CN106356899A - Layered optimization model applicable to large-scale wind power control process - Google Patents

Abstract

The invention provides a layered optimization model applicable to a large-scale wind power control process. The layered optimization model comprises a generator unit type division unit, a rolling optimized planning layer, a real-time base point tracking layer and a feedback correction layer. An economic optimized plan output value of the optimized planning layer serves as an adjusting base point of a base point tracking layer scheduling process, and a scheduling instruction after adjustment of the base point tracking layer serves as a modification base point of the feedback correction layer; the feedback correction layer issues a modified control instruction to a unit to form a forward planning issuing data process. Every layer is responsible for modifying deviation of a previous layer, and the deviation left by every layer is modified by a subsequent layer, so that the concept of multi-stage coordination and stage-by-stage refinement is embodied, and the influence of inaccuracy in wind power forecasting on the control process can be weakened effectively. Actual operation results show that the layered optimization model is used for repeated rolling optimization and update of control policies, so that system processing disturbance and uncertainties can be enhanced and control robustness is improved.

Description

It is applied to the bilevel optimization model of large-scale wind power control process

Technical field

The invention belongs to power scheduling technical field is and in particular to a kind of layering being applied to large-scale wind power control process Optimized model.

Background technology

Economic Dispatch refers to meeting network security, under generation load equilibrium condition, with most economical fortune Row cost realizes the reasonable distribution of generation load between unit, and ensures a kind of dispatching method to user's reliable power supply.By optimization The difference of period, power system optimal dispatch problem can be divided into two aspects: static optimization scheduling and dynamically optimized scheduling.

Static optimization scheduling refers to: the economic load Optimizing Allocation to power system single run time section. Static optimization scheduling is algorithmically broadly divided into two classes: classic economic dispatch based on equal consumed energy ratio and with optimum tide Safety economy scheduling based on stream.

Because power system is the dynamical system in a continuous service, change when larger workload demand occurs in system When, limited by electromotor adjustment capability, the getted over ability between each static scheduling result cannot ensure.Accordingly, it would be desirable to grind Study carefully the continuous feasibility problems of economic load dispatching result, i.e. dynamic economic dispatch problem.

Conventional open-loop dynamic scheduling mode carried out a suboptimization and will solve sequence in the optimization starting stage to whole optimization cycle Row all issue execution, due to load prediction precision higher so that this scheduling method applies effect in traditional power system Fruit disclosure satisfy that requirement substantially.However, after large-scale wind power accesses, wind-powered electricity generation precision of prediction is far below the prediction essence of traditional load Degree, and the prolongation with the time of optimization, wind-powered electricity generation forecast error significantly increases, this feature being difficult to Accurate Prediction of wind-powered electricity generation make according to The result of conventional open-loop dynamic scheduling mode of Lai Yu wind-powered electricity generation prediction a few days ago is larger with actual electric network demand disruption, needs badly to this Scheduling method improves.

Content of the invention

The defect existing for prior art, the present invention provides a kind of layering being applied to large-scale wind power control process excellent Change model, can effectively solving the problems referred to above.

The technical solution used in the present invention is as follows:

The present invention provides a kind of bilevel optimization model being applied to large-scale wind power control process, including electromotor unit class Type division unit, rolling optimal plan layer, real-time basic point tracking layer and feedback compensation layer；

Electromotor unit Type division unit, for incorporating the whole network unit for roller press set type first into；Then, according to Electromotor unit responding ability, is in turn divided into manually adjusting unit, Real-Time Scheduling unit, real-time control unit by rolling unit And agc unit；Wherein, the described unit that manually adjusts refers to: does not have self-checking device, needs to rely on power plant operator manual Adjust out force tracking, with the unit of the rolling optimal plan curve for interval for the 15min；Described Real-Time Scheduling unit refers to: has certainly Dynamic adjusting means, can from motion tracking with 5min for interval Real-Time Scheduling correction Planning Directive unit；Described real-time control Unit refers to: there are the agc unit of very fast regulating power and the Wind turbines with self-regulation ability, can be from motion tracking 1min The real-time control at interval is exerted oneself adjust instruction；Described agc unit refers to: the holding of responsible system frequency and dominant eigenvalues plan；

Described rolling optimal plan layer is used for: described rolling optimal plan layer adopts Coarse grain model, to plan to be a few days ago Basis, the load prediction results according to electrical network Extended short-term model and Extended short-term wind-powered electricity generation predict the outcome rolling amendment unit a few days ago Generation schedule goes out activity of force so that system generating gross capability power is gradually approached with actual power demand, obtains economic optimum Unit generation plan is exerted oneself, and the unit generation plan of described economic optimum is exerted oneself act on described in manually adjust unit；

Described real-time basic point tracking layer is used for: described real-time basic point tracking layer adopts fine granularity rolling optimization model, with institute The unit generation plan stating economic optimum is exerted oneself as basic point power, according to electrical network ultra-short term result and ultra-short term wind Electricity predicts the outcome adjustment unit output, generates the real-time tune that the system mode in the plan period is carried out with minimum unit output adjustment Degree revises Planning Directive, and described Real-Time Scheduling correction Planning Directive is acted on described Real-Time Scheduling unit, and then eliminates machine The amount of unbalance that the change at random of group execution economic optimum calculated unbalanced power amount and wind-powered electricity generation load causes；

