CN105720574A - SPSA-based data drive control method for single region load frequency of power system - Google Patents
SPSA-based data drive control method for single region load frequency of power system Download PDFInfo
- Publication number
- CN105720574A CN105720574A CN201610095937.0A CN201610095937A CN105720574A CN 105720574 A CN105720574 A CN 105720574A CN 201610095937 A CN201610095937 A CN 201610095937A CN 105720574 A CN105720574 A CN 105720574A
- Authority
- CN
- China
- Prior art keywords
- power
- controller
- power system
- control
- value
- 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.)
- Granted
Links
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]
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses an SPSA-based data drive control method for a single region load frequency of a power system. The data drive control method comprises the following steps: obtaining a frequency deviation increment of a controlled power system, an increment change of an output power of a turbine generator and data information of the increment change of an adjusting valve position in real time; building an online neural network identification model of the system; building a neural network controller; updating a connection weight of the neural network controller; calculating a control signal, and feeding back the control signal to the controlled system; and stabilizing the frequency of the system at a set value again by controlling the opening of an air intake valve or a steam valve of the turbine generator or a water inlet valve of a water turbine and improving the unbalance condition of a prime mover power and a load power. The method disclosed by the invention is simple in calculation, few in parameter and easy to implement, can effectively avoid the complicated modeling process of the power system, and can solve the problems of a poor control effect and the like cased by inaccurate or uncertain parameters of the system model.
Description
Technical field
The present invention relates to a kind of data drive control method of power system list region LOAD FREQUENCY based on SPSA, belong to power system load FREQUENCY CONTROL field.
Background technology
LOAD FREQUENCY control is one of problem important in electric power system design and operation, is topmost system frequency control device.For power system, load is always continually changing, and it would furthermore be possible to various fault can be there is at any time, it is necessary to designs a LOAD FREQUENCY and controls system so that the load of electromotor is controlled by power system dependent Frequency.Therefore, for the power system with Parameter uncertainties, how by FREQUENCY CONTROL in an acceptable scope, it is an extremely challenging research topic all the time.
Power system is complicated nonlinear dynamic system, owing to power system only exposes less load variations when properly functioning, is frequently used linearizing model in the conventional way to represent the system dynamics near operating point.In the research that LOAD FREQUENCY controls, the LOAD FREQUENCY that existing AGC research is both for greatly Kirchmayer proportional integral (PI) control mode proposed controls (LFC) model [1] and improves, but PI controller is difficult to meet tracking fixed valure and Disturbance Rejection requirement in performance simultaneously；And bad adaptability during control object characteristic gradual change, it will cause the contradiction [2] between rapidity and the overshoot that system responds.In recent years, Intelligent Control Strategy is widely used in the LOAD FREQUENCY control research of power system gradually.Document [3] utilizes the dynamic model of neural network identification power system, determines optimum PID controller parameter by dynamic optimization, makes the LOAD FREQUENCY in two regions control existing self study adaptive capacity, has again the PID broad applicability controlled.But this method is easily subject to the impact of local minimum point when optimizing.Document [4] provides a kind of New PID intelligent controller based on fuzzy genetic algorithm.The method adopts genetic algorithm PID value as the initial value of fuzzy controller, then relevant parameter genetic algorithm optimization to fuzzy control again, it is to avoid the blindness that parameter selects, it is achieved that the optimum control to system.Though but genetic algorithm has stronger ability of searching optimum, being easily absorbed in local optimum situation, and search speed be slower.Additionally, AGC research field it is also proposed some other new control mode, such as Sliding mode variable structure control [5], fuzzy control [6], Self Adaptive Control [7] etc., but these algorithms have again respective defect, such as the complexity etc. of chatter phenomenon intrinsic when Sliding mode variable structure control switches at control point, fuzzy quantization limited precision and tradition adaptive control algorithm.
Analyzed from above, complex characteristics due to power system, at present the control method of LOAD FREQUENCY is need further to improve and innovation, it is necessary in conjunction with features such as the Nonlinear Dynamic of its complexity, uncertain parameter, invent novel system model information is relied on less and algorithm is relatively easy, be easily achieved, in systems in practice can the control algolithm of stable operation, thus the LOAD FREQUENCY being effectively realized power system controls.
Based on the non-model control method [8] of SPSA, there is computing simple, relate to parameter few, fast convergence rate, adjustable parameter are few, system model information is relied on the advantages such as few, algorithm has been successfully applied [9～12] since proposing in numerous areas, and its application in control field also has been achieved for initial achievements [13～15].Due to it application widely, in its application, constantly have new discovery.
[list of references]
[1] Liu Mengxin, Wang Jie, Chen Chen. power system frequency control theory and development [J]. electrotechnics journal, 2007,22 (11): 135-145.
