CN117318553B - Low-wind-speed permanent magnet direct-driven wind turbine control method based on TD3 and Vienna rectifier - Google Patents
Low-wind-speed permanent magnet direct-driven wind turbine control method based on TD3 and Vienna rectifier Download PDFInfo
- Publication number
- CN117318553B CN117318553B CN202311276120.XA CN202311276120A CN117318553B CN 117318553 B CN117318553 B CN 117318553B CN 202311276120 A CN202311276120 A CN 202311276120A CN 117318553 B CN117318553 B CN 117318553B
- Authority
- CN
- China
- Prior art keywords
- permanent magnet
- network
- actor
- wind
- output
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 63
- 238000012549 training Methods 0.000 claims abstract description 50
- 230000006870 function Effects 0.000 claims abstract description 29
- 238000011217 control strategy Methods 0.000 claims abstract description 23
- 230000007613 environmental effect Effects 0.000 claims abstract description 19
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 18
- 230000002787 reinforcement Effects 0.000 claims abstract description 18
- 230000008859 change Effects 0.000 claims abstract description 14
- 238000013461 design Methods 0.000 claims abstract description 11
- 230000001360 synchronised effect Effects 0.000 claims description 63
- 230000009471 action Effects 0.000 claims description 34
- 239000003795 chemical substances by application Substances 0.000 claims description 23
- 239000003990 capacitor Substances 0.000 claims description 16
- 230000008569 process Effects 0.000 claims description 16
- 230000002457 bidirectional effect Effects 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 230000003068 static effect Effects 0.000 claims description 6
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000004907 flux Effects 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000010355 oscillation Effects 0.000 claims description 3
- 230000000452 restraining effect Effects 0.000 claims description 3
- 238000000638 solvent extraction Methods 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000007935 neutral effect Effects 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 238000005457 optimization Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 230000007774 longterm Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000004804 winding Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P21/00—Arrangements or methods for the control of electric machines by vector control, e.g. by control of field orientation
- H02P21/0003—Control strategies in general, e.g. linear type, e.g. P, PI, PID, using robust control
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
- H02M1/00—Details of apparatus for conversion
- H02M1/0048—Circuits or arrangements for reducing losses
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
- H02M1/00—Details of apparatus for conversion
- H02M1/12—Arrangements for reducing harmonics from ac input or output
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P21/00—Arrangements or methods for the control of electric machines by vector control, e.g. by control of field orientation
- H02P21/22—Current control, e.g. using a current control loop
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P2101/00—Special adaptation of control arrangements for generators
- H02P2101/15—Special adaptation of control arrangements for generators for wind-driven turbines
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02P—CONTROL OR REGULATION OF ELECTRIC MOTORS, ELECTRIC GENERATORS OR DYNAMO-ELECTRIC CONVERTERS; CONTROLLING TRANSFORMERS, REACTORS OR CHOKE COILS
- H02P2103/00—Controlling arrangements characterised by the type of generator
- H02P2103/20—Controlling arrangements characterised by the type of generator of the synchronous type
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Control Of Eletrric Generators (AREA)
Abstract
The invention discloses a low-wind-speed permanent magnet direct-driven wind turbine control method based on a TD3 and a Vienna rectifier, and belongs to the field of wind power. The machine side converter adopts a Vienna rectifier and adopts a PID design rotation speed tracking controller; adopting a deep reinforcement learning TD3 algorithm to design a current tracking and midpoint voltage balance controller: constructing an intelligent body based on a TD3 algorithm according to the environmental state of the wind turbine, designing a reward function according to the wind speed, training the intelligent body, transplanting a trained intelligent body model into a main control chip of a side converter, establishing an actual control system based on the main control chip, and realizing current tracking and neutral point voltage balance control; and a feedforward decoupling control strategy and an SVPWM control strategy are adopted to control the work of the generator. The invention has strong self-adaptive capability and robustness to wind speed fluctuation and self-structural parameter change of the unit, low stator current harmonic ratio and low system power consumption, and ensures the effective control and efficient operation of the low wind speed permanent magnet direct drive wind turbine.