Described feedback compensation layer includes agc unit generation balancing the load correction unit and real-time control unit is exerted oneself and corrected list Unit；

Described real-time control unit wind power output correction unit is used for: the real-time tune being given with described real-time basic point tracking layer Degree revises Planning Directive as controlling basic point, and the stochastic prediction error that advanced prediction link is produced is revised in real time, generates Real-time control for being controlled to real-time control unit is exerted oneself adjust instruction, and described real-time control is exerted oneself adjust instruction It is issued to described real-time control unit, realize the control to real-time control unit；

Described agc unit generation balancing the load correction unit is used for: the real-time tune being given with described real-time basic point tracking layer Degree revises Planning Directive as controlling basic point, and the stochastic prediction error that advanced prediction link is produced is revised in real time, generates For the agc unit output adjust instruction that agc unit is controlled, and described agc unit output adjust instruction is issued to Described agc unit, realizes the control to agc unit.

Preferably, described rolling optimal plan layer includes: Extended short-term wind-powered electricity generation predicting unit, Extended short-term load prediction list Unit and economic optimum model；

Described Extended short-term wind-powered electricity generation predicting unit is used for: Extended short-term wind-powered electricity generation is predicted；

Described Extended short-term load estimation unit is used for: Extended short-term load is predicted；

Described economic optimum model is used for: with described Extended short-term wind-powered electricity generation predicting unit and described Extended short-term load prediction Unit predict the outcome for input, based on rolling optimization control strategy rolling amendment, unit generation plan goes out activity of force a few days ago；

Wherein, described rolling optimization control strategy is: using the economic optimum scheduling model abandoned based on minimum on the basis of wind, See formula (1):

$f_1\left(p_{it}\right)=min\underset{t=t_0+1}{\overset{t_0+t_h}{\σ}}\left(\underset{i=1}{\overset{n}{\σ}}\right(a_i{p_{it}}^2+b_i{p}_{it}+c_i)+\underset{j\&element;g_{wind}}{\σ}{\mathrm{\λ}}_j(p_{jt}^f-p_{jt}^w\left)\right)---\left(1\right)$

Wherein, f₁(p_it) it is scheduling model object function；T is the optimization time；t₀For optimizing start periods；t_hFor optimum meter Draw layer and optimize Period Length；N is conventional power unit number；a_i、b_i、c_iCoal consumption coefficient for conventional power unit i；p_itFor conventional power unit i The active plan of exerting oneself of t period；g_windFor Wind turbines number；λ_jFor abandoning wind cost factor；Pre- for Wind turbines Extended short-term Measure power；For Wind turbines j the t period the active plan of exerting oneself.

Preferably, described real-time basic point tracking layer includes: ultra-short term wind-powered electricity generation predicting unit, ultra-short term unit with And basic point tracing model；

Described ultra-short term wind-powered electricity generation predicting unit is used for: ultra-short term wind-powered electricity generation is predicted；

Described ultra-short term unit is used for: super short period load is predicted；

Described basic point tracing model is used for: with described ultra-short term wind-powered electricity generation predicting unit and described ultra-short term unit Predict the outcome as input, minimum adjustment is carried out based on basic point pursive strategy to the system mode in the plan period；

Described basic point pursive strategy is: using the scheduling model of formula (2):

$f_1\left(p_{it}\right)=min\underset{t=t_0+1}{\overset{t_0+t_l}{\σ}}\left(\underset{i=1}{\overset{n}{\σ}}\right(a_i{\left(p_{it}^{roll}+{\mathrm{\δp}}_{it}\right)}^2+b_i\left(p_{it}^{roll}+{\mathrm{\δp}}_{it}\right)c_i)+\underset{j\&element;g_{wind}}{\σ}{\mathrm{\λ}}_j(p_{jt}^f-p_{jt}^w\left)\right)---\left(2\right)$

f₁(p_it) it is scheduling model object function；T is the optimization time；t₀For optimizing start periods；t_lExcellent for basic point tracking layer Change Period Length；N is conventional power unit number；a_i、b_i、c_iCoal consumption coefficient for conventional power unit i；p_itFor conventional power unit i in the t period The active plan of exerting oneself；For i-th unit the t period rolling optimal plan layer plan；δp_itExist for i-th unit The basic point of t period follows the trail of plan adjustment amount, for controlling output；g_windFor Wind turbines number；λ_jFor abandoning wind cost factor； Exert oneself for the prediction of Wind turbines Extended short-term；For Wind turbines j the t period the active plan of exerting oneself.

Preferably, described rolling optimal plan layer and described real-time basic point tracking layer are all using the rolling based on mpc principle Optimization Solution strategy；Wherein, described rolling optimal plan layer starts once every 15min, provides 4 hours futures Period Length Planning Directive；The every 5min of described real-time basic point tracking layer starts once, and the plan providing following 15min minute length every time refers to Order.

The bilevel optimization model being applied to large-scale wind power control process that the present invention provides has the advantage that

The bilevel optimization model being applied to large-scale wind power control process that the present invention provides, with the economy of optimal plan layer Optimal plan output valve, as the adjustment basic point of basic point tracking layer scheduling process, is made with the dispatch command after the adjustment of basic point tracking layer Correction basic point for feedback compensation layer.Finally, revised control instruction is handed down to unit by feedback compensation layer, before constituting to Plan issues data flow.Each layer is responsible for revising the deviation of last layer, and the deviation left to be revised by next layer, embodies one The thought of kind of " multilevel coordination, step by step refine ", can effectively reduce the impact to control process of the inaccuracy of wind-powered electricity generation prediction.Actual Operation result shows, carries out repeatedly the control strategy of rolling optimization renewal using bilevel optimization model of the present invention, can strengthen system System processes disturbance and probabilistic ability, improves the robustness controlling.

