameter are the (fluctuation of nonlinear trend term and stationarity of different decomposition domain by wavelet decomposition
) subsequence；3) AR prediction model is used for the fluctuation item of stationarity, least square is used for nonlinear trend term
The prediction model of support vector machines；4) reconstruction signal realizes set state trend prediction.It is embodied as follows：
1) time series modeling based on least square method supporting vector machine constructs input vector：
Original time series are one group of one-dimensional measured values, need to carry out the timing before establishing LSSVM model mutually empty
Between reconstruct, obtain its input vector of the corresponding phase space matrix as LSSVM model.
Wherein X is input vector, and n is constant, and k is some value in 1~n, and m is space dimensionality, and τ is that delay time is normal
Number.
2) AR is modeled
The random difference equation (AR model) of linear regression model (LRM) is expressed as：
In formula：For auto-regressive parameter, i=1 ... p, p are order；atFor white noise, residual error is indicated.
It needs to be determined that the order of model and carrying out the parameter Estimation of model when establishing AR model.Determine the common of model order
Criterion has AIC criterion：
In formula：P is model order；N is data amount check；For the prediction error under different orders.
First setting models order, prediction model parameter are answered in modeling process, available each rank model finally takes AIC (p)
The corresponding order of first minimum determines the best order of model, may finally determine AR model.
Combined prediction modeling
If Hydropower Unit quantity of state sequence is Xt={ X1…Xn, prediction modeling procedure is as follows：
I, multilayer decomposition is carried out using wavelet transformation, obtain description trend term and fluctuates each layer sub-sequence of item：
X=ak+d1+d2+…+dk
In II, formula, akFor trend term, diTo fluctuate item, i=1,2 ..., k are the wavelet decomposition number of plies.
III, for trend term akLSSVM prediction model is established, model training and new value prediction is carried out, result is carried out
Exact evaluation.
IV, for each fluctuation item di, i=1,2 ..., k carry out AR modeling respectively, carry out precision evaluation to result.
V, superposition calculation is carried out after obtaining the prediction result of each section sequence, obtains the forecasting sequence of original oscillating sequenceThe error criterion of measured value and predicted value is calculated, precision evaluation is carried out.High combination forecasting
Realization process is as shown in Figure 3.
VI, estimated performance evaluation is carried out to result, that is, the precision of prediction result is evaluated, analyze prediction result
With the departure degree of actual value, common error criterion below can be used to the evaluation of prediction effect：
VII, average relative error：
VIII, root-mean-square error：
Wherein k is some value in 1~N,For average value.
The mobile terminal APP of the embodiment of the present invention, can be monitored set state by mobile device and trend is divided
Analysis, while alarm, early warning and device fault information are pushed into mobile device in real time.
To realize mobile application, it is additionally arranged relevant device on the basis of former pump-up power station Unit State Monitor System, mainly
Including WEB server, lateral isolation device, hardware firewall and the area the III network switch.WEB server is deployed in safe III
Area is isolated by unilateral network isolating device and the area II, data from the area II be one-way transmitted to WEB server be managed and
Storage.WEB server is connected by firewall with local area network, and by data publication to local area network, the user on local area network can lead to
Special-purpose software is crossed to be monitored data, analyze.WEB server is connected through firewalls Internet and (using special line or leads to
Cross LAN optimization), be connected to the mobile application private server being deployed on Ali's cloud, by server by status data with
And alarm, warning information are distributed to mobile terminal, mobile terminal user can be browsed and be analyzed data by dedicated APP.Data
It is stored in WEB server, is issued by the private server on Ali's cloud, any data are not stored on Ali's cloud.
For the safety for ensuring data, the transmission process of data is encryption, is shown after mobile terminal is again decrypted data.To ensure
The safety of WEB server, WEB server is only connect with mobile application private server on internet (is pacified by firewall
Full strategy is realized), the connection on other computers will be considered illegal and be rejected.
