CN103904657A - Regional power grid reactive voltage control method based on support vector machine - Google Patents

Abstract

The invention discloses a regional power grid reactive voltage control method based on a support vector machine. Voltage values of main nodes of a regional power grid and power factors of the main nodes of the regional power grid are used as sample elements, corresponding action strategies are used as sample values, state classification is conducted on a certain number of historical data of the regional power grid and the action conditions of adjusting equipment after classification are recorded, learning training is conducted by using a one-to-one multi-classification mathematical model algorithm of the support vector machine, a reactive voltage control model is generated, the model is corrected in the control process according to the gradual learning idea. Thus, online real-time control over power grid reactive voltage is realized, and coordinate actions of adjusting equipment between transformer substations of the same subsystem is effectively realized.

Description

A kind of area power grid reactive power/voltage control method based on SVMs

Technical field

The present invention relates to reactive power optimization of power system field, relate in particular to a kind of area power grid reactive power/voltage control method based on SVMs.

Background technology

SVMs (Support Vector Machines, SVM) is a kind of general learning method being based upon on Statistical Learning Theory basis.Its theoretical foundation is quite ripe, has mathematical form and reasoning process, clearly geometric interpretation and the good generalization ability of standard.Algorithm of support vector machine computational process is that a convex programming problem is solved.Convex programming refers to: if the constraint set X of problem (MP) is convex set, target function f is the convex function on X, (MP) to be called be Non-Linear Programming.At present, SVMs existing more successful application in the fields such as Power System Reliability Analysis, Transient Stability Evaluation, short-term load forecasting.

Voltage is the important indicator of weighing power system operation situation and the quality of power supply, and quality of voltage and reactive power distribution direct relation economy and the safety of electric power system.Along with electric power networks is interconnected, electrical network scale constantly expands, and relies on operations staff artificially electric power system to be carried out to reactive power/voltage control and has been difficult to accomplish to eliminate in time the impact of the change at random of loading on System Reactive Power distribution and quality of voltage.In existing Reactive Power Optimization Algorithm for Tower research, be mainly divided into classic algorithm and the large class of intelligent algorithm two.In classic algorithm, mainly contain linear programming technique (Linear Programming, LP), Nonlinear Programming Method (Non Linear Programming, NLP), mixed integer programming method (Mixed-integer Programming), dynamic programming (Dynamic Programming).Classic algorithm have computational speed fast, restrain the advantages such as reliable, but the continuity to majorized function, non-convexity, differentiability have higher requirements, and exist and be easy to be absorbed in the shortcomings such as locally optimal solution, the Reactive Power Optimazation Problem that solves complicated electric power system is more difficult.Intelligent algorithm, processing the limitation that has overcome classical way in non-linear, multiple constraint, multivariable, discontinuous, the non-optimization problem such as protruding, has highlighted very strong optimizing ability.Expert system (Expert System, ES), artificial neural net (Artificial Neural Network, ANN), fuzzy theory (Fuzzy Theory, FT), genetic algorithm (Genetic Algorithm, GA) and the application of intelligent algorithm in power system reactive power voltage control such as Agent system (Multi-Agent System, MAS) obtained certain achievement.

At present, the idle work optimization of electrical network roughly can be divided into two classes solution thinkings.One class is the method that adopts global optimization, and another kind of is the optimization method on the spot adopting based on local message.For first kind method, its subject matter is that algorithm robustness is not strong, very different for changeable power system operation environment and metric data quality, often convergence difficulties of algorithm on-line operation.For Equations of The Second Kind method, its design principle is mainly based on multi-region chart, although simple, there is not convergence problem, due to based on single-point information, be difficult to coordinate between multiple stations.

Summary of the invention

The object of this invention is to provide a kind of area power grid reactive power/voltage control method based on SVMs, can carry out reasonably optimizing coordination to the adjustable device between transformer station in each subsystem of area power grid, realize the real-time online control of area power grid reactive voltage.