[2]OysalY.Acomparativestudyofadaptiveloadfrequencycontrollerdesignsinapowersystemwithdynamicneuralnetworkmodels[J].EnergyConversionandManagement,2005,46(15-16):2656-2668.
[3]JuangC.F.,LuC.F.Powersystemloadfrequencycontrolwithfuzzygainschedulingdesignedbygeneticalgorithms[C].IEEEInternationalConferenceonPlasmaScience.Honolulu,HI,USA:2002.
[4] Shi Haigang. Automatic Generation Control (AGC) combines static test and thermal experimental study [J]. protecting electrical power system and control, 2010,38 (13): 65-70.
[5] Meng Xiangping, Xue Changfei, Zhang Huaguang. the PI sliding formwork LOAD FREQUENCY of multi-area Interconnected Power System controls [J]. Proceedings of the CSEE, 2001,21 (3): 6-11.
[6] Li Yuanyuan, Song Yonghua, Wei lacks the Chaos Genetic Algorithm research [J] of rock .AGC Units Allotment. protecting electrical power system and control, and 2008,36 (16): 17-21.
[7] Yu Tao, Zhou Bin. based on interconnected network CPS Self-correc ting control [J] of intensified learning. protecting electrical power system and control, 2009,37 (10): 33-38.
[8] SpallJ.C., CristionJ.A.Model-freecontrolofgeneraldiscrete-timesyste ms [C] .Proceedingsofthe32ndIEEEConferenceonDecisionandControl, SanAntonio, Texas, USA, 1993:2792～2797.
[9] Cheema, J.S., Sankpal, N.V., Tambe, S.S., andKulkarni, B.D.GeneticProgrammingAssistedStochasticOptimizationStra tegiesforOptimizationofGlucosetoGluconicAcidFermentation [J] .BiotechnolgyProgress, 2002,18:1356～1365.
[10] Cipriani, E.Florian, M., Mahut, M., andNigro, M.AGradientApproximationApproachforAdjustingTemporalOrig in-DestinationMatrices [J] .TransportationResearch, PartC-EmergingTechnologies, 2011,19 (2): 270～282.
[11] Das, S., Spall, J.C., andGhanem, R.EfficientMonteCarloComputationofFisherInformationMatri xUsingPriorInformation [J] .ComputationalStatisticsandDataAnalysis, 2010,54 (2): 272～289.
[12] Finck, S.andBeyer, H.G.PerformanceAnalysisoftheSimultaneousPerturbationStoc hasticApproximationAlgorithmontheNoisySphereModel [J] .TheoreticalComputerScience, 2012,419:50～72.
[13] Wu Zhiwei, Wu Yongjian, Chai Tianyou. based on the electric melting magnesium furnace intelligence control system [J] of simultaneous perturbation stochastic approximation algorithm. Shanghai Communications University journal .2011,45 (8): 1095～1100.
[14] Bhatnagar, S.andKumar, S.ASimultaneousPerturbationStochasticApproximation-Based Actor CriticAlgorithmforMarkovDecisionProcesses [J] .IEEETransactionsonAutomaticControl, 2004,49:592～598.
[15] Radac, M., Precup, R., Petriu, E.M., andPreitl, S.ApplicationofIFTandSPSAtoServoSystemControl [J] .IEEETransactionsonNeuralNetworks, 2011,22 (12): 2363～2375.
Summary of the invention
For problems of the prior art, the present invention proposes a kind of data drive control method of power system list region LOAD FREQUENCY based on simultaneous perturbation stochastic approximation (SPSA).The method calculates simple, and parameter is few, easily realizes, and can be effectively prevented from the modeling process that power system is complicated, also is able to avoid system model inaccurate or the problem such as control weak effect that uncertain parameter may result in simultaneously.
