CN104239970A - Power transmission line gallop risk early-warning method based on Adaboost - Google Patents

Abstract

The invention provides a power transmission line gallop risk early-warning method based on Adaboost. The method comprises the following steps that internal reasons of gallop of power transmission lines are classified, and statistics is carried out on meteorological feature factors in historical gallop accidents of the power transmission lines according to the classification result; according to the information of a power transmission line to be predicted, the class, in the classification result of the internal reasons of gallop, corresponding to the power transmission line is selected, the meteorological feature factors under the conditions of the historical gallop accidents of the class are recorded to form a training sample set, a classifier is formed with an Adaboost ensemble learning algorithm, forecast data of the meteorological feature factors of the gallop of the power transmission line serve as input, and a gallop early-warning result of the power transmission line is obtained through the classifier; according to the early-warning result, the early-warning level of the gallop of the power transmission line is obtained through judgment. According to the power transmission line gallop risk early-warning method based on Adaboost, the internal reasons and the external reasons influencing the gallop of the power transmission line are comprehensively considered, the historical gallop information and the weather forecast information of the power transmission line are made full use of, and the method meets the actual conditions better; the algorithm in use is high in generalization ability, easy to encode and high in early-warning result accuracy.

Description

A kind of conductor galloping method for prewarning risk based on Adaboost

Technical field

The invention belongs to the failure risk early warning technology field of the overhead transmission line of electric system, specifically a kind of conductor galloping method for prewarning risk based on Adaboost algorithm.

Background technology

Conductor galloping be guide line wind-force and (or) low frequency, the significantly autovibration that cause under the effect of asymmetric icing, be a kind of Aerodynamic Instability phenomenon.Conductor galloping mostly occurs in the winter time, its energy is very large, and the duration is long, easily physical damage and electric fault are caused to transmission line of electricity, light then cause alternate flashover, damage wires, ground wire and gold utensil etc., heavy then cause disconnected stock, broken string, even fall the severe accident such as tower, the safe and stable operation of transmission line of electricity in serious threat.

Operation and observation and statistical data show, China is one of country that conductor galloping disaster is the most serious.Along with development and the boisterous frequent appearance of China's electrical network scale, the occurrence frequency of conductor galloping accident and the extent of injury all have obvious increase, and conductor galloping region is not only confined within the scope of minority, also throughout the most area to national grid.Therefore, to the research of conductor galloping and preventive measure thereof, there is important theory significance and engineering practical value.

In the last few years, Chinese scholars had carried out many-sided research to conductor galloping excitation mechanism, computer sim-ulation and power transmission line Anti-Oscillation Measures etc., achieved a lot of important achievement and was applied to Practical Project.Regrettably, because power transmission line and air-flow interact the geometrical non-linearity etc. caused of significantly moving of the coupling that causes and power transmission line, make conductor galloping problem become very complicated, there is no unified, pervasive conductor galloping excitation theory up to now.

Existing power transmission line Anti-Oscillation Measures sums up and can be divided into three major types: one is consider from meteorological condition, avoids being easy to forming the icing region and line alignment of waving; Two is that design improves the anti-ability of waving of line system from machinery and electric angle; Three is take various anti-dancing device, suppresses the generation of waving, for built circuit, and the way of unique feasible especially.But it should be noted that current conductor galloping defensive measure still has following deficiency:

1) consider the economy designing requirement of saving line corridor and the factors such as cheap property of constructing, make part transmission line of electricity cannot avoid waving district completely;

2) in actual applications, power transmission line quality enhancement techniques and the full and accurate still not and specification of anti-dance design, economy and operability poor, simultaneously also lack practical experience;

3) anti-dance device obtains based on different conductor galloping mechanism exploitation, and cause several anti-dancing device that application is more at present, all have its obvious design feature and application limitation, anti-dance effect also exists very big-difference.