Mobile terminal APP function mainly includes following several respects：
1) real-time monitoring, TN8000APP can put vibration, throw, axial displacement, pressure fluctuation, the air gap, part
Electricity and duty parameter etc. carry out real-time monitoring, and system offer curves figure, numerical tabular, rod figure, waveform diagram, spectrogram etc. are a variety of can
Depending on change chart for user select, can from different perspectives, different levels set state is monitored, whenever and wherever possible hold unit fortune
Row state.
2) data analysis and trend prediction, system carry out forecast analysis to set state by vibration trend prediction model,
Pump-up power station operation maintenance personnel is facilitated to assess set state.
3) event pushes, and early warning, warning information can be pushed in real time mobile terminal by system, reminds user to pay close attention to machine in time
Group state change, generation of preventing accident.System also provides query history event information function simultaneously.
4) state of Unit State Monitor System itself can be monitored in real time by mobile APP in system autodiagnosis, including
Sensor states, acquisition device state, grid state, software operation state etc. are found in system operation in time
Defect, it is ensured that system is long-term, stable operation.
Claims (10)
1. a kind of pumped storage unit vibration trend forecasting method, which is characterized in that include the following steps：
A1, online data obtain, and by using 485 communications protocol, obtain online from the monitoring of pump-up power station runout and monitoring system
The history and real time data of unit nonstationary vibration；
The history of the unit nonstationary vibration obtained online and real time data are converted into the vibration of unit non-stationary by b1, data transmitting
Dynamic signal, and successively user terminal is passed to by network isolating device, WEB server, hardware firewall and power plant's local area network；
C1, signal analysis, user terminal shake to unit non-stationary for unit Non-stationary vibration signal, using experience wavelet decomposition
Dynamic signal carries out Time-Frequency Analysis；
D1, feature extraction, on the basis of carrying out Time-Frequency Analysis to unit Non-stationary vibration signal using experience wavelet decomposition,
The comprehensive characteristics that description unit vibration state is extracted from unit Non-stationary vibration signal, obtain Energy-Entropy and singular value；
E1, vibration association and quantitative analysis set empirical mode decomposition is utilized to the data acquisition system of Energy-Entropy and singular value
(EEMD) or experience wavelet decomposition method (EWT) decomposes vibration signal, obtains a series of unifrequency component, then carry out from
It is associated analysis with unit operating condition after dispersion processing, carries out frequent-item, parsing unit vibration using Apriori algorithm
The temporal correlation of dynamic data characteristics and unit failure, marks off unit safety operation region by association analysis result；
F1, building time serial model, multiple linear regression model and arma modeling, using time series trend prediction side
Method predicts the development trend in the following finite time, and then operating states of the units trend is predicted and assessed, to implement machine
Group repair based on condition of component provides technical support.
2. a kind of pumped storage unit vibration trend forecasting method according to claim 1, it is characterised in that：In step f1 when
Between Sequence Trend prediction technique, include the following steps：
A2, the time series for extracting unit state parameter；
B2, using experience wavelet decomposition by the Time Series of set state parameter be different decomposition domain subsequence, obtain flat
The fluctuation item and nonlinear trend term of stability；
C2, AR prediction model is used for the fluctuation item of stationarity, signal A is obtained, for nonlinear trend term using minimum two
The prediction model for multiplying support vector machines obtains signal B；
D2, signal A and signal B are reconstructed to signal, realize set state trend prediction.
3. a kind of pumped storage unit vibration trend forecasting method according to claim 1, it is characterised in that：In step b1, number
According to transmitting, the history of the unit nonstationary vibration obtained online and real time data can also be converted into unit nonstationary vibration letter
Number, successively passed in the APP on mobile terminal by network isolating device, WEB server, hardware firewall and Ali's cloud.
4. a kind of pumped storage unit vibration trend forecasting method according to claim 2, it is characterised in that：The least square
The prediction model of support vector machines is expressed as
Wherein X is input vector, and n is constant, and k is some value in 1~n, and m is space dimensionality, and τ is delay time constant.
5. a kind of pumped storage unit vibration trend forecasting method according to claim 2, it is characterised in that：The AR model table
It is shown as
In formula：For auto-regressive parameter, i=1 ... p, p are order；atFor white noise, residual error is indicated.
