CN106056466A - Large-power-grid key line identification method based on FP-growth algorithm - Google Patents

Abstract

The invention provides a large-power-grid key line identification method based on an FP-growth algorithm. The method is characterized by comprising the following steps: obtaining a line active relation/dependence matrix by utilizing a DC power flow method, and establishing a line breaking probability model based on a power grid operation state, a line power flow constraint condition and the line active relation/dependence matrix; performing a Monte Carlo simulation for cascading failures based on the line breaking probability model, and generating a cascading failure set; and, on the basis of the FP-growth algorithm, excavating a frequent item for the cascading failure set, and determining a key line. Through an application of the technical scheme, a few lines having sharp changes in line active power flow can be identified and controlled in a short time at an early failure stage, and the technical scheme has great significances for inhibiting the cascading failures and guaranteeing safe and stable operation of an electric power system.

Description

Bulk power grid critical circuits recognition methods based on FP-growth algorithm

Technical field

The invention belongs to Operation of Electric Systems and fault diagnosis technology field, particularly relate to a kind of based on FP-growth The bulk power grid critical circuits identification technical scheme of algorithm.

Background technology

In recent years, in world wide, large-scale blackout takes place frequently, and causes substantial amounts of Socie-economic loss.For bulk power grid, I.e. electric pressure is 220kV and above power transmission network, and research shows, most large-scale blackout are all to be caused by cascading failure 's.In the cascading failure early stage of development, fault element mainly causes other element generation circuit by causing power flow transfer Overload or the fault such as busbar voltage offrating is excessive.In early days, only minority circuit effective power flow occurs drastically to change fault, The change of most of circuit effective power flow is little.Therefore, be there is minority line jumpy in circuit effective power flow at short notice Road is identified and is controlled by, and suppression cascading failure being ensured, the safe and stable operation of power system is significant.

At present, for the research of critical circuits identification achieved with remarkable progress.Part experts and scholars are with Power System Analysis Theory is core, utilizes definitiveness or the identification of probabilistic approach research critical circuits；Also have experts and scholars from complex network angle Degree sets out, and electrical network turns to abstract topological network, studies the impact of the topological parameter such as average path length, node degree, and then The structurally critical circuits in identification network.Along with developing rapidly of Internet technology and computer technology, big data are So become one of the most popular current vocabulary, at numerous areas such as astronomy, atmospheric science, genomics, biogeochemistrys The most well applied.

The weak link of the power grids circuits of identification 220kV and above thereof, the research in this field is still in Starting stage, is also the innovation place of the present invention.

Summary of the invention

Consider that existing research in terms of electrical network critical circuits identification is mainly according to power system theory and complex web The knowledge that network is theoretical, lacks the relevant programme of the current big data technique risen of application, and the present invention proposes a kind of based on covering spy Carlow simulation and the bulk power grid critical circuits identification new method of FP-growth algorithm.

Technical solution of the present invention provides a kind of bulk power grid critical circuits recognition methods based on FP-growth algorithm, including Utilize DC power flow algorithm to obtain circuit to gain merit incidence relation matrix, based on operation of power networks state, Line Flow constraints and line Road incidence relation matrix of gaining merit sets up line disconnection probabilistic model；Carry out covering spy to cascading failure based on line disconnection probabilistic model Carlow is simulated, and generates cascading failure collection；Based on FP-growth algorithm, cascading failure collection is carried out frequent-item, determines key Circuit；

Cascading failure carries out Monte Carlo simulation, and to realize process as follows,

1) line disconnection probabilistic model parameter is set, including route searching termination condition, described route searching termination condition For control measure parameter E_n, E_nRepresent maximum fault chain length；

2) electrical network initial operating state is determined, including Line Flow and topological structure；

3) selecting primary fault circuit and disconnect, note circuit disconnects bar number N_T=1, if meeting route searching termination condition, If N_T≥E_n, then 8 are proceeded to), otherwise enter 4)；

4) calculate the corresponding circuit of each branch road to gain merit incidence relation matrix, calculate each branch road according to line disconnection probabilistic model Cut-off probability；

5) carry out Monte Carlo simulation, determine and cut-off branch road；If without newly cut-offfing branch road, then proceeding to 8)；Otherwise, next is entered Step 6)；

6) if newly cut-offfing circuitry number to be more than one, then only select one according to roulette method, make N_T=N_T+1；

7) judge whether to meet route searching termination condition, if N_T≥E_n, then 8 are proceeded to)；Otherwise, disconnect and newly cut-off branch road, Carry out optimal load flow calculating, update operation of power networks state, proceed to 4)；

8) single cascading failure simulation process terminates, and record cut-offs branch road.