Obviously, want to subdue the effort that conductor galloping also needs a very long time completely, at present, power transmission line Anti-galloping aid decision-making method that range of application is wider stronger in the urgent need to a kind of initiative, what alleviate that transmission system suffers waves disaster.Conductor galloping on-Line Monitor Device and method obtain flourish in recent years, and the successful operation of conductor galloping on-line monitoring system is the meteorological data that research conductor galloping have accumulated preciousness; In addition, the degree that becomes more meticulous of weather forecast in recent years and accuracy all have a distinct increment, more and more closer with electrical network cooperation, make to utilize weather information to realize the approach that conductor galloping Risk-warning is a kind of feasible, science.

Summary of the invention

For the deficiency of the existing measure of above-mentioned analysis, the invention provides a kind of conductor galloping method for prewarning risk based on Adaboost, the computing to related datas such as the forecast information of conductor galloping Meteorological Characteristics factor and the structural parameters of transmission line of electricity can be realized, and export the conductor galloping disaster alarm analysis result of region.

In order to solve the technical problem, the technical solution used in the present invention is:

Based on a conductor galloping method for prewarning risk of Adaboost, comprise the following steps;

(1), by the internal cause of conductor galloping classify, and by classification results, the Meteorological Characteristics factor that power transmission line history is waved under accident is added up;

Wherein internal cause classification results includes: the 1. type of circuit, is divided into S.C. and split conductor; 2. the sectional area of circuit, is divided into large, medium and small cross section circuit; 3. line span, is divided into large, medium and small span circuit;

Meteorological Characteristics factor includes: 1. wind speed; 2. wind direction to traverse shaft to angle; 3. temperature; 4. relative humidity;

(2), according to predicted transmission line information, select a class corresponding with conductor galloping internal cause classification results, and the Meteorological Characteristics factor information record composing training sample set waved with history in such under emergency conditions, sorter is formed with Adaboost Ensemble Learning Algorithms, again using the forecast data of conductor galloping Meteorological Characteristics factor that describes in the step 1 that obtains from meteorological department as input, obtain transmission line galloping early warning Output rusults and degree of confidence margin value by sorter;

(3), according to the early warning Output rusults described in step 2, according to the degree of confidence margin value that conductor galloping predicts the outcome, judge to obtain conductor galloping Risk-warning grade.

Wherein said step (2) is specific as follows:

2.1), input: training sample set X, wherein should comprise sample class label; N is number of training; T is frequency of training, is also Weak Classifier number; Wherein Weak Classifier sorting algorithm is C, and specify in sample class label that 1 represents that power transmission line is waved ,-1 represents that power transmission line is waved;

2.2), initialization: sample weights distribution w ₁(i)=1/N, i=1,2 ..., N;

2.3), t=1 is worked as, 2 ..., T:

1. according to the sample weights distribution w of the t time _ti () carries out there is the sampling of putting back to from original sample collection X, generate new sample set X _t;

2. at X _tupper training Weak Classifier C _t, and use C (X) _t(X) original sample collection X is classified;

3. Weak Classifier C is calculated _t(X) classification error rate;

${\mathrm{\ϵ}}_t=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_t\left(i\right)I(C_t\left(x_i\right)\≠y_i)---\left(1\right)$

In formula (1), work as C _t(x _i) ≠ y _itime, I (g) is 1, and all the other are then 0;

4. Weak Classifier C is calculated _t(X) coefficient;

$a_t=\frac{1}{2}\ln \left(\frac{1-{\mathrm{\ϵ}}_t}{{\mathrm{\ϵ}}_t}\right)---\left(2\right)$

5. weights distribution is upgraded;

$w_{t+1}\left(i\right)=\frac{w_t\left(i\right)}{Z_t}\×\left\{\begin{array}{cc}{e}^{-a_t},& C_t\left(x_i\right)=y_i\\ e^{a_t},& C_t\left(x_i\right)\≠y_i\end{array}\right.=\frac{w_t\left(i\right)}{Z_t}\·e^{-a_t{y}_i{C}_t\left(x_i\right)},i=\mathrm{1,2},L,N---\left(3\right)$

In formula (3), $Z_t=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_t\left(i\right)\·e^{-a_t{y}_i{C}_t\left(x_i\right)}$ Be normalized factor, make $\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{w}_{t+1}\left(i\right)=1;$