The technical solution used in the present invention is:

An area power grid reactive power/voltage control method based on SVMs, comprises the following steps:

A: area power grid system decomposition is become to multiple subsystems, each subsystem is control unit independently, gather transformer low voltage, middle pressure bus voltage value and high voltage side of transformer power factor, and voltage is examined higher limit, lower limit, power factor examination higher limit, lower limit;

B: according to high voltage side of transformer power factor examination higher limit λ in subsystem _max, lower limit λ _minand in subsystem, transformer operates in the one group of power factor λ ' that our station capacitor obtains that puts into operation under power factor in limited time _min, excise the power factor λ ' that one group of our station capacitor obtains when transformer fortune ' row merit ' rate factor higher limit _max, wherein λ ' _minand λ ' _maxtry to achieve by formula (1), (2):

$\left\{\begin{array}{c}\frac{\mathrm{<figure-callout\; id="2"\; label="P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">P</figure-callout>}}{\sqrt{P^{}}}={\mathrm{\&lambda;}}_{\mathrm{<figure-callout\; id="2"\; label="min\; P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">min</figure-callout>}}& \\ \frac{P}{\sqrt{P^{}}}\frac{\sqrt{{}^2+{(Q_{\max }+Q_c)}^2}}{}={\mathrm{\&lambda;}}_{\min }^{\&prime;}\end{array}\right.---\left(1\right)$

$\left\{\begin{array}{c}\frac{\mathrm{<figure-callout\; id="2"\; label="P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">P</figure-callout>}}{\sqrt{P^{}}}={\mathrm{\&lambda;}}_{\mathrm{<figure-callout\; id="2"\; label="max\; P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">max</figure-callout>}}& \\ \frac{P}{\sqrt{P^{}}}\frac{\sqrt{{}^2+{(Q_{\min }+Q_c)}^2}}{}={\mathrm{\&lambda;}}_{\max }^{\&prime;}\end{array}\right.---\left(2\right)$

In formula: λ _min, λ _maxbe respectively power factor lower limit and the higher limit definite according to the regulation of electrical network, P is the active power that flows through transformer, Q _maxfor flowing through the reactive power higher limit of transformer, Q under current P _minfor flowing through the reactive power lower limit of transformer, Q under current P _cfor our station low-voltage bus bar one group capacitor capacity; Power factor is divided into 5 states, and separation is followed successively by λ _min, λ ' _min, λ ' _max, λ _max:

State 1: the defective state of lower limit, power factor is lower than lower limit λ _min;

State 2: lower limit transition eligible state, power factor Jie ' in λ _minand λ ' _minbetween;

State 3: eligible state completely, power factor is between λ ' _minand λ ' _max;

State 4: upper limit transition eligible state, power factor is between λ ' _maxand λ _maxbetween;

State 5: the defective state of the upper limit, power factor is higher than higher limit λ _max;

C: according to transformer bus voltage examination higher limit V in subsystem _max, lower limit V _min, and V ' _min, V ' _max, wherein V ' _minand V ' _maxtry to achieve by formula (3):

$\left\{\begin{array}{c}{V}_{\min }^{\&prime;}=V_{\min }+\frac{V_{\max }-V_{\min }}{4}\\ V_{\max }^{\&prime;}=V_{\max }-\frac{V_{\max }-V_{\min }}{4}\end{array}\right.---\left(3\right)$

In formula: V _min, V _maxlower voltage limit value, the higher limit of electrical network to relevant voltage grade regulation, V ' _minthe separation of state 2 and state 3, V ' _maxthe separation of state 3 and state 4; Magnitude of voltage is divided into 5 states, and separation is followed successively by V _min, V ' _min, V ' _max, V _max:

State 1: the defective state of lower limit, magnitude of voltage is lower than lower limit V _min;

State 2: lower limit transition eligible state, magnitude of voltage is between V _minand V ' _minbetween;

State 3: eligible state completely, magnitude of voltage is between V ' _minand V ' _max;

State 4: upper limit transition eligible state, magnitude of voltage is between V ' _maxand V _maxbetween;

State 5: the defective state of the upper limit, magnitude of voltage is higher than higher limit V _max;

D: need to examine the magnitude of voltage of area power grid and/or power factor to carry out state classification according to step b, c, and according to the historical action record of idle conditioning equipment in area power grid subsystem, be organized into state vector and give label;

E: utilize the state vector training of tape label to generate the many disaggregated models of SVMs, the following formula of the many disaggregated models of SVMs (4):

