CN104915534A - Deformation analysis and decision-making method of electric power tower based on sequence learning - Google Patents

Abstract

The invention belongs to the technical field of maintenance and monitoring of electric equipment, and relates to a deformation analysis and decision-making method of an electric power tower based on sequence learning. Dynamic displacement monitoring is realized through a Beidou navigation and positioning RTK resolving method. Static displacement monitoring is realized through a static resolving method. The posture of the electric power tower is monitored through an inclination sensor. Horizontal and vertical relative displacement monitoring of the electric power tower is realized through an acceleration sensor. Windage yaw and icing monitoring is realized through a meteorological sensor. Multi-dimensional combined application is realized through the monitoring method, and intelligent monitoring is realized through early warning of front-end software and analysis processing of central-end software.

Description

Based on electric power tower deformation analysis and the decision-making technique of Sequence Learning

Technical field

The invention belongs to power equipment to safeguard and monitoring technical field, relate to a kind of electric power tower deformation analysis based on Sequence Learning and decision-making technique.The present invention utilizes the sustainable raising of Sequence Learning artificial interpretation measurement data efficiency, reduces the error in judgement of human factor.

Background technology

Tradition deformation monitoring generally adopts ground routine measuring technique, terrestrial photogrammetry technology, GPS space orientation technique to apply, use total powerstation, camera, GPS navigation positioning equipment, the artificial interpretation measurement data of general employing, analyzes and judges steel tower deformation state.GPS navigation technology for deformation monitoring adopts single Time series analysis method modeling, cannot judge the burst displacements such as windage yaw by accurate analysis.Artificial experience analysis cannot process large data and sampling error.Single Time series analysis method, can not solve the deformation of multifactor generation, multifactorly generally comprises horizontal sedimentation, windage yaw, icing etc.

Summary of the invention

The object of the invention is to the above-mentioned deficiency overcoming prior art, the present invention utilizes Sequence Learning theoretical, set up built-up pattern, for power tower deformation analysis, this method solve the modeling that conventional electric power tower deformation monitoring mode exists single coarse, the problems such as artificial judgment precision is not high, efficiency is low.

The technical solution adopted for the present invention to solve the technical problems is: first use Kalman filtering algorithm to carry out raw data screening, secondly one-variable linear regression and least square method is used to carry out the check of screening rear data, reuse Wavelet Analysis Theory and Filtering Analysis is carried out to data after check, then set up autoregressive moving average (Auto-Regressive and Moving Average is called for short ARMA) model and carry out tentative prediction, finally set up built-up pattern and analyze iron tower deformation state.

The described electric power tower deformation analysis based on Sequence Learning and decision-making technique, described use Kalman filtering algorithm carries out raw data screening and comprises following concrete steps:

The first step: utilize Kalman filtering algorithm, determines the observation equation of the whole network, the mathematical model of Kalman filtering;

Second step: employing Kalman filtering is carried out filtering, prediction to the coordinate of each monitoring point, analyzes the consistance of real-time monitored value and predicted value vector.

The described electric power tower deformation analysis based on Sequence Learning and decision-making technique, the check that described use one-variable linear regression and least square method carry out screening rear data specifically comprises following step:

The first step: according to steel tower distortion image data, calculate acceleration and inclination angle, acceleration and the data resolved by real-time dynamic system algorithm (Real-time kinematic is called for short RTK), the related coefficient of acceleration and wind speed;

Second step: set up acceleration and inclination angle, acceleration and RTK resolved data, the equation of acceleration and wind speed;

3rd step: use least square method to calculate the regression parameter of above-mentioned three equations;

4th step: use equation to calculate data, contrast image data, find problem data.

The described electric power tower deformation analysis based on Sequence Learning and decision-making technique, it is as follows that described use Wavelet Analysis Theory carries out Filtering Analysis concrete steps to data after check:

The first step: utilize wavelet algorithm to decompose the data collected;

Second step: quantification treatment is carried out to the wavelet decomposition high frequency coefficient threshold value of image data useful information and noise;

3rd step: utilize wavelet algorithm to reconstruct image data.The n-th layer low frequency coefficient of wavelet decomposition and rearranging through the 1st layer of high frequency coefficient to n-th layer of threshold value quantizing process, can obtain the observation data sequence estimation value after denoising, i.e. accuracy of observation estimated value.

The described electric power tower deformation analysis based on Sequence Learning and decision-making technique, described set up arma modeling to carry out tentative prediction concrete steps as follows:

The first step: the determination of ARMA Mixed Regression Model factor exponent number;

Second step: the selection of the independent variable factor;

3rd step: the pre-service of original observational data;

4th step: result of calculation analysis.

