CN102594244B - Joint control method of primary frequency modulation for doubly-fed wind power generation set - Google Patents

Abstract

The invention discloses a joint control method of a primary frequency modulation for a doubly-fed wind power generation set, which comprises the steps of performing the on-line optimal design on the PD (proportional derivative) controller parameter by using continuous Hopfield neural network so as to build a neutral controller with relatively strong self-adaptability and realize the joint control of the rotor kinetic energy and backup power; using the minimum frequency change as the target function so that the network weight corresponds to the variable of the system state, and using the output of the neuron as the parameter of the PD controller; obtaining the change rule of the parameter by combining the target function expression and the energy function expression, and further seeking for the stable output according to the rule. By using the PSCAD (power system computer aided design)/EMTDC (electromagnetic transient in DC system) simulation platform, the method of the invention simulates and researches the neural joint control strategy in details, and compares with the traditional frequency control strategy; and the result shows that the Hopfield neural joint control has a better primary frequency modulation control effect.

Description

Double-fed fan motor unit primary frequency modulation combination control method

Technical field

The present invention relates to double-fed fan motor unit primary frequency modulation control method, the control raising power system frequency stability in order to the research and utilization wind-driven generator, belong to technical field of wind power generation.

Background technology

In recent years, the exploitation of regenerative resource more and more are subject to the extensive attention of countries in the world, and wind energy, as a kind of new forms of energy of sustainable development, with its nonstaining property and recyclability, becomes a kind of rising green energy resource.Wind energy is compared with tidal energy, solar energy equal energy source, and its utilance is the highest, has the competitiveness that can compare with the conventional power generation usage mode.Therefore the wind power generation industry of China has obtained development rapidly in recent years, and installed capacity of wind-driven power improves constantly.

In wind generating technology, wind generator system based on variable speed constant frequency doubly-fed induction machine (DFIG) is compared and is had obvious advantage with traditional constant-speed and constant-frequency wind generator system based on common asynchronous generator, therefore becomes gradually the mainstream model of wind-power market.Because the double-fed fan motor unit control system makes the decoupling zero of decoupling zero, rotating speed and the system frequency of its mechanical output and system electromagnetic power, the wind mill rotor mechanical part can't change and to make response fast and effectively system frequency, so its rotation function does not almost have the contribution of system inertia.A large amount of double-fed fan motor unit access electrical network Substitute For Partial conventional power generation usage units, the inertia of whole system will inevitably be affected and relatively reduce.The inertia of known system is relevant with the rate of change that frequency reduces, and when serious frequency reduction accident occurs electrical network, inertia is lower, and system frequency reduces sooner, so the frequency of system is by more difficult control.

Nowadays, increasing Utilities Electric Co. has proposed strict wind farm grid-connected technology guide rule, and the FREQUENCY CONTROL ability is one of wherein important specification requirement.Therefore, require wind power generation can as conventional power generation usage factory, there is the ability that participates in electrical network one secondary frequencies and become an important and urgent task.

The research of double-fed fan motor unit frequency control system mainly comprised to three kinds of methods both at home and abroad: 1) rotor kinetic energy is controlled.Contain a large amount of rotation functions in the rotor of double-fed fan motor unit, by increasing additional frequency control link, control the rotor torque reference value, can realize reducing rotating speed when frequency descends, thereby the kinetic energy discharged in blade provides frequency to support.2) non-firm power is controlled.Need to have certain reserve capacity while being similar to the operation of synchronous generator unit, in order to guarantee the margin of power of speed-changing draught fan, blower fan has to operate at the working point that is not that maximal wind-energy is followed the trail of.Generally can make blower fan unloading operation by controlling propeller pitch angle or regulating power speed curves, make the active power reference value lower than the best power curve, thereby guarantee the primary frequency modulation reserve capacity.When system frequency significantly reduces, thereby reduce propeller pitch angle or move on the optimal power speed curves to increase meritorious output, participate in primary frequency modulation.3) jointly control.Consider rotor kinetic energy and non-firm power control simultaneously.