And, described utilize DC power flow algorithm to obtain circuit to gain merit incidence relation matrix, obtain initially including according to following formula Circuit after faulty line cut-offs is gained merit incidence relation matrix,

${\mathrm{\ΔP}}_{rest}^{ij}=P_{ij}^0{\mathrm{SB}}^{-1}{M}^{ij}$

Wherein,Each branch road active power incremental vector after disconnecting for branch road between node i, j,For branch breaking Front active power, S is branch node incidence matrix, and B is node susceptance matrix, M^ijFor with to cut-off branch road the most vectorial.

And, described set up line disconnection probabilistic model, describe trend including the pattern utilizing sectional curve and stop with circuit Relation between fortune probability,

If L_limitFor the trend limit value that circuit is properly functioning, L_maxThe trend limit value run for circuit, P_HGeneral for hidden failure Rate, P_TFor line disconnection probability during Line Flow over-limit condition, the functional relationship of homologous thread is as follows,

$P=\left\{\begin{array}{cc}{P}_H& L<L_{\lim it}\\ P_H+\frac{P_T-P_H}{L_{max}-L_{limit}}(L-L_{\lim it})& L_{\lim it}\&le;L\&le;L_{max}\\ P_T& L>L_{\max }\end{array}\right.$

Wherein, P is line disconnection probability, and L is circuit Real-time Power Flow.

Monte Carlo simulation and big data algorithm FP-growth, based on power system theory, are applied to by the present invention Cascading failure critical circuits identification, and utilize the basic thought of FP-growth algorithm, with the cascading failure collection that generates as basic number According to, building FP tree, and FP tree is carried out frequent-item, the frequent episode that available cascading failure is concentrated, this frequent episode is electrical network When breaking down, cut-off the line set that probability is high, be defined as the critical circuits of electrical network, i.e. electrical network weak link place.This Bright operational efficiency is high, and recognition accuracy is high, saves artificial, can occur circuit effective power flow drastically within the fault short time in early days The minority circuit of change is identified and is controlled by, and suppression cascading failure being ensured, the safe and stable operation of power system has Significance, has important market value.

Accompanying drawing explanation

Fig. 1 is the line disconnection probability segmented line model of the embodiment of the present invention.

Fig. 2 is the cascading failure modeling process chart of the embodiment of the present invention.

Fig. 3 is the FP tree diagram of the embodiment of the present invention.

Detailed description of the invention

Combine accompanying drawing below by embodiment, technical scheme is described in further detail.

The technical scheme that the embodiment of the present invention is used provides a kind of point cloud segmentation method based on cluster, including following Step:

1. set up circuit to gain merit incidence relation matrix model: include proposing circuit and gain merit the generation side of incidence relation matrix Formula, i.e. based on electrical network running status, topological structure and line parameter circuit value, utilize DC power flow algorithm to obtain circuit and gain merit incidence relation Matrix.

Present invention assumes that the running status of electrical network, topological structure and line parameter circuit value it is known that based on DC power flow algorithm, push away Lead the line disconnection between node i, j and cause the change of residue circuit effective power flow, build the incidence relation of circuit effective power flow Matrix, this matrix and network topology structure, each branch parameters and the position cut-offfing circuit and meritorious through-put power are relevant.

The step 1 of embodiment implements and includes following sub-step:

1.1 DC power flow algorithm derivation node voltage phase angle increments.

Ignore branch road reactive power, branch resistance and over the ground susceptance time, the fundamental equation of DC power flow algorithm is:

P=B θ (1)

In formula (1), P is the meritorious injection column vector of node；B is node susceptance matrix；θ be node voltage phase angle arrange to Amount.