$\begin{array}{cc}\underset{\mathrm{\&alpha;}}{\min }& \frac{1}{2}{\left(w^{\mathrm{ij}}\right)}^T{w}^{\mathrm{ij}}-C\underset{t}{\mathrm{\&Sigma;}}{\left({\mathrm{\&xi;}}^{\mathrm{ij}}\right)}_t\\ \mathrm{subject\; to}& {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&GreaterEqual;{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=i,\\ & {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&le;-1+{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=j,\\ & {\mathrm{\&xi;}}_t^{\mathrm{ij}}\&GreaterEqual;0,\end{array}---\left(4\right)$

In formula: w ^ijbe illustrated in x _tcarry out i, the coefficient vector while classification between j two classes, b ^ijfor constant amount, φ (x _t) be by x _tbe mapped to high-dimensional feature space, C is regularization parameter,

for slack variable, y _trepresent classification;

F: the state vector of the each subsystem of the required measurement area power grid of Real-time Obtaining, utilize the model generating to carry out label to the state vector of each subsystem, can obtain the control strategy of this area's electrical network;

G: the result of the action of control strategy is carried out to correctness detection, if there is underproof state, measuring value and the classification results thereof of this examination electric parameters are recorded, give rational label by empirical value to it, training generates supporting vector machine model again;

H: for the supporting vector machine model regenerating, repeat step f, g, until all examine the measuring value of electric parameters in acceptability limit.

Described each subsystem comprises transformer station take 220kv as root and transformer station of subordinate and the subsidiary capacitor of subsidiary capacitor and electrical connection thereof.

The present invention is take the magnitude of voltage of area power grid main node and power factor as sample elements, take corresponding action policy as sample value, the area power grid historical data of some is carried out to the action situation of conditioning equipment after state classification record sort, utilize the many classification of SVMs " one to one " mathematics model algorithm to carry out learning training, generate reactive power/voltage control model, and in control procedure with the thought correction model of progressive learning, realize the on line real time control to electric network reactive-load voltage, effectively realize the coordination of adjustable device between transformer station in same subsystem.

Method advantage of the present invention is: (1), the present invention do not exist convergence problem, must obtain a certain result; (2) the present invention can effectively consider cooperatively interacting of adjustable device between transformer station, makes its coordination, provides Optimal Control Strategy; (3) control method of the present invention, based on progressive learning the Theory Construction, can, by error result is revised and learnt, improve model accuracy.

Control method of the present invention has solved lacking of current area power grid reactive voltage On-Line Control Method

Fall into, both guaranteed convergence, also reached the global optimization object that takes into full account controllable device coordination between transformer station simultaneously.

Accompanying drawing explanation

Fig. 1 is power factor state classification figure of the present invention;

Fig. 2 is magnitude of voltage state classification figure of the present invention;

Fig. 3 is that supporting vector machine model of the present invention is optimized block diagram;

Fig. 4 embodiment of the present invention transformer station subsystem schematic diagram.

Embodiment

As shown in Figure 1, 2, a kind of area power grid reactive power/voltage control method based on SVMs of the present invention, comprises the following steps:

A: area power grid system decomposition is become to multiple subsystems, each subsystem is control unit independently, each subsystem comprises transformer station take 220kv as root and transformer station of subordinate and the subsidiary capacitor of subsidiary capacitor and electrical connection thereof, gather transformer low voltage, middle pressure bus voltage value and high voltage side of transformer power factor, and voltage is examined higher limit, lower limit, power factor examination higher limit, lower limit;

B: according to high voltage side of transformer power factor examination higher limit λ in subsystem _max, lower limit λ _minand in subsystem, transformer operates in the one group of power factor λ ' that our station capacitor obtains that puts into operation under power factor in limited time _min, excise the power factor λ ' that one group of our station capacitor obtains when transformer fortune ' row merit ' rate factor higher limit _max, wherein λ ' _minand λ ' _maxtry to achieve by formula (1), (2):

$\left\{\begin{array}{c}\frac{\mathrm{<figure-callout\; id="2"\; label="P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">P</figure-callout>}}{\sqrt{P^{}}}={\mathrm{\&lambda;}}_{\mathrm{<figure-callout\; id="2"\; label="min\; P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">min</figure-callout>}}& \\ \frac{P}{\sqrt{P^{}}}\frac{\sqrt{{}^2+{(Q_{\max }+Q_c)}^2}}{}={\mathrm{\&lambda;}}_{\min }^{\&prime;}\end{array}\right.---\left(1\right)$

