CN1929234A - Parallel computation method for large-scale electrical power system network tidal current segmentation - Google Patents

Abstract

This invention provides one power system flow computation method, which comprises the following steps: dividing power network into proper sub network to form flow computation tree according to combination type; according to division results, power network each sub net relates to each leaf of computation tree; first layer of network division relates to flow joint; first division selected network line relates to flow computation tree middle point. This invention adopts Newton overlap method to work out above nonlinear equation for variable as inner and outer boundary voltage to compute upper substitution and down computation operation.

Description

Parallel computation method for large-scale electrical power system network tidal current segmentation

[technical field]: the present invention relates to power system operation and safety on line analysis technical field, particularly a kind of large-scale electrical power system trend parallel calculating method.

[background technology]: it is one of calculation task the most basic during power system operation and safety on line are analyzed that trend is calculated.It is that trend according to given electric power networks structure and service conditions are calculated whole network distributes that trend is calculated, the most basic electric parameters is the steady-state value of the complex voltage (being called system mode) of each bus of system in the result of calculation, and its content belongs to power system mesomeric state (or static) and analyzes.Mathematical Modeling that it is conventional and method for solving briefly introduce as follows:

The electric power system tide calculation mathematic model represents that by non-linear algebraic equation group shown in the formula (1) wherein vectorial X represents separating of non-linear algebraic equation group usually.Trend is calculated and to be belonged to ultra-large non-linear algebraic equation and find the solution problem on the Mathematical Modeling large-scale electrical power system.

G(X)＝0 (1)

The predicted value X of given variable X ⁰After (being called initial value again), can find the solution the numerical solution of X by Newton iteration equation (2) after the Newton iteration method linearisation and formula (3), wherein subscript k represents iterations, J (X ^k) be the k time Jacobian matrix (Jacobian) matrix in the iteration.

J(X ^k)ΔX ^k＝-G(X ^k) (2)

X ^k+1＝X ^k+ΔX ^k (3)

As correction amount X ^kMaximum value in the element is less than a certain threshold value ε, for example ε＜10 ^-5, (3) formula gained X ^K+1Numerical solution as X.

Above-mentioned electric power system tide calculates and need carry out on computers, and the scale of its amount of calculation and electric power system is relevant, and the electric power system scale is big more, and the amount of calculation during trend is calculated is big more, and the computing time that needs is many more.For improving the speed that electric power system tide calculates, can adopt (on parallel or cluster computer, carrying out) Distributed Calculation or parallel processing technique on the technological layer; On the algorithm aspect, then need to study the new algorithm that amount of calculation is few, be suitable for parallel processing.

Newton method iterative method shown in formula (2) and the formula (3) is a kind of electric power system tide computational methods commonly used, and this method has the advantage that the handing-over error is introduced in convergence fast and not.But, all non-linear algebraic equations that conventional simultaneous solution method will obtain concentrate in together and carry out disposed of in its entirety, setting up a whole Jacobian matrix equation is also found the solution, owing to do not consider the network configuration characteristics of electric power system, the variation of each localized network all can have influence on whole Jacobian matrix equation and revise, so extensibility and flexibility deficiency.And existing various simultaneous solution method all needing its whole Jacobian matrix is carried out algebraic manipulation in each iteration, and amount of calculation is bigger, can not satisfy the requirement that extensive interconnected electric power system is calculated in real time on computational efficiency well.In addition, Chang Gui simultaneous solution method is difficult for realizing parallelization calculating.

[summary of the invention]: the objective of the invention is to solve the amount of calculation that existing simultaneous solution algorithm exists big, can not satisfy problem such as the real-time calculation requirement of extensive interconnected electric power system well, a kind of parallel computation method for large-scale electrical power system network tidal current segmentation is provided.