Brief description

The structural representation of the bilevel optimization model being applied to large-scale wind power control process that Fig. 1 provides for the present invention.

Rolling optimal plan layer that Fig. 2 provides for the present invention and the starting time of real-time basic point tracking layer and optimize step-length Figure.

Specific embodiment

In order that technical problem solved by the invention, technical scheme and beneficial effect become more apparent, below in conjunction with Drawings and Examples, the present invention will be described in further detail.It should be appreciated that specific embodiment described herein only in order to Explain the present invention, be not intended to limit the present invention.

Power system in the Dispatching Control System after design large-scale wind power access it is necessary first to after to introducing wind-powered electricity generation Operation characteristic is analyzed, and designs corresponding control structure framework for operation characteristic, is further directed to the need of control process Seek the corresponding decision-making mechanism of design.

Based on this, the present invention is analyzed to the concrete feature of the power system after wind power integration first, for control Probabilistic feature proposes using model predictive control method reply process by force；Processing greatly for model predictive control method The Global Optimality of control process existing during scale complexity On-line Control problem and the contradiction of real-time anti-interference, will divide Solution-control method for coordinating is dissolved in Model Predictive Control structure, defines the layering being applied to large-scale wind power control process Model predictive control method, explain multilevel coordination active power dispatch pattern control mechanism and principle of grading the problems such as.

In order to realize to large-scale wind power reliable scheduling controlling purpose in real time online, the present invention passing decomposition-coordination Rank control structure is dissolved in model predictive control method, and rolling optimization layer is further layered by " change granularity " method, will Control process is decomposed into the rolling optimal plan layer based on Coarse grain model and the real-time basic point tracking layer based on finely granular access control Two-layer, the former stresses economy, and the latter stresses safety, forms the Hierarchy that internal layer-outer layer is nested.Specifically , roll optimal plan layer and real-time basic point tracking layer every layer of inside all comprise complete controls in advance process model prediction and Rolling optimization link, execute single mpc optimal control respectively for different control targes (economy, safety) it is ensured that The realization of every layer of respective control targe.Pass through the coordination optimization of the economy between different layers and safety further, realize whole Body control effect.This hierarchy optimization control problem by execute some sub- optimization process realize, upper strata optimization process defeated Go out and pass to lower floor's optimization process in the form of Plan rescheduling basic point, so that it is full to control requirement to obtain successively by rank height Foot.

Roll optimal plan layer and adopt Coarse grain model, based on the Extended short-term prediction knot to wind power output and workload demand Fruit is optimized to the process of longer period of time, and be given is systematic economy optimal plan ideally, and this is intended to be base The Plan rescheduling basic point of point tracking layer is executing.

Because the reasons such as deviation are followed the trail of in the random fluctuation of wind-powered electricity generation and plan, the optimal control target of system will constantly become Change.Basic point tracking layer adopts fine granularity rolling optimization model, and it is right to be predicted the outcome based on the ultra-short term of wind power output and workload demand Plan is done and is refined further, to ensure the anti-interference of control process.Meanwhile, for ensureing the economy of system, basic point tracking layer The generator output optimal plan that should be given immediately following last layer of adjustment, when disturbance occurs, take the control model of deviation minimum.

Wind-powered electricity generation prediction is the basis realizing hierarchical mode PREDICTIVE CONTROL.Under current wind-powered electricity generation precision of prediction, advanced mould Type PREDICTIVE CONTROL process cannot avoid the deviation existing between wind power output plan and actual power ability, causes certain to abandon wind Economic loss initiating system safety problem.The present invention proposes to introduce by the feedback compensation process in mpc link lag time Wind-powered electricity generation is exerted oneself feedback control link (abbreviation real-time control link) in real time, controls using based on standby follow the trail of etc. of wind-fire joint debugging or backspin Strategy, is revised in real time to this plan deviation.Form the double-deck control of the feedforward compensation-feedback compensation to wind-powered electricity generation control process Structure processed, constitutes closed loop, realizes the lead compensation to wind-powered electricity generation forecast error and delayed correction, reaches and overcome wind-powered electricity generation probabilistic Impact, the purpose of wind-powered electricity generation of at utmost dissolving.

The optimal plan layer main purpose of the superiors is the economy of guarantee system, based on planning, according to electricity a few days ago Net Extended short-term load prediction and Extended short-term wind-powered electricity generation predict the outcome rolling amendment a few days ago unit generation plan so that system generates electricity Gross capability is gradually approached with actual power demand, reduces the uncertainty a few days ago planned, and alleviates basic point tracking layer and feedback The pressure of correcting layer unit.The basic point tracking layer of the second layer is intended to be basic point power with the economic optimum that upper strata provides, according to Electrical network ultra-short term and ultra-short term wind-powered electricity generation predict the outcome and adjust related unit output, to the system mode in the plan period The change at random carrying out minimum adjustment to eliminate unbalanced power amount during unit executes optimal plan and wind-powered electricity generation load is made The amount of unbalance becoming is it is ensured that the safety of system operation.The control that the feedback compensation layer of third layer is given with basic point tracking layer refers to Order is revised in real time as adjustment basic point, the stochastic prediction error that advanced prediction link is produced, to overcome system not know Property impact, obtain closed loop stability.