$\left\{\begin{array}{c}\frac{\mathrm{<figure-callout\; id="2"\; label="P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">P</figure-callout>}}{\sqrt{P^{}}}={\mathrm{\&lambda;}}_{\mathrm{<figure-callout\; id="2"\; label="max\; P\; P"\; filenames="140324165025.png,140324165031.png"\; state="\{\{state\}\}">max</figure-callout>}}& \\ \frac{P}{\sqrt{P^{}}}\frac{\sqrt{{}^2+{(Q_{\min }+Q_c)}^2}}{}={\mathrm{\&lambda;}}_{\max }^{\&prime;}\end{array}\right.---\left(2\right)$

In formula: λ _min, λ _maxbe respectively power factor lower limit and the higher limit definite according to the regulation of electrical network, P is the active power that flows through transformer, Q _maxfor flowing through the reactive power higher limit of transformer, Q under current P _minfor flowing through the reactive power lower limit of transformer, Q under current P _cfor our station low-voltage bus bar one group capacitor capacity; Power factor is divided into 5 states, and separation is followed successively by λ _min, λ ' _min, λ ' _max, λ _max:

State 1: the defective state of lower limit, power factor is lower than lower limit λ _min;

State 2: lower limit transition eligible state, power factor Jie ' in λ _minand λ ' _minbetween;

State 3: eligible state completely, power factor is between λ ' _minand λ _max;

State 4: upper limit transition eligible state, power factor is between λ ' _maxand λ _maxbetween;

State 5: the defective state of the upper limit, power factor is higher than higher limit λ _max;

C: according to transformer bus voltage examination higher limit V in subsystem _max, lower limit V _min, and V ' _min, V ' _max, wherein V ' _minand V ' _maxtry to achieve by formula (3):

$\left\{\begin{array}{c}{V}_{\min }^{\&prime;}=V_{\min }+\frac{V_{\max }-V_{\min }}{4}\\ V_{\max }^{\&prime;}=V_{\max }-\frac{V_{\max }-V_{\min }}{4}\end{array}\right.---\left(3\right)$

In formula: V _min, V _maxrespectively lower voltage limit value, the higher limit of electrical network to relevant voltage grade regulation, V ' _minthe separation of state 2 and state 3, V ' _maxthe separation of state 3 and state 4; Magnitude of voltage is divided into 5 states, and separation is followed successively by V _min, V ' _min, V ' _max, V _max:

State 1: the defective state of lower limit, magnitude of voltage is lower than lower limit V _min;

State 2: lower limit transition eligible state, magnitude of voltage is between V _minand V ' _minbetween;

State 3: eligible state completely, magnitude of voltage is between V ' _minand V ' _max;

State 4: upper limit transition eligible state, magnitude of voltage is between V ' _maxand V _maxbetween;

State 5: the defective state of the upper limit, magnitude of voltage is higher than higher limit V _max;

D: need to examine the magnitude of voltage of area power grid and/or power factor to carry out state classification according to step b, c, and according to the historical action record of idle conditioning equipment in area power grid subsystem, be organized into state vector and give label;

E: utilize the state vector training of tape label to generate the many disaggregated models of SVMs, the following formula of the many disaggregated models of SVMs (4):

$\begin{array}{cc}\underset{\mathrm{\&alpha;}}{\min }& \frac{1}{2}{\left(w^{\mathrm{ij}}\right)}^T{w}^{\mathrm{ij}}-C\underset{t}{\mathrm{\&Sigma;}}{\left({\mathrm{\&xi;}}^{\mathrm{ij}}\right)}_t\\ \mathrm{subject\; to}& {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&GreaterEqual;{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=i,\\ & {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&le;-1+{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=j,\\ & {\mathrm{\&xi;}}_t^{\mathrm{ij}}\&GreaterEqual;0,\end{array}---\left(4\right)$

In formula: w ^ijbe illustrated in x _tcarry out i, the coefficient vector while classification between j two classes, b ^ijfor constant amount, φ (x _t) be by x _tbe mapped to high-dimensional feature space, C is regularization parameter,

for slack variable, y _trepresent classification;

F: the state vector of the each subsystem of the required measurement area power grid of Real-time Obtaining, utilize the model generating to carry out label to the state vector of each subsystem, can obtain the control strategy of this area's electrical network;

G: the result of the action of control strategy is carried out to correctness detection, if there is underproof state, measuring value and the classification results thereof of this examination electric parameters are recorded, give rational label by empirical value to it, training generates supporting vector machine model again;

H: for the supporting vector machine model regenerating, repeat step f, g, until all examine the measuring value of electric parameters in acceptability limit.