The present invention is a kind of large-scale electrical power system trend parallel computing, calculating the simultaneous solution method that belongs to classics on the principle based on Newton method, but on technology realizes, it is divided by network, a plurality of equation group that whole electric power system higher-dimension equation group is decomposed into low-dimensional are handled, reduced amount of calculation effectively.The inventive method may further comprise the steps:

Step 101: extensive or ultra-large electric power networks is gathered the one group of subnet that is divided on a fairly large scale according to selected interconnection.(for example China's electrical network is formed by net level grid interconnect to consider to have the interconnected characteristics of layering by large scale electric network, net level electrical network is formed by the provincial power network interconnection, provincial power network is formed by the region grid interconnect), when selecting the network division interconnection, should consider the interconnected characteristics of electrical network layering.

Step 102:, make up electric network swim and calculate tree according to the network division result.Trend is calculated tree one " root node ", " middle root node " and a plurality of undermost " leaf node " (tip node) formation by the superiors.The corresponding electron network of each leaf node, the corresponding a kind of network division of root node and middle root node.

Step 103:, promptly wait to ask the busbar voltage variable to classify to the variable in the tidal current analysis in the electron network of leaf node correspondence.According to the interconnection relation of its bus and its network division of participation, wherein all unknown variables are divided into three classes: border output busbar voltage variable, border inlet highway voltage quantities and internal bus voltage quantities.

Step 104: the initial value of given system busbar voltage (being called state variable).For example, all buses (comprising the reference bus in the tidal current analysis) voltage-mode value initial value is 1 (perunit value unit), and the phase angle initial value is 0 (Rad).

Step 105: row are write the trend algebraic equation (being the power balance equation of PQ bus and the meritorious and balance of voltage equation of PV node) of each leaf node map network, it is carried out Taylor (Taylor) series expansion, omit second order and above higher order term in the Taylor progression, obtain the mathematical Jacobian matrix equation of subsystem.The Jacobian matrix equation is made up of the uneven vector of the correction vector sum of Jacobian matrix, busbar voltage, and in each iteration, Jacobian matrix and uneven vector all are real number matrix and the vectors of determining, what need find the solution is the correction vector of busbar voltage.

Step 106: the correction vector that calculates busbar voltage.Concrete steps are:

106-1. up replacement

For leaf node: this up replacement is from all leaf nodes.For leaf node, by the Jacobian matrix equation being carried out elementary algebra conversion (decomposing and back substitution calculating), all write the correction of output of the border in the leaf node and builtin voltage variable as represent linear representation by the correction of border input voltage variable as LU.

For middle root node: (if any): the up replacement that middle root node will be done is a border busbar voltage correction linear representation of accepting lower floor's node, disappear the border busbar voltage correction of network division of this centre root node correspondence obtains the correction linear representation of the border busbar voltage that the last layer root node needs.

For root node, merge the border output of its all lower floor's leaf nodes, the system of linear equations that the input voltage correction is represented.And find the solution this equation group, solve border output, input voltage correction.

106-2. descending calculating

This operation is calculated all leaf nodes of tree by root node to trend and is carried out.The boundary voltage correction that solves is brought into the linear representation of up replacement gained inside (or following one deck border input and output voltage correction) and border input voltage correction and tried to achieve each sub-network builtin voltage correction vector at last.

Step 107: restrain judgement (if whether maximum in the correction vector of judgment variable or the maximum in the uneven vector sure, judged calculate restrained less than given acceptable threshold).If do not restrain, use the variable correction of obtaining that variable is revised, and change step 105 iterative computation;

Step 108: the electric power system tide result of calculation after the output convergence.

Step 109: finish trend and calculate.

[advantage of the present invention and good effect]: the inventive method has following advantage,

A, significantly reduced and use Newton method the large-scale electrical power system power flow equation to be carried out the amount of calculation of simultaneous solution.This algorithm utilize the interconnected characteristics of Network Layering with the sparse equation group of higher-dimension of the whole network find the solution work allocation in each network calculations node, be the computational problem of a plurality of low-dimensional equation group with the integrated solution PROBLEM DECOMPOSITION of higher-dimension electric power system equation group.Block the sparse equation group of higher-dimension and found the solution the problem that unit is newly filled out in continuous generation, improved computational efficiency.