Traditional control model, comprise that rotor kinetic energy is controlled, non-firm power is controlled and commonly jointly control, all need to set up a system mathematic model effectively, and for DFIG wind-powered electricity generation unit, due to the uncertainty of wind speed and the complexity of power electronics model, model trends towards non-linear and time variation, and the Mathematical Modeling of setting up a set of detailed complete is very difficult; On the other hand, the control parameter of traditional control method generally artificially determines to have very large subjectivity according to experience, and the variation of control parameter is larger on the impact of system control characteristic, so gained control parameter robustness is poor.

Summary of the invention

Technical problem to be solved by this invention is to provide a kind of double-fed fan motor unit control method for frequency, get final product execution control function without accurate Mathematical Modeling, and it controls parameter is to calculate in real time according to the operating structure of system, has strong adaptability and robustness.

For solving the problems of the technologies described above, the invention provides a kind of double-fed fan motor unit primary frequency modulation combination control method, it is characterized in that, comprise the following steps:

(1) set up respectively rotor kinetic energy control module and non-firm power control module in the double-fed blower fan control system, the input variable of rotor kinetic energy control module and non-firm power control module is frequency departure, the index word that the output variable of rotor kinetic energy control module is torque reference value, the index word that the input variable of non-firm power control module is the propeller pitch angle reference value;

(2) set up the Hopfield neural net, determine the control object system model, and when target function is the skew of electrical network occurrence frequency, system frequency changes minimum, makes the state variable of control object corresponding to network weight, and the parameter using neuronic output as the PD controller;

(3) target function expression formula and standard energy function expression are mapped, derive and obtain the inclined to one side value expression of connection weight matrix and network;

(4) the connection weight matrix and the network that obtain are inputted in the dynamical equation of inclined to one side value matrix substitution Hopfield network, and the nonlinear characteristic of getting neuron output is symmetric form S nonlinear interaction function the Changing Pattern of the controlled device parameter of deriving.

The beneficial effect that the present invention reaches: the double-fed fan motor unit primary frequency modulation combination control method based on the Hopfield neural net of invention, for target adopts the Hopfield neural net, the parameter of controller is optimized to design according to frequency decline is minimum, can more reasonably arranges the associating frequency modulation that rotor kinetic energy is controlled and non-firm power is controlled.When the system occurrence frequency is offset, blower fan can be realized discharging and is stored in the rotation function in rotor blade, and increases meritorious exerting oneself and provide frequency to support, and the frequency that can further alleviate system descends.The Hopfield nerve network controller adopted has been realized parameter adaptive function, is not subject to the impact that external environment changes, the variation can very fast adaptive system parameter etc. occurred.

The accompanying drawing explanation

Fig. 1 is that the Hopfield nerve jointly controls schematic diagram;

Fig. 2 is four machine two simulation of domain systems;

Fig. 3 is neuron controller A and the neuron controller B schematic diagram in PSCAD;

Fig. 4 is the Hopfield neural net schematic diagram in neuron controller A;

Fig. 5 is that under the various strategies of power disturbance, the frequency modulation curve compares schematic diagram;

Fig. 6 is that under the various strategies of wind speed disturbance, the frequency modulation curve compares schematic diagram.

Embodiment

Below in conjunction with accompanying drawing, the present invention is specifically described as follows:

(1) first of technical scheme: consider the control of rotor kinetic energy and non-firm power and control, set up respectively rotor kinetic energy control module and non-firm power control module (seeing accompanying drawing 1) in the double-fed blower fan control system.The core of controlling design is Hopfield neuron controller A and Hopfield neuron controller B, and rotor kinetic energy control module and non-firm power control module input variable are frequency deviation f, and output variable is respectively the index word Δ T of torque reference value _refindex word Δ β with the propeller pitch angle reference value _ref, purpose is to optimize the frequency characteristics control jointly controlled.

(2) second portion of technical scheme: set up the Hopfield neural net, determine the control object system model, and make network weight known, corresponding to the state variable of control object, using neuronic output as the PD controller parameter, determine that target function is the frequency change minimum.