In order to solve above-mentioned technical problem, the data drive control method of a kind of power system list region LOAD FREQUENCY based on SPSA that the present invention proposes, the control for power system can depart from the dependence for its model information, and this control method comprises the following steps:
The data message that step one, in real time the acquisition frequency departure increment of controlled power system, the increment change of steam turbine generator output and the increment of throttle position change；
Step 2, set up the online nerve network recognition model of system: the state value of former P moment real system, i.e. { x (k), x (k-1), ..., x (k-P+1) } and the control input value { u (k-1) of front Q moment real system, u (k-2), ..., u (k-Q) } as the input of neural network model, P and Q is time window length, using the state estimation x* (k+1) of current time power system as the output of neutral net；Choosing three-decker, i.e. an input layer, a hidden layer and an output layer, radial basis function neural network is as the on-line identification model of system, and the basic function of network hidden node adopts Euclidean distance function, and uses Gaussian function as activation primitive；Determine network Hidden nodes, each RBF data center and extension constant, will with error functionAs the object function of neural network learning, wherein, S is the number of training of input, βjFor forgetting factor, ejError for network output with real system outlet chamber；Adopt gradient training method, regulate the data center of each hidden node, extension constant by minimizing object function and connect weights, so that the power system that neural network identification Model approximation is actual controlled；
Step 3, set up nerve network controller: the state value of former M moment real system, i.e. { x (k), x (k-1), ..., x (k-M+1) }, the control input value of front n-hour real system, i.e. { u (k-1), u (k-2), ..., u (k-N) } as the input of nerve network controller, with the control signal of system, namely the required Power Systems value adjusted is as the output of nerve network controller, choose the BP network of four-layer structure as nerve network controller, i.e. an input layer, two hidden layers and an output layer；
Step 4, controller parameter update: adopt SPSA algorithm, chooseFor optimization object function, and use online neural network identifier model to replace real system to calculate the state value of system when adding random disturbance signal, then complete nerve network controller and connect the renewal of weights；
Step 5, calculating obtain control signal and feed back in controlled system: use the connection weights of updated nerve network controller, calculate the control signal u (k) obtaining system, i.e. the required system power Δ P adjustedcValue, and according to Δ PcAdjust the power of system prime mover: namely as Δ Pc> 0 time, increase the aperture of the inlet valve of the admission valve of steam turbine generator or the hydraulic turbine, and then increase the power of prime mover；As Δ Pc< when 0, reduce the steam valve of steam turbine generator or the aperture of the inlet valve of the hydraulic turbine, and then reduce the power of prime mover；Thus improving the unbalance condition of original machine power and load power so that the frequency of system is stabilized to set-point again.
Further, in step 4, the method connecting weights updating nerve network controller is as follows: computing controller parameterControlled quentity controlled variable produced by nerve network controller, is denoted as u (k)±, wherein θkFor nerve network controller connection weights,Controller for a upper moment connects the estimated value of weights, ckSequence or value for tending to 0 are equal to the scalar factor of constant, Δk=[Δk1,Δk2,...,ΔkL]TIt is a random vector, and by u (k)±Input signal respectively as twice to add to the online nerve network recognition model of system, record it and export accordingly, be denoted as x (k+1)±；By x (k+1)±Value be updated to the optimization object function of controllerIn, calculating obtainsAgain willSubstitute into into gradient estimation formulasIn, wherein, l=1,2 ..., L, L is the number of controller parameter, and then obtains the estimated value of gradientFinally willIt is updated to controller parameter more new formulaIn, wherein akFor scalar factor, calculating obtains k moment controller and connects the estimated value of weightsThus completing controller parameter and the renewal of nerve network controller connection weights.
Compared with prior art, the invention has the beneficial effects as follows:
First, it is difficult to the characteristics such as acquisition for the Complex Dynamic of power system, parameter uncertainty and its mathematical models, present invention employs the data drive control algorithm based on SPSA and realize the LOAD FREQUENCY control of single regional power system, the traditional control algorithm Dependence Problem to system model can be efficiently solved, the problems such as the control performance difference simultaneously avoiding system model inaccurate and to cause, better control effect thus being finally reached；
Second, select the method for SPSA to be controlled the parameter estimation in process, there is parameter simple, amount of calculation is little, it is easy to accomplish etc. advantage, be effectively improved the performance of data drive control device；
3rd, selecting neutral net as controller, it is possible to more neatly controller architecture to be adjusted, thus more efficiently coordinating the characteristic of concrete controlled power system, and then obtaining the control effect of the best.
Accompanying drawing explanation
Fig. 1 is that power system individually controls system block diagram；
Fig. 2 is power system overall system control block diagram；
Fig. 3 is the data drive control algorithm structure block diagram based on SPSA；
Fig. 4 is online nerve network recognition device structured flowchart；
Fig. 5 is the frequency changes delta f of the power system response curve when three groups of different system parameters are arranged.
Detailed description of the invention
LOAD FREQUENCY control usual adopted be that PI controller is to keep system frequency constant and to control the exchange plan of interconnection.But power system is as a very huge and complicated system, it is difficult to realize it is analyzed and on-line control.It addition, the dynamic characteristic of complexity and parameter uncertainty also make to hardly result in practice the mathematical models of power system, thus causing very big based on the control strategy mostly expense designed by approximate mathematical model and usual reliability is not high.Traditional LOAD FREQUENCY control design case method is only applicable to certain specific operating point of system mostly, but consider in modern power systems the impact of the scale of generator unit constantly expanded, changeable system structure, the access of regenerative resource, the access of Large Copacity/centralized load and other some new uncertain factors so that the method for designing that traditional LOAD FREQUENCY controls is no longer able to realize the effective adjustment to modern power systems.