B, owing to adopt tree-like computation structure, do not rely on mutually with the voltage correction computational process on the layer different node during trend is calculated, be convenient to adopt that the distribution integration computer is parallel carries out.

C, this algorithm are set up Mathematical Modeling separately to each electric power system sub-network, and row are write Jacobi equation separately, therefore have stronger modularity, better expansibility, be more convenient for programming and realize than conventional simultaneous solution algorithm.

The network division parallel computing that d, this invention propose, but the electric power system Nonlinear System of Equations expanded application is found the solution differencing in large-scale electrical power system emulation after realizes the online overtime real emulation of large-scale electrical power system.

[description of drawings]:

Fig. 1 is that electric power system tide of the present invention calculates the rudimentary algorithm schematic flow sheet;

Fig. 2 is a kind of partitioning scheme of target power network among the present invention;

Fig. 3 is that electric power networks trend of the present invention is calculated tree formation schematic diagram;

Fig. 4 is typical New England NETWORK STRUCTURE PRESERVING POWER SYSTEM figure;

Fig. 5 is typical New England electric power system network division result for the first time;

Fig. 6 is that the trend that typical New England electric power system obtains after the network division is for the first time calculated tree;

Fig. 7 is result after the typical New England electric power system network division second time;

Fig. 8 is that the trend that typical New England electric power system obtains after the network division is for the second time calculated tree.

[embodiment]:

Embodiment 1:

Electric power system is made up of the various electric equipments of production, conveying, distribution and consumption electric energy.Electric power networks is made up of transmission line and other power transmitting devices of transformer, different electric pressures.Electric power networks has stronger region graded properties, and an electric power system can be considered as being made up of by interconnection is interconnected a plurality of subnets.

Below in conjunction with accompanying drawing the present invention is described in further detail, then by its parallel execution mode of cluster machine system introduction.Accompanying drawing 1 is that electric power system tide of the present invention calculates the rudimentary algorithm schematic flow sheet.

Feature of the present invention is may further comprise the steps:

Step 101: electric power networks is divided, based on large scale electric network be area power grid through the interconnected characteristics that form of interconnection, the present invention proposes by interconnection and electric power networks is divided into the method for a plurality of subnets on a fairly large scale.Electrical network is divided the result and should be met the following conditions:

(1) the electric power subnet after the division can be organized as the multilayer tree structure by its network joint mode, and the superiors have only a root node, and orlop leaf node map network is cut apart the all-ones subnet network of back gained.The interconnection and the bus related with it of root node and middle root node map network division, generally its network graphic is disconnected.

(2) each subnet is connected with one or more subnets by selected interconnection.

(3) bus that is associated with the interconnection of selecting is called the border bus of this subnet.

Fig. 2 has provided a kind of electric power networks partitioning scheme.As shown in Figure 2, target grid is divided into totally 6 electric power subnets.Label B1 ~ B11 represents in the primitive network and 6 11 border buses that selected interconnection is related.These 6 interconnections (comprising its bus) constitute the root node that trend is calculated tree.Five sub-networks of all the other N1 ~ N5 constitute five leaf nodes.When application was of the present invention, under the situation that electric power networks does not change, network partition only need carry out once getting final product.The subregion result reuses in trend is calculated.

Step 102:, make up with trend and calculate tree according to the network division result.Contrast the trend of cutting apart foundation shown in Figure 2 and calculate tree as shown in Figure 3.If five sub-networks of N1 ~ N5 are still huge, each subnet can be by method shown in Figure 1, and selected interconnection carries out sub-network to be divided, and sets up the sub-network trend and calculates tree, and the root node that the sub-network trend is calculated tree is middle root node.

Step 103: the voltage quantities to be asked in the corresponding sub-network of each node is classified.Contrast system shown in Figure 3 trend is calculated tree, and we are elected to be the border input/output variable with B1 ~ B11 busbar voltage, and all the other busbar voltages are each sub-network built-in variable.