Below in conjunction with Fig. 1---the present invention will be described in detail for Fig. 4:

Method of the present invention is the area power grid reactive power/voltage control method based on SVMs, is the process to the training of known sample vector.For current area power grid, often adopt the operation of 500kV looped network, the pattern of 220kV electrical network open loop operation, avoids the existence of electromagnetic looped network.Therefore, an area power grid system decomposition is become multiple independent subsystems by the present invention, each subsystem is take 220kV transformer station as root, comprise 220kV transformer station and subsidiary capacitor and be electrically connected the transformer station of subordinate that connects and subsidiary capacitor with it, and, suppose that the condenser capacity of same transformer station is identical.Each subsystem is the unit of control independently, and in acquisition subsystem, main electric parameters, as sample elements, comprises in transformer, low-voltage bus bar magnitude of voltage and high voltage side of transformer power factor.Obtain some tape labels sample according to on-the-spot historical record data empirical value, for the tape label sample set of each subsystem K, generate SVM model model[k].

Wherein, before generating SVM model, electric parameters is divided into 5 states by the present invention, as depicted in figs. 1 and 2, Fig. 1 is the schematic diagram that the power factor of area power grid is divided into 5 states, according to the on-the-spot ruuning situation of electrical network, 5 states can embody the nargin that upper and lower level power factor regulation allows.The state defective state of 1 lower limit and the defective state of state 5 upper limit are the states that need to regulate, and need to provide to make it recover the action policy of eligible state.State 2 lower limit transition eligible state and state 4 upper limit transition eligible state be in same subsystem, need to take into account when the action of other transformer equipment state, to avoid conflict.The complete eligible state of state 3, regulates with regard to not needing.Fig. 2 is the schematic diagram that area power grid voltage is divided into 5 states, and Consideration and power factor are classified roughly the same, no longer describes in detail.Voltage is divided into different conditions, has taken into full account the impact that the action of adjustable device can bring the performance assessment criteria of the superior and the subordinate.State 2 lower limit transition eligible state and state 4 upper limit transition eligible state are the states that need to take into account when other transformer equipment moves in same subsystem, prevent that it from changing to defective state.Boundary between state has clear and definite separation, and the separation of power factor can be tried to achieve according to formula (1), (2), and the separation of voltage can be tried to achieve according to formula (3).Wherein, for Fig. 1, performance assessment criteria higher limit and lower limit in 2, slightly difference of the requirement between each area power grid, should coordinate according to field condition in procedure of the present invention implementing.

After this, the electric parameters (electric parameters refers to magnitude of voltage and/or power factor) that area power grid need to be examined is carried out state classification according to step b, c, according to the historical action record of idle conditioning equipment in area power grid subsystem, be organized into state vector and give label, utilize the state vector training of tape label to generate the many disaggregated models of SVMs.

First the content of SVMs is briefly introduced:

(1) SVM bis-disaggregated models

Given training sample vector x _i∈ R ⁿ, i=1,2 ... l, is divided into two classes, and given label vector is y ∈ R ^l, y _i∈ 1, and-1}, the former problem model that need to solve is formula (5)

$\begin{array}{cc}\underset{w,b,\mathrm{\&xi;}}{\min }& \frac{1}{2}{w}^{T}w+C\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}{\mathrm{\&xi;}}_i\\ \mathrm{subject\; to}& y_i(w^T\mathrm{\&phi;}\left(x_i\right)+b)\&GreaterEqual;1-{\mathrm{\&xi;}}_i,\\ & {\mathrm{\&xi;}}_i\&GreaterEqual;0,i=1,...,l\end{array}---\left(5\right)$

In formula, W is coefficient parameter, φ (x _i) be x _ibe mapped to high-dimensional feature space, C is regularization parameter, ζ _ifor slack variable;