The present invention is directed to the complicated dynamic characteristic of power system, parameter uncertainty and its mathematical models and be difficult to the impact of the factors such as acquisition, design a kind of model information is relied on less, parameter is simple, the data drive control method based on SPSA that is easily achieved, thus realizing the LOAD FREQUENCY fast and effectively of single regional power system is controlled.
The design process of the present invention mainly considers the following aspects:
(1) single regional power system model
Figures 1 and 2 show that single regional power system model framework chart.Wherein, Δ f is frequency departure increment (Hz), Δ PGIt is increment change (P.u., MW) of steam turbine generator output, Δ PcIt is control signal (P.u., MW), Δ XGIt is increment change (P.u., MW) of throttle position, Δ PdIt is load increment (P.u., MW), TSIt is speed regulator time constant (s), TTIt is steam turbine generator time constant (s), TpIt is power plant model time constant (s) in power system, KpBeing the power plant's gain in power system, R is the rate adaptation (HzP.u., MW-1) caused by speed regulator action.Single regional power system model meets below equation:
The form being write as matrix equation is:
X=AX+BU+Fd, wherein:
XT=[Δ f Δ PGΔXG]T, U=[Δ Pc], d=[Δ Pd]
(2) data drive control of single regional power system LOAD FREQUENCY
Being difficult to the characteristics such as acquisition for the complicated dynamic characteristic of power system, parameter uncertainty and its mathematical models, the present invention devises, and adopts the control algolithm based on SPSA to realize the effective control to single regional power system LOAD FREQUENCY.Concrete control strategy is as follows:
Neutral net is used to produce control signal uk.The structure (number of plies and nodal point number) of neutral net is fixed, but allows it to connect weights and change over and be constantly updated.Here, LOAD FREQUENCY control problem is solved as Optimal Control Problem.In control process, it is necessary to producing ukThe connection weights of neutral net (be designated as parameter vector θk) estimate.As controller neutral net with (1) formula for input:
x(k),x(k-1),...,x(k-M+1),u(k-1),u(k-2),...,u(k-N)(1)
Namely input as the controlled quentity controlled variable in regular length time window before current time and quantity of state.Wherein, the state of etching system when x (k) is for k；The control that u (k) is the controlled device of moment k inputs, and namely needs the system power Δ P adjustedc；The time window length of M and N respectively data.Control performance standard can represent by (2) formula:
Wherein, Q is positive semi-definite weight matrix, and G is the weight matrix of positive definite.The purpose of controller design is exactly at each moment k, finds and enables to the optimal controller parameter that Control performance standard (2) is minimum
Data drive control method based on SPSA adopts the algorithm of simultaneous perturbation stochastic approximation to solve the optimization problem of unknown plant model, utilizes recurrence formula:
Carry out estimated sequence { θk, wherein,For the calculated estimated value of current iteration, akFor scalar factor,Can being calculated by (4) formula and obtain, it isSimultaneous perturbation estimated value.
Wherein L is the number of controller parameter.ForEstimated value:
Require over x (k+1)±With u (k)±Measured value calculate obtain, x (k+1)±Control for controlled device inputs equal to u (k)±Time system mode measured value, u (k)±It it is then controller parameterTime nerve network controller produced by controlled quentity controlled variable, ckSequence or value for tending to 0 is equal to the scalar factor of constant, Δk=[Δk1,Δk2,...,ΔkL]TIt is a random vector, and ΔklObey the distribution of independent bounded symmetric.
So, the operating each iteration of algorithm only need to by twice closed-loop experiment, it is possible to obtains system mode x (k+1)±Measured value, be updated to (4) and (5) formula and just can obtain gradient estimated valueJust can calculate the control parameter obtaining now controller again through (3) formula, i.e. the connection weights of nerve network controller, and then obtain control input u (k) of etching system time this.
(3) the enforcement step of single regional power system LOAD FREQUENCY data drive control
When single regional power system is carried out the data drive control of LOAD FREQUENCY, mainly include following link:
A, when power system frequency changes, read in real time system frequency fluctuation and system state amount Δ PGWith Δ XGConcrete numerical value；
LOAD FREQUENCY control problem, based on the data control method of SPSA, is solved by b, utilization as Optimal Control Problem, is now needed the system power Δ P adjustedc；
The system power Δ P of the required adjustment that c, basis are tried to achievecValue, adjust the power of system prime mover, make every effort to improve the unbalance condition of original machine power and load power.That is as Δ Pc> 0 time, the aperture of the admission valve of steam turbine generator or the inlet valve of the hydraulic turbine will increase, increase prime mover power；As Δ Pc< when 0, the aperture of the steam valve of steam turbine generator or the inlet valve of the hydraulic turbine will reduce, and reduces the power of prime mover；
Below in conjunction with the drawings and specific embodiments, technical solution of the present invention being described in further detail, the present invention is only explained by described specific embodiment, not in order to limit the present invention.
