CN107069708B - Extreme learning machine-based transmission network line active safety correction method - Google Patents

Extreme learning machine-based transmission network line active safety correction method Download PDF

Info

Publication number
CN107069708B
CN107069708B CN201710151119.2A CN201710151119A CN107069708B CN 107069708 B CN107069708 B CN 107069708B CN 201710151119 A CN201710151119 A CN 201710151119A CN 107069708 B CN107069708 B CN 107069708B
Authority
CN
China
Prior art keywords
generator
power
load
sensitivity
active
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN201710151119.2A
Other languages
Chinese (zh)
Other versions
CN107069708A (en
Inventor
李淼
周强明
周悦
鲁鸿毅
曾鹏
姜盛波
杨军
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
State Grid Hubei Electric Power Co Ltd
Original Assignee
Wuhan University WHU
State Grid Hubei Electric Power Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU, State Grid Hubei Electric Power Co Ltd filed Critical Wuhan University WHU
Priority to CN201710151119.2A priority Critical patent/CN107069708B/en
Publication of CN107069708A publication Critical patent/CN107069708A/en
Application granted granted Critical
Publication of CN107069708B publication Critical patent/CN107069708B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]

Abstract

The invention belongs to the field of operation and control of an electric power system, and relates to a power transmission network line active safety correction method based on an extreme learning machine. When the existing sensitivity-based heuristic active safety correction method is considered, the invention provides the extreme learning machine-based transmission line active safety correction method in consideration of rapidity and accuracy of adjustment strategy formulation, and balances the rapidity and accuracy of the active safety correction method. The method comprises the steps that firstly, power grid operation data under different states are analyzed and learned on the basis of an extreme learning machine, and the sensitivity of injection power of each node to a power transmission line is obtained; then establishing an active safety correction model based on the sensitivity of the node injection power to the power transmission line; and finally, solving the active safety correction model by utilizing a particle swarm intelligent algorithm to obtain a generator and load adjustment scheme.