Because converters is controlled fast to electrical power, can think between torque reference value index word that wind-powered electricity generation unit medium frequency controlling unit integral body is regulated and propeller pitch angle reference value index word and actual output not dynamically:

$\mathrm{\&Delta;}{T}_{\mathrm{ref}}=-{\mathrm{<figure-callout\; id="1"\; label="K\; f"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">K</figure-callout>}}_f\mathrm{\&Delta;f}-K_{\mathrm{in}1}\frac{\mathrm{df}}{\mathrm{dt}}---\left(1\right)$

$\mathrm{\&Delta;}{\mathrm{\&beta;}}_{\mathrm{ref}}=-K_{f2}\mathrm{\&Delta;f}-K_{\mathrm{in}2}\frac{\mathrm{df}}{\mathrm{dt}}$

The object that controller A controls can be thought $\left\{\begin{array}{c}\frac{\mathrm{df}}{\mathrm{dt}}=K_A\mathrm{\&Delta;}{T}_{\mathrm{ref}}\\ y=f\end{array}\right.,$ The object that controller B controls can be thought $\left\{\begin{array}{c}\frac{\mathrm{df}}{\mathrm{dt}}=K_B\mathrm{\&Delta;}{\mathrm{\&beta;}}_{\mathrm{ref}}\\ y=f\end{array}\right.,$ Wherein the B parameter of the parameter A of neuron controller A and neuron controller B can obtain by simple rotor kinetic energy control, with the operational factor of simple non-firm power control, carrying out simple identification.Wherein f is system frequency, and Δ f is frequency departure, Δ T _reffor the torque reference value index word, Δ β _reffor propeller pitch angle reference value index word, K _f1, K _in1be respectively proportional control factor and the differential adjustment factor of neuron controller A, K _f2, K _in2be respectively proportional control factor and the differential adjustment factor of neuron controller B, the output that y is controller, K _afor the parameter of the control object of controller A, K _bparameter for the control object of controller B.

To the system model y (t) of the ordinary circumstance of controller A, B=C[Ax (t)+Bu (t)] to discuss, the conclusion of deriving like this has generality, and model is as follows:

$\left\{\begin{array}{c}\frac{\mathrm{dx}\left(t\right)}{\mathrm{dt}}=\mathrm{Bu}\left(t\right)\\ y\left(t\right)=C[\mathrm{Ax}\left(t\right)+\frac{\mathrm{dx}\left(t\right)}{\mathrm{dt}}]\end{array}\right.---\left(2\right)$

In formula, y (t) is the system output variable, and x (t) is system state variables, and u (t) is the system input variable, and A, B, C are the undetermined coefficient of descriptive system model.

Controller adopts the PD controller, and it is output as:

$u\left(t\right)=k_{p}e\left(t\right)+k_d\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}---\left(3\right)$

In formula, k _p, k _dfor controller proportional control factor and differential adjustment factor, e (t) is parallel algorithm, and r (t) is rated frequency, gets constant; Y (t) is current frequency), that is:

e(t)＝r(t)-y(t)(4)

The target function of control system is got:

$E=\frac{1}{2}{e}^2\left(t\right)---\left(5\right)$

Formula (2), (3), (4) substitution formula (5) are launched:

$E\left(t\right)=\frac{1}{2}{e}^2\left(t\right)=\frac{1}{2}{\{r\left(t\right)-C[\mathrm{Ax}\left(t\right)+\mathrm{Bu}\left(t\right)\left]\right\}}^2$

$=\frac{1}{2}{\{r\left(t\right)-C[\mathrm{Ax}\left(t\right)+B(k_p\left(t\right)e\left(t\right)+k_d\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}})\left]\right\}}^2---\left(6\right)$

Because the parameter that two controllers are arranged is P, D parameter, getting Hopfield network output neuron number is 2, and CHNN is comprised of two neurons, and CHNN is output as at t constantly:

V(t)＝[V ₁(t)，V ₂(t)] ^T＝[k _p(t)，k _d(t)] ^T(7)

The third part of technical scheme: target function expression formula and standard energy function expression are mapped, derive and obtain connection weight matrix and the inclined to one side value expression of network.