In the present invention, the LOAD FREQUENCY that the data drive control strategy based on SPSA realizes single regional power system is utilized to control, in control process, the inputoutput data of real-time acquisition system, and it is passed in nerve network controller, utilize the algorithm that disturbance approaches to come controller parameter, that is the connection weights of nerve network controller carry out real-time update, and utilize the nerve network controller after renewal to produce control signal, feed back in controlled power system, complete to control.Here it is the LOAD FREQUENCY control problem of power system is converted into Optimal Control Problem solve, when frequency of occurrences fluctuation in electrical network, utilize the data drive control strategy based on SPSA to be optimized to solve, obtain the system power value needing to adjust, and according to this value, the power of system prime mover is adjusted, and then improve the unbalance condition of original machine power and load power, make the frequency retrieval of system to steady statue.The explanation of this wherein involved key operation link is as follows:
Based on the control algolithm of SPSA, each iteration in calculating process only needs the measurement data of twice closed-loop experiment can be obtained byEstimated valueAnd then producing control signal u (k), whole control process is without the model information of controlled device.The overall control structure block diagram of algorithm is presented in Fig. 3, in control process single regional power system that real-time storage is controlled status information and input control signal, utilize the historical data input as the data drive control device based on SPSA of stored single regional power system, calculate and produce control signal u (k), and return it in controlled single regional power system, and then complete to control.
Control algolithm based on SPSA needs addition random disturbance signal in controlled system in parameter estimation procedure, and controlled system can be produced interference in various degree by such disturbance, thus affecting control effect.In the present invention, in order to avoid the generation of this kind of situation, devise the process utilizing online nerve network recognition device to be controlled device parameter estimation, thus being effectively prevented from the random disturbance signal impact on controlled power system.In control algolithm running, for the neural network identifier model that controlled system training is online, as shown in Figure 4, wherein TDLm and TDLn respectively length is the time window of m and n.
The parameter estimation procedure of controller calculates gradient estimated valueUsed system mode measured value x (k)±Namely be random disturbance signal is joined in neural network identifier model calculated.So, the random disturbance signal that algorithm to be used when each step is run has been added in virtual neural network identifier, thus efficiently avoid the disturbing signal impact on real system.
Analyzing through above, in the present invention, the data drive control of single regional power system LOAD FREQUENCY based on SPSA specifically can carry out in accordance with the following steps:
The data message that step one, in real time the acquisition frequency departure increment of controlled power system, the increment change of steam turbine generator output and the increment of throttle position change；
Step 2, set up the online nerve network recognition model of system: the state value of former P moment real system, i.e. { x (k), x (k-1), ..., x (k-P+1) } and the control input value { u (k-1) of front Q moment real system, u (k-2), ..., u (k-Q) } as the input of neural network model, P and Q is time window length, using the state estimation x* (k+1) of current time power system as the output of neutral net；Choosing three-decker, i.e. an input layer, a hidden layer and an output layer, radial basis function neural network is as the on-line identification model of system, and the basic function of network hidden node adopts Euclidean distance function, and uses Gaussian function as activation primitive；Determine network Hidden nodes, each RBF data center and extension constant, will with error functionAs the object function of neural network learning, wherein, S is the number of training of input, βjFor forgetting factor, ejError for network output with real system outlet chamber；Adopt gradient training method, regulate the data center of each hidden node, extension constant by minimizing object function and connect weights, so that the power system that neural network identification Model approximation is actual controlled；
Step 3, set up nerve network controller: the state value of former M moment real system, i.e. { x (k), x (k-1), ..., x (k-M+1) }, the control input value of front n-hour real system, i.e. { u (k-1), u (k-2), ..., u (k-N) } as the input of nerve network controller, with the control signal of system, namely the required Power Systems value adjusted is as the output of nerve network controller, choose the BP network of four-layer structure as nerve network controller, i.e. an input layer, two hidden layers and an output layer；
Step 4, controller parameter update: adopt SPSA algorithm, chooseFor optimization object function, and use online neural network identifier model to replace real system to calculate the state value of system when adding random disturbance signal, then complete nerve network controller and connect the renewal of weights；
The method connecting weights updating nerve network controller is as follows: computing controller parameterControlled quentity controlled variable produced by nerve network controller, is denoted as u (k)±, wherein θkFor nerve network controller connection weights,Controller for a upper moment connects the estimated value of weights, ckSequence or value for tending to 0 are equal to the scalar factor of constant, Δk=[Δk1,Δk2,...,ΔkL]TIt is a random vector, and by u (k)±Input signal respectively as twice to add to the online nerve network recognition model of system, record it and export accordingly, be denoted as x (k+1)±；By x (k+1)±Value be updated to the optimization object function of controllerIn, calculating obtainsAgain willSubstitute into into gradient estimation formulasIn, wherein, l=1,2 ..., L, L is the number of controller parameter, and then obtains the estimated value of gradientFinally willIt is updated to controller parameter more new formulaIn, wherein akFor scalar factor, calculating obtains k moment controller and connects the estimated value of weightsThus completing controller parameter and the renewal of nerve network controller connection weights.