The described electric power tower deformation analysis based on Sequence Learning and decision-making technique, described built-up pattern of setting up analyzes iron tower deformation state, and concrete steps are as follows:

The first step: utilize finite element analysis theoretical, dividing elements is carried out to problem steel tower, be different from traditional space rigid model, for steel tower deformation factor, devise trusses mixture model especially, be defined as beam element by the main structure of steel tower and tabula structure, slant-pull structure is defined as bar unit, closer to the iron tower structure of reality, and the masty problem of employing truss model can also be solved easily;

Second step: structure steel tower Bulk stiffness matrix, computation structure equivalent joint load, determines constitutional balance equation;

3rd step: solve steel tower nodal equilibrium equation, obtain nodal displacement, calculates each cellular construction stress, carries out forecast analysis to iron tower deformation, here will in conjunction with weather data, inertia measurement data verification.

The present invention has following beneficial effect:

The invention provides a kind of electric power tower deformation analysis based on Sequence Learning and decision-making technique, first the method establishes built-up pattern based on sequence of events, solves artificial interpretation measurement data efficiency low, analyzes and judges the problem that deformation state error is larger; Meanwhile, built-up pattern differentiates for windage yaw burst misalignment energy accurate analysis, improves danger warning accuracy, and can realize automatically processing measurement data in enormous quantities, carry out the problem of Multifactor Combination analysis, improve accuracy and the reliability of monitoring system.

Accompanying drawing explanation

Fig. 1 is the implementing procedure of power tower deformation analysis based on Sequence Learning and decision-making technique.

Embodiment

Below in conjunction with embodiment, the specific embodiment of the present invention is described in further detail.Following examples for illustration of the present invention, but are not used for limiting the scope of the invention.

Based on the power tower deformation analysis of Sequence Learning and decision-making technique implementing procedure as shown in Figure 1.Specific implementation method is as follows:

1, basic data Kalman filtering process

Present stage, Beidou satellite navigation was in deformation observation, normally in the state of the instant static research monitoring point determined, and did not consider that monitoring point changes the dynamic perfromance of displacement in time.And in observation process, observation data is often discontinuous, so just real-time accurate description cannot be carried out to dynamic deformation.Kalman filtering has minimum unbiased variance.In deformation monitoring, if deformable body to be considered as the motion conditions that dynamically just can be used for describing this deformable body, and the funtcional relationship between state parameter and its observed reading representing them can be found out in this system respectively, Kalman filtering so just can be utilized to weaken the interference of random noise, and then reach the object improving deformation observation data precision.

Based in the steel tower deformation monitoring of Beidou satellite navigation, with the position of monitoring point, rate parameter for state vector, in state equation, add the noise of system again, construct a typical motion model.Deformation Control Net is made up of n monitoring point, and with monitoring point three-dimensional position and three-dimensional speed for state vector, the i that sets up an office is ξ in the position vector of moment t _it (), its momentary rate is λ _i(t), and by instantaneous rate of acceleration Ω _it () regards a kind of random disturbance as, then have following differential relationship:

$\left.\begin{array}{c}{\mathrm{\ξ}}_i\left(t\right)={\mathrm{\λ}}_i\left(t\right)\\ {\mathrm{\λ}}_i\left(t\right)={\mathrm{\Ω}}_i\left(t\right)\end{array}\right\}---\left(1\right)$

Getting i dotted state vector is X _it (), the whole network state vector is X (t), has n point to be located, namely in net

$\begin{array}{c}{X}_i\left(t\right)={\left[\begin{array}{cc}{\mathrm{\ξ}}_i\left(t\right)& {\mathrm{\λ}}_i\left(t\right)\end{array}\right]}^r\\ ={\left[\begin{array}{cccccc}{X}_i\left(t\right)& Y_i\left(t\right)& Z_i\left(t\right)& {\stackrel{\·}{X}}_i\left(t\right)& {\stackrel{\·}{Y}}_i\left(t\right)& {\stackrel{\·}{Z}}_i\left(t\right)\end{array}\right]}^T\end{array}---\left(2\right)$

Wherein,

X _i(t)＝[X ₁(t) X ₂(t)…X _n(t)] ^r

${\mathrm{\Ω}}_i\left(t\right){\left[\begin{array}{ccc}{\stackrel{\·\·}{X}}_i\left(t\right)& {\stackrel{\·\·}{Y}}_i\left(t\right)& {\stackrel{\·\·}{Z}}_i\end{array},\left(t\right)\right]}^T---\left(3\right)$