Description

Extreme learning machine-based transmission network line active safety correction method
Technical Field
The invention belongs to the field of operation and control of an electric power system, and relates to a power transmission network line active safety correction method based on an extreme learning machine.
Background
The development of the interconnection technology of the smart power grid and the large power grid plays an important role in solving the problem of uneven energy distribution pattern in China and improving the robustness of the power grid, but simultaneously leads to unprecedented improvement of the interconnection degree and complexity of the power grid in China, and puts higher requirements on safe and stable operation of the power grid. Research shows that cascading failures are important causes of major power failure accidents, and normal operation lines are changed into heavy-load and overload operation states due to power flow transfer caused by failed elements, which are important factors for cascading failures. Therefore, the active safety correction is carried out on the operating power transmission line in a heavy load and overload state, and a dispatching department needs to take rapid and effective power generator output adjustment and load shedding measures to enable the power transmission line to recover normal operation, so that the method has important significance for restraining cascading failure development and ensuring safe and stable operation of a power system.
At the present stage, in the field of active safety correction of power transmission lines, expert scholars mainly adopt an optimization planning algorithm and a sensitivity algorithm. The optimization planning algorithm obtains correction measures by establishing an active safety correction mathematical model, setting a target function and various constraint conditions and solving the mathematical model by using a mathematical analysis method, and the method has comprehensive consideration and better safety and economy, but may have the problems of overlong solving time, excessive adjusting equipment, difficulty in meeting load flow convergence conditions and the like; the sensitivity method can quickly obtain the adjustment measures to eliminate the overload by solving the sensitivity of the generator to the overload line and further obtaining the correction measures according to the overload capacity of the line, and the method has no problem of tidal current convergence, but the accuracy of the sensitivity directly influences the effectiveness of the correction scheme.
Disclosure of Invention
The technical problem of the invention is mainly solved by the following technical scheme:
a power transmission line active safety correction method based on an extreme learning machine is characterized by comprising the following steps:
step 1, acquiring a generator sensitivity sample set and a load sensitivity sample set, and performing learning training on the acquired power grid sample set based on an extreme learning machine to obtain a mapping relation between the sensitivity of node injection power to a power transmission line and power grid operation data;
step 2, establishing an active safety correction model of the power transmission line based on the sensitivity analysis in the step 1, wherein the active safety correction model is based on a target function:
Figure GDA0001296865170000021
Figure GDA0001296865170000022
in order to reduce the adjustment amount of the power generator,
Figure GDA0001296865170000023
for adding a regulation of the power generator, Δ PG,zIn order to adjust the amount of the balancing machine,
Figure GDA0001296865170000024
the load shedding amount is the load shedding amount, M is a penalty coefficient and is a normal number far larger than 1, the output of the generator is preferentially adjusted when the line overload is eliminated, then the load shedding is carried out, and the objective function takes the minimum of the generator and the load adjustment amount as the target:
and 3, acquiring power grid data to be controlled, solving the power transmission line active safety correction model based on the active safety correction model in the step 2, and determining a control scheme according to the adjustment quantity of the generator and the load.
In the above method for correcting the active safety of the power transmission line based on the extreme learning machine, the step 1 specifically includes:
step 1.1, generating an extreme learning machine training sample set comprising a generator sensitivity sample set and a load sensitivity sample set, firstly determining the initial operation state and the network topology structure of a power grid in any power grid system, and recording the active power output P of a generator in the initial state of the power gridG 0Active power of load
Figure GDA0001296865170000025
And the active power of each transmission line
Figure GDA0001296865170000026
The method comprises the following steps:
set one, generator sensitivity sample set: randomly changing the active output of any generator (except balance node)
Figure GDA0001296865170000031
The output is randomly changed to ensure that the obtained sample data can comprehensively reflect different operating states of the power grid. In order to ensure the active power balance of the system, the active power output of the balancing machine needs to be reversely adjusted while the active power output of the generator is changed
Figure GDA0001296865170000032
After the generated output is adjusted, the tidal current information of the power grid is updated, and the active power of each transmission line is recorded
Figure GDA0001296865170000033
And active variation
Figure GDA0001296865170000034
The sensitivity of the generator to the transmission line can be calculated from the recorded information
Figure GDA0001296865170000035
A set of sample data (x) is available1,t1) Wherein
Figure GDA0001296865170000036
This process is repeated N times, each time yielding a set (x)i,ti) Wherein
Figure GDA0001296865170000037
Finally, a generator sensitivity training sample set can be obtained
Figure GDA0001296865170000038
Wherein n is 3 and m is 1.
Set two, load sensitivity sample set: randomly changing the active load aiming at any load node except the balance node
Figure GDA0001296865170000039
Output of balance machine adjusted in same direction
Figure GDA00012968651700000310
Updating power flow information of the power grid and recording active power of each transmission line
Figure GDA00012968651700000311
And active variation
Figure GDA00012968651700000312
Calculating the sensitivity of the adjusted load node to the transmission line
Figure GDA00012968651700000313
A set of sample data (x) is available1,t1) Wherein
Figure GDA00012968651700000314
This process is repeated N times, one group (x) being obtained each timei,ti) Wherein
Figure GDA00012968651700000315
Finally, a load sensitivity training sample set can be obtained
Figure GDA00012968651700000316
Wherein n is 3 and m is 1.
Step 1.2, training a sample set based on an extreme learning machine: sensitivity training sample set given based on step 1.1
Figure GDA00012968651700000317
And if the number of the nodes of the selected hidden layer is L, expressing the mathematical model of the extreme learning machine as follows:
Figure GDA00012968651700000318
in the formula (1), giAs a function of excitation wi＝[wi1,wi2,...,win]As weight vectors of the input layer, betai＝[βi1,βi2,...,βim]As output weight vectors, i.e. the concatenation weight vectors of the hidden layer to the output layer, biIs the bias of the ith hidden layer.
The matrix expression is of the form:
Hβ＝Y (2)
in the formula (2), the reaction mixture is,
Figure GDA0001296865170000046
Figure GDA0001296865170000041
h is the hidden layer output of the neural network, and the ith column vector corresponds to the output of the ith hidden layer node.