Step 104: the initial value of given all voltage quantities.That is, it is 1.0 (perunit value units) that busbar voltage mould value initial value is asked in all waiting, the phase angle initial value is 0.0 (Rad).

Step 105: the Mathematical Modeling of setting up each subnet.Adopt right angle or polar form (narration of rectangular coordinate form is pressed in following narration), each subnet power flow equation can be expressed as

G _Ni(V _in，V _bi，V _bo)＝0 (4)

V wherein _InThe vector (hereinafter to be referred as the busbar voltage vector) that expression sub-network internal bus voltage real part and imaginary part constitute, V _BiBe the vector (be called the sub-network trend and calculate output variable) of network internal border busbar voltage real part and imaginary part formation, V _BoVector (be called the sub-network trend and calculate input variable) for network-external border busbar voltage real part and imaginary part formation.Make vectorial V _In, V _BiAnd V _BoDimension be respectively n _In, n _BiAnd n _Bo, then the dimension of equation (4) is n _In+ n _BiTo the value V in the iteration _In ^k, V _Bi ^kAnd V _Bo ^kNewton iteration increment equation after its linearisation as the formula (5), wherein k represents iterations, J (V _In ^k, V _Bi ^k, V _Bo ^k) be the Jacobian matrix of sub-network power flow equation in the k time iteration.

$J$ $(V_{\mathrm{in}}^k,V_{\mathrm{bi}}^k,V_{\mathrm{bo}}^k)\left[\begin{array}{c}{\mathrm{\ΔV}}_{\mathrm{in}}^k\\ {\mathrm{\ΔV}}_{\mathrm{bi}}^k\\ {\mathrm{\ΔV}}_{\mathrm{bo}}^k\end{array}\right]=-G(V_{\mathrm{in}}^k,V_{\mathrm{bi}}^k,V_{\mathrm{bo}}^k)\·\·\·\left(5\right)$

Step 106: computed correction.

106-1. up replacement

All leaf nodes that this operation is calculated tree at trend carry out.Contrast formula (5) by the Jacobian matrix equation being carried out elementary algebra conversion (decomposing and back substitution calculating as LU), all as is write the correction of output of the border in the leaf node and built-in variable the following canonical form of being represented by border input correction.

${\mathrm{\ΔV}}_{\mathrm{in}}^k+P^k{\mathrm{\ΔV}}_{\mathrm{bo}}^k=Q^k\·\·\·\left(6\right)$

${\mathrm{\ΔV}}_{\mathrm{bi}}^k+R^k{\mathrm{\ΔV}}_{\mathrm{bo}}^k=S^k\·\·\·\left(7\right)$

Formula (7) is called up replacement equation, and formula (6) is called descending back substitution equation.P in each iteration ^k, R ^k, Q ^kAnd S ^kAll be real number matrix and the vector of determining, what need find the solution is the correction vector of voltage quantities.

106-2. border output, input correction are calculated

For root node and middle root node, the canonical form equation (7) that merges its all lower floor's leaf nodes obtains calculating the system of linear equations of all border outputs, input correction, finds the solution this equation group, can solve all border inputs, output correction amount V _Bi ^kWith Δ V _Bo ^k

106-3. descending calculating

All leaf nodes that this operation is calculated tree at trend carry out.With the border input correction amount V that solves _Bo ^kBring up replacement back gained equation (6) into and get the inner vectorial Δ V of correction of sub-network _In ^k

In order to accelerate the speed that large-scale electrical power system calculates, people often adopt the group system of being made up of a plurality of computers as the hardware facility that calculates.Group system has become a kind of new mainstream technology that realizes parallel computation in recent years.Group system is formed by express network is interconnected by a plurality of computing machines (comprising a main frame and some handsets), adopts data communication mechanism to intercommunicate between each computing machine.Introduce the parallel execution mode of the present invention on clustered computing system below.