Make V ₁=k _p, V ₂=k _d, by E _n(t) launch:

$E_N\left(t\right)=-\frac{1}{2}[w_{11}\left(t\right)k_p\left(t\right)k_p\left(t\right)+w_{12}\left(t\right)k_p\left(t\right)k_d\left(t\right)+w_{21}\left(t\right)k_d\left(t\right)k_p\left(t\right)---\left(8\right)$

$+w_{22}\left(t\right)k_d\left(t\right)k_d\left(t\right)]-k_p\left(t\right)I_1\left(t\right)-k_d\left(t\right)I_2\left(t\right)$

V wherein ₁, V ₂be respectively the output of neuron 1 and neuron 2, w _ijfor the connection weights of neuron i and neuron j, I ₁(t), I ₂(t) be respectively the threshold value of neuron 1 and neuron 2.

When the Hopfield network during in poised state, energy function minimum, w ₁₂=w ₂₁, now:

$\frac{\&PartialD;E_N}{\&PartialD;k_p}=\frac{\&PartialD;E}{\&PartialD;k_p}=0---\left(9\right)$

$\frac{{\&PartialD;E}_N}{\&PartialD;k_d}=\frac{\&PartialD;E}{\&PartialD;k_d}=0$

E wherein _nfor the standard energy function of network, E is target function, and e is system deviation, the x system mode, and r is constant input.

By $\frac{{\&PartialD;E}_N}{\&PartialD;k_p}=\frac{\&PartialD;E}{\&PartialD;k_p}=0$ :

$\frac{{\&PartialD;E}_N}{\&PartialD;k_p}=\frac{1}{2}({-2w}_{11}{k}_p-2w_{12}{k}_d)-{\mathrm{<figure-callout\; id="1"\; label="I"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">I</figure-callout>}}_1=0---\left(10\right)$

$\frac{\&PartialD;E}{\&PartialD;k_p}=\frac{1}{2}({2B}^2{C}^2{e}^2{k}_p+{2B}^{2}C{k}_{d}e\frac{\mathrm{de}}{\mathrm{dt}}+2{\mathrm{ABC}}^2\mathrm{xe}-2\mathrm{BCre})=0---\left(11\right)$

By top two formulas, obtained:

ω ₁₁＝-B ²C ²e ²， ${\mathrm{\&omega;}}_{12}={\mathrm{\&omega;}}_{21}=-2B^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}},---\left(12\right)$

I ₁＝-2ABC ²ex+2BCre

Same reason $\frac{{\&PartialD;E}_N}{\&PartialD;k_d}=\frac{\&PartialD;E}{\&PartialD;k_d}=0$ :

${\mathrm{\&omega;}}_{22}=-B^2{C}^2{\left(\frac{\mathrm{de}}{\mathrm{dt}}\right)}^2,$ ${\mathrm{\&omega;}}_{12}={\mathrm{\&omega;}}_{21}=-2B^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}},$ (13)

$I_2=-2{\mathrm{ABC}}^2\frac{\mathrm{de}}{\mathrm{dt}}x+2\mathrm{BCr}\frac{\mathrm{de}}{\mathrm{dt}}$

Inclined to one side value (threshold value) I that obtains connection weight matrix W and network by top derivation is as follows:

$W=-\left[\begin{array}{cc}{\left[\mathrm{CBe}\left(t\right)\right]}^2& 2B^2{C}^{2}e\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}\\ 2B^2{C}^{2}e\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}& {\left[\mathrm{CB}\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}\right]}^2\end{array}\right]---\left(14\right)$

$I={\left[\begin{array}{c}-2A{\mathrm{BC}}^{2}e\left(t\right)x\left(t\right)+2\mathrm{BCr}\left(t\right)e\left(t\right)\\ -2{\mathrm{ABC}}^2\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}x\left(t\right)+2\mathrm{BCr}\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}\end{array}\right]}^T---\left(15\right)$

(3) the 4th part of technical scheme: will obtain connection weight matrix and network and input in the dynamical equation of inclined to one side value matrix substitution Hopfield network, and the nonlinear characteristic of getting neuron output is symmetric form S nonlinear interaction function, the Changing Pattern of the controlled device parameter of deriving.