With reference to Fig. 1, for being applied to the structural representation of the bilevel optimization model of large-scale wind power control process, main inclusion Electromotor unit Type division unit, rolling optimal plan layer, real-time basic point tracking layer and feedback compensation layer.Below to this four Part is discussed in detail respectively:

(1) electromotor unit Type division unit

The time scale that optimal plan, basic point are followed the trail of, feedback compensation three controls requires different, to unit allocation performance Require also different, need the performance classification according to available unit, bring in different key-courses.

In actual application, in system, the difference of machine set type, model, aging conditions pressed by electromotor, to plan Ability of tracking there is also different restrictions, and the present invention is divided to machine set type by unit responding ability.First, for ensureing system Economy, needs each unit of the whole network is all provided with corresponding rolling optimal plan.Therefore, the whole network unit is firstly the need of incorporating into For roller press set type.Further, by unit responding ability, it is in turn divided into manually adjusting unit, adjusts in real time to rolling unit Degree unit, real-time control unit and agc unit.

Manually adjust unit and refer to there is no self-checking device, need rely on power plant operator manually adjust out force tracking with 15min is the unit of the rolling optimal plan curve at interval.

Real-Time Scheduling unit refers to self-checking device, can be from motion tracking with 5min for the Real-Time Scheduling at interval Revise the unit of Planning Directive.

Real-time control unit belongs to the agc unit with very fast regulating power and the wind turbine with self-regulation ability Group, can exert oneself adjust instruction from the real-time control at motion tracking 1min interval.

Agc unit is then responsible for the holding of system frequency and dominant eigenvalues plan.

The basic principle of computer-assisted classification is: it is the slowest that unit the slowest of regulating the speed is limited tracking adjustment speed by adjustment capability Rolling optimal plan layer plan, to ensure overall economy as target；Slightly fast unit of regulating the speed is responsible for revising optimum control Plan deviation processed, changes in balance cycle slower system interference, exerted oneself with the economic optimum on foundation for security and be adjusted to target, with The real-time correction plan of track basic point tracking layer；The agc unit regulating the speed the fastest executes the feedback compensation instruction of feedback compensation layer, Main consideration safety, responsible balance high frequency interference, on the premise of safety ensure that, pass through real-time control ring further Section increases wind-powered electricity generation, improves system economy.

In practical application, electromotor unit Type division unit is also referred to as unit role's distribute module, in real time to complete The role of net unit is allocated, and unit role includes planning unit, rolling planning unit, in real time plan unit and agc a few days ago Unit；Plan machine as manually adjusts unit a few days ago；Rolling planning unit is Real-Time Scheduling unit；Plan unit is in real time Real-time control unit.

Described unit role's distribute module adopts following methods, determines the role of each unit of the whole network every Preset Time:

Step 1, according to the departure degree of electrical network real-time frequency and normal setpoint frequency, by from irrelevance from light suitable to weight Sequence, is in turn divided into 4 control zones, is respectively as follows: dead band, normal area, auxiliary region and coordinated regions；

Specifically, it is effectively realized the Real-time Balancing of power supply and demand for ensureing agc to smooth, stablize, it is to avoid reducing ace During the situation of toning or less stress occurs, need to divide control zone (controlzone).Control zone is used for representing The order of severity of ace, is also referred to as command area (command including dead band (dead band zone), normal area (normal zone) Zone), auxiliary region (assist zone) is also referred to as and allows area (permissive zone), coordinated regions (cooperation zone) Also referred to as urgent area (emergency zone).

Step 2, determines deadband boundaries value ace respectively by below equation_d, normal area boundary value ace_n, auxiliary region boundary value ace_aWith coordinated regions boundary value ace_e:

$\left\{\begin{array}{c}ac{e}_d=20b_i{\mathrm{\ϵ}}_1\\ {\mathrm{ace}}_n=2l_{10}\\ {\mathrm{ace}}_a=3l_{10}\\ {\mathrm{ace}}_e=0.8loss\end{array}\right.$

Wherein: b_i: the frequency bias coefficient that control area sets, unit mw/0.1hz, take positive sign；

ε₁: the root mean square control targe to annual one minute frequency averaging deviation for the interconnected network；

l₁₀: the control limit of the absolute value of ten minutes ace meansigma methodss；

Loss: unstability power；

According to the occurrence of ace, the coordination control strategy of each agc unit is as shown in the table:

Agc unit cooperative control scheme list

In table: " do not do and control " expression does not carry out any regulation；" bias adjustment " expression need to leave basic point value, participates in Ace is adjusted, and promotes ace to reduce." basic point is close " expression directly carries out basic point regulation, does not consider the impact to ace；" condition is returned Return " represent carry out basic point adjust when, consider whether ace is impacted, if approach to basic point value can lead to ace increase, Remain stationary as；If approaching and ace will be promoted to reduce to basic point value, changed.

Basic point value be set with various ways, the basic point value that the present invention adopts is plan basic point, and that is, basic point value is a company Continuous curve.