As can be seen, the Mathematical Modeling of calculating object of the present invention has tree-like structure, finds the solution the expression formula (6) of the up replacement of each leaf node and the process and the descending formula (6) of using of (7) and calculates the inner correction amount V of sub-network _In ^kComputational process does not rely on mutually, can walk abreast to carry out.

The distributed parallel trend is calculated and is described:

Contrast trend shown in Figure 3 and calculate tree, can satisfy electric power system safety on line monitoring needs if each processor of clustered computing system is born (comprising a main frame and handset) speed (considering the data communication time) of up replacement of each sub-network of N1 to N5 and descending calculation task, then parallel computation can be finished with a main frame and 5 handsets.That the up replacement of communication transmits between main frame and the handset is R ^k(n _Bi* n _BoTie up matrix), and S ^k(n _BoDimensional vector), general dimension n _BoVery little, that descending calculating transmits is border busbar voltage correction amount V _Bi ^k(n _BiDimensional vector) and Δ V _Bo ^k(n _BoDimensional vector), the conceptual data conveying capacity is very limited.Another characteristics are root node and each subnet boundaries busbar voltage correction amount V of middle root node simultaneous solution _Bi ^kWith Δ V _Bo ^kThe amount of calculation of (seeing formula (7)) is very limited.

If a certain processor of clustered computing system born (comprising a main frame and handset) speed (considering the data communication time) of up replacement of a certain sub-network and descending calculation task and do not reached electric power system safety on line monitoring needs, then this sub-network further to be cut apart by dividing method of the present invention.This moment original sub-network node become do in the middle of root node, the sub-network after cutting apart is directly as the leaf node of this centre root node.The cutting procedure of above-mentioned sub-network is until the processor of clustered computing system reaches the safety on line analysis speed in trend is calculated till.

The sub-network that the distributed parallel trend is calculated is described (topological structure and component parameters) and is stored on each processor.

Embodiment 2:

Below calculate the method for tree with typical New England shown in Figure 4 system specialization network division of the present invention and structure trend.With interconnection " 16-17 ", " 15-16 ", " 1-39 " and " 4-14 " is elected to be the network division interconnection, the network division result of gained as shown in Figure 5, corresponding trend is calculated tree as shown in Figure 6.

Adopt the distributed parallel trend to calculate, 3 sub-computers are born N1 respectively, N2 and up replacement of N3 sub-network and descending calculation task.To N1 ~ N3 subnet, set up power flow equation, the k time iteration carried out LU to the Jacobian matrix equation and decomposed that the output of gained border and built-in variable standard update equation are after the elementary algebra conversion:

N1: descending back substitution equation: ${\mathrm{\ΔV}}_{\mathrm{in},N1}^k+P_{N1}^k{\mathrm{\ΔV}}_{\mathrm{bo},N1}^k=Q_{N1}^k;$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N1}^k+R_{N1}^k{\mathrm{\ΔV}}_{\mathrm{bo},N1}^k=S_{N1}^k$ (6 equations of three inner boundary buses)

N2: descending back substitution equation: ${\mathrm{\ΔV}}_{\mathrm{in},N2}^k+P_{N2}^k{\mathrm{\ΔV}}_{\mathrm{bo},N2}^k=Q_{N2}^k$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N2}^k+R_{N2}^k{\mathrm{\ΔV}}_{\mathrm{bo},N2}^k=S_{N2}^k$ (2 equations of an inner boundary bus)

N3: descending back substitution equation: ${\mathrm{\ΔV}}_{\mathrm{in},N3}^k+P_{N3}^k{\mathrm{\ΔV}}_{\mathrm{bo},N3}^k=Q_{N3}^k$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N3}^k+R_{N3}^k{\mathrm{\ΔV}}_{\mathrm{bo},N3}^k=S_{N3}^k$ (6 equations of three inner boundary buses)