Branch road between disconnected node i and j, then node susceptance matrix and voltage phase angle vector all will change:

B '=B+ Δ B；θ '=θ+Δ θ (2)

In formula (2), Δ B is the part after branch breaking to original susceptance matrix correction；Δ θ is the increasing of node voltage phase angle Amount vector.

Assume that the active power of each node is injected constant, then the equation deviateing basic status is:

P=(B+ Δ B) (θ+Δ θ) (3)

Formula (3) is launched, omits two increments and be multiplied item, obtain:

Δ θ=-B^-1ΔBθ (4)

Wherein,b_ijFor branch admittance matrix between node i j.

Therefore, formula (4) can also be write as: Δ θ=b_ij(θ_j-θ_i)B^-1M^ij (5)

Wherein, θ_jFor the voltage phase angle of node j, θ_iVoltage phase angle for node i；M^ijBe one with to cut-off branch road relevant Vector, is expressed as:

M^ij=[0 ..., 1 ... ,-1 ..., 0]^T (6)

M^ijThe i-th row element be 1, jth row element is-1.

Each node voltage phase angle increment after can get branch breaking according to (5) formula.

1.2 determine meritorious incidence relation matrix

Definition branch node incidence matrix S=[S_lb]_A×B, for A × B matrix, A is grid branch quantity, and B is nodes Mesh.

X_ijLine impedance for branch road l.

By S_lbSimultaneously in formula (5) both sides premultiplication, can obtain

S Δ θ=b_ij(θ_j-θ_i)SB^-1M^ij (7)

Visible:

Formula (7) the right b_ij(θ_j-θ_i) item is the active power before branch breaking, it is designated asLeft item is between node i, j Each branch road active power incremental vector after branch road disconnection, is designated as

Therefore, formula (7) can be rewritten as:

${\mathrm{\&Delta;P}}_{rest}^{ij}=P_{ij}^0{\mathrm{SB}}^{-1}{M}^{ij}---\left(8\right)$

Circuit after can obtaining primary fault line disconnection according to formula (8) is gained merit incidence relation matrix.

2. set up line disconnection based on operation of power networks state, Line Flow constraints, circuit incidence relation matrix of gaining merit Probabilistic model.

Upgrade in time operation of power networks state, and the circuit that integrating step 1 generates is gained merit incidence relation matrix, general to branch breaking Rate is modeled.Probabilistic model considers the impact between each circuit, and meter and the malfunction of protective relaying device, tripping are to chain simultaneously The impact of fault, introduces hidden failure probability.

The step 2 of embodiment implements and includes following sub-step:

2.1 consider branch road incidence relation and the line disconnection probabilistic model of hidden failure.

The probability of malfunction of traditional element uses the meansigma methods of long-time statistical data, have ignored the time-varying of element fault information Property and element in relatedness, it is impossible to explain, assessment system event of failure under abnormal running mode.The present invention considers The impact that cascading failure is developed by faulty line by impact and the hidden failure of normal working line, in conjunction with the running status of electrical network With Line Flow constraints, set up line disconnection probabilistic model.This model utilizes the pattern of sectional curve to describe trend and line Relation between road stoppage in transit probability, as shown in Figure 1.

In Fig. 1, L_limitFor the trend limit value that circuit is properly functioning；L_maxThe trend limit value run for circuit；P_HFor recessive event Barrier probability；P_TFor line disconnection probability during Line Flow over-limit condition.The functional relationship of homologous thread is:

$P=\left\{\begin{array}{cc}{P}_H& L<L_{\lim it}\\ P_H+\frac{P_T-P_H}{L_{\max }-L_{\lim it}}(L-L_{\lim it})& L_{\lim it}\&le;L\&le;L_{\max }\\ P_T& L>L_{\max }\end{array}\right.---\left(9\right)$

In formula, P is line disconnection probability；L is circuit Real-time Power Flow；

The method that 2.2 present invention propose to calculate each branch breaking probability.

Before assuming line disconnection, each branch road effective power flow is P⁰, each branch road effective power flow after the circuit beginning between node i j For P ', then

Known electric network operation trend constraints, i.e. L_limitAnd L_maxFor known quantity, can be according to formula (9) to each branch breaking Probability calculates, and obtains each branch breaking probability vector P_k, k=1,2,3 ... m.