6. a kind of pumped storage unit vibration trend forecasting method according to claim 5, it is characterised in that：The AR model is built
The order of model is determined by AIC criterion when mould and carries out the parameter Estimation of model.
7. a kind of pumped storage unit vibration trend forecasting method according to claim 2, it is characterised in that：It is reconstructed in step d2
The method of signal, includes the following steps：
A3, multilevel wavelet decomposition is carried out to the time series of set state parameter, obtain description trend term and fluctuates each layer of item
Subsequence：
X=ak+d1+d2+…+dk
In formula, akFor trend term, diTo fluctuate item, i=1,2 ..., k are the wavelet decomposition number of plies；
B3, for trend term akLSSVM prediction model is established, model training and new value prediction is carried out, result is accurately commented
Valence；
C3, for each fluctuation item di, i=1,2 ..., k carry out AR modeling respectively, carry out precision evaluation to result；
D3, superposition calculation is carried out after obtaining the prediction result of trend term and each fluctuation item, obtains the prediction of original oscillating sequence
SequenceThe error criterion of measured value and predicted value is calculated, precision evaluation is carried out；
E3, the precision of the prediction result of original oscillating sequence is evaluated, analyzes the departure degree of prediction result and actual value,
Following common error criterion is used to the evaluation of prediction effect.
8. a kind of pumped storage unit vibration trend forecasting method according to claim 7, it is characterised in that：The common error
Index includes
Average relative error：With
Root-mean-square error：
Wherein k is some value in 1~N,For average value.
9. a kind of pumped storage unit vibration trend forecasting method according to claim 1, it is characterised in that：The signal analysis
It shows as, the mode function f of N+1 single-frequency ingredient is broken down into for a signal f (t), EWTk(t)。
Wherein, fk(t) AM/FM amplitude modulation/frequency modulation signal (AM-FM) can be defined as：
fk(t)=Fk(t)cos(φk(t))
Wherein, k is some value in 0~N, FkFor frequency-domain function, φkAngularly to measure.
The decomposition method of non-stationary signal is intended to that the f of AM-FM ingredient can be decomposited from original signalk(t).EWT passes through certainly
The segmentation Fourier spectrum wavelet structure filter of adaptation extracts the AM-FM ingredient of signal.Assuming that by Fourier support [0,
π] it is divided into N number of continuous part.Every section with ∧nIt indicates：
∧n=[ωn-1, ωn], n=1,2 ..., N
Wherein, ωnFor each section boundary, ωnThere are many mode of selection, general to take two adjacent maximum in Fourier spectrum
Terminal between point, and ω0=0, ωn=π.With ωnCentered on to define width be Tn=2 τnChangeover portion.According to Meyer
The make of small echo is to each ∧nConstruct bandpass filter, experience wavelet functionWith experience scaling functionIt is defined as：
Wherein,
β (x)=x4(35-84x+70x2-20x3)。
Assuming that Fourier's variation is denoted as F [], inverse Fourier transform is denoted as F-1[·]；Then according to the solution of traditional wavelet
Mode defines experience wavelet transformation, and the detail coefficients of experience wavelet transformation are by experience wavelet function Ψn(t) and in signal f (t)
Product acquires：
Experience wavelet approximation coefficients are by signal f (t) and scaling function Ψ1(t) inner product acquires：
Wherein,WithIt is the Fourier transformation of experience wavelet function and scaling function respectively；WithIt is the conjugate complex number of experience wavelet function and scaling function respectively.
Original signal can rule of thumb wavelet function and scaling function reconstruct：
Wherein, * indicates convolution；∧ indicates Fourier transformation.