Ω _i(t)[Ω ₁(t) Ω ₂(t)…Ω _n(t)] ^T

The former differential equation can be written as:

${\stackrel{\·}{X}}_i\left(t\right)=\left[\begin{array}{cc}0& E\\ 0& 0\end{array}\right]X_i\left(t\right)+\left[\begin{array}{c}0\\ E\end{array}\right]{\mathrm{\Ω}}_i\left(t\right)---\left(4\right)$

In formula, O and E represents three rank null matrix and three rank unit matrixs respectively, establishes Δ t after discrete _k=t _k-1-t _k, t _kand t _k-1be respectively the observation moment of kth phase and kth-1 phase, be respectively X at t _k, t _k-1time speed, for t _kacceleration.State equation according to n point obtains scale form, and the state equation that can obtain the whole network is

(5)

Matrix representation is

X _k+1＝φ _k+1，kX _k+Γ _k+1，kΩ _k(6)

The observation equation of the whole network is:

L _k+1＝B _k+1X _k+1+Δ _k+1(7)

Finally obtain, the mathematical model that dynamic equation and measurement equation form steel tower Deformation Control Net Kalman filtering is

$\left.\begin{array}{c}{X}_{k+1}={\mathrm{\φ}}_{k,k}{X}_{k-1}+{\mathrm{\Γ}}_{k,k-1}{\mathrm{\Ω}}_{k-1}\\ L_k=B_k{X}_k+{\mathrm{\Δ}}_k\end{array}\right\}---\left(8\right)$

Two, one-variable linear regression and least square method data check

According to steel tower distortion image data, calculate acceleration and inclination angle, acceleration and the data resolved by real-time dynamic system algorithm (Real-time kinematic is called for short RTK), formula (9) is utilized to calculate acceleration and inclination angle, acceleration and RTK resolved data, the related coefficient of acceleration and wind speed

With reference to table 1, when what calculate be greater than analog value in table, can think that correlativity meets the condition of configuration regression straight line, least square method can be used to set up and calculate acceleration and inclination angle, acceleration and RTK resolved data, the regression parameter equation of related coefficient three equations of acceleration and wind speed.

Table 1 related-coefficient test method critical value

Three, wavelet theory carries out data hierarchy filtering

Electric power tower deformation monitoring is in actual observation application, the data recorded often are subject to multiple random or uncertain factor impact and produce error interference, these error interferences are general less, and there is randomness, the combined influence of signal is shown as and superposes stochastic error in the signal, namely contain white Gaussian noise in observation sequence, and stochastic error is disturbed with systematicness, also can contain mutation disturbance or rough error simultaneously, need to adopt the deformation time series Data Denoising based on wavelet package transforms.

The first step: utilize wavelet algorithm to decompose the data collected;

Decomposition level is larger, more by the noise filtered, and the distortion of synchronous signal is also larger, so must select a best decomposition level J, under the distortionless prerequisite of guarantee signal, farthest filters noise.

Whether the signal to noise ratio (S/N ratio) of the deformation observation data of actual measurement is unknown in advance, method of estimation can be adopted to determine best decomposition level, increase decomposition level gradually, then tend towards stability according to the change of square error (RMSE) and determine best decomposition level J.

$\mathrm{RMSE}\left(j\right)=\sqrt{\frac{1}{n}\underset{n}{\mathrm{\Σ}}{[f\left(n\right)-{\hat{f}}_j\left(n\right)]}^2}---\left(10\right)$

And calculate successively:

$r_{j+1}=\frac{\mathrm{RMSE}(j+1)}{\mathrm{RMSE}\left(j\right)}---\left(11\right)$

Generally always there is r>l, when r is close to l, generally can think r≤1.1, then think that noise is removed substantially.Best decomposition level j be make r close to 1 time j or j+l.By test, j generally gets 3 to 5.

Second step: quantification treatment is carried out to the wavelet decomposition high frequency coefficient threshold value of image data useful information and noise;

Threshold estimation is one of key of the wavelet packet threshold denoising of steel tower Deformation Monitoring Signal, if threshold value is too little, the signal after de-noising still exists noise; And threshold value is too large, important signal characteristic will be filtered again, causes deviation.With reference to SNR and the RMSE contrast table of various threshold estimation criterion de-noising.