M is that fault branch cut-offs rear electrical network residue branch road quantity

3. the generation of cascading failure fault chains based on Monte Carlo simulation.

The line disconnection probabilistic model set up based on step 2, carries out Monte Carlo simulation to cascading failure.At the beginning of randomly selecting Beginning faulty line, given search end condition E_n(present invention limits maximum fault chain length, i.e. control measure parameter), the most right Cascading failure chain scans for.This process repeatedly, generates cascading failure collection.Cascading failure search procedure of the present invention considers The impact of hidden failure, and the impact that dispatcher is on anthropic factors such as power grid regulation.

The step 3 of embodiment implements and includes following sub-step:

3.1 cascading failure simulation processes

Carrying out cascading failure fault chains simulation, comprise the following steps that, flow chart is as shown in Figure 2.

1) line disconnection probabilistic model parameter is set: hidden failure probability P_H, Line Flow is out-of-limit cut-offs probability P_T, circuit Properly functioning trend limit value L_1imit, the trend limit value L of circuit operation_max, control measure parameter E_n。

2) electrical network initial operating state is determined, including Line Flow and topological structure.

3) selecting primary fault circuit and disconnect, note circuit disconnects bar number N_T=1, if meeting route searching termination condition, Even N_T≥E_n, proceed to 8), otherwise enter 4).

4) utilize 1.1 and 1.2 offer methods calculate the corresponding circuits of each branch road and gain merit incidence relation matrix, utilize 2.2 Thered is provided method, calculates each branch breaking probability.When being embodied as, can in step 1 before, 2 calculate in advance, it is also possible to In step 1,2 establishing methods, calculate when carrying out 3.1 and needing.

5) carry out Monte Carlo simulation, determine and cut-off branch road.If without newly cut-offfing branch road, then proceeding to 8)；Otherwise, next is entered Step 6).The method of Monte Carlo simulation line disconnection is prior art, and it will not go into details for the present invention.

6) if newly cut-offfing circuitry number to be more than one, then only select one according to roulette method, make N_T=N_T+1。

7) judge whether to meet route searching termination condition, even N_T≥E_n, then 8 are proceeded to)；Otherwise, disconnect newly cut-offfing and prop up Road, carries out optimal load flow calculating, updates operation of power networks state, proceeds to 4).When being embodied as, available MATLAB, PSASP etc. are soft Part carries out optimal load flow calculating, and it will not go into details for the present invention.

8) single cascading failure simulation process terminates, and record cut-offs branch road.

3.2 generate cascading failure fault chains collection.

If according to the cascading failure simulation process shown in 3.1, carry out n simulation, obtain comprising the chain event of n bar fault chain Barrier collection C={c₁,c₂,c₃…c_n, wherein c_iFor frequent item set, the i.e. set of grid branch numbering；Arbitrary frequent item set c_iIt is one Group set of digits, c_i={ l₁,l₂…l_v..., l_vFor branch number, v is 1,2,3 ... Any Digit value between m.

4., based on FP-growth algorithm, cascading failure collection is carried out frequent-item.

FP-growth (Frequent Pattern-growth) algorithm is the discovery algorithm of a kind of frequent item set, in search The numerous areas of engine and the Internet is used widely.FP-growth algorithm is broadly divided into two steps: FP-tree structure Build, recurrence excavates FP-tree.Present invention application FP-growth algorithm carries out data mining to cascading failure fault chains collection, obtains Frequent episode in fault set, the high probability after frequent episode is electrical network generation random fault cut-offs circuit.By Mining Frequent item, The weak link of operation of power networks can be found, find to affect the critical circuits of operation of power networks.

4.1 build FP-tree, i.e. FP tree (frequent pattern tree (fp tree)) based on cascading failure collection C.