It is hereby achieved that the mode f that experience small echo is decomposedk(t)：
。
10. a kind of pumped storage unit vibration trend forecasting method according to claim 1, it is characterised in that：The feature mentions
It takes and shows as, it is assumed that original signal obtains a series of mode, c after experience wavelet transformations(t), s=1,2 ..., K, K are point
The number of plies after solution, E1,E2..., EKIt is the corresponding energy value of every layer of mode.The calculating process of energy feature is as follows：
The energy feature of every layer of mode is calculated first：
Then the total energy value of all mode is calculated：
Finally calculate Energy-Entropy feature；
The singular value features of all mode composition matrixes will be also extracted after signal experience wavelet transformation.Singular value features are to matrix
Carry out the series of features vector obtained after singular value decomposition.According to matrix theory principle, singular value is to reflect a matrix to consolidate
There is a value of attribute, and the value has stability.Meanwhile singular value has scale invariability and rotational invariance.Therefore
Singular value features are that a very reliable evaluation index is used to distinguish different failures
For original signal after experience wavelet transformation, every layer of energy has first layer successively to successively decrease to last one layer, Q layers current
When mode almost can include most energy informations of original signal, preceding Q layers of mode is just taken to construct initial feature
Matrix A：
A=[c1, c2..., cQ]T
Assuming that the size of real number matrix A is PQ, while meeting the following conditions：
A=U ∧ VT
Wherein U and V is respectively the orthogonal matrix that size is what PP of QQ, and ∧ is the fault signature matrix that a size is QP, and
All component σi(i=1,2 ..., Q) (Q≤P) is arranged from low to high according to the mode number of plies.These ffault matrix elements
It is some singular values of matrix A.Therefore by vector [σ1, σ2..., σQ] as another fault signature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810692295.1A CN108875841A (en) | 2018-06-29 | 2018-06-29 | A kind of pumped storage unit vibration trend forecasting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810692295.1A CN108875841A (en) | 2018-06-29 | 2018-06-29 | A kind of pumped storage unit vibration trend forecasting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108875841A true CN108875841A (en) | 2018-11-23 |
Family
ID=64296816
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810692295.1A Withdrawn CN108875841A (en) | 2018-06-29 | 2018-06-29 | A kind of pumped storage unit vibration trend forecasting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108875841A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109946075A (en) * | 2018-12-25 | 2019-06-28 | 东北大学 | A kind of bearing condition monitoring and method for diagnosing faults |
| CN110175541A (en) * | 2019-05-13 | 2019-08-27 | 武汉大学 | A kind of method that sea level variability nonlinear trend is extracted |
| CN110298132A (en) * | 2019-07-05 | 2019-10-01 | 国家电网有限公司 | A kind of turbine-generator units bearing throw trend forecasting method |
| CN110673997A (en) * | 2019-09-06 | 2020-01-10 | 苏州浪潮智能科技有限公司 | Disk failure prediction method and device |
| CN110943481A (en) * | 2019-12-04 | 2020-03-31 | 重庆大学 | Unit combination method considering wind turbine generator safety domain with frequency response control |
| CN110991055A (en) * | 2019-12-08 | 2020-04-10 | 中国航空综合技术研究所 | Residual life prediction system for rotary mechanical products |
| CN111489027A (en) * | 2020-04-08 | 2020-08-04 | 五凌电力有限公司近尾洲水电厂 | Hydroelectric generating set waveform data trend prediction method and system |
| CN111695248A (en) * | 2020-05-27 | 2020-09-22 | 淮阴工学院 | Rapid early warning method for degradation state trend of pumped storage unit |
| CN112668419A (en) * | 2020-12-17 | 2021-04-16 | 太原理工大学 | Engine emission prediction method based on vibration signal |