Step 5, calculating obtain control signal and feed back in controlled system: use the connection weights of updated nerve network controller, calculate the control signal u (k) obtaining system, i.e. the required system power Δ P adjustedcValue, and according to Δ PcAdjust the power of system prime mover: namely as Δ Pc> 0 time, increase the aperture of the inlet valve of the admission valve of steam turbine generator or the hydraulic turbine, and then increase the power of prime mover；As Δ Pc< when 0, reduce the steam valve of steam turbine generator or the aperture of the inlet valve of the hydraulic turbine, and then reduce the power of prime mover；Thus improving the unbalance condition of original machine power and load power so that the frequency of system is stabilized to set-point again.
Introducing the LOAD FREQUENCY control problem with the probabilistic single regional power system of systematic parameter below and do simulation analysis, the standard parameter value of system is: Tp=20sec, Kp=120sec, TT=0.3sec, TG=0.08sec, R=2.4.The parameter variation range of system is to obtain on the basis that its standard parameter value changes 30% and 50% simultaneously.Make TpChange 50%, TT, TGChange 30% with R, obtain the constant interval of systematic parameter: Standardized systematic parameter is to obtain by parameters is averaged, then standardized system model can be written as:
B0=[0013.736]T, F0=[-800]T
In order to verify that in the present invention, designed control strategy processes the effectiveness of the LOAD FREQUENCY control problem with the probabilistic power system of systematic parameter, is provided with systematic parameter three groups different to do simulation analysis:
First group: modular system parameter；
Second group:
3rd group:
In simulation process, selecting a neutral net with two intermediate layers as data drive control device, its structure is N5-3-2-1, in the parameter estimation procedure of controller, parameter akAnd ckIt is chosen for: ak=0.05/k0.602, ck=0.15/k0.101.Above power system is controlled by the data drive control strategy utilizing single regional power system LOAD FREQUENCY proposed by the invention, and the system response condition when three groups of different parameters are arranged is as shown in Figure 5.
Be can be seen that by simulation result, parameter three groups different is arranged, system can both reach stable state within 500 steps, also further related to the data drive control strategy based on single regional power system LOAD FREQUENCY of SPSA designed in the present invention and can obtain good control effect, and the stability of whole system can have been ensured in the scope of the system parameter variations allowed.
Although above in conjunction with accompanying drawing, invention has been described; but the invention is not limited in above-mentioned detailed description of the invention; above-mentioned detailed description of the invention is merely schematic; rather than it is restrictive; those of ordinary skill in the art is under the enlightenment of the present invention; without deviating from the spirit of the invention, it is also possible to make many variations, these belong within the protection of the present invention.
Claims (2)
1. the data drive control method based on the power system list region LOAD FREQUENCY of SPSA, it is characterised in that the control for power system can depart from the dependence for its model information, comprises the following steps:
The data message that step one, in real time the acquisition frequency departure increment of controlled power system, the increment change of steam turbine generator output and the increment of throttle position change；
Step 2, set up the online nerve network recognition model of system: the state value of former P moment real system, i.e. { x (k), x (k-1), ..., x (k-P+1) } and the control input value { u (k-1) of front Q moment real system, u (k-2), ..., u (k-Q) } as the input of neural network model, P and Q is time window length, using the state estimation x* (k+1) of current time power system as the output of neutral net；Choosing three-decker, i.e. an input layer, a hidden layer and an output layer, radial basis function neural network is as the on-line identification model of system, and the basic function of network hidden node adopts Euclidean distance function, and uses Gaussian function as activation primitive；Determine network Hidden nodes, each RBF data center and extension constant, will with error functionAs the object function of neural network learning, wherein, S is the number of training of input, βjFor forgetting factor, ejError for network output with real system outlet chamber；Adopt gradient training method, regulate the data center of each hidden node, extension constant by minimizing object function and connect weights, so that the power system that neural network identification Model approximation is actual controlled；
Step 3, set up nerve network controller: the state value of former M moment real system, i.e. { x (k), x (k-1), ..., x (k-M+1) }, the control input value of front n-hour real system, i.e. { u (k-1), u (k-2), ..., u (k-N) } as the input of nerve network controller, with the control signal of system, namely the required Power Systems value adjusted is as the output of nerve network controller, choose the BP network of four-layer structure as nerve network controller, i.e. an input layer, two hidden layers and an output layer；
Step 4, controller parameter update: adopt SPSA algorithm, chooseFor optimization object function, and use online neural network identifier model to replace real system to calculate the state value of system when adding random disturbance signal, then complete nerve network controller and connect the renewal of weights；
Step 5, calculating obtain control signal and feed back in controlled system: use the connection weights of updated nerve network controller, calculate the control signal u (k) obtaining system, i.e. the required system power Δ P adjustedcValue, and according to Δ PcAdjust the power of system prime mover: namely as Δ Pc> 0 time, increase the aperture of the inlet valve of the admission valve of steam turbine generator or the hydraulic turbine, and then increase the power of prime mover；As Δ Pc< when 0, reduce the steam valve of steam turbine generator or the aperture of the inlet valve of the hydraulic turbine, and then reduce the power of prime mover；Thus improving the unbalance condition of original machine power and load power so that the frequency of system is stabilized to set-point again.