It is known that when the excitation function g is infinitely differentiable, not all network parameters need to be adjusted. Thus randomly setting the input weight wiAnd bias of the hidden layer biThus, for a given set of training samples, the output matrix H of the hidden layer can be uniquely determined, with only the variable β present in the model.
Constructing an error function of the extreme learning machine about beta:
Figure GDA0001296865170000042
at this time, the training process of the extreme learning machine on the sample set can be converted into a least square solution problem for solving an output weight matrix:
minβ||Hβ-T|| (4)
solving equation (4) yields:
Figure GDA0001296865170000043
in the formula (5)
Figure GDA0001296865170000044
Is the Moore-Penrose generalized inverse of matrix H.
Finally, obtaining a mapping relation between the sensitivity and the power grid operation information:
Figure GDA0001296865170000045
in the above method for correcting the active safety of the power transmission line based on the extreme learning machine, the step 2 is based on the definition: line L of the grid in a certain operating statemAn overload of
Figure GDA0001296865170000051
Wherein P ismIs a line LmThe actual active power of the power plant,
Figure GDA0001296865170000052
is a line LmThe active transmission limit of (a) is that the line L needs to be put away in order to eliminate the overload and to keep a certain marginmActive power reduction of
Figure GDA0001296865170000053
Active output P of each generator of known power gridGAnd the active power of the load PLThus, the input vector x of equation (6) can be determinedG＝[ΔPc,Pm,PG]And xL＝[ΔPc,Pm,PL]Finding the generator and load pair line LmThe sensitivity of (c); specifically, the method includes an objective function having constraint conditions, where the constraint conditions include:
and (b) constraint condition a, system active power balance constraint, and the generator adjustment quantity and the load adjustment quantity are equal.
Figure GDA0001296865170000054
And (c) eliminating line overload constraint to ensure that the overload phenomenon is eliminated after adjustment.
Figure GDA0001296865170000055
In the formula (9), the reaction mixture is,
Figure GDA0001296865170000056
for reducing the generator pair overload line LmThe sensitivity of (a) to (b) is,
Figure GDA0001296865170000057
for adding power generator to overload line LmThe sensitivity of (a) to (b) is,
Figure GDA0001296865170000058
for load to overload line LmThe sensitivity of (2).
And c, the active operation safety constraint of the branch circuit guarantees the safe operation of other adjusted circuits.
Figure GDA0001296865170000059
In the formula (10), t is a line number other than the overloaded line,
Figure GDA00012968651700000510
the initial active power, the minimum active power limit value and the maximum active power limit value of the line t are respectively.
And (5) constraining the output adjustment quantity of the generator set.
Figure GDA00012968651700000511
In the formula (11), the reaction mixture is,
Figure GDA00012968651700000512
respectively the initial active power output and the active power output lower limit of the reduced-power generator,
Figure GDA00012968651700000513
are respectively added outAn initial active power output and an active power output upper limit of the power generator.
And e, constraint condition e, load adjustment amount constraint.
Figure GDA0001296865170000061
In the formula (12), the reaction mixture is,
Figure GDA0001296865170000062
is the initial active load.
In the above method for correcting the active safety of the power transmission line based on the extreme learning machine, the step 4 is based on defining a certain line L in the power gridmDuring overload, the generator to be adjusted has u platforms, and the load to be cut off has v. And solving the active safety correction model established in the step 2.2 based on a particle swarm intelligent algorithm to obtain the adjustment quantity of the generator and the load, and determining a control scheme. The method comprises the following specific steps:
step 4.1, setting the maximum iteration times and initializing particle swarm: setting the maximum iteration number as MaxD, the particle swarm size as N, the particle dimension as u + v, and setting the initial position x of each particle according to the constraint conditions shown in the formulas (8), (11) and (12)iAnd an initial velocity viCarry out initialization, xi＝[ΔPG,1,ΔPG,2,...,ΔPG,u,ΔPL,1,ΔPL,2,...ΔPL,v]，vi＝[vi1,vi2,...vi(u+v)]。
Step 4.2, calculating the initial fitness value of each particle, and determining the individual extreme value p of the particleiAnd the global extremum p of the particle swarmgThe fitness function is shown in equation (13).
Figure GDA0001296865170000063
Delta P in the formula (13)G,iFor the adjustment of the generator to be adjusted, Δ PL,jAnd M is a penalty coefficient for the cutting amount of the load to be cut.
Step 4.3, updating each granuleVelocity v of the seediAnd position xiGuarantee xiThe update is performed according to equation (14) while satisfying the constraints expressed by equations (8), (11), and (12).
vi＝w*vi+c1r1(pi-xi)+c2r2(pg-xi),xi＝xi+vi (14)
W in the formula (14) is the inertial weight, c1And c2Is a learning factor, r1And r2Is [0,1 ]]A uniform random number within the range.
Step 4.4, calculating each particle x after updating the positioniIs a fitness value F (x)i) And is in parallel with the individual extremum piFitness value F (p)i) By comparison, if F (x)i)＜F(pi) Then use xiIn place of pi(ii) a For each particle xiIts fitness value F (x)i) And a global extreme value F (p)g) By comparison, if F (x)i)＜F(pg) Then use xiIn place of pg。
Step 4.5, for each particle xiAnd (4) carrying out overload elimination verification according to the formula (9) and carrying out line safety operation constraint verification according to the formula (10). If the verification is satisfied, the local optimum p is savediAnd global optimum pg(ii) a And updating the particles which do not meet the verification according to the step 1.
And 4.6, checking whether the search termination condition is met, namely whether the iteration times are greater than the maximum iteration times. If the termination condition is met, outputting the global optimum pgObtaining a control scheme; otherwise, returning to the step 3.
In the above-mentioned power transmission line active safety correction method based on the extreme learning machine, in step 4, the control scheme is a generator and load adjustment strategy based on an equivalent reverse pairing principle, and it is a common active safety correction measure by adjusting the output and the load shedding of the generator, and in the adjustment process, it is necessary to ensure the power balance of the system, preferentially adjust the output of the generator and then perform load adjustment, and at the same time, the objective of minimum adjustment quantity of the generator and the load is achieved;for overload lines L according to generator and loadmThe generator and the load are grouped: the generator sensitivity is positive, and then the generator is classified as a reduced output unit G-(ii) a The sensitivity of the generator is negative, and the generator is classified as an added power unit G+(ii) a The sensitivity of the generator is zero, and the generator is classified as a balance unit G0(ii) a For the load nodes, selecting the nodes with negative sensitivity as the load shedding set L-. And then, the generators and the loads in the set are respectively arranged in a descending order according to the absolute value. The specific adjustment strategy comprises the following steps:
and 5.1, adjusting the added power according to the absolute value sequence in the unit with negative sensitivity, and adjusting the subtracted power according to the absolute value sequence in the unit with negative sensitivity to ensure that the subtracted power of the other unit exists when one unit adds the power and the absolute values of the adjusted power are equal.