Table 2 SNR and RMSE contrast table

Birge-massart threshold value criterion is selected to calculate:

$\mathrm{crit}\left(t\right)=-\mathrm{sum}({c\left(k\right)}^2,k\≤t)+2\×{\mathrm{\σ}}^2\×t\×(\mathrm{alpha}+\log \left(\frac{n}{2}\right))---\left(12\right)$

In formula, c (k) is wavelet packet coefficient, it be by absolute value successively decrease tactic, n is coefficient number; Alpha be adjustment parameter, must be greater than 1 real number, its value is larger, and the wavelet packet of de-noising signal represents more sparse, and the representative value of alpha is 2.

3rd step: utilize wavelet algorithm to reconstruct image data.The n-th layer low frequency coefficient of wavelet decomposition and rearranging through the 1st layer of high frequency coefficient to n-th layer of threshold value quantizing process, can obtain the observation data sequence estimation value after denoising, i.e. accuracy of observation estimated value.

Four, arma modeling tentative prediction steel tower deformation data

ARMA mixing recurrence need return according to actual steel tower deformation observation data, obtains the regression coefficient of each dependent variable, independent variable and each independent variable history value factor under sum of square of deviations minimal condition.

The first step: the determination of ARMA Mixed Regression Model factor exponent number

Steel tower deformation monitoring arma modeling is selected because of the period of the day from 11 p.m. to 1 a.m, and independent variable exponent number can not be unlimited many, must do in practice and suitably choose, and general AIC or the BIC criterion of adopting reaches minimum model as best model to make criterion function.

$\mathrm{AIC}\left(n\right)=\lg {\left({\mathrm{\σ}}_{\mathrm{\ϵ}}^{\left(n\right)}\right)}^2+\frac{2n}{N}---\left(13\right)$

$\mathrm{BIC}\left(n\right)=\lg {\left({\mathrm{\σ}}_{\mathrm{\ϵ}}^{\left(n\right)}\right)}^2+\frac{n\left(\mathrm{lgN}\right)}{N}---\left(14\right)$

In formula, for the regression criterion variance of n rank model; N is exponent number (number of parameters), and N subtracts maximum delayed step number for observing number.

Second step: the selection of the independent variable factor

The selection of the independent variable factor should construct according to electric power tower, affect the factor of steel tower distortion and observational data is determined, different tower is different with the factor that structure influence steel tower is out of shape.Also need the multiple effect such as linear, non-linear, delayed considering influence factor simultaneously.

(1) the windage yaw displacement component factor.Wind-force change has linear, nonlinear effect to electric power tower distortion, thus select monthly wind-force mean value and two, cube (P, P2, P3) three factors are as the windage yaw displacement component factor.

(2) temperature movement component factor.Temperature variation has linear processes effect to steel tower distortion, thus get monthly temperature mean value and two, cube (T, T2, T3) is as temperature movement component factor.

(3) the dependent variable factor.Displacement observation result, as the dependent variable factor, monthly gets an observed reading.

(4) determination of delayed step number.From field data graph, the periodicity of this steel tower distortion is not obvious.Rule of thumb, the lag-effect of steel tower influence factor is generally less than 6 months, therefore gets delayed step number 6 and can meet actual needs.

3rd step: the pre-service of original observational data

ARMA mixing regretional analysis is that order carries out analyzing and processing to electric power tower deformation observation data with time, require that observational data is continuous, sequence stationary, and the observation interval selected should be equal, just can guarantee that deformation analysis is reliable, Deformation Prediction is accurate.But be difficult in practice reach above-mentioned requirements, should first carry out lacking survey, steady and standardization to original deformation observation data when therefore analyzing steel tower distortion.

4th step: result of calculation analysis

Before adopting, the observational data of 10a is set up distorted pattern, is calculated regression coefficient, and the observational data of rear 2a is as forecasting verifying data and comparing with predicted value.Be below the result of calculation of 3 horizontal shifts in left, center, right at steel tower top.

Left: σ _ε=± 3.198mm

In: σ _ε=± 3.143mm

Right: σ _ε=± 3.331mm

Five, steel tower finite element model analysis deformation state

The first step: utilize finite element analysis theoretical, dividing elements is carried out to problem steel tower, be different from traditional space rigid model, for steel tower deformation factor, devise trusses mixture model especially, be defined as beam element by the main structure of steel tower and tabula structure, slant-pull structure is defined as bar unit, closer to the iron tower structure of reality, and the masty problem of employing truss model can also be solved easily.

Second step: structure steel tower Bulk stiffness matrix, computation structure equivalent joint load, determines constitutional balance equation.