Because the dimension of feature space is very high, or even infinite, most methods is the former problem of direct solution problem not, but solves their dual problem formula (6)

$\begin{array}{cc}\underset{\mathrm{\&alpha;}}{\min }& \frac{1}{2}\mathrm{\&alpha;TQ\&alpha;}-e^T\mathrm{\&alpha;}\\ \mathrm{subject\; to}& y^T\mathrm{\&alpha;}=0,\\ & 0\&le;{\mathrm{\&alpha;}}_i\&le;C,i=1,...,l,\end{array}---\left(6\right)$

In formula: α=[α ₁, α ₂... α _l] ^t, α _iinequality constraints y in former problem _i(w ^tφ (x _i)+b)>=1-ξ _icorresponding Lagrangian Product-factor, Q is positive semi-definite l dimension square formation, Q _ij=y _iy _jk (x _i, x _j), K (x _i, x _j) ≡ φ (x _i) ^tφ (x _i) be kernel function.

After above-mentioned dual problem is solved, obtain two grader formula (7)

$y\left(x\right)=\mathrm{sgn}(\underset{i=1}{\overset{l}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_{i}K(x_i,x)+b)---\left(7\right)$

If α _i=0, sample x _ibe called non-support vector; If α _i>0, sample x _ibe called support vector, if α _i=C, sample x _ibe called bounded support vector, if 0< is α _i<C, sample x _ibe called non-bounded support vector.

(2) the many disaggregated models of SVM

" one to one " SVM multi-categorizer need to be between each classification structural classification device, need to construct n (n-1)/2 grader to n classification, the training sample of each classifier functions belongs to two relevant classes, combine these binary classifiers and use ballot method, who gets the most votes's class is the class under sample point.Model is formula (4):

$\begin{array}{cc}\underset{\mathrm{\&alpha;}}{\min }& \frac{1}{2}{\left(w^{\mathrm{ij}}\right)}^T{w}^{\mathrm{ij}}-C\underset{t}{\mathrm{\&Sigma;}}{\left({\mathrm{\&xi;}}^{\mathrm{ij}}\right)}_t\\ \mathrm{subject\; to}& {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&GreaterEqual;{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=i,\\ & {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&le;-1+{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=j,\\ & {\mathrm{\&xi;}}_t^{\mathrm{ij}}\&GreaterEqual;0,\end{array}---\left(4\right)$

In formula: w ^ijbe illustrated in x _tcarry out i, the coefficient vector while classification between j two classes, b ^ijfor constant amount, φ (x _t) be by x _tbe mapped to high-dimensional feature space, C is regularization parameter,

for slack variable, y _trepresent classification.

For a certain sample, if it belongs to i class, i number of votes obtained adds one; If it belongs to j class, j number of votes obtained adds one, and final who gets the most votes's class is sample point x _taffiliated class.

Provide embodiment 1 below, what embodiment 1 adopted is the C-SVM model based on polynomial kernel function.

As shown in Figure 3, to utilize the SVM model having generated to carry out reactive power/voltage control to area power grid, for a certain subsystem K, utilizing its model model[k] state (sample) to current subsystem classifies, obtain, after classification results (sample label), regulating corresponding control appliance according to instruction.The correctness that detects the result of the action in optimizing process is that after moving by checkout facility, whether each examines the measuring value of electric parameters at qualified state, if there is underproof state, this sample and classification results thereof are recorded, give the most rational label by empirical value to it again, put into training set and regenerate SVM model.

As shown in Figure 4, illustrate subsystem situation, for example subsystem contains three transformer stations, and controllable equipment is T1, T2, T3, C1, C2, C3, and the index of examination has high voltage side of transformer power factor λ ₁, λ ₂, λ ₃with low-pressure side bus voltage V ₁, V ₂, V ₃.Getting sample vector is x _i=[V ₁, λ ₁, V ₂, λ ₂, V ₃, λ ₃] T, corresponding label vector y ∈ R ¹³, y _i={ 0,1,2,3,4,5,6,7,8,9,10,11,12}, the corresponding relation of label value and action policy is in table 1.