And 5.2, one unit can be sequentially matched and adjusted with a plurality of units until the adjustment capability is fully exerted.
Step 5.3, when set G-Set G when the middle unit has no capacity of reducing power0The power of the machine set can be reduced; when set G+When the middle unit has no capacity of exerting force, the unit G in the set0A force may be applied.
And 5.4, when the generator cannot eliminate the overload of the line by adjustment, performing load shedding processing until the overload of the branch line is eliminated.
Therefore, the invention has the following advantages: the method is based on the electric power system analysis theory, comprehensively considers the advantages and disadvantages of a planning algorithm and a sensitivity algorithm, establishes a sensitivity-based heuristic active safety correction model, simultaneously considers the problem of insufficient accuracy of direct current sensitivity, efficiently learns a historical data set to obtain an accurate sensitivity value based on an extreme learning machine method, and ensures the accuracy of the active safety correction scheme while realizing the rapid generation of the active safety correction scheme.
Drawings
FIG. 1 is a diagram of an extreme learning machine network architecture.
FIG. 2 is a flow chart of an optimization model solution.
Detailed Description
The technical scheme of the invention is further specifically described by the following embodiments and the accompanying drawings.
Example (b):
the following specifically describes the steps of the method of the present invention, which specifically include:
1. and performing learning training on the power grid operation data by using an extreme learning machine to obtain a mapping relation between the sensitivity of the node injection power to the power transmission line and the power grid operation data.
1.1 Generation of extreme learning machine training sample sets
The method comprises the steps of giving a certain power grid system, firstly determining the initial operation state and the network topology structure of the power grid, and recording the active power output P of a generator in the initial state of the power gridG 0Active power of load
Figure GDA0001296865170000081
And the active power of each transmission line
Figure GDA0001296865170000082
(1) Generator sensitivity sample set
Randomly changing the active output of any generator (except balance node)
Figure GDA0001296865170000091
The output is randomly changed to ensure that the obtained sample data can comprehensively reflect different operating states of the power grid. In order to ensure the active power balance of the system, the active power output of the balancing machine needs to be reversely adjusted while the active power output of the generator is changed
Figure GDA0001296865170000092
After the generated output is adjusted, the tidal current information of the power grid is updated, and the active power of each transmission line is recorded
Figure GDA0001296865170000093
And active variation
Figure GDA0001296865170000094
The sensitivity of the generator to the transmission line can be calculated from the recorded information
Figure GDA0001296865170000095
A set of sample data (x) is available1,t1) Wherein
Figure GDA0001296865170000096
This process is repeated N times, each time yielding a set (x)i,ti) Wherein
Figure GDA0001296865170000097
Finally, a generator sensitivity training sample set can be obtained
Figure GDA0001296865170000098
Wherein n is 3 and m is 1.
(2) Load sensitivity sample set
Similarly to the generation of the generator sensitivity sample set, the active load is randomly changed for any one load node (except the balance node)
Figure GDA0001296865170000099
Output of balance machine adjusted in same direction
Figure GDA00012968651700000910
Updating power flow information of the power grid and recording active power of each transmission line
Figure GDA00012968651700000911
And active variation
Figure GDA00012968651700000912
Calculating the sensitivity of the adjusted load node to the transmission line
Figure GDA00012968651700000913
A set of sample data is available(x1,t1) Wherein
Figure GDA00012968651700000914
This process is repeated N times, one group (x) being obtained each timei,ti) Wherein
Figure GDA00012968651700000915
Finally, a load sensitivity training sample set can be obtained
Figure GDA00012968651700000916
Wherein n is 3 and m is 1.
1.2 training a sample set based on an extreme learning machine
The extreme learning machine is a neural network algorithm for solving the problem proposed by Huang Guang and is shown in a network structure diagram in fig. 1. Compared with the traditional neural network, particularly a single hidden layer feedforward neural network, the extreme learning machine has the greatest characteristic that the training speed is higher, the input weight and the offset do not need to be adjusted iteratively in the training process, only the output weight beta needs to be obtained, and the extreme learning machine is very suitable for on-line calculation and updating.
Sensitivity training sample set given based on section 1.1
Figure GDA00012968651700000917
If the number of the nodes of the selected hidden layer is L, the mathematical model of the extreme learning machine can be expressed as follows:
Figure GDA0001296865170000101
in the formula (1), giAs a function of excitation wi＝[wi1,wi2,...,win]As weight vectors of the input layer, betai＝[βi1,βi2,...,βim]As output weight vectors, i.e. the concatenation weight vectors of the hidden layer to the output layer, biIs the bias of the ith hidden layer.
The matrix expression is of the form:
Hβ＝Y (2)
in the formula (2), the reaction mixture is,
Figure GDA0001296865170000102
Figure GDA0001296865170000103
h is the hidden layer output of the neural network, and the ith column vector corresponds to the output of the ith hidden layer node.
It is known that when the excitation function g is infinitely differentiable, not all network parameters need to be adjusted. Thus randomly setting the input weight wiAnd bias of the hidden layer biThus, for a given set of training samples, the output matrix H of the hidden layer can be uniquely determined, with only the variable β present in the model.
Constructing an error function of the extreme learning machine about beta:
Figure GDA0001296865170000104
at this time, the training process of the extreme learning machine on the sample set can be converted into a least square solution problem for solving an output weight matrix:
minβ||Hβ-T|| (4)
solving equation (4) yields:
Figure GDA0001296865170000105
in the formula (5)
Figure GDA0001296865170000111
Is the Moore-Penrose generalized inverse of matrix H.
Finally, obtaining a mapping relation between the sensitivity and the power grid operation information:
Figure GDA0001296865170000112
2. and establishing an active safety correction model of the power transmission line based on sensitivity analysis.
Assuming that the grid is in a certain operating state, the line LmAn overload of
Figure GDA0001296865170000113
Wherein P ismIs a line LmThe actual active power of the power plant,
Figure GDA0001296865170000114
is a line LmThe active transmission limit of (a) is that the line L needs to be put away in order to eliminate the overload and to keep a certain marginmActive power reduction of
Figure GDA0001296865170000115
Active output P of each generator of known power gridGAnd the active power of the load PLThus, the input vector x of equation (6) can be determinedG＝[ΔPc,Pm,PG]And xL＝[ΔPc,Pm,PL]Finding the generator and load pair line LmThe sensitivity of (2).
2.1 Generator and load adjustment strategy based on equal reverse pairing principle