Step 3, according to unit performance, to the whole network, all units are ranked up, according to unit performance order from high to low, Unit is designated as respectively: unit 1, unit 2 ..., obtain sequencing table；

Meet the minimum m value of following condition in selected and sorted table, obtain numbering and be followed successively by: unit 1, unit 2 ... unit m M agc unit；Agc unit is adjusted according to the control strategy of bias adjustment:

${\mathrm{cap}}_{agc}=\underset{i=1}{\overset{m}{\σ}}(p_{i,max}-\frac{p_{i,max}+p_{i,min}}{2})\&greaterequal;{\mathrm{ace}}_n$

Wherein: cap_agc: agc bias adjustment capacity；

p_i,max: the EIAJ value of unit i；

p_i,min: the minimum load value of unit i；

Step 4, selects numbering to be followed successively by: unit m+1, the n platform unit of unit m+2 ... unit n plan unit as real-time, Plan unit is controlled according to following the tracks of real-time planning strategy in real time, and wherein, n is the minima meeting following constraint:

${\mathrm{cap}}_{agc}+\underset{i=1}{\overset{n}{\σ}}(p_{i,max}-p_i)\&greaterequal;{\mathrm{ace}}_a$

Due to p_iTo exert oneself be continually varying value, therefore, n is also the numerical value of a dynamic change；

Step 5, selects numbering to be followed successively by unit n+1, the k platform unit of unit n+2 ... unit n+k as rolling planning machine Group, rolling planning unit is controlled according to following the tracks of rolling planning strategy, and wherein, k is the minima meeting following constraint:

${\mathrm{cap}}_{agc}+\underset{i=1}{\overset{n}{\σ}}(p_{i,max}-p_i)+\underset{i=n+1}{\overset{n+k}{\σ}}(p_{i,max}-p_i)\&greaterequal;{\mathrm{ace}}_e$

Step 6, the remaining unit in sequencing table is to plan unit a few days ago, and plan unit is planned according to tracking a few days ago a few days ago Control strategy is adjusted.

The distribution of above unit role can carry out statistics by Automatic Program and select, for avoiding frequently changing role, can One subseries is carried out with δ t at set intervals.For the unit by bias adjustment, according to the requirement of energy-saving distribution, with coal consumption The size proportional assignment imbalance power of coefficient.

In addition, unit role's distribute module can also determine unit role using following methods:

Step 1, counts ace according to historical data first and falls in the probability of each control zone, might as well make in dead band Probability be pro1, the probability falling within normal area is pro2, and the probability falling in coordinated regions is pro3, fall in urgent area and Probability outside urgent area is pro4, then have:

$\underset{j=1}{\overset{4}{\σ}}{\mathrm{pro}}_j=1$

Step 2, the agc unit participating in adjusting ace must also meet the requirement of total spinning reserve in each period, might as well If the agc unit (1≤n≤n) that in the n platform unit in system, total n Radix codonopsis pilosulae is controlled with ace, and be made up of this n platform unit Set is designated as seta；The lower limit of spinning reserve should give according to the practical operation situation of electrical network, might as well be set to srt, its value is necessary More than acee (namely 0.8loss)；

Belonging to the n platform conventional rack of set seta, to have 4 kinds of roles available:

Unit role and control model

Variable roleid represents unit role, value be 1,2,3,4, corresponding control model be respectively as follows: bias adjustment, with Track plans in real time, follow the tracks of rolling planning and follow the tracks of and plan a few days ago；Its corresponding unit role is respectively as follows: agc unit, real-time machine Group, rolling planning unit and a few days ago plan unit；

The 4 vector role (1) mapping therewith, role (2), role (3), role can be constructed according to roleid vector (4), for preserving the agc unit subscript that role is respectively 1,2,3,4；

The object function being thus building up to following optimization problem is:

Wherein: p_it: unit i goes out force value in t；

a_i: the secondary term coefficient of non-linear relation；

b_i: the Monomial coefficient of non-linear relation；

c_i: the constant term of non-linear relation；

D: worth correction factor of currently exerting oneself；

Above-mentioned object function ensures that ace in a day for all agc units belonging to role (j) adjusts the expectation of total cost Minimum.

And after unit role distribution, need to ensure ace to fall in regional have enough agc to adjust nargin, therefore, Create following constraint:

$s.t.\left\{\begin{array}{c}\underset{i\&element;role\left(1\right)}{\mathrm{\σ}}{s}_{it}\&greaterequal;{\mathrm{ace}}_d\\ \underset{i\&element;role\left(1\right)}{\mathrm{\σ}}{s}_{it}+\underset{i\&element;role\left(2\right)}{\σ}{s}_{it}\&greaterequal;{\mathrm{ace}}_n\\ \underset{i\&element;role\left(1\right)}{\mathrm{\σ}}{s}_{it}+\underset{i\&element;role\left(2\right)}{\σ}{s}_{it}+\underset{i\&element;role\left(3\right)}{\mathrm{\σ}}{s}_{it}\&greaterequal;{\mathrm{ace}}_a\\ \underset{i\&element;role\left(1\right)}{\mathrm{\σ}}{s}_{it}+\underset{i\&element;role\left(2\right)}{\σ}{s}_{it}+\underset{i\&element;role\left(3\right)}{\mathrm{\σ}}{s}_{it}+\underset{i\&element;role\left(4\right)}{\mathrm{\σ}}{s}_{it}\&greaterequal;{\mathrm{ace}}_e\\ t=1,2,...,t\end{array}\right.$

s_itFor i-th unit t spinning reserve；

Under above-mentioned constraints, object function is solved, the agc role finally being determined.