FP-tree is constructed by twice data scanning, by the Redundant Transaction Compression idea in initial data to a FP-tree tree, and should FP-tree is similar to prefix trees, and the path of same prefix can share, thus reaches to compress the purpose of data.Concrete steps are such as Under:

1) travel through fault set C for the first time, record each element entry l_vFrequency f occurred_v.By each element entry according to f_vFall Sequence arranges, then by each element entry l_vAnd frequency f_vIt is saved in head pointer table Tab_head, head pointer table points to preserved element entry First element entry, this element entry also can be linked to its follow-up element entry；Pre-set minimum support f_minIf, f_v＜ f_min, the most from the beginning pointer gauge removes this element entry.

When being embodied as, head pointer table comprises all elements item, and each element entry can be with this element entry place in FP tree Position establish the link relation, so a certain element entry in head pointer table first links first of this element in FP tree, first Individual link second again, arrive successively this element last, set up chained list.A certain element entry appears in the different branches of FP tree On, by branch's order from left to right, respectively first, second, subsequent third, the 4th ... can be found in Fig. 3.

2) second time traversal fault set C, each subset c to C_iCarry out screening and sequence process.According to head pointer table Tab_headIn element entry put in order to c_iIt is ranked up, and by f_v＜ f_minElement entry delete, obtain new fault set C_N, C_N={ c₁,c₂,c₃…c_n, wherein c_iFor frequent item set.This step realizes rescaning data, at each frequent item set Reason, deletes f in each frequent item set_v＜ f_minElement entry.

3) by C_NBuild FP tree.Start C from empty set_NIn each frequent item set c_iAdd to successively in tree.In if tree There is existing element l_v, then make the value+1 of existing element, the most often add a frequent item set, the element entry+1 that frequent episode contains； If existing element l_vDo not exist, then add a branch to tree.Ultimately generate in FP tree, containing new fault set C in tree_NContained Element entry and quantity information.

Assume that fault set C comprises 5 fault chains, wherein: c₁={ l₁,l₂,l₃,l₄},c₂={ l₂,l₄,l₅,l₆},c₃= {l₂,l₃,l₄},c₄={ l₁,l₂,l₃,l₄,l₅,l₆},c₅={ l₁,l₃,l₆},c₆={ l₂,l₃,l₆},c₇={ l₁,l₃,l₄},c₈= {l₁,l₂,l₃,l₄,l₇},c₉={ l₁,l₂,l₄,l₇}。

Building FP tree according to above-mentioned steps, FP tree sees Fig. 3.FP tree most starts to be an empty setGrowth is started from empty set. In figure, head pointer table comprises the element entry contained by fault set and occurrence number thereof, and by all similar in curve link FP tree Element entry；Straight line connection element, represents parent and descendant relations, and parent is upper.

4.2 from FP tree Mining Frequent Itemsets Based.

The conditional pattern base of each element entry, condition FP-tree is found out, excavation condition FP-of recurrence by FP-tree Tree obtains all of frequent item set.Definition conditional pattern base be with lookups element entry be end up set of paths, each Path is all the elements between searched element entry and root vertex.Definition condition FP-tree is with conditional pattern base As the FP-tree entering data to generation.Set up frequent item set list, for preserving the element entry that frequent item set comprises.Tool Body step is as follows:

1) conditional pattern base obtained from FP tree.Pass up to the root node of tree with element entry for starting point, record traced back The element entry run in journey, obtains a prefix path, and is this prefix path assignment, and this value takes the counting of starting elemental item Value.The chained list corresponding to element entry in traversal head pointer table, often arrives an element in chained list, the most once traces back, Can obtain a plurality of prefix path, prefix path set is conditional pattern base.

2) condition FP tree is built by conditional pattern base.By l_vAdd in frequent item set list, then with l_vFor root node, l_v Corresponding conditional pattern base carries out building in 4.1 FP tree root in the process (note: step 3 in 4.1) of FP tree and saves as input data Point is empty set, and root node is l here_v).Element entry (node that digs up the roots is outer) in condition FP tree is added separately to frequent item set row In table.

3) element entry (node that digs up the roots is outer) in condition FP tree is repeated step 1) and 2), until tree comprises an element entry Till, obtain final frequent item set list, in store containing element entry l in this list_vFrequent item set and the frequency of occurrences.

4) element entry in correct pointer gauge carries out ascending order arrangement according to its frequency occurred, then to each element entry Carry out 1) to 3) in the process of step, obtain the frequent item set of each element entry.