The output and the load shedding of the generator are common active safety correction measures, in the adjustment process, the power balance of the system needs to be ensured, the output of the generator is preferentially adjusted, then the load adjustment is carried out, and meanwhile the aim of minimizing the adjustment quantity of the generator and the load is achieved.
For overload lines L according to generator and loadmThe generator and the load are grouped: the generator sensitivity is positive, and then the generator is classified as a reduced output unit G-(ii) a The sensitivity of the generator is negative, and the generator is classified as an added power unit G+(ii) a The sensitivity of the generator is zero, and the generator is classified as a balance unit G0(ii) a For the load nodes, selecting the nodes with negative sensitivity as the load shedding set L-. And then, the generators and the loads in the set are respectively arranged in a descending order according to the absolute value. The specific adjustment strategy is as follows:
a. and (3) adding force adjustment is carried out in the order of magnitude of absolute values from the unit with negative sensitivity, force reduction of the other unit is ensured when force is added by one unit, and the absolute values of the adjustment amounts are equal.
b. One unit can be matched and adjusted with a plurality of units in sequence until the adjustment capability is fully exerted.
c. When set G-Set G when the middle unit has no capacity of reducing power0The power of the machine set can be reduced; when set G+When the middle unit has no capacity of exerting force, the unit G in the set0A force may be applied.
d. When the generator adjustment cannot eliminate the line overload, the load shedding processing is carried out until the overload of the branch line is eliminated.
2.2 establishing an active safety correction model of the transmission line
Aiming at the power generator and load adjustment strategy introduced in 2.1 and based on the principle of equivalent reverse pairing, a mathematical formula is used for expression, and an active power safety correction model of the power transmission line is established.
(1) Objective function
The objective function targets minimum generator and load adjustment:
Figure GDA0001296865170000121
in the formula (7), the reaction mixture is,
Figure GDA0001296865170000122
in order to reduce the adjustment amount of the power generator,
Figure GDA0001296865170000123
for adding a regulation of the power generator, Δ PG,zIn order to adjust the amount of the balancing machine,
Figure GDA0001296865170000124
and M is a punishment coefficient which is a normal number far larger than 1, so that the output of the generator is preferentially adjusted when the overload of the line is eliminated, and then the load is cut.
(2) Constraint conditions
a. And the active power balance constraint of the system ensures that the generator adjustment quantity is equal to the load adjustment quantity.
Figure GDA0001296865170000125
b. And the overload restraint of the line is eliminated, and the overload phenomenon after adjustment is guaranteed to be eliminated.
Figure GDA0001296865170000126
In the formula (9), the reaction mixture is,
Figure GDA0001296865170000127
for reducing the generator pair overload line LmThe sensitivity of (a) to (b) is,
Figure GDA0001296865170000128
for adding power generator to overload line LmThe sensitivity of (a) to (b) is,
Figure GDA0001296865170000129
for load to overload line LmThe sensitivity of (2).
c. The branch circuit has active operation safety constraint to ensure the safe operation of other circuits after adjustment.
Figure GDA00012968651700001210
In the formula (10), t is a line number other than the overloaded line,
Figure GDA0001296865170000131
the initial active power, the minimum active power limit value and the maximum active power limit value of the line t are respectively.
d. And (5) restraining the output adjustment amount of the generator set.
Figure GDA0001296865170000132
In the formula (11), the reaction mixture is,
Figure GDA0001296865170000133
respectively the initial active power output and the active power output lower limit of the reduced-power generator,
Figure GDA0001296865170000134
the initial active output and the upper limit of the active output of the added-output generator are respectively.
e. And (5) restraining the load adjustment amount.
Figure GDA0001296865170000135
In the formula (12), the reaction mixture is,
Figure GDA0001296865170000136
is the initial active load.
3. And solving the transmission line active safety correction model based on a particle swarm intelligent algorithm.
The particle swarm optimization realizes the search of the global optimum point in a complex search space through competition and cooperation among particles, has the characteristics of easy realization and strong global search capability, and has unique advantages in the aspect of solving the problem of optimizing global optimization.
Suppose a line L in the gridmDuring overload, the generator to be adjusted has u platforms, and the load to be cut off has v. And solving the active safety correction model established in the step 2.2 based on a particle swarm intelligent algorithm to obtain the adjustment quantity of the generator and the load, and determining a control scheme. The specific steps are as follows, and the flow chart is shown in fig. 2.
1. Setting the maximum iteration times and initializing particle swarms: setting the maximum iteration number as MaxD, the particle swarm size as N, the particle dimension as u + v, and setting the initial position x of each particle according to the constraint conditions shown in the formulas (8), (11) and (12)iAnd an initial velocity viThe initialization is carried out such that,
xi＝[ΔPG,1,ΔPG,2,...,ΔPG,u,ΔPL,1,ΔPL,2,...ΔPL,v]，vi＝[vi1,vi2,...vi(u+v)]。
2. calculating initial fitness value of each particle, and determining individual extreme value p of each particleiAnd the global extremum p of the particle swarmgThe fitness function is shown in equation (13).
Figure GDA0001296865170000137
Delta P in the formula (13)G,iFor the adjustment of the generator to be adjusted, Δ PL,jAnd M is a penalty coefficient for the cutting amount of the load to be cut.
3. Updating the velocity v of each particleiAnd position xiGuarantee xiThe update is performed according to equation (14) while satisfying the constraints expressed by equations (8), (11), and (12).
vi＝w*vi+c1r1(pi-xi)+c2r2(pg-xi),xi＝xi+vi (14)
W in the formula (14) is the inertial weight, c1And c2Is a learning factor, r1And r2Is [0,1 ]]A uniform random number within the range.
4. Calculating each particle x after updating the positioniIs a fitness value F (x)i) And is in parallel with the individual extremum piFitness value F (p)i) By comparison, if F (x)i)＜F(pi) Then use xiIn place of pi(ii) a For each particle xiIts fitness value F (x)i) And a global extreme value F (p)g) By comparison, if F (x)i)＜F(pg) Then use xiIn place of pg。
5. For each particle xiAnd (4) carrying out overload elimination verification according to the formula (9) and carrying out line safety operation constraint verification according to the formula (10). If the verification is satisfied, the local optimum p is savediAnd global optimum pg(ii) a For unsatisfied schoolThe particles tested were particle refreshed as in step 1.
6. And (4) checking whether a search termination condition is met, namely whether the iteration number is greater than the maximum iteration number. If the termination condition is met, outputting the global optimum pgObtaining a control scheme; otherwise, returning to the step 3.
The specific embodiments described herein are merely illustrative of the spirit of the invention. Various modifications or additions may be made to the described embodiments or alternatives may be employed by those skilled in the art without departing from the spirit or ambit of the invention as defined in the appended claims.