Description
Technical Field
The invention relates to a control method, in particular to a control method of a low-wind-speed permanent magnet direct-driven wind turbine based on a TD3 and Vienna rectifier, and belongs to the technical field of wind power.
Background
Permanent magnet synchronous wind generators (PMSG) have been widely used in wind power generation systems, whose converter system consists of a Machine Side Converter (MSC) and a grid side converter.
In a low-wind-speed wind power plant, the turbulence is strong, the wind direction is changeable, the wind speed fluctuation is large, and the uncertainty of external interference of a wind turbine is often caused; moreover, the long-term operation of the wind turbine can also cause the parameter of the motor to change, and the parameters all provide great challenges for the control of the low-wind-speed permanent magnet direct-driven wind turbine. On the other hand, for low wind speed power generation, the self loss of the whole wind turbine generator is necessarily required to be low, but the reduction of the loss of the wind turbine generator and the converter thereof is one of the difficulties. Therefore, stable control of the low-wind-speed permanent magnet direct-driven wind turbine generator is realized, and effective reduction of PMSG stator current harmonic and self-loss of a machine side converter is one of important measures for realizing efficient operation of the wind turbine generator, so that the topology structure and control scheme of the MSC are very important.
Currently, MSC topologies are classified as VSC and VIENNA (VIENNA) rectifiers. The latter is increasingly gaining attention in the wind power field because of its low voltage stress, good power quality and no dead time. However, the vienna rectifier needs to balance the midpoint voltage, and as the wind speed increases, the midpoint voltage imbalance becomes more and more severe, and if the control is not good, the vienna rectifier will eventually be damaged.
For a direct-drive permanent magnet wind generating set (PGVR) adopting a VIENNA rectifier, a currently commonly used control method comprises PID control and Model Predictive Control (MPC). The PID control has high reliability and is easy to program, but because the parameters are fixed, the control effect is difficult to be optimized at different wind speeds. The MPC realizes the optimal performance of the PGVR machine set by designing a proper cost function. However, the MPC selects the action of the bidirectional switch tube at the next moment by comparing the cost of different switch states, so the MPC needs to have a modulation function, which requires that the MPC be implemented by hardware with higher calculation speed.
The deep reinforcement learning control algorithm effectively combines the perception capability of deep learning and the decision capability of reinforcement learning, can control according to the input information, and is an artificial intelligent control method which is closer to a human thinking mode. The reward function of Deep Reinforcement Learning (DRL) is similar to the cost function of MPC and therefore has the same advantages as MPC. However, DRL control corresponds to a special PID control where parameters can vary with the external environment and no modulation is required, so its computational cost is far lower than that of MPC. In addition, the control effect of the MPC depends on an accurate mathematical model and equipment parameters, and the optimal action of the DRL control is obtained by searching and optimizing in random and uncertain environments, so that the DRL control has strong robustness to internal parameter mismatch and external environment disturbance. This robustness is very important for PGVR control systems because their performance is affected by wind randomness and the greater the wind speed, the greater the impact. As one of the most advanced training algorithms of DRL, a dual-delay depth deterministic strategy gradient (TD 3) is used for power electronics and motor control, and the result shows that the TD3 control has a better control effect in an uncertain environment, so that the method is particularly suitable for a low-wind-speed permanent magnet direct-driven wind turbine generator with unknown time-varying interference (such as fluctuation and uncertainty of wind speed and wind direction), change of self structural parameters after long-term operation, and nonlinear and strong coupling characteristics. However, the application research of the deep reinforcement learning control on the aspect of the low-wind-speed permanent magnet direct-driven wind turbine generator is very little at present.
Disclosure of Invention
The main purpose of the invention is that: aiming at the defects of the prior art, the invention provides a low-wind-speed permanent magnet direct-drive wind turbine generator control method based on a deep reinforcement learning TD3 algorithm and a Vienna rectifier, wherein the Vienna rectifier is adopted by a machine side converter, and the optimal control of multi-objective tasks such as reducing the current harmonic content of a stator side, reducing the self-loss of the machine side converter, balancing the midpoint voltage and the like is realized according to the operation characteristics of PGVR under different wind speeds by adopting the deep reinforcement learning TD3 algorithm, so that the effective control and the efficient operation of the low-wind-speed permanent magnet synchronous wind turbine generator and the machine side converter thereof are ensured.