The coefficient matrix of 3 up replacement equations and right-hand member vector all are real number matrix, are sent to main frame by data communication.The number of 3 up replacement equations is 14, and waiting to ask border busbar voltage correction also is 14, finds the solution three up replacement equations and obtains 14 border busbar voltage corrections.To respectively: a work song computer is returned in 1,3 and 17 busbar voltage corrections (6) communication; Two work song computers are returned in 16 busbar voltage corrections (2) communication; Three work song computers are returned in 39,4,15 busbar voltage corrections (6) communication; Border busbar voltage correction is brought three sub-computers into and is just calculated N1, all internal bus voltage corrections of three sub-networks of N2 and N3.After repeating said process judgement convergence as shown in Figure 1, trend is calculated and is finished.

Suppose that the N3 subnet is larger, consider that the needs of actual time safety analysis speed are proceeded to cut apart, select circuit 6-11 and 4-14 as cutting apart interconnection, the network division result of gained as shown in Figure 7, two kinds of corresponding trends are calculated tree as shown in Figure 8.A kind of with N1, N2, N5 and N6 merge N0 and N4 node as root node as leaf node; Another kind of N2, N5 and N6 are as leaf node with N1, and the NO node is as root node, and the N4 node is as middle root node.

Adopt the distributed parallel trend to calculate, bear N1 respectively by 4 sub-computers, N2, N5 and up replacement of N6 sub-network and descending calculation task.To N1, N2, N5 and N6 sub-network are set up power flow equation, and the k time iteration carries out to the Jacobian matrix equation that LU decomposes the border output of elementary algebra conversion gained and built-in variable standard update equation is:

N1: descending back substitution equation: ${\mathrm{\ΔV}}_{\mathrm{in},N1}^k+P_{N1}^k{\mathrm{\Δ}}_{\mathrm{bo},N1}^k=Q_{N1}^k;$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N1}^k+R_{N1}^k{\mathrm{\ΔV}}_{\mathrm{bo},N1}^k=S_{N1}^k$ (6 equations of three inner boundary buses)

N2: descending back substitution equation:: ${\mathrm{\ΔV}}_{\mathrm{in},N2}^k+P_{N2}^k{\mathrm{\ΔV}}_{\mathrm{bo},N2}^k=Q_{N2}^k;$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N2}^k+R_{N2}^k{\mathrm{\ΔV}}_{\mathrm{bo},N2}^k=S_{N2}^k$ (2 equations of an inner boundary bus)

N5: descending back substitution equation: ${\mathrm{\ΔV}}_{\mathrm{in},N5}^k+P_{N5}^k{\mathrm{\ΔV}}_{\mathrm{bo},N5}^k=Q_{N5}^k;$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N5}^k+R_{N5}^k{\mathrm{\ΔV}}_{\mathrm{bo},N5}^k=S_{N5}^k$ (6 equations of three inner boundary buses)

N6: descending back substitution equation: ${\mathrm{\ΔV}}_{\mathrm{in},N6}^k+P_{N6}^k{\mathrm{\ΔV}}_{\mathrm{bo},N6}^k=Q_{N6}^k;$

Up replacement equation: ${\mathrm{\ΔV}}_{\mathrm{bi},N6}^k+R_{N6}^k{\mathrm{\ΔV}}_{\mathrm{bo},N6}^k=S_{N6}^k$ (6 equations of three inner boundary buses)

N1, N2, the coefficient matrix and the right-hand member vector of 4 up replacement equations of N5 and N6 sub-network all are real number matrix, are sent to main frame by data communication.The number of 4 up replacement equations is 20, waits to ask border busbar voltage correction by 14 original 6 voltage corrections that increased from 6,11 and 14 buses, and waiting to ask border busbar voltage correction sum also is 20.

Merge N0 and N4 node as the network division of root node to first kind, directly find the solution 4 up replacement equations and obtain 20 border busbar voltage corrections.Aforementioned calculation is carried out at main frame.