Claims (4)

1. A power transmission line active safety correction method based on an extreme learning machine is characterized by comprising the following steps:
step 1, acquiring a generator sensitivity sample set and a load sensitivity sample set, and performing learning training on the acquired power grid sample set based on an extreme learning machine to obtain a mapping relation between the sensitivity of node injection power to a power transmission line and power grid operation data;
step 2, establishing an active safety correction model of the power transmission line based on the sensitivity analysis in the step 1, wherein the active safety correction model is based on a target function:
Figure FDA0002731259490000011
Figure FDA0002731259490000012
in order to reduce the adjustment amount of the power generator,
Figure FDA0002731259490000013
for adding a regulation of the power generator, Δ PG,zIn order to adjust the amount of the balancing machine,
Figure FDA0002731259490000014
the load shedding amount is the load shedding amount, M is a penalty coefficient and is a normal number far larger than 1, the output of the generator is preferentially adjusted when the line overload is eliminated, then the load shedding is carried out, the objective function takes the minimum of the generator and the load adjustment amount as the target, the sensitivity of the generator is positive, and the generator is classified as a generator set G with the output shedding amount-(ii) a The sensitivity of the generator is negative, and the generator is classified as an added power unit G+(ii) a The sensitivity of the generator is zero, and the generator is classified as a balance unit G0(ii) a For the load nodes, selecting the nodes with negative sensitivity as the load shedding set L-：
Step 3, collecting power grid data to be controlled, solving the power transmission line active safety correction model based on the active safety correction model in the step 2 to obtain the adjustment quantity of the generator and the load, and determining a control scheme;
the step 1 specifically comprises:
step 1.1, generating an extreme learning machine training sample set comprising a generator sensitivity sample set and a load sensitivity sample set, firstly determining the initial operation state and the network topology structure of a power grid in any power grid system, and recording the active power output P of a generator in the initial state of the power gridG 0Active power of load
Figure FDA0002731259490000015
And the active power of each transmission line
Figure FDA0002731259490000016
The method comprises the following steps:
set one, generator sensitivity sample set: randomly changing the active power output of any generator
Figure FDA0002731259490000017
The output is randomly changed to ensure that the obtained sample data can comprehensively reflect different operating states of the power grid; in order to ensure the active power balance of the system, the active power output of the balancing machine needs to be reversely adjusted while the active power output of the generator is changed
Figure FDA0002731259490000021
After the generated output is adjusted, the tidal current information of the power grid is updated, and the active power of each transmission line is recorded
Figure FDA0002731259490000022
And active variation
Figure FDA0002731259490000023
The sensitivity of the generator to the transmission line can be calculated from the recorded information
Figure FDA0002731259490000024
A set of sample data (x) is available1,t1) Wherein
Figure FDA0002731259490000025
This process is repeated N times, each time yielding a set (x)i,ti) Wherein
Figure FDA0002731259490000026
Finally, a generator sensitivity training sample set can be obtained
Figure FDA0002731259490000027
Wherein n is 3 and m is 1;
set two, load sensitivity sample set: randomly changing the active load delta P aiming at any load node except the balance nodeL 1While adjusting the output of the balancing machine in the same direction
Figure FDA0002731259490000028
Updating power flow information of the power grid and recording active power of each transmission line
Figure FDA0002731259490000029
And active variation
Figure FDA00027312594900000210
Calculating the sensitivity of the adjusted load node to the transmission line
Figure FDA00027312594900000211
A set of sample data (x) is available1,t1) Wherein
Figure FDA00027312594900000212
This process is repeated N times, one group (x) being obtained each timei,ti) Wherein
Figure FDA00027312594900000213
Finally, a load sensitivity training sample set can be obtained
Figure FDA00027312594900000214
Wherein n is 3 and m is 1;
step 1.2, training a sample set based on an extreme learning machine: sensitivity training sample set given based on step 1.1
Figure FDA00027312594900000215
And if the number of the nodes of the selected hidden layer is L, expressing the mathematical model of the extreme learning machine as follows:
Figure FDA00027312594900000216
in the formula (1), giAs a function of excitation wi＝[wi1,wi2,...,win]As weight vectors of the input layer, betai＝[βi1,βi2,...,βim]As output weight vectors, i.e. the concatenation weight vectors of the hidden layer to the output layer, biBias for the ith hidden layer;
the matrix expression is of the form:
Hβ＝Y(2)
in the formula (2), the reaction mixture is,
Figure FDA0002731259490000031
Figure FDA0002731259490000032
h is the hidden layer output of the neural network, and the ith column vector corresponds to the output of the ith hidden layer node;
when the excitation function w is knowniWhen infinite or micro, the network parameters do not need to be adjusted completely; thus randomly setting the input weight wiAnd bias of the hidden layer biThus, for a given set of training samples, the output matrix H of the hidden layer can be uniquely determined, with only the variable β present in the model;
constructing an error function of the extreme learning machine about beta:
Figure FDA0002731259490000033
at this time, the training process of the extreme learning machine on the sample set can be converted into a least square solution problem for solving an output weight matrix:
minβ||Hβ-T||(4)
solving equation (4) yields:
Figure FDA0002731259490000034
in the formula (5)
Figure FDA0002731259490000035
Moore-Penrose generalized inverse of matrix H;
finally, obtaining a mapping relation between the sensitivity and the power grid operation information:
Figure FDA0002731259490000036
2. the method for correcting the active safety of the power transmission line based on the extreme learning machine according to claim 1, wherein the step 2 is based on definition: line L of the grid in a certain operating statemAn overload of
Figure FDA0002731259490000037
Wherein P ismIs a line LmThe actual active power of the power plant,
Figure FDA0002731259490000041
is a line LmThe active transmission limit of (a) is that the line L needs to be put away in order to eliminate the overload and to keep a certain marginmActive power reduction of
Figure FDA0002731259490000042
Active output P of each generator of known power gridGAnd the active power of the load PLThus, the input vector x of equation (6) can be determinedG＝[ΔPc,Pm,PG]And xL＝[ΔPc,Pm,PL]Finding the generator and load pair line LmThe sensitivity of (c); specifically, the method includes an objective function having constraint conditions, where the constraint conditions include:
the constraint condition a is that the active power of the system is balanced and constrained, and the adjustment quantity of the generator is equal to the adjustment quantity of the load;
Figure FDA0002731259490000043
the constraint condition b is to eliminate the line overload constraint and ensure the elimination of the overload phenomenon after the adjustment;
Figure FDA0002731259490000044
in the formula (9), the reaction mixture is,
Figure FDA0002731259490000045
for reducing the generator pair overload line LmThe sensitivity of (a) to (b) is,
Figure FDA0002731259490000046
for adding power generator to overload line LmThe sensitivity of (a) to (b) is,
Figure FDA0002731259490000047
for load to overload line LmThe sensitivity of (c);
the constraint condition c is that the branch circuit has active operation safety constraint to ensure the safe operation of other adjusted circuits;
Figure FDA0002731259490000048
in the formula (10), t is a line number other than the overloaded line, Pt o,Pt min,Pt maxRespectively the initial active power, the minimum active power limit value and the maximum active power limit value of the line t;
the constraint condition d is that the output adjustment quantity of the generator set is constrained;
Figure FDA0002731259490000049
in the formula (11), the reaction mixture is,
Figure FDA00027312594900000410
respectively the initial active power output and the active power output lower limit of the reduced-power generator,
Figure FDA00027312594900000411
the initial active output and the upper limit of the active output of the added-output generator are respectively;
constraint condition e, load adjustment amount constraint;
Figure FDA00027312594900000412
in the formula (12), the reaction mixture is,
Figure FDA00027312594900000413
is the initial active load.
3. The transmission line active safety correction method based on the extreme learning machine as claimed in claim 2, characterized in that step 2 is based on defining a certain line L in the power gridmWhen the generator is overloaded, u generators to be adjusted are provided, and v loads to be cut off are provided; solving the established active power safety correction model based on a particle swarm intelligent algorithm to obtain the adjustment quantity of the generator and the load, and determining a control scheme; the method comprises the following specific steps:
step 4.1, setting the maximum iteration times and initializing particle swarm: setting the maximum iteration number as MaxD, the particle swarm size as N, the particle dimension as u + v, and the initial position x 'of each particle according to the constraint conditions shown in the formulas (8), (11) and (12)'iAnd an initial velocity viInitialization is carried out, x'i＝[ΔPG,1,ΔPG,2,...,ΔPG,u,ΔPL,1,ΔPL,2,...ΔPL,v]，vi＝[vi1,vi2,...vi(u+v)]；
Step 4.2, calculating the initial fitness value of each particle, and determining the individual extreme value p of the particleiAnd the global extremum p of the particle swarmgThe fitness function is shown as a formula (13);
Figure FDA0002731259490000051
delta P in the formula (13)G,iFor the adjustment of the generator to be adjusted, Δ PL,jThe amount of the load to be cut off is obtained, and M is a penalty coefficient;
step 4.3, update the velocity v of each particleiAnd position x'iProtection ofX'iUpdating according to equation (14) while satisfying the constraint conditions expressed by equations (8), (11) and (12);
vi＝w*vi+c1r1(pi-x′i)+c2r2(pg-x′i),x′i＝x′i+vi (14)
w in the formula (14) is the inertial weight, c1And c2Is a learning factor, r1And r2Is [0,1 ]]A uniform random number within a range;
step 4.4, calculating each particle x 'after updating the position'iIs a fitness value F (x)i) And is in parallel with the individual extremum piFitness value F (p)i) By comparison, if F (x)i)＜F(pi) X 'is then'iIn place of pi(ii) a X 'for each particle'iIts fitness value F (x)i) And a global extreme value F (p)g) By comparison, if F (x)i)＜F(pg) X 'is then'iIn place of pg；
Step 4.5, for each particle x'iPerforming overload elimination verification according to the formula (9), and performing line safety operation constraint verification according to the formula (10); if the verification is satisfied, the local optimum p is savediAnd global optimum pg(ii) a Updating the particles which do not meet the verification according to the step 1;
step 4.6, checking whether the search termination condition is met, namely whether the iteration times are greater than the maximum iteration times; if the termination condition is met, outputting the global optimum pgObtaining a control scheme; otherwise, the step 4.3 is returned.
4. The active safety correction method for power transmission line based on extreme learning machine as claimed in claim 1, wherein in step 2, the control scheme is generator and load adjustment strategy based on equal reverse pairing principle, the adjustment of generator output and load shedding is the common active safety correction measure, during the adjustment process, the power balance of the system needs to be ensured, the generator output is preferentially adjusted and then the load adjustment is performed,the aim of minimizing the generator and the load adjustment amount is achieved simultaneously; for overload lines L according to generator and loadmThe generator and the load are grouped: then, the generators and the loads in the set are respectively arranged in a descending order according to the absolute value; the specific adjustment strategy comprises the following steps:
step 5.1, adding force adjustment is carried out on the machine set with negative sensitivity according to the absolute value sequence, when force reduction is carried out on the machine set with negative sensitivity according to the absolute value sequence, force reduction of the other machine set is ensured when force is added to one machine set, and the absolute values of the adjustment quantities are equal;
step 5.2, one unit can be matched and adjusted with a plurality of units in sequence until the adjustment capability is fully exerted;
step 5.3, when the output reducing unit G-Balancing the unit G when the middle unit has no capacity of reducing power0The power of the machine set can be reduced; when adding force unit G+When the middle unit has no capacity of exerting force, the balancing unit G in the set0Force can be added;
and 5.4, when the generator cannot eliminate the overload of the line by adjustment, performing load shedding processing until the overload of the branch line is eliminated.
CN201710151119.2A 2017-03-14 2017-03-14 Extreme learning machine-based transmission network line active safety correction method Expired - Fee Related CN107069708B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710151119.2A CN107069708B (en) 2017-03-14 2017-03-14 Extreme learning machine-based transmission network line active safety correction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710151119.2A CN107069708B (en) 2017-03-14 2017-03-14 Extreme learning machine-based transmission network line active safety correction method