In order to achieve the above purpose, the low wind speed permanent magnet direct drive wind turbine generator set of the present invention comprises: permanent magnet synchronous wind power generator, machine side converter, net side converter, etc.; one end of the machine side converter is connected with the stator of the permanent magnet synchronous wind driven generator, and the other end of the machine side converter is connected with the grid side converter.
The machine side converter is a Vienna rectifier, and the Vienna rectifier comprises three filter inductors, three bidirectional switch units and two output capacitors; one end of each of the three filter inductors is connected with a stator three-phase winding of the permanent magnet synchronous wind driven generator respectively; the two output capacitors are connected in series, wherein the output capacitor connected with the positive electrode of the direct current output end of the Vienna rectifier is called an upper capacitor, and the output capacitor connected with the negative electrode of the direct current output end of the Vienna rectifier is called a lower capacitor.
The invention discloses a low-wind-speed permanent magnet direct-driven wind turbine control method based on a TD3 and a Vienna rectifier, which comprises the following steps of:
step 1, adopting PID control to design an outer ring rotating speed tracking controller of the machine side converter, wherein the output of the outer ring rotating speed tracking controller is the q-axis component i of the stator current of the permanent magnet synchronous wind driven generator q Reference value i of (2) q * 。
And 2, designing an inner loop current tracking and midpoint voltage balance controller of the machine side converter by adopting a deep reinforcement learning TD3 algorithm, wherein the inner loop current tracking and midpoint voltage balance controller is specifically as follows:
21 Constructing an intelligent agent based on a TD3 algorithm according to the environmental state of the low-wind-speed permanent magnet direct-driven wind turbine generator:
the environmental conditions include wind speed v w The actual rotating speed n (t) of the permanent magnet synchronous wind driven generator and the reference value n (t) thereof * Dq-axis component i of stator current of said permanent magnet synchronous wind power generator d I d Reference value i of (2) d * 、i q I q Reference value i of (2) q * Upper capacitance voltage U of the machine side converter 1 And lower capacitance voltage U 2 The method comprises the steps of carrying out a first treatment on the surface of the The environmental state is recorded as S t Then there is S t ＝[v w n(t)n(t) * i d i d * i q i q * U 1 U 2 ]。
The intelligent agent comprises an Actor network structure and a Critic network structure, contains 6 depth neural networks in total, and specifically comprises the following components:
the Actor network structure consists of a strategy network Actor main network and a Target network Actor Target network thereof.
The input of the Actor main network is the environmental state S t Adding action search noise N to the output of the Actor main network t Then obtain action A t (S t θ), where θ is a weight parameter of the Actor main network, and motion exploration noise N t Obeying normal distribution;
the input of the Actor Target network is the environmental state S at the next moment t+1 Adding a Target smooth noise epsilon to the output of the Actor Target network to obtain an action estimated value A of the next moment t+1 (S t+1 |θ m ) Wherein θ m Epsilon obeys normal distribution for the weight parameters of the Actor Target network;
the Critic network structure consists of two sets of evaluation networks for cost function estimation, namely Critic main networkAnd its Target network Critic Target network +.>Where s, a are general representations of motion input values to a state, deep neural network, θ m Respectively the weight parameters of the Actor network and the Target network Actor network of the Actor network; /> The weight parameters of the Critic main network and the Target network Critic Target network are respectively.
The Critic main network inputs S according to the current Actor main network t And output action A t (S t I θ), respectively output a cost functionThe Critic Target network is based on the Actor Target networkInput S t+1 And outputting the motion estimation value A t+1 (S t+1 |θ m ) Output objective cost function respectively>
22 Design a bonus function):
the bonus function R t The design is as follows:
wherein r is 1 、r 2 、r 3 The weight coefficients are used for restraining stator current harmonic waves, midpoint voltage balance and direct current output voltage respectively and are used for carrying out priority division on a plurality of control tasks, and the sum of the weight coefficients is equal to 1; u (U) dc A direct current output voltage for the vienna rectifier; h is a 1 