The dynamical equation of standard Hopfield network is:

$\left\{\begin{array}{c}{C}_i\frac{{\mathrm{du}}_i}{\mathrm{dt}}=\underset{j}{\mathrm{\&Sigma;}}{w}_{\mathrm{ij}}{V}_j+I_i\\ V_i=f\left(u_i\right)\end{array}\right.---\left(16\right)$

Get C _i=1.0, required W and I substitution above formula are obtained:

$\frac{{\mathrm{<figure-callout\; id="1"\; label="du"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">du</figure-callout>}}_1}{\mathrm{dt}}=w_{11}{\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">V</figure-callout>}}_1+w_{12}{V}_2+{\mathrm{<figure-callout\; id="1"\; label="I"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">I</figure-callout>}}_1$

$=-B^2{C}^2{e}^{2}g\left(u_1\right)+2B^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}}g\left(u_2\right)-2{\mathrm{ABC}}^2\mathrm{ex}+2\mathrm{BCre}---\left(17\right)$

$\frac{{\mathrm{du}}_2}{\mathrm{dt}}=w_{21}{\mathrm{<figure-callout\; id="1"\; label="V"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">V</figure-callout>}}_1+w_{22}{V}_2+I_2$

$=2B^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}}g\left(u_1\right)-B^2{C}^2{\left(\frac{\mathrm{de}}{\mathrm{dt}}\right)}^{2}g\left(u_2\right)-2{\mathrm{ABC}}^2\frac{\mathrm{de}}{\mathrm{dt}}x+2\mathrm{BCr}\frac{\mathrm{de}}{\mathrm{dt}}$

W wherein _ijbe i neuron and j the neuronic weights that are connected, I _i, u _i, V _ibe respectively i neuronic inclined to one side value, quantity of state and output, f (u _i) be action function.

The nonlinear characteristic of getting neuron output is symmetric form S nonlinear interaction function (gain K):

$g\left(u_i\right)=K_i\frac{{1-e}^{-{\mathrm{\&beta;}}_i{u}_i}}{1+e^{-{\mathrm{\&beta;}}_i{u}_i}}=\frac{{2\mathrm{<figure-callout\; id="1"\; label="K\; i"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">K</figure-callout>}}_i}{1+e^{-{\mathrm{\&beta;}}_i{u}_i}}-K_i,i=\mathrm{1,2}---\left(18\right)$

G (u wherein _i) be action function, K _ifor gain, β _ifor parameter.

The actual of network is output as:

k _p＝g(u ₁)

(19)

k _d＝g(u ₂)

Due to $1+e^{-{\mathrm{\&beta;}}_1{\mathrm{<figure-callout\; id="1"\; label="u"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">u</figure-callout>}}_1}=\frac{{2K}_1}{k_p+{\mathrm{<figure-callout\; id="1"\; label="K"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">\; <figure-callout\; id="1"\; label="K"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">K</figure-callout>\; </figure-callout>}}_1},$ $1+e^{-{\mathrm{\&beta;}}_2{u}_2}=\frac{{2K}_2}{k_d+K_2},$ Have:

$\frac{{\mathrm{dk}}_p}{{\mathrm{du}}_1}=\frac{-2K_1{e}^{-{\mathrm{\&beta;}}_1{u}_1}\left({-\mathrm{\&beta;}}_1\right)}{{(1+e^{-{\mathrm{\&beta;}}_1{u}_1})}^2}=2{\mathrm{\&beta;}}_1{K}_1\frac{K_1-k_p}{k_p+K_1}\frac{{(k_p+K_1)}^2}{{\left({2K}_1\right)}^2}=\frac{{\mathrm{\&beta;}}_1({{\mathrm{<figure-callout\; id="1"\; label="K"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">K</figure-callout>}}_1}^2-{k_p}^2)}{2{\mathrm{<figure-callout\; id="1"\; label="K"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">K</figure-callout>}}_1}$

$\frac{{\mathrm{dk}}_p}{\mathrm{dt}}=\frac{{\mathrm{dk}}_{\mathrm{<figure-callout\; id="1"\; label="p\; d\; u"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">p</figure-callout>}}}{d{u}_{}}\frac{{\mathrm{du}}_1}{\mathrm{dt}}$

(20)

$=\frac{{\mathrm{\&beta;}}_1({{\mathrm{<figure-callout\; id="1"\; label="K"\; filenames="HDA0000136750190000013.png,HDA0000136750190000021.png"\; state="\{\{state\}\}">K</figure-callout>}}_1}^2-{k_p}^2)}{2K_1}(-B^2{C}^2{e}^{2}g\left(u_1\right)+2B^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}}g\left(u_2\right)-2{\mathrm{ABC}}^2\mathrm{ex}+2\mathrm{BCre})$