As root node, the N4 node is as the network division of middle root node to second kind of N0 node, and computational process is described below: the up replacement equation of simultaneous N5 and N6 at first, totally 18 of the border busbar voltage corrections that this equation relates to are respectively Δ V _{Bi, N3} ^k(6 variable vectors), Δ V _{Bo, N3} ^k(6 variable vectors) and Δ V _N5-N6 ^k(secondary splitting produces 6 voltage corrections of (6,11 and 14 bus)).The line linearity conversion of going forward side by side of arrangement N5 and the up replacement simultaneous equations of N6 can get: the up replacement equation of N4 node (this equation is the up replacement equation of N3 subnet):

${\mathrm{\ΔV}}_{\mathrm{bi},N3}^k+R_{N3}^k{\mathrm{\ΔV}}_{\mathrm{bo},N3}^k=S_{N3}^k\·\·\·\left(7\right)$

With descending accounting equation:

${\mathrm{\&Delta;<figure-callout\; id="5"\; label="V\; N"\; filenames="A20061001571800151.png,A20061001571800152.png"\; state="\{\{state\}\}">V</figure-callout>}}_N^{5+N6}+{\mathrm{<figure-callout\; id="3"\; label="R\; N"\; filenames="A20061001571800151.png,A20061001571800152.png"\; state="\{\{state\}\}">R</figure-callout>}}_N^3{\mathrm{\&Delta;V}}_{\mathrm{bo},N3}^k={\mathrm{<figure-callout\; id="3"\; label="S\; N"\; filenames="A20061001571800151.png,A20061001571800152.png"\; state="\{\{state\}\}">S</figure-callout>}}_N^3\&CenterDot;\&CenterDot;\&CenterDot;\left(8\right)$

The up replacement equations simultaneousness of formula (7) and N1 and N2 is tried to achieve divided ownership boundary voltage correction for the first time, and incites somebody to action Δ V wherein _{Bo, N3} ^kBring formula (8) into and calculate Δ V _N5+N6 ^kCan get 20 border busbar voltage corrections equally.The work of 20 borders of aforementioned calculation busbar voltage correction is all carried out at main frame.

To respectively: 1 work song computer is returned in 1,3 and 17 busbar voltage corrections (6) communication; 2 work song computers are returned in 16 busbar voltage corrections (2) communication; 3 work song computers are returned in 15,4,11 busbar voltage corrections (6) communication; 4 work song computers are returned in 39,4,6 busbar voltage corrections (6) communication; Soon border busbar voltage correction is brought 4 sub-computers into and is just calculated N1, N2, all internal bus voltage corrections of four sub-networks of N5 and N6.After repeating said process judgement convergence as shown in Figure 1, trend is calculated and is finished.

Claims (8)

1, a kind of parallel computation method for large-scale electrical power system network tidal current segmentation is characterized in that this method comprises:

1), extensive or ultra-large electric power networks is divided into one group of subnet according to selected interconnection set;

2), according to the network division result, make up electric network swim and calculate tree; Trend is calculated tree " root node ", " middle root node " and undermost " leaf node " formation by the node of the superiors;

3), the variable in the electron network tidal current analysis after cutting apart is classified;

4), row are write each electron network power flow equation;

5), adopt Newton iteration method solution procedure 4) power flow equation of each sub-network of obtaining, the calculating of variable correction is operated by " up replacement " and " descending calculating " and is finished in the solution procedure;

6), the corrected Calculation of above-mentioned variable till the Newton iteration convergence, output result of calculation finishes trend and calculates;

7), above-mentioned large-scale electrical power system tidal current computing method parallel computation on clustered computing system.

2, trend parallel calculating method according to claim 1 is characterized in that in the step 1), and the characteristics of large-scale electrical power system network division are:

A), according to the interconnected characteristics of large scale electric network layering, layering selects the network division interconnection that electrical network is divided into one group of sub-network;

B), the electrical link between sub-network and sub-network is by the interconnection set description of layering selection.