2.4), final sorter:

$C\left(X\right)=\mathrm{sgn}\left[\underset{t=1}{\overset{T}{\mathrm{\Σ}}}{a}_t{C}_t\left(X\right)\right]---\left(4\right)$

In formula (4), function sgn () is sign function, and its concrete mathematic(al) representation is

$\mathrm{sgn}\left(x\right)=\left\{\begin{array}{cc}1,& x\&GreaterEqual;0\\ -1,& x<0\end{array}\right.---\left(5\right)$

Then conductor galloping early warning output packet to predict the outcome y and degree of confidence margin (x, y) containing sorter:

$y=C\left(x\right)=\mathrm{sgn}\left[\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{a}_t{C}_t\left(x\right)\right]---\left(6\right)$

$m\arg \mathrm{in}(x,y)=\frac{y\underset{t}{\mathrm{\&Sigma;}}{a}_t{C}_t\left(x\right)}{\underset{t}{\mathrm{\&Sigma;}}\left|a_t\right|}---\left(7\right)$

In formula, x is the forecast data of conductor galloping Meteorological Characteristics factor; Y ∈-1 ,+1}, for sorter predicts the outcome: 1 for predicting that this power transmission line will be waved, and-1 for predicting that this power transmission line will not be waved; Margin ∈ [-1 ,+1], it is with a high credibility that larger positive boundary represents that this circuit of prediction is waved, and it is with a high credibility that larger negative edge represents that this circuit of prediction is not waved, and less border then represents that the confidence level predicted the outcome is lower.

Because meteorological condition affects the most important extraneous factor that conductor galloping excites, when internal cause is relatively constant, the change of things will be determined by external cause.Narrow in order to make up existing conductor galloping defensive measure range of application, initiative difference and the deficiency such as utilization is lacked to weather information, the present invention proposes a kind of conductor galloping method for prewarning risk from operation angle, conductor galloping early warning problem arises is the classification forecasting problem under supervised learning by the present invention, by conductor galloping internal cause, statistical treatment is disaggregatedly carried out to the Meteorological Characteristics factor information that power transmission line history is waved in situation, strong classifier is set up by Adaboost Ensemble Learning Algorithms, the related datas such as the location parameter of COMPREHENSIVE CALCULATING process Weather Forecast Information and transmission line of electricity, what provide region transmission line of electricity waves risk class, realize the conductor galloping disaster alarm of science.Early warning result can be operation of power networks dispatcher and carries out the decision support that Anti-Oscillation Measures provides science, carry out in advance targetedly wind resistance, except power transmission line Anti-Oscillation Measures such as ice-melts, avoid taking precautions against deficiency cause accident and excessively take precautions against waste resource, transmission line galloping failure rate can be reduced, ensure the safe and stable operation of transmission line of electricity.

The present invention is generally applicable to the early warning of electric system conductor galloping, particularly conductor galloping Frequent Accidents area, and compared to existing technology, tool of the present invention has the following advantages:

1) the present invention has considered the internal cause and external cause that affect conductor galloping, makes full use of power transmission line history and waves information, more realistic;

2) the present invention adopts Adaboost Ensemble Learning Algorithms, have generalization ability (namely from sample data learning to rule can be applicable to the ability of new data) strong, the easy advantage such as coding, early warning result reliability is high.

Accompanying drawing explanation

Fig. 1 is the conductor galloping Risk-warning process flow diagram based on Adaboost;

Fig. 2 is decision-making pile (Weak Classifier) process flow diagram splitting criterion based on Gini.

Embodiment

Because the physical model of existing conductor galloping is accurate not, and the Some Parameters in model is difficult to obtain by measuring on actual track, make to utilize physical model to carry out the practicality of conductor galloping early warning and accuracy lower, and now machine Learning Theory just for we providing good method for early warning.Machine learning obtains based on observation before and predicts more accurately, it provide a kind of from observation data up till now still not by the rule that principle analysis obtains, and then utilize the method for these law forecasting Future Datas.