(2) roll optimal plan layer

Roll optimal plan layer to be used for: described rolling optimal plan layer adopts Coarse grain model, based on planning a few days ago, Load prediction results according to electrical network Extended short-term model and Extended short-term wind-powered electricity generation predict the outcome rolling amendment unit generation a few days ago It is planned out activity of force so that system generating gross capability power is gradually approached with actual power demand, obtain the unit of economic optimum Generation schedule is exerted oneself, and the unit generation plan of described economic optimum is exerted oneself act on described in manually adjust unit.

Wherein, roll optimal plan layer include: Extended short-term wind-powered electricity generation predicting unit, Extended short-term load estimation unit and Economic optimum model；

Described Extended short-term wind-powered electricity generation predicting unit is used for: Extended short-term wind-powered electricity generation is predicted；

Described Extended short-term load estimation unit is used for: Extended short-term load is predicted；

Described economic optimum model is used for: with described Extended short-term wind-powered electricity generation predicting unit and described Extended short-term load prediction Unit predict the outcome for input, based on rolling optimization control strategy rolling amendment, unit generation plan goes out activity of force a few days ago；

Wherein, the major function of optimal plan layer is to ensure that the optimality of plan, and therefore, optimal plan layer is using based on The little economic optimum scheduling model abandoned on the basis of wind, is shown in formula (1):

$f_1\left(p_{it}\right)=min\underset{t=t_0+1}{\overset{t_0+t_h}{\σ}}\left(\underset{i=1}{\overset{n}{\σ}}\right(a_i{p_{it}}^2+b_i{p}_{it}+c_i)+\underset{j\&element;g_{wind}}{\σ}{\mathrm{\λ}}_j(p_{jt}^f-p_{jt}^w\left)\right)---\left(1\right)$

Wherein, f₁(p_it) it is scheduling model object function；T is the optimization time；t₀For optimizing start periods；t_hFor optimum meter Draw layer and optimize Period Length；N is conventional power unit number；a_i、b_i、c_iCoal consumption coefficient for conventional power unit i；p_itFor conventional power unit i The active plan of exerting oneself of t period；g_windFor Wind turbines number；λ_jFor abandoning wind cost factor；Pre- for Wind turbines Extended short-term Measure power；For Wind turbines j the t period the active plan of exerting oneself.

Require higher control process generally, for control effect, larger optimization cycle should be taken, to expand reflection The quantity of information of process future trends, covers the main dynamic response of controlled device, such as load climbing process is it is ensured that optimize knot The seriality of fruit, strengthens the ability overcoming various uncertain and complicated change impact, so that closed loop system is had desired steady Qualitative.Meanwhile, the cost-effectiveness requirement of active power dispatch is also required to the cumulative process of a long period has longer adopting it is desirable to control The sample cycle.Therefore, the optimization cycle of optimal plan layer is taken as the maximum cycle 4 hours of current wind energy turbine set wind-powered electricity generation ultra-short term prediction, Sampling step length is taken as 15min.

(3) real-time basic point tracking layer

Described real-time basic point tracking layer is used for: described real-time basic point tracking layer adopts fine granularity rolling optimization model, with institute The unit generation plan stating economic optimum is exerted oneself as basic point power, according to electrical network ultra-short term result and ultra-short term wind Electricity predicts the outcome adjustment unit output, generates the real-time tune that the system mode in the plan period is carried out with minimum unit output adjustment Degree revises Planning Directive, and described Real-Time Scheduling correction Planning Directive is acted on described Real-Time Scheduling unit, and then eliminates machine The amount of unbalance that the change at random of group execution economic optimum calculated unbalanced power amount and wind-powered electricity generation load causes.

Basic point tracking layer includes in real time: ultra-short term wind-powered electricity generation predicting unit, ultra-short term unit and basic point are followed the trail of Model；

Described ultra-short term wind-powered electricity generation predicting unit is used for: ultra-short term wind-powered electricity generation is predicted；

Described ultra-short term unit is used for: super short period load is predicted；

Described basic point tracing model is used for: with described ultra-short term wind-powered electricity generation predicting unit and described ultra-short term unit Predict the outcome as input, minimum adjustment is carried out based on basic point pursive strategy to the system mode in the plan period；

Described basic point pursive strategy is: using the scheduling model of formula (2):

$f_1\left(p_{it}\right)=min\underset{t=t_0+1}{\overset{t_0+t_l}{\σ}}\left(\underset{i=1}{\overset{n}{\σ}}\right(a_i{\left(p_{it}^{roll}+{\mathrm{\δp}}_{it}\right)}^2+b_i\left(p_{it}^{roll}+{\mathrm{\δp}}_{it}\right)c_i)+\underset{j\&element;g_{wind}}{\σ}{\mathrm{\λ}}_j(p_{jt}^f-p_{jt}^w\left)\right)---\left(2\right)$

f₁(p_it) it is scheduling model object function；T is the optimization time；t₀For optimizing start periods；t_lExcellent for basic point tracking layer Change Period Length；N is conventional power unit number；a_i、b_i、c_iCoal consumption coefficient for conventional power unit i；p_itFor conventional power unit i in the t period The active plan of exerting oneself；For i-th unit the t period rolling optimal plan layer plan；δp_itExist for i-th unit The basic point of t period follows the trail of plan adjustment amount, for controlling output；g_windFor Wind turbines number；λ_jFor abandoning wind cost factor； Exert oneself for the prediction of Wind turbines Extended short-term；For Wind turbines j the t period the active plan of exerting oneself.