| CN112949986A (en) * | 2021-02-01 | 2021-06-11 | 中国人民解放军海军工程大学 | Steam power system operation stability evaluation method and system, electronic equipment and storage medium |
| CN113077172A (en) * | 2021-04-19 | 2021-07-06 | 瑞湖智科数据(苏州)有限公司 | Equipment state trend analysis and fault diagnosis method |
| CN114442543A (en) * | 2021-10-29 | 2022-05-06 | 南京河海南自水电自动化有限公司 | Computer monitoring method suitable for early warning of hydropower station fault |
| CN114863651A (en) * | 2021-01-15 | 2022-08-05 | 湖南五凌电力科技有限公司 | Intelligent early warning method for monitoring state of auxiliary machine |
| CN115511880A (en) * | 2022-11-07 | 2022-12-23 | 长江勘测规划设计研究有限责任公司 | Method for identifying and measuring vibration amplitude and frequency of two shafts by using machine vision |
| CN115685949A (en) * | 2022-12-29 | 2023-02-03 | 深圳市信润富联数字科技有限公司 | Method and device for adjusting data sampling frequency in discrete machining production process |
| CN115795999A (en) * | 2022-10-26 | 2023-03-14 | 国网新源控股有限公司 | Performance abnormity early warning method for long-term service pumped storage unit |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104268658A (en) * | 2014-09-29 | 2015-01-07 | 招商局重庆交通科研设计院有限公司 | Bridge structure safety monitoring data prediction method |
| CN104899432A (en) * | 2015-05-19 | 2015-09-09 | 上海大学 | Kernel function combination-based PSO-LSSVM fluctuating wind speed prediction method |
| CN106529145A (en) * | 2016-10-27 | 2017-03-22 | 浙江工业大学 | ARIMA-BP neutral network-based bridge monitoring data prediction method |
-
2018
- 2018-06-29 CN CN201810692295.1A patent/CN108875841A/en not_active Withdrawn
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104268658A (en) * | 2014-09-29 | 2015-01-07 | 招商局重庆交通科研设计院有限公司 | Bridge structure safety monitoring data prediction method |
| CN104899432A (en) * | 2015-05-19 | 2015-09-09 | 上海大学 | Kernel function combination-based PSO-LSSVM fluctuating wind speed prediction method |
| CN106529145A (en) * | 2016-10-27 | 2017-03-22 | 浙江工业大学 | ARIMA-BP neutral network-based bridge monitoring data prediction method |
Non-Patent Citations (5)
| Title |
|---|
| YUNXIAO FU等: ""Product function correntropy and its application in rolling bearing fault identification"", 《MEASUREMENT》 * |
| 姜鑫 等: ""数据挖掘技术在水电厂主设备状态检修中的应用研究"", 《水电自动化与大坝监测》 * |
| 孙慧芳 等: ""基于时间序列组合模型的水电机组状态趋势预测"", 《水力发电学报》 * |
| 罗萌: ""水电机组振动故障诊断与趋势预测研究"", 《万方数据库》 * |
| 陆冠鹏: ""深圳抽蓄机组在线监测系统设计"", 《吉林水利》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109946075A (en) * | 2018-12-25 | 2019-06-28 | 东北大学 | A kind of bearing condition monitoring and method for diagnosing faults |
| CN110175541A (en) * | 2019-05-13 | 2019-08-27 | 武汉大学 | A kind of method that sea level variability nonlinear trend is extracted |
| CN110175541B (en) * | 2019-05-13 | 2022-06-14 | 武汉大学 | Method for extracting sea level change nonlinear trend |
| CN110298132A (en) * | 2019-07-05 | 2019-10-01 | 国家电网有限公司 | A kind of turbine-generator units bearing throw trend forecasting method |
| CN110673997B (en) * | 2019-09-06 | 2023-01-10 | 苏州浪潮智能科技有限公司 | Disk failure prediction method and device |
| CN110673997A (en) * | 2019-09-06 | 2020-01-10 | 苏州浪潮智能科技有限公司 | Disk failure prediction method and device |
| CN110943481A (en) * | 2019-12-04 | 2020-03-31 | 重庆大学 | Unit combination method considering wind turbine generator safety domain with frequency response control |
| CN110991055A (en) * | 2019-12-08 | 2020-04-10 | 中国航空综合技术研究所 | Residual life prediction system for rotary mechanical products |
| CN110991055B (en) * | 2019-12-08 | 2022-09-16 | 中国航空综合技术研究所 | Residual life prediction system for rotary mechanical products |
| CN111489027A (en) * | 2020-04-08 | 2020-08-04 | 五凌电力有限公司近尾洲水电厂 | Hydroelectric generating set waveform data trend prediction method and system |
| CN111695248B (en) * | 2020-05-27 | 2024-01-05 | 淮阴工学院 | Quick early warning method for degradation state trend of pumped storage unit |
| CN111695248A (en) * | 2020-05-27 | 2020-09-22 | 淮阴工学院 | Rapid early warning method for degradation state trend of pumped storage unit |
| CN112668419A (en) * | 2020-12-17 | 2021-04-16 | 太原理工大学 | Engine emission prediction method based on vibration signal |