The present invention proposes a kind of conductor galloping method for prewarning risk based on Adaboost algorithm, described Adaboost algorithm is that the self-adaptation in Ensemble Learning Algorithms strengthens (Adaptive boosting, be called for short Adaboost) algorithm, its basic thought is the Weak Classifier utilizing a large amount of classification capacities general, by certain method superposition, gather, form the final sorter (strong classifier) that a classification capacity is stronger.The present invention can realize the computing to related datas such as the forecast information of conductor galloping Meteorological Characteristics factor and the structural parameters of transmission line of electricity, and exports the conductor galloping disaster alarm analysis result of region.

Below in conjunction with embodiment and accompanying drawing, the inventive method is done and describes clearly and completely further, but embodiments of the present invention are not limited to this.

As shown in Figure 1, the present invention includes following steps:

(1), by affect that conductor galloping excites internal cause (conductor galloping internal cause refers to the internal factor affecting conductor galloping and excite)---line construction and parameter are classified by shown in table 1, amount to 18 kinds of combinations, such as: S.C., small bore, little span circuit are a kind of combination.

Table 1 conductor galloping internal cause categorised statistical form

And to power transmission line history wave this external cause of Meteorological Characteristics factor information carry out statistics sort out, conductor galloping Meteorological Characteristics factor refers to the external meteorological factor affecting conductor galloping and excite, and conductor galloping Meteorological Characteristics factor mainly comprises wind speed, wind direction to wire axis angle, temperature and relative humidity.It should be noted that the change due to height can have impact to wind speed, and weather station observe the wind speed that obtains and weather forecast wind speed be generally be defaulted as the high wind speed of liftoff 10m, therefore should by the unified conversion of following formula to the wind speed v at conductor height place _l.

$v_l=v_q\&CenterDot;{\left(\frac{H}{10}\right)}^{\mathrm{\&mu;}}---\left(1\right)$

In formula, v _qfor liftoff 10m eminence wind speed, H is conductor height (m); μ is ground roughness exponent, is respectively 0.12,0.15,0.22 and 0.30 according to the ground roughness exponent of marine, rural area, city and center, big city 4 class.

(2), according to the actual conditions of tested transmission line of electricity, select a class corresponding in above-mentioned 18 kinds of combinations, the Meteorological Characteristics factor information record composing training sample set under accident is waved with the power transmission line history described in step 1, form strong classification learning device with the Adaboost Ensemble Learning Algorithms that the present embodiment is following, and the forecast data of the Meteorological Characteristics factor provided according to meteorological department carries out conductor galloping early warning.Specific as follows:

2.1), input: conductor galloping training sample set X={ (x ₁, y ₁), (x ₂, y ₂) ..., (x _n, y _n); Wherein, x _iit is the Meteorological Characteristics factor vector of i-th conductor galloping sample; y _i={-1, the 1} category label representing i-th sample :-1 represents that power transmission line is waved, and 1 represents that power transmission line is waved; N is number of training; T is frequency of training (being also Weak Classifier number).Weak Classifier sorting algorithm C adopts existing method individual layer decision tree, also can adopt the existing method that support vector machine etc. can realize.

2.2), initialization: sample weights distribution w ₁(i)=1/N, i=1,2 ..., N.

2.3), t=1 is worked as, 2 ..., T:

1. according to the sample weights distribution w of the t time _ti () carries out there is the sampling of putting back to from original sample collection X, generate new sample set X _t;

2. at X _tupper training Weak Classifier C _t, and use C (X) _t(X) original sample collection X is classified;

3. Weak Classifier C is calculated _t(X) classification error rate;

${\mathrm{\&epsiv;}}_t=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_t\left(i\right)I(C_t\left(x_i\right)\&NotEqual;y_i)---\left(2\right)$

In formula (2), work as C _t(x _i) ≠ y _itime, I (g) is 1, and all the other are then 0.