In like manner can obtain:

$\frac{{\mathrm{dk}}_d}{{\mathrm{du}}_2}=\frac{{\mathrm{\&beta;}}_2({K_2}^2-{k_d}^2)}{2K_2}---\left(21\right)$

$\frac{{\mathrm{dk}}_d}{\mathrm{dt}}=\frac{{\mathrm{dk}}_d}{d{u}_2}\frac{{\mathrm{du}}_2}{\mathrm{dt}}$

$=\frac{{\mathrm{\&beta;}}_2({K_2}^2-{k_d}^2)}{2K_2}({2B}^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}}g{\left(u_1\right)-B^2{C}^2\left(\frac{\mathrm{de}}{\mathrm{dt}}\right)}^{2}g\left(u_2\right)-2{\mathrm{ABC}}^2\frac{\mathrm{de}}{\mathrm{dt}}x+2\mathrm{BCr}\frac{\mathrm{de}}{\mathrm{dt}})---\left(22\right)$

Solve differential equation (20) and formula (22), the k after can being optimized _p, k _dthereby, realize adjusting of PD parameter.

For example adopt 4 machine 2 regional models of KUNDUR, transformation forms analogue system of the present invention control strategy in literary composition is carried out to simulating, verifying.As shown in Figure 2, analogue system comprises 4 of synchronous machines, and power output is respectively 120MW, 120MW, 124MW, 120MW; The wherein otherness of each typhoon group of motors has been ignored in the double-fed fan motor field, and with the equivalence of a double-fed blower fan, gross output is 80MW; System loading L1, L2 are respectively 206MW, 342MW.System is when 5s, and synchronous machine G2 is because the step-out fault is out of service, or disturbance appears in the wind speed of simulating area, and by rated wind speed, 12m/s becomes 10m/s, analyzes now system frequency situation of change.Build Simulation Model in the PSCAD platform, respectively to rotor kinetic energy control, non-firm power is controlled, jointly control, the Hopfield nerve jointly controls and carries out simulation analysis, system frequency modulation characteristic difference under each control strategy relatively.

Set up Hopfield neuron controller A and Hopfield neuron controller B module as shown in Figure 3 in PSCAD.Two inside modules all comprise two neuronic Hopfield neural nets, take neuron controller A as example, as shown in Figure 4, weights and threshold value by the input of the state of system corresponding to network, neuronic output is as the parameter of controller, the solution procedure of nonlinear differential equation completes automatically, and wherein the sfunction module is S type function module.

By simulation result, as Fig. 5,6, relatively can find out, double-fed fan motor unit primary frequency modulation combination control method based on the Hopfield neural net proposed by the invention can be realized parameter adaptive function, thereby more can reasonable arrangement rotor kinetic energy control and the frequency modulation ratio of non-firm power control, that compares has adaptive advantage in jointly controlling of permanent parameter, descend so can further alleviate the frequency of system, improve low-limit frequency and have and more preferably control effect improving double-fed fan motor unit primary frequency modulation characteristic.

Claims (1)

1. a double-fed fan motor unit primary frequency modulation combination control method, is characterized in that, comprises the following steps:

(1) set up respectively rotor kinetic energy control module and non-firm power control module in the double-fed blower fan control system, the input variable of rotor kinetic energy control module and non-firm power control module is frequency departure, the index word that the output variable of rotor kinetic energy control module is torque reference value, the index word that the input variable of non-firm power control module is the propeller pitch angle reference value;

(2) set up the Hopfield neural net, determine the control object system model, model is as follows:

$\left\{\begin{array}{c}\frac{\mathrm{dx}\left(t\right)}{\mathrm{dt}}=\mathrm{Bu}\left(t\right)\\ y\left(t\right)=C[\mathrm{Ax}\left(t\right)+\frac{\mathrm{dx}\left(t\right)}{\mathrm{dt}}]\end{array}\right.---\left(2\right)$