2., according to claim 1 based on the data drive control method of the power system list region LOAD FREQUENCY of SPSA, it is characterised in that in step 4, the method connecting weights updating nerve network controller is as follows: computing controller parameterControlled quentity controlled variable produced by nerve network controller, is denoted as u (k)±, wherein θkFor nerve network controller connection weights,Controller for a upper moment connects the estimated value of weights, ckSequence or value for tending to 0 are equal to the scalar factor of constant, Δk=[Δk1,Δk2,...,ΔkL]TIt is a random vector, and by u (k)±Input signal respectively as twice to add to the online nerve network recognition model of system, record it and export accordingly, be denoted as x (k+1)±；By x (k+1)±Value be updated to the optimization object function of controller In, calculating obtainsAgain willSubstitute into into gradient estimation formulasIn, wherein, l=1,2 ..., L, L is the number of controller parameter, and then obtains the estimated value of gradientFinally willIt is updated to controller parameter more new formulaIn, wherein akFor scalar factor, calculating obtains k moment controller and connects the estimated value of weightsThus completing controller parameter and the renewal of nerve network controller connection weights.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610095937.0A CN105720574B (en) | 2016-02-22 | 2016-02-22 | The data drive control method of electric system list region LOAD FREQUENCY based on SPSA |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610095937.0A CN105720574B (en) | 2016-02-22 | 2016-02-22 | The data drive control method of electric system list region LOAD FREQUENCY based on SPSA |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105720574A true CN105720574A (en) | 2016-06-29 |
| CN105720574B CN105720574B (en) | 2018-06-08 |
Family
ID=56156968
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610095937.0A Expired - Fee Related CN105720574B (en) | 2016-02-22 | 2016-02-22 | The data drive control method of electric system list region LOAD FREQUENCY based on SPSA |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105720574B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530118A (en) * | 2016-10-17 | 2017-03-22 | 温州大学 | Constrained GPC optimal control method for load frequency of multi-zone interconnection power system |
| CN107769273A (en) * | 2017-11-03 | 2018-03-06 | 杭州澎康自动化科技有限公司 | A kind of steam turbine load distribution control method and system |
| CN112666423A (en) * | 2020-12-03 | 2021-04-16 | 广州电力通信网络有限公司 | Testing device for power communication network |
| CN112821424A (en) * | 2021-01-29 | 2021-05-18 | 国网辽宁省电力有限公司大连供电公司 | Power system frequency response analysis method based on data-model fusion drive |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5668717A (en) * | 1993-06-04 | 1997-09-16 | The Johns Hopkins University | Method and apparatus for model-free optimal signal timing for system-wide traffic control |
| CN104019520A (en) * | 2014-05-20 | 2014-09-03 | 天津大学 | Data drive control method for minimum energy consumption of refrigerating system on basis of SPSA |
-
2016
- 2016-02-22 CN CN201610095937.0A patent/CN105720574B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5668717A (en) * | 1993-06-04 | 1997-09-16 | The Johns Hopkins University | Method and apparatus for model-free optimal signal timing for system-wide traffic control |
| CN104019520A (en) * | 2014-05-20 | 2014-09-03 | 天津大学 | Data drive control method for minimum energy consumption of refrigerating system on basis of SPSA |
Non-Patent Citations (5)
| Title |
|---|
| J.C.SPALL等: "MODEL-FREE CONTROL OF GENERAL DISCRETE-TIME SYSTEMS", 《PROCEEDINGS OF THE 32ND CONFERENCE ON DECISION AND CONTROL》 * |
| 侯忠生等: "数据驱动控制理论及方法的回顾和展望", 《自动化学报》 * |
| 彭建春 等: "单区域负荷频率的神经模糊自适应PID控制", 《计算技术与自动化》 * |
| 董娜 等: "改进的非线性离散系统自适应数据驱动控制", 《控制理论与应用》 * |
| 董娜: "非线性自适应数据驱动控制器", 《PROCEEDINGS OF THE 33RD CHINESE CONTROL CONFERENCE》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530118A (en) * | 2016-10-17 | 2017-03-22 | 温州大学 | Constrained GPC optimal control method for load frequency of multi-zone interconnection power system |
| CN106530118B (en) * | 2016-10-17 | 2019-08-02 | 温州大学 | The constraint GPC optimal control method of multi-region interconnected electric power system LOAD FREQUENCY |