4. Weak Classifier C is calculated _t(X) coefficient;

$a_t=\frac{1}{2}\ln \left(\frac{1-{\mathrm{\&epsiv;}}_t}{{\mathrm{\&epsiv;}}_t}\right)---\left(3\right)$

5. weights distribution is upgraded;

$w_{t+1}\left(i\right)=\frac{w_t\left(i\right)}{Z_t}\&times;\left\{\begin{array}{cc}{e}^{-a_t},& C_t\left(x_i\right)=y_i\\ e^{a_t},& C_t\left(x_i\right)\&NotEqual;y_i\end{array}\right.=\frac{w_t\left(i\right)}{Z_t}\&CenterDot;e^{-a_t{y}_i{C}_t\left(x_i\right)},i=\mathrm{1,2},L,N---\left(4\right)$

In formula (4), $Z_t=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_t\left(i\right)\&CenterDot;e^{-a_t{y}_i{C}_t\left(x_i\right)}$ Be normalized factor, make $\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{w}_{t+1}\left(i\right)=1.$

2.4), final sorter:

$C\left(X\right)=\mathrm{sgn}\left[\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{a}_t{C}_t\left(X\right)\right]---\left(5\right)$

In formula (5), function sgn () is sign function, and its concrete mathematic(al) representation is

$\mathrm{sgn}\left(x\right)=\left\{\begin{array}{cc}1,& x\&GreaterEqual;0\\ -1,& x<0\end{array}\right.---\left(6\right)$

Further, Weak Classifier described in step 2 selects conventional individual layer decision tree, and this decision tree only adopts threshold value division methods to do decision-making based on single input feature vector, namely a node is only had, and set for once fission process due to this, be similar to stub, therefore it is also referred to as decision-making pile.The most critical issue of decision-making pile structure is the quality of how judgment threshold division result, to select optimal partition point.The present embodiment adopts Gini impurity level index to assess the good and bad degree of segmentation rule, and for the data set S comprising c classification, it is defined as follows:

$\mathrm{gini}\left(S\right)=1-\underset{j=1}{\overset{c}{\mathrm{\&Sigma;}}}{p}_j^2---\left(7\right)$

In formula, p _jrepresent the ratio shared by sample of classification j in S set.If split regular rule S is divided into S ₁and S ₂two subsets, then the Gini assessed value of this rule is designated as:

$\mathrm{gini}(S,\mathrm{rule})=\frac{n_1}{n}\&CenterDot;\mathrm{gini}\left(S_1\right)+\frac{n_2}{n}\&CenterDot;\mathrm{gini}\left(S_2\right)---\left(8\right)$

In formula, n ₁for subset S ₁number of samples, n ₂for subset S ₂number of samples, n is the number of samples of S set.

For a Numeric Attributes, the two categorised decision stake segmentation thoughts based on Gini index are: after all possible dividing method of traversal, select the rule of the optimal dividing as this Nodes making assessed value gini (S, rule) reach minimum.Its flow process as shown in Figure 2, is described as:

1) sample value of logarithm value type attribute sorts, and supposes that the result after sorting is (x ₁, y ₁), (x ₂, y ₂) ..., (x _n, y _n); y _i(i=1,2 ..., n) be category label.

2) because segmentation only occurs between two data points, so usually get mid point (x _i+ x _i+1different cut-points, as cut-point, is then got from small to large successively in)/2, and calculates the gini value of each segmentation rule by formula (7) and formula (8).

3) get make gini value minimum point as optimal partition point, and with this cut-point for threshold value constructs decision-making pile.

Further, the conductor galloping early warning output packet described in step (2) to predict the outcome y and degree of confidence margin (x, y) containing sorter:

$y=C\left(x\right)=\mathrm{sgn}\left[\underset{t=1}{\overset{T}{\mathrm{\&Sigma;}}}{a}_t{C}_t\left(x\right)\right]---\left(9\right)$

$m\arg \mathrm{in}(x,y)=\frac{y\underset{t}{\mathrm{\&Sigma;}}{a}_t{C}_t\left(x\right)}{\underset{t}{\mathrm{\&Sigma;}}\left|a_t\right|}---\left(10\right)$

In formula, x is the forecast data of conductor galloping Meteorological Characteristics factor; Y ∈-1 ,+1}, for sorter predicts the outcome: 1 for predicting that power transmission line will be waved, and-1 for predicting that power transmission line will not be waved; Margin ∈ [-1, + 1], it is with a high credibility that larger positive boundary (namely from 1 close to) represents that this circuit of prediction is waved, it is with a high credibility that larger negative edge (namely from-1 close to) represents that this circuit of prediction is not waved, and less border (namely from 0 close to) then represents that the confidence level predicted the outcome is lower.