(1) global stiffness matrix of unit

By shifting the stiffness matrix of the beam element of iron tower model onto, drawing the stiffness matrix of bar unit, considering axial displacement, torsional displacement, bending displacement simultaneously, the global stiffness matrix deriving unit is as follows:

Wherein,

E is the elastic modulus of unit material; A is the sectional area of unit; L is the length of unit; R is the distance from unit neutral axis to edge; G is the modulus of shearing of unit material.

3rd step: solve steel tower nodal equilibrium equation, obtain nodal displacement, calculates each cellular construction stress, carries out forecast analysis to iron tower deformation, here will in conjunction with weather data, inertia measurement data verification.

Application context of methods, first the internal force calculating test steel tower determine the load of steel tower, herein the deadweight of consideration steel tower and centre-point load.For " 90 degree of strong wind " operating mode.It is as shown in table 3 below that software calculates load(ing) point load.

Table 3 software calculates load(ing) point load

Claims (6)

1. based on electric power tower deformation analysis and the decision-making technique of Sequence Learning, it is characterized in that, first Kalman filtering algorithm is used to carry out raw data screening, secondly use one-variable linear regression and least square method to carry out the check of screening rear data, reuse Wavelet Analysis Theory and Filtering Analysis is carried out to data after check; Then set up autoregressive moving average (Auto-Regressive and Moving Average is called for short ARMA) model and carry out tentative prediction, finally set up built-up pattern and analyze iron tower deformation state.

2. the electric power tower deformation analysis based on Sequence Learning according to claim 1 and decision-making technique, is characterized in that, described use Kalman filtering algorithm carries out raw data screening and comprises following concrete steps:

The first step: utilize Kalman filtering algorithm, determines the observation equation of the whole network, the mathematical model of Kalman filtering;

Second step: employing Kalman filtering is carried out filtering, prediction to the coordinate of each monitoring point, analyzes the consistance of real-time monitored value and predicted value vector.

3. the electric power tower deformation analysis based on Sequence Learning according to claim 1 and decision-making technique, is characterized in that, the check that described use one-variable linear regression and least square method carry out screening rear data specifically comprises following step:

The first step: according to steel tower distortion image data, calculate acceleration and inclination angle, acceleration and the data resolved by real-time dynamic system algorithm (Real-time kinematic is called for short RTK), the related coefficient of acceleration and wind speed;

Second step: set up acceleration and inclination angle, acceleration and RTK resolved data, the equation of acceleration and wind speed;

3rd step: use least square method to calculate the regression parameter of above-mentioned three equations;

4th step: use equation to calculate data, contrast image data, find problem data.

4. the electric power tower deformation analysis based on Sequence Learning according to claim 1 and decision-making technique, is characterized in that, it is as follows that described use Wavelet Analysis Theory carries out Filtering Analysis concrete steps to data after check:

The first step: utilize wavelet algorithm to decompose the data collected;

Second step: quantification treatment is carried out to the wavelet decomposition high frequency coefficient threshold value of image data useful information and noise;

3rd step: utilize wavelet algorithm to reconstruct image data;

Of wavelet decomposition nlayer low frequency coefficient and through the 1st layer of threshold value quantizing process to the nthe high frequency coefficient of layer rearranges, and can obtain the observation data sequence estimation value after denoising, i.e. accuracy of observation estimated value.

5. the electric power tower deformation analysis based on Sequence Learning according to claim 1 and decision-making technique, is characterized in that, described set up arma modeling to carry out tentative prediction concrete steps as follows:

The first step: the determination of ARMA Mixed Regression Model factor exponent number;

Second step: the selection of the independent variable factor;

3rd step: the pre-service of original observational data;

4th step: result of calculation analysis.

6. the electric power tower deformation analysis based on Sequence Learning according to claim 1 and decision-making technique, is characterized in that, described built-up pattern of setting up analyzes iron tower deformation state, and concrete steps are as follows:

The first step: utilize finite element analysis theoretical, dividing elements is carried out to problem steel tower, be different from traditional space rigid model, for steel tower deformation factor, devise trusses mixture model especially, be defined as beam element by the main structure of steel tower and tabula structure, slant-pull structure is defined as bar unit, closer to the iron tower structure of reality, and the masty problem of employing truss model can also be solved easily;

Second step: structure steel tower Bulk stiffness matrix, computation structure equivalent joint load, determines constitutional balance equation;

3rd step: solve steel tower nodal equilibrium equation, obtain nodal displacement, calculates each cellular construction stress, carries out forecast analysis to iron tower deformation, here will in conjunction with weather data, inertia measurement data verification.