Based on the 4.1 FP trees built, according to step 1)-4) carry out frequent-item, the results are shown in Table 1.Can from table Going out, the frequent item set that frequency is the highest is { l₄,l₂, electrical network i.e. cut-offs the line combination that probability is the highest, is defined as the key of electrical network Circuit.

Table 1 frequent item set list

Frequent episode comprises element	Frequency
		l4,l2	6
l4,l3	5
		l2,l3	5
l4,l1	5
		l3,l1	5

When being embodied as, those skilled in the art can use computer software technology to realize the automatic operation of above flow process.

It is emphasized that embodiment of the present invention is illustrative rather than determinate.Bag the most of the present invention Include the embodiment being not limited to described in detailed description of the invention, every by those skilled in the art according to technical scheme Other embodiments drawn, also belong to the scope of protection of the invention.

Claims (3)

1. a bulk power grid critical circuits recognition methods based on FP-growth algorithm, it is characterised in that: include utilizing direct current tide Stream method obtains circuit and gains merit incidence relation matrix, associates based on operation of power networks state, Line Flow constraints and circuit are meritorious Relational matrix sets up line disconnection probabilistic model；Based on line disconnection probabilistic model, cascading failure is carried out Monte Carlo simulation, Generate cascading failure collection；Based on FP-growth algorithm, cascading failure collection is carried out frequent-item, determines critical circuits；

Cascading failure carries out Monte Carlo simulation, and to realize process as follows,

1) setting line disconnection probabilistic model parameter, including route searching termination condition, described route searching termination condition is control Measure parameter E processed_n, E_nRepresent maximum fault chain length；

2) electrical network initial operating state is determined, including Line Flow and topological structure；

3) selecting primary fault circuit and disconnect, note circuit disconnects bar number N_T=1, if meeting route searching termination condition, if N_T≥ E_n, then 8 are proceeded to), otherwise enter 4)；

4) calculate the corresponding circuit of each branch road to gain merit incidence relation matrix, calculate each branch breaking according to line disconnection probabilistic model Probability；

5) carry out Monte Carlo simulation, determine and cut-off branch road；If without newly cut-offfing branch road, then proceeding to 8)；Otherwise, next step is entered

6)；

6) if newly cut-offfing circuitry number to be more than one, then only select one according to roulette method, make N_T=N_T+1；

7) judge whether to meet route searching termination condition, if N_T≥E_n, then 8 are proceeded to)；Otherwise, disconnect and newly cut-off branch road, carry out Optimal load flow calculates, and updates operation of power networks state, proceeds to 4)；

8) single cascading failure simulation process terminates, and record cut-offs branch road.

Point cloud segmentation method based on cluster the most according to claim 1, it is characterised in that: described utilize DC power flow algorithm to obtain Circuit is gained merit incidence relation matrix, gain merit incidence relation square including the circuit after obtaining primary fault line disconnection according to following formula Battle array,

${\mathrm{\&Delta;P}}_{rest}^{ij}=P_{ij}^0{\mathrm{SB}}^{-1}{M}^{ij}$

Wherein,Each branch road active power incremental vector after disconnecting for branch road between node i, j,Before branch breaking Active power, S is branch node incidence matrix, and B is node susceptance matrix, M^ijFor with to cut-off branch road the most vectorial.

Point cloud segmentation method based on cluster the most according to claim 2, it is characterised in that: described set up line disconnection probability Model, describes the relation between trend and line outage probability including the pattern utilizing sectional curve,

If L_limitFor the trend limit value that circuit is properly functioning, L_maxThe trend limit value run for circuit, P_HFor hidden failure probability, P_T For line disconnection probability during Line Flow over-limit condition, the functional relationship of homologous thread is as follows,

$P=\left\{\begin{array}{cc}{P}_H& L<L_{\lim it}\\ P_H+\frac{P_{T-}{P}_H}{L_{\max }-L_{\lim it}}(L-L_{\lim it})& L_{\lim it}\&le;L\&le;L_{\max }\\ P_T& L>L_{\max }\end{array}\right.$

Wherein, P is line disconnection probability, and L is circuit Real-time Power Flow.