3, trend parallel calculating method according to claim 1 is characterized in that step 2) in, the electric network swim that constructs calculates tree and has following characteristics:

A), each " leaf node " corresponding electron network;

B), " root node " correspondence is carried out the interconnection set of ground floor network division;

C), the corresponding trend of selected interconnection of cutting apart that ground floor is later is calculated " the middle root node " set;

D), electric network swim calculates the logical relation that tree has been described tidal current analysis, between layer node, except with the calculating mutually of its upper strata node outside the Pass, be mutually independent.

4, trend parallel calculating method according to claim 1 is characterized in that in the step 3), and the unknown variable classification in the sub-network tidal current analysis has following characteristics:

A), the unknown variable that uses in the sub-network tidal current analysis is its busbar voltage variable;

B), according to its bus with participate in the interconnection relation of its network division, wherein all unknown variables are divided into three classes: border output busbar voltage variable, border inlet highway voltage quantities and internal bus voltage quantities;

C), internal bus is the sub-network bus not related with the network division interconnection;

D), the border bus is the interconnection bus related with network division, its interconnection busbar voltage that belongs to this sub-network is called border output busbar voltage, its interconnection busbar voltage that does not belong to this sub-network is called border inlet highway voltage.

5, trend parallel calculating method according to claim 1 is characterized in that in the step 4), and electron network power flow equation has following characteristics:

A), be listed as the trend algebraic equation of writing each leaf node map network, the i.e. meritorious and balance of voltage equation of the power balance equation of PQ bus and PV node;

B), all-ones subnet network power flow equation simultaneous promptly is: the power flow equation of the whole network trend;

C), each sub-network power flow equation is carried out Taylor (Taylor) series expansion, omit second order and above higher order term in the Taylor series, obtain the Jacobian matrix equation of subsystem; The Jacobian matrix equation is made up of the uneven vector of the correction vector sum of Jacobian matrix, busbar voltage, and in each iteration, Jacobian matrix and uneven vector all are real number matrix and the vectors of determining, what need find the solution is the correction vector of busbar voltage.

6, trend parallel calculating method according to claim 1 is characterized in that in the step 5), being calculated as follows of Jacobian matrix equation variable correction:

The calculating of the correction vector of the whole network busbar voltage is finished by " up replacement " and " descending calculating " operation that electric network swim calculates tree, wherein,

A), up replacement operation:

This up replacement operation is from all leaf nodes, for leaf node, by the Jacobian matrix equation is carried out the elementary algebra conversion, all write the correction of output of the border in the leaf node and builtin voltage variable as represent linear representation by the correction of border input voltage variable;

For middle root node: the up replacement that middle root node will be done is a border busbar voltage correction linear representation of accepting lower floor's node, the border busbar voltage correction of the network division of this centre root node correspondence of cancellation, the linear expression of correction of the border busbar voltage that acquisition last layer root node needs;

For root node, the system of linear equations that the border of its all lower floor's nodes of simultaneous output, input voltage correction are represented, and find the solution this equation group, gained promptly: border output, input voltage correction;

B), descending calculating operation:

Descending calculating operation calculates all leaf nodes of tree by root node to trend to carry out, bring the boundary voltage correction that solves the linear representation of the inner or following one deck border input of up replacement gained, output voltage correction and border input voltage correction into, try to achieve each sub-network internal bus voltage correction vector at last.

7, trend parallel calculating method according to claim 1 is characterized in that in the step 6), and the criterion of Newton iteration convergence is:

The absolute value of all-ones subnet network busbar voltage correction is all less than a certain threshold value ε, ε＜10 ^-5

8, trend parallel calculating method according to claim 1 is characterized in that in the step 7), and large-scale electrical power system tidal current computing method method of parallel computation on clustered computing system is:

Realize on the sub-computer of clustered computing system that respectively the up of leaf node substitutes and descending calculating, realize that on main frame the up of each layer root node substitutes and descending calculating; Calculating parallel synchronous on the sub-computer is carried out.

Images