Table 1 label value table corresponding to action policy

Suppose that transformer T1 high-pressure side active power is P1=35MW, reactive power is Q1=9MVar, and low-pressure side bus voltage is V1=36.8kV; Transformer T2 high-pressure side active power is P2=16MW, and reactive power is Q2=4MVar, and low-pressure side bus voltage is V2=10.4kV; Transformer T3 high-pressure side active power is P3=18MW, and reactive power is Q3=4.5MVar, and low-pressure side bus voltage is V3=10.3kV; Condenser capacity Q _c1=3MVar, Q _c2=2MVar, Q _c3=2MVar.According to electrical network regulation, for 35kV bus, lower voltage limit value is V _min=34kV, upper voltage limit value is V _max=37.5kV, for 10kV bus, lower voltage limit value is V _min=10kV, upper voltage limit value is V _max=10.7kV, three transformer efficiency factor lower limits are λ _min=0.92, power factor higher limit is λ _max=1.0.

Can obtain according to formula (1), (2), (3)

λ′ _min1=0.947,λ′ _max1=0.961

λ′ _min2=0.958,λ′ _max2=0.992

λ′ _min3=0.954,λ′ _max3=0.994

V′ _min1=34.875kV，V′ _max1=36.625kV

V′ _min2=10.175kV，V′ _max2=10.525kV

V′ _min3=10.175kV，V′ _max3=10.525kV

Embodiment neutron systematic sample vector and label thereof are as following table 2.It is qualified that the state that this sample vector represents is, y _i=0 shows not need action.

Table 2 exemplar

V 1	λ 1	V 2	λ 2	V 3	λ 3	y i
							4	4	3	3	3	3	0

The present invention utilizes the many classification of SVMs " one to one " mathematics model algorithm K to get [10] sample vector of 175 of this subsystem tape labels is trained, and has generated the many disaggregated models of SVM.Then utilize the many disaggregated models that generate to classify to 200 random unknown sample vectors, rate of accuracy reached to 92.5%.Obviously there is certain error, but can pass through feedback mechanism, the sample vector of classification error is given joining after correct sample in training sample, regenerate the many disaggregated models of SVM, with this continuous raising accuracy.In table 3, listing the classification results of some sample vectors analyzes.Sample vector classification results in table is all correct, can find out from the result of optimizing: this kind of reactive power/voltage control mode can be considered from the running status of whole subsystem, provides optimum action policy.

Table 3 unknown sample vector classification results

Sample	V 1	λ 1	V 2	λ 2	V 3	λ 3	y i
								1	3	4	1	3	3	3	9
2	3	2	2	3	2	2	0

By to classification results, analysis shows: can take into full account the operation conditions of each examination node of whole subsystem on the basis of multi-region chart control principle based on many category theories of SVM reactive power/voltage control mode, coordination, provides optimum action policy.

Claims (2)

1. the area power grid reactive power/voltage control method based on SVMs, is characterized in that: comprise the following steps:

A: area power grid system decomposition is become to multiple subsystems, each subsystem is control unit independently, gather transformer low voltage, middle pressure bus voltage value and high voltage side of transformer power factor, and voltage is examined higher limit, lower limit, power factor examination higher limit, lower limit;

B: according to high voltage side of transformer power factor examination higher limit λ in subsystem _max, lower limit λ _minand in subsystem, transformer operates in the one group of power factor λ ' that our station capacitor obtains that puts into operation under power factor in limited time _min, transformer excises the power factor λ ' that one group of our station capacitor obtains while operating in power factor higher limit _max, wherein λ ' _minand λ ' _maxtry to achieve by formula (1), (2):

$\left\{\begin{array}{c}\frac{P}{\sqrt{P^2+{Q_{\max }}^2}}={\mathrm{\&lambda;}}_{\min }\\ \frac{P}{\sqrt{P^2+{(Q_{\max }+Q_c)}^2}}={\mathrm{\&lambda;}}_{\min }^{\&prime;}\end{array}\right.---\left(1\right)$

$\left\{\begin{array}{c}\frac{P}{\sqrt{P^2+{Q_{\min }}^2}}={\mathrm{\&lambda;}}_{\max }\\ \frac{P}{\sqrt{P^2+{(Q_{\min }+Q_c)}^2}}={\mathrm{\&lambda;}}_{\max }^{\&prime;}\end{array}\right.---\left(2\right)$