| CN107769273A (en) * | 2017-11-03 | 2018-03-06 | 杭州澎康自动化科技有限公司 | A kind of steam turbine load distribution control method and system |
| CN107769273B (en) * | 2017-11-03 | 2020-04-17 | 杭州澎康自动化科技有限公司 | Steam turbine load distribution control method and system |
| CN112666423A (en) * | 2020-12-03 | 2021-04-16 | 广州电力通信网络有限公司 | Testing device for power communication network |
| CN112821424A (en) * | 2021-01-29 | 2021-05-18 | 国网辽宁省电力有限公司大连供电公司 | Power system frequency response analysis method based on data-model fusion drive |
| CN112821424B (en) * | 2021-01-29 | 2023-07-14 | 国网辽宁省电力有限公司大连供电公司 | Power system frequency response analysis method based on data-model fusion driving |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105720574B (en) | 2018-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108512258B (en) | Wind power plant active scheduling method based on improved multi-agent consistency algorithm | |
| Verma et al. | Intelligent automatic generation control of two-area hydrothermal power system using ANN and fuzzy logic | |
| CN105720574A (en) | SPSA-based data drive control method for single region load frequency of power system | |
| CN107437815B (en) | The optimal control method and relevant device of governor in Hydropower Unit | |
| Azizipanah-Abarghooee et al. | Modified shuffled frog leaping algorithm for multi-objective optimal power flow with FACTS devices | |
| WO2022121446A1 (en) | Control system, reactive voltage control method and device, medium, and calculation device | |
| CN111030141A (en) | Source-load cooperative distributed optimization regulation and control method based on consistency algorithm | |
| Jaber et al. | A new parameters identification of single area power system based LFC using Segmentation Particle Swarm Optimization (SePSO) algorithm | |
| CN113852098B (en) | Automatic power generation control scheduling method based on multi-target dragonfly algorithm | |
| CN112012875B (en) | Optimization method of PID control parameters of water turbine regulating system | |
| Hosseini et al. | Improving response of wind turbines by pitch angle controller based on gain-scheduled recurrent ANFIS type 2 with passive reinforcement learning | |
| CN114722693A (en) | Optimization method of two-type fuzzy control parameter of water turbine regulating system | |
| Ren et al. | Feedforward feedback pitch control for wind turbine based on feedback linearization with sliding mode and fuzzy PID algorithm | |
| CN109066728A (en) | The online damp coordinated control method of the multiple inter-area oscillation modes of extra-high voltage grid | |
| CN111245032A (en) | Voltage prediction control method considering loss reduction optimization of wind power plant current collection line | |
| CN105896547B (en) | A kind of bulk power grid hierarchical voltage control method under wind power integration | |
| Zadeh et al. | Load frequency control in interconnected power system by nonlinear term and uncertainty considerations by using of harmony search optimization algorithm and fuzzy-neural network | |
| CN105207220B (en) | A kind of tapping voltage regulation and control method based on progressive learning | |
| CN113270867B (en) | Automatic adjustment method for weak power grid tide without solution | |
| Maurya et al. | Performance of Load Frequency Control of Two-Area Power System by Using Proportional Integral Derivative Controller | |
| CN115133540A (en) | Power distribution network model-free real-time voltage control method | |
| CN111509724B (en) | Hierarchical power distribution network voltage control method combining decentralized time sequence and centralized model prediction | |
| Lei | Application improved particle swarm algorithm in parameter optimization of hydraulic turbine governing systems | |
| Xiangyu et al. | Regional power grid AGC control strategy research based on Q-learning algorithm under WACPB mode | |
| Li et al. | Application of Improved Artificial Bee Colony Algorithm in constant pressure water supply system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210719 Address after: No.216 Tangli Road, Chaoyang District, Beijing 100020 Patentee after: GUOTIE GONGTIE (BEIJING) TECHNOLOGY Co.,Ltd. Address before: 300072 Tianjin City, Nankai District Wei Jin Road No. 92 Patentee before: Tianjin University |
|
| 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: 20180608 |