Publications (2)

Publication Number Publication Date
CN107069708A CN107069708A (en) 2017-08-18
CN107069708B true CN107069708B (en) 2021-01-15

Family

ID=59617849

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710151119.2A Expired - Fee Related CN107069708B (en) 2017-03-14 2017-03-14 Extreme learning machine-based transmission network line active safety correction method

Country Status (1)

Country Link
CN (1) CN107069708B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107967497A (en) * 2017-08-28 2018-04-27 苏州明逸智库信息科技有限公司 Manuscripted Characters Identification Method based on convolutional neural networks and extreme learning machine
CN108879665B (en) * 2018-07-03 2021-07-27 河海大学 Power system safety correction optimization method aiming at minimum number of adjusting equipment
CN110676852B (en) * 2019-08-26 2020-11-10 重庆大学 Improved extreme learning machine rapid probability load flow calculation method considering load flow characteristics
CN114243690A (en) * 2021-12-15 2022-03-25 国网河北省电力有限公司 Power grid active safety correction method and device, electronic equipment and storage medium

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104821605A (en) * 2015-04-13 2015-08-05 国家电网公司 Active safety correction method based on improved particle swarm optimization algorithm
CN106356856A (en) * 2016-09-18 2017-01-25 国电南瑞科技股份有限公司 Safety correction calculating method based on regional load control

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104821605A (en) * 2015-04-13 2015-08-05 国家电网公司 Active safety correction method based on improved particle swarm optimization algorithm
CN106356856A (en) * 2016-09-18 2017-01-25 国电南瑞科技股份有限公司 Safety correction calculating method based on regional load control

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
基于广域信息量的潮流转移问题研究;姜臻;《CNKI中国优秀硕士学位论文全文库》;20120715;第I，23-33页 *
基于粒子群优化的核极限学习机模型的风电功率区间预测方法;杨锡运 等;《中国电机工程学报》;20150930;第35卷;第146-153页 *
姜臻.基于广域信息量的潮流转移问题研究.《CNKI中国优秀硕士学位论文全文库》.2012, *

Also Published As

Publication number Publication date
CN107069708A (en) 2017-08-18

Similar Documents

Publication Publication Date Title
CN107069708B (en) Extreme learning machine-based transmission network line active safety correction method
CN108694467B (en) Method and system for predicting line loss rate of power distribution network
Sivalingam et al. A modified whale optimization algorithm-based adaptive fuzzy logic PID controller for load frequency control of autonomous power generation systems
Raglend et al. Comparison of AI techniques to solve combined economic emission dispatch problem with line flow constraints
CN109861202B (en) Dynamic optimization scheduling method and system for flexible interconnected power distribution network
CN110932281B (en) Multi-section cooperative correction method and system based on quasi-steady-state sensitivity of power grid
CN114362196B (en) Multi-time-scale active power distribution network voltage control method
CN105048479B (en) A kind of idle packet adjusting method of photovoltaic plant
CN109638815B (en) Method for determining safety and stability prevention control strategy of medium-and-long-term voltage of power system
CN110738344A (en) Distributed reactive power optimization method and device for load prediction of power system
CN106684885B (en) Wind turbine generator system power distribution network reactive power optimization method based on multi-scene analysis
CN113300380B (en) Load curve segmentation-based power distribution network reactive power optimization compensation method
CN102129259A (en) Neural network proportion integration (PI)-based intelligent temperature control system and method for sand dust environment test wind tunnel
CN108551175B (en) Energy storage capacity configuration method for power distribution network
CN107436971A (en) Suitable for the improvement Latin Hypercube Sampling method of non-positive definite form correlation control
CN107516892A (en) The method that the quality of power supply is improved based on processing active optimization constraints
CN115085202A (en) Power grid multi-region intelligent power collaborative optimization method, device, equipment and medium
CN111030089B (en) Method and system for optimizing PSS (Power System stabilizer) parameters based on moth fire suppression optimization algorithm
CN108830451A (en) A kind of the convergence potential evaluation method and system of user side distributed energy storage
Sun et al. Global state estimation for whole transmission and distribution networks
CN113872213B (en) Autonomous optimization control method and device for power distribution network voltage
CN111900767B (en) Method and system for controlling multi-section active power flow of power system
CN110661264B (en) Safety constraint optimal power flow calculation method based on particle swarm algorithm with inertial weight
CN109193662A (en) The most dangerous load margin calculation method and system that meter and imbalance power are shared
CN105576653A (en) 220kV district power grid power supply capacity optimization method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210115