In formula, y (t) is the system output variable, and x (t) is system state variables, and u (t) is the system input variable, and A, B, C are the undetermined coefficient of descriptive system model;

And when target function is the skew of electrical network occurrence frequency, system frequency changes minimum, makes the state variable of control object corresponding to network weight, and the parameter using neuronic output as the PD controller, and target function is:

$E\left(t\right)=\frac{1}{2}{e}^2\left(t\right)=\frac{1}{2}{\{r\left(t\right)-C[\mathrm{Ax}\left(t\right)+\mathrm{Bu}\left(t\right)\left]\right\}}^2$ $=\frac{1}{2}{\{r\left(t\right)-C[\mathrm{Ax}\left(t\right)+B(k_p\left(t\right)e\left(t\right)+k_d\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}})\left]\right\}}^2---\left(6\right)$

K _p, k _dfor controller proportional control factor and differential adjustment factor, e (t) is parallel algorithm, and r (t) is rated frequency, gets constant;

(3) target function expression formula and standard energy function expression are mapped, derive and obtain the inclined to one side value expression of connection weight matrix and network, the inclined to one side value I of connection weight matrix W and network:

$W=-\left[\begin{array}{cc}{\left[\mathrm{CBe}\left(t\right)\right]}^2& {2B}^2{C}^{2}e\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}\\ {2B}^2{C}^{2}e\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}& {\left[\mathrm{CB}\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}\right]}^2\end{array}\right]---\left(14\right)$

$I={\left[\begin{array}{c}-2\mathrm{AB}{C}^{2}e\left(t\right)x\left(t\right)+2\mathrm{BCr}\left(t\right)e\left(t\right)\\ -2\mathrm{AB}{C}^2\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}x\left(t\right)+2\mathrm{BCr}\left(t\right)\frac{\mathrm{de}\left(t\right)}{\mathrm{dt}}\end{array}\right]}^T---\left(15\right)$

(4) the connection weight matrix and the network that obtain are inputted in the dynamical equation of inclined to one side value matrix substitution Hopfield network, and the nonlinear characteristic of getting neuron output is symmetric form S nonlinear interaction function:

$g\left(u_i\right)=K_i\frac{1-e^{{\mathrm{\&beta;}}_i{u}_i}}{{1+e}^{-{\mathrm{\&beta;}}_i{u}_i}}=\frac{{2K}_i}{{1+e}^{{\mathrm{\&beta;}}_i{u}_i}}-K_i$ i=1,2(18)

G (u wherein _i) be action function, K _ifor gain, β _ifor parameter; u _ibe i neuronic quantity of state;

The Changing Pattern of the controlled device parameter of deriving:

$\frac{{\mathrm{dk}}_p}{\mathrm{dt}}=\frac{{\mathrm{dk}}_p}{{\mathrm{du}}_1}\frac{{\mathrm{du}}_1}{\mathrm{dt}}$

$=\frac{{\mathrm{\&beta;}}_1({K_1}^2-{k_p}^2)}{{2K}_1}({-B}^2{C}^2{e}^{2}g\left(u_1\right)+{2B}^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}}g\left(u_2\right)-2\mathrm{AB}{C}^2\mathrm{ex}+2\mathrm{BCre})$

(20)

$\frac{{\mathrm{dk}}_d}{\mathrm{dt}}=\frac{{\mathrm{dk}}_d}{{\mathrm{du}}_2}\frac{{\mathrm{du}}_2}{\mathrm{dt}}$

$=\frac{{\mathrm{\&beta;}}_2({K_2}^2-{k_d}^2)}{{2K}_2}({2B}^2{C}^{2}e\frac{\mathrm{de}}{\mathrm{dt}}g\left(u_1\right){-B}^2{C}^2{\left(\frac{\mathrm{de}}{\mathrm{dt}}\right)}^{2}g\left(u_2\right)-2\mathrm{AB}{C}^2\frac{\mathrm{de}}{\mathrm{dt}}x+2\mathrm{BCr}\frac{\mathrm{de}}{\mathrm{dt}})$

(22)

Solve differential equation (20) and formula (22), the controller proportional control factor after being optimized and differential adjustment factor k _p, k _dthereby, realize adjusting of PD parameter.

Images