(3) the early warning Output rusults, according to step (2), judges to obtain conductor galloping Risk-warning grade.

The margin value that can predict the outcome according to conductor galloping, judges the Risk-warning grade of conductor galloping shown according to the form below.

Table 2 conductor galloping advanced warning grade table

Above-described embodiment is the present invention's preferably embodiment; but embodiments of the present invention are not limited by the examples; change, the modification done under other any does not deviate from Spirit Essence of the present invention and principle, substitute, combine, simplify; all should be the substitute mode of equivalence, be included within protection scope of the present invention.

Claims (2)

1., based on a conductor galloping method for prewarning risk of Adaboost, it is characterized in that, comprise the following steps;

(1), by the internal cause of conductor galloping classify, and by classification results, the Meteorological Characteristics factor that power transmission line history is waved under accident is added up;

Wherein internal cause classification results includes: the 1. type of circuit, is divided into S.C. and split conductor; 2. the sectional area of circuit, is divided into large, medium and small cross section circuit; 3. line span, is divided into large, medium and small span circuit;

Meteorological Characteristics factor includes: 1. wind speed; 2. wind direction to traverse shaft to angle; 3. temperature; 4. relative humidity;

(2), according to predicted transmission line information, select a class corresponding with conductor galloping internal cause classification results, and the Meteorological Characteristics factor information record composing training sample set waved with history in such under emergency conditions, sorter is formed with Adaboost Ensemble Learning Algorithms, again using the forecast data of conductor galloping Meteorological Characteristics factor that describes in the step 1 that obtains from meteorological department as input, obtain transmission line galloping early warning Output rusults and degree of confidence margin value by sorter;

(3), according to the early warning Output rusults described in step 2, according to the degree of confidence margin value that conductor galloping predicts the outcome, judge to obtain conductor galloping Risk-warning grade.

2. the conductor galloping method for prewarning risk based on Adaboost according to claim 1, is characterized in that: described step (2) is specific as follows:

2.1), input: training sample set X, wherein should comprise sample class label; N is number of training; T is frequency of training, is also Weak Classifier number; Wherein Weak Classifier sorting algorithm is C, and specify in sample class label that 1 represents that power transmission line is waved ,-1 represents that power transmission line is waved;

2.2), initialization: sample weights distribution w ₁(i)=1/N, i=1,2 ..., N;

2.3), t=1 is worked as, 2 ..., T:

1. according to the sample weights distribution w of the t time _ti () carries out there is the sampling of putting back to from original sample collection X, generate new sample set X _t;

2. at X _tupper training Weak Classifier C _t, and use C (X) _t(X) original sample collection X is classified;

3. Weak Classifier C is calculated _t(X) classification error rate;

In formula (1), work as C _t(x _i) ≠ y _itime, I (g) is 1, and all the other are then 0;

4. Weak Classifier C is calculated _t(X) coefficient;

5. weights distribution is upgraded;

In formula (3), be normalized factor, make

2.4), final sorter:

In formula (4), function sgn () is sign function, and its concrete mathematic(al) representation is

Then conductor galloping early warning output packet to predict the outcome y and degree of confidence margin (x, y) containing sorter:

In formula, x is the forecast data of conductor galloping Meteorological Characteristics factor; Y ∈-1 ,+1}, for sorter predicts the outcome: 1 for predicting that this power transmission line will be waved, and-1 for predicting that this power transmission line will not be waved; Margin ∈ [-1 ,+1], it is with a high credibility that larger positive boundary represents that this circuit of prediction is waved, and it is with a high credibility that larger negative edge represents that this circuit of prediction is not waved, and less border then represents that the confidence level predicted the outcome is lower.