(when sampling step length value is less, the anti-interference of system is stronger with 5min as sampling step length for basic point tracking layer.But it is subject to Unit adjusts the impact of speed, and sampling step length value is less, then be more difficult to ensure the ability of tracking to planned value for the unit output, control Effect is also poorer.Therefore, the selection of sampling step length is considered as the time lag response of controlled unit so that sampling step length at least should be greater than The response time of unit, because general unit response time is in more than 2min, therefore, basic point tracking layer at least will ensure sampling step Length, in more than 2min, is chosen as 5min here to ensure the implementation effect to basic point tracking plan for the different response speed units), 15min is the sampling period, the optimal power being given using optimal plan layer as adjustment basic point, by the system in the plan period State carries out minimum adjustment to follow the tracks of overall steady-state optimization result it is ensured that the safety of system operation.

(4) feedback compensation layer

Feedback compensation layer includes agc unit generation balancing the load correction unit and real-time control unit is exerted oneself correction unit；

Described real-time control unit wind power output correction unit is used for: the real-time tune being given with described real-time basic point tracking layer Degree revises Planning Directive as controlling basic point, and the stochastic prediction error that advanced prediction link is produced is revised in real time, generates Real-time control for being controlled to real-time control unit is exerted oneself adjust instruction, and described real-time control is exerted oneself adjust instruction It is issued to described real-time control unit, realize the control to real-time control unit.

Rolling optimal plan layer for the present invention and real-time basic point tracking layer, optimal plan layer and basic point tracking layer adopt Rolling optimization solution strategies, figure it is seen that optimal plan layer starts once every 15min, provide following 4 hour period The Planning Directive of length；And the every 5min of basic point tracking layer starts once, the plan providing following 15min minute length every time refers to Order, both adopt different size granularity, realize having complementary functions.

As can be seen here, the bilevel optimization model being applied to large-scale wind power control process that the present invention provides, in terms of optimum Draw the adjustment basic point as basic point tracking layer scheduling process for the economic optimum plan output valve of layer, after the adjustment of basic point tracking layer Dispatch command is as the correction basic point of feedback compensation layer.Finally, revised control instruction is handed down to unit by feedback compensation layer, Constitute before to plan issue data flow.Each layer is responsible for revising the deviation of last layer, and the deviation left to be repaiied by next layer Just, embody a kind of thought of " multilevel coordination, step by step refinement ", the inaccuracy that can effectively reduce wind-powered electricity generation prediction is to control process Impact.The actual running results show, carry out repeatedly the control strategy of rolling optimization renewal using bilevel optimization model of the present invention, Strengthening system can process disturbance and probabilistic ability, improve the robustness controlling.

The above is only the preferred embodiment of the present invention it is noted that ordinary skill people for the art For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should Depending on protection scope of the present invention.

Claims (4)

1. a kind of bilevel optimization model being applied to large-scale wind power control process is it is characterised in that include electromotor unit class Type division unit, rolling optimal plan layer, real-time basic point tracking layer and feedback compensation layer；

Electromotor unit Type division unit, for incorporating the whole network unit for roller press set type first into；Then, according to generating Machine unit responding ability, by roll unit be in turn divided into manually adjusting unit, Real-Time Scheduling unit, real-time control unit and Agc unit；Wherein, the described unit that manually adjusts refers to: does not have self-checking device, needs to rely on power plant operator to adjust manually Whole go out force tracking, with 15min for interval rolling optimal plan curve unit；Described Real-Time Scheduling unit refers to: has automatically Adjusting means, can from motion tracking with 5min for interval Real-Time Scheduling correction Planning Directive unit；Described real time control machine Group refers to: there are the agc unit of very fast regulating power and the Wind turbines with self-regulation ability, can be between motion tracking 1min Every real-time control exert oneself adjust instruction；Described agc unit refers to: the holding of responsible system frequency and dominant eigenvalues plan；

Described rolling optimal plan layer is used for: described rolling optimal plan layer adopts Coarse grain model, based on planning a few days ago, Load prediction results according to electrical network Extended short-term model and Extended short-term wind-powered electricity generation predict the outcome rolling amendment unit generation a few days ago It is planned out activity of force so that system generating gross capability power is gradually approached with actual power demand, obtain the unit of economic optimum Generation schedule is exerted oneself, and the unit generation plan of described economic optimum is exerted oneself act on described in manually adjust unit；

Described real-time basic point tracking layer is used for: described real-time basic point tracking layer adopts fine granularity rolling optimization model, with described warp The optimum unit generation plan of Ji is exerted oneself as basic point power, pre- according to electrical network ultra-short term result and ultra-short term wind-powered electricity generation Survey result adjustment unit output, generate the Real-Time Scheduling that minimum unit output adjustment is carried out to the system mode in the plan period and repair Positive Planning Directive, and described Real-Time Scheduling correction Planning Directive is acted on described Real-Time Scheduling unit, and then elimination unit is held The amount of unbalance that the change at random of the unbalanced power amount in Ji of passing through optimal plan and wind-powered electricity generation load causes；