| CN112668419B (en) * | 2020-12-17 | 2022-05-24 | 太原理工大学 | Engine emission prediction method based on vibration signal |
| CN114863651A (en) * | 2021-01-15 | 2022-08-05 | 湖南五凌电力科技有限公司 | Intelligent early warning method for monitoring state of auxiliary machine |
| CN112949986B (en) * | 2021-02-01 | 2022-06-28 | 中国人民解放军海军工程大学 | Steam power system operation stability evaluation method and system, electronic equipment and storage medium |
| CN112949986A (en) * | 2021-02-01 | 2021-06-11 | 中国人民解放军海军工程大学 | Steam power system operation stability evaluation method and system, electronic equipment and storage medium |
| CN113077172A (en) * | 2021-04-19 | 2021-07-06 | 瑞湖智科数据(苏州)有限公司 | Equipment state trend analysis and fault diagnosis method |
| CN114442543A (en) * | 2021-10-29 | 2022-05-06 | 南京河海南自水电自动化有限公司 | Computer monitoring method suitable for early warning of hydropower station fault |
| CN115795999A (en) * | 2022-10-26 | 2023-03-14 | 国网新源控股有限公司 | Performance abnormity early warning method for long-term service pumped storage unit |
| CN115511880A (en) * | 2022-11-07 | 2022-12-23 | 长江勘测规划设计研究有限责任公司 | Method for identifying and measuring vibration amplitude and frequency of two shafts by using machine vision |
| CN115685949A (en) * | 2022-12-29 | 2023-02-03 | 深圳市信润富联数字科技有限公司 | Method and device for adjusting data sampling frequency in discrete machining production process |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108875841A (en) | A kind of pumped storage unit vibration trend forecasting method | |
| CN106059661B (en) | A kind of optical transport network trend forecasting method based on Time-Series analysis | |
| US20190288904A1 (en) | Network Detection Method and Apparatus | |
| CN102063651B (en) | Urban power grid risk evaluation system based on on-line data acquisition | |
| CN1862278B (en) | Method and system for predicting remaining life for motors featuring on-line insulation condition monitor | |
| CN102867132B (en) | The online fault combination forecasting method of aviation DC converter based on fractional wavelet transform | |
| AU2012284459B2 (en) | Method of sequential kernel regression modeling for forecasting and prognostics | |
| CN108647891A (en) | Data exception classification, Reasons method and device | |
| KR101214765B1 (en) | System and method for monitoring electrical power equipment | |
| JP2015516530A (en) | Method and system for real-time performance recovery recommendations for centrifugal compressors | |
| CN115454778B (en) | Timing sequence index abnormity intelligent monitoring system in large-scale cloud network environment | |
| JP6210867B2 (en) | Data relationship analysis system and data management device | |
| CN112953738A (en) | Root cause alarm positioning system, method and device and computer equipment | |
| CN115642706A (en) | Power distribution load monitoring system in power grid | |
| US11657148B2 (en) | Event analysis in an electric power system | |
| Chen et al. | Real-time forecasting and visualization toolkit for multi-seasonal time series | |
| Aikhuele | Intuitionistic fuzzy model for reliability management in wind turbine system | |
| Jia et al. | Software aging analysis and prediction in a web server based on multiple linear regression algorithm | |
| CN111767193A (en) | Server data anomaly detection method and device, storage medium and equipment | |
| CN108845285B (en) | Electric energy metering device detection method and system | |
| CN102012902A (en) | Website visitor value estimation system and method | |
| CN114615134B (en) | IT intelligent operation and maintenance monitoring system and operation and maintenance method | |
| Bhoi et al. | Advanced edge computing framework for grid power quality monitoring of industrial motor drive applications | |
| CN109450103A (en) | Condition detection method, device and the intelligent terminal of pressing plate | |
| CN109784712A (en) | Lightning protection real-time monitoring and risk Prediction System and method based on Internet of Things |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20181123 |
|
| WW01 | Invention patent application withdrawn after publication |