In formula: λ _min, λ _maxbe respectively power factor lower limit and the higher limit definite according to the regulation of electrical network, P is the active power that flows through transformer, Q _maxfor flowing through the reactive power higher limit of transformer, Q under current P _minfor flowing through the reactive power lower limit of transformer, Q under current P _cfor our station low-voltage bus bar one group capacitor capacity; Power factor is divided into 5 states, and separation is followed successively by λ _min, λ ' _min, λ ' _max, λ _max:

State 1: the defective state of lower limit, power factor is lower than lower limit λ _min;

State 2: lower limit transition eligible state, power factor is between λ _minand λ _minbetween;

State 3: eligible state completely, power factor is between λ ' _minand λ ' _max;

State 4: upper limit transition eligible state, power factor is between λ ' _maxand λ _maxbetween;

State 5: the defective state of the upper limit, power factor is higher than higher limit λ _max;

C: according to transformer bus voltage examination higher limit V in subsystem _max, lower limit V _min, and V ' _min, V ' _max, wherein V ' _minand V ' _maxtry to achieve by formula (3):

$\left\{\begin{array}{c}{V}_{\min }^{\&prime;}=V_{\min }+\frac{V_{\max }-V_{\min }}{4}\\ V_{\max }^{\&prime;}=V_{\max }-\frac{V_{\max }-V_{\min }}{4}\end{array}\right.---\left(3\right)$

In formula: V _min, V _maxlower voltage limit value, the higher limit of electrical network to relevant voltage grade regulation, V ' _minthe separation of state 2 and state 3, V ' _maxthe separation of state 3 and state 4; Magnitude of voltage is divided into 5 states, and separation is followed successively by V _min, V ' _min, V ' _max, V _max:

State 1: the defective state of lower limit, magnitude of voltage is lower than lower limit V _min;

State 2: lower limit transition eligible state, magnitude of voltage is between V _minand V ' _minbetween;

State 3: eligible state completely, magnitude of voltage is between V ' _minand V ' _max;

State 4: upper limit transition eligible state, magnitude of voltage is between V ' _maxand V _maxbetween;

State 5: the defective state of the upper limit, magnitude of voltage is higher than higher limit V _max;

D: need to examine the magnitude of voltage of area power grid and/or power factor to carry out state classification according to step b, c, and according to the historical action record of idle conditioning equipment in area power grid subsystem, be organized into state vector and give label;

E: utilize the state vector training of tape label to generate the many disaggregated models of SVMs, the following formula of the many disaggregated models of SVMs (4):

$\begin{array}{cc}\underset{\mathrm{\&alpha;}}{\min }& \frac{1}{2}{\left(w^{\mathrm{ij}}\right)}^T{w}^{\mathrm{ij}}-C\underset{t}{\mathrm{\&Sigma;}}{\left({\mathrm{\&xi;}}^{\mathrm{ij}}\right)}_t\\ \mathrm{subject\; to}& {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&GreaterEqual;{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=i,\\ & {\left(w^{\mathrm{ij}}\right)}^T\mathrm{\&phi;}\left(x_t\right)+b^{\mathrm{ij}}\&le;-1+{\mathrm{\&xi;}}_t^{\mathrm{ij}},y_t=j,\\ & {\mathrm{\&xi;}}_t^{\mathrm{ij}}\&GreaterEqual;0,\end{array}---\left(4\right)$

In formula: w ^ijbe illustrated in x _tcarry out i, the coefficient vector while classification between j two classes, b ^ijfor constant amount, φ (x _t) be by x _tbe mapped to high-dimensional feature space, C is regularization parameter,

for slack variable, y _trepresent classification;

F: the state vector of the each subsystem of the required measurement area power grid of Real-time Obtaining, utilize the model generating to carry out label to the state vector of each subsystem, can obtain the control strategy of this area's electrical network;

G: the result of the action of control strategy is carried out to correctness detection, if there is underproof state, measuring value and the classification results thereof of this examination electric parameters are recorded, give rational label by empirical value to it, training generates supporting vector machine model again;

H: for the supporting vector machine model regenerating, repeat step f, g, until all examine the measuring value of electric parameters in acceptability limit.

2. the area power grid reactive power/voltage control method based on SVMs according to claim 1, is characterized in that: described each subsystem comprises transformer station take 220kv as root and transformer station of subordinate and the subsidiary capacitor of subsidiary capacitor and electrical connection thereof.

Images