Described feedback compensation layer includes agc unit generation balancing the load correction unit and real-time control unit is exerted oneself correction unit；

Described real-time control unit wind power output correction unit is used for: is repaiied with the Real-Time Scheduling that described real-time basic point tracking layer is given Positive Planning Directive as controlling basic point, revised in real time, generates and is used for by the stochastic prediction error that advanced prediction link is produced The real-time control that real-time control unit is controlled is exerted oneself adjust instruction, and described real-time control adjust instruction of exerting oneself is issued To described real-time control unit, realize the control to real-time control unit；

Described agc unit generation balancing the load correction unit is used for: is repaiied with the Real-Time Scheduling that described real-time basic point tracking layer is given Positive Planning Directive as controlling basic point, revised in real time, generates and is used for by the stochastic prediction error that advanced prediction link is produced The agc unit output adjust instruction that agc unit is controlled, and described agc unit output adjust instruction is issued to described Agc unit, realizes the control to agc unit.

2. the bilevel optimization model being applied to large-scale wind power control process according to claim 1 is it is characterised in that institute State rolling optimal plan layer to include: Extended short-term wind-powered electricity generation predicting unit, Extended short-term load estimation unit and economic optimum mould Type；

Described Extended short-term wind-powered electricity generation predicting unit is used for: Extended short-term wind-powered electricity generation is predicted；

Described Extended short-term load estimation unit is used for: Extended short-term load is predicted；

Described economic optimum model is used for: with described Extended short-term wind-powered electricity generation predicting unit and described Extended short-term load estimation unit Predict the outcome for input, based on rolling optimization control strategy rolling amendment, unit generation plan goes out activity of force a few days ago；

Wherein, described rolling optimization control strategy is: using the economic optimum scheduling model abandoned based on minimum on the basis of wind, sees formula (1):

$f_1\left(p_{it}\right)=\min \underset{t=t_0+1}{\overset{t_0+t_h}{\σ}}\left(\underset{i=1}{\overset{n}{\σ}}\right(a_i{p_{it}}^2+b_i{p}_{it}+c_i)+\underset{j\&element;g_{wind}}{\σ}{\mathrm{\λ}}_j(p_{jt}^f-p_{jt}^w\left)\right)---\left(1\right)$

Wherein, f₁(p_it) it is scheduling model object function；T is the optimization time；t₀For optimizing start periods；t_hFor optimal plan layer Optimize Period Length；N is conventional power unit number；a_i、b_i、c_iCoal consumption coefficient for conventional power unit i；p_itFor conventional power unit i in t The active plan of exerting oneself of section；g_windFor Wind turbines number；λ_jFor abandoning wind cost factor；Predict for Wind turbines Extended short-term Power；For Wind turbines j the t period the active plan of exerting oneself.

3. the bilevel optimization model being applied to large-scale wind power control process according to claim 1 is it is characterised in that institute State real-time basic point tracking layer to include: ultra-short term wind-powered electricity generation predicting unit, ultra-short term unit and basic point tracing model；

Described ultra-short term wind-powered electricity generation predicting unit is used for: ultra-short term wind-powered electricity generation is predicted；

Described ultra-short term unit is used for: super short period load is predicted；

Described basic point tracing model is used for: pre- with described ultra-short term wind-powered electricity generation predicting unit and described ultra-short term unit Surveying result is input, based on basic point pursive strategy, the system mode in the plan period is carried out with minimum adjustment；

Described basic point pursive strategy is: using the scheduling model of formula (2):

$f_1\left(p_{it}\right)=\min \underset{t=t_0+1}{\overset{t_0+t_l}{\σ}}\left(\underset{i=1}{\overset{n}{\σ}}\right(a_i{\left(p_{it}^{roll}+{\mathrm{\δp}}_{it}\right)}^2+b_i\left(p_{it}^{roll}+{\mathrm{\δp}}_{it}\right)+c_i)+\underset{j\&element;g_{wind}}{\σ}{\mathrm{\λ}}_j(p_{jt}^f-p_{jt}^w\left)\right)---\left(2\right)$

f₁(p_it) it is scheduling model object function；T is the optimization time；t₀For optimizing start periods；t_lWhen optimizing for basic point tracking layer Segment length；N is conventional power unit number；a_i、b_i、c_iCoal consumption coefficient for conventional power unit i；p_itFor conventional power unit i having in the t period The work(plan of exerting oneself；For i-th unit the t period rolling optimal plan layer plan；δp_itFor i-th unit in t The basic point of period follows the trail of plan adjustment amount, for controlling output；g_windFor Wind turbines number；λ_jFor abandoning wind cost factor；For The prediction of Wind turbines Extended short-term is exerted oneself；For Wind turbines j the t period the active plan of exerting oneself.

4. the bilevel optimization model being applied to large-scale wind power control process according to claim 1 is it is characterised in that institute State rolling optimal plan layer and described real-time basic point tracking layer all using the rolling optimization solution strategies based on mpc principle；Wherein, Described rolling optimal plan layer starts once every 15min, provides the Planning Directive of following 4 hours Period Length；Described real-time The every 5min of basic point tracking layer starts once, provides the Planning Directive of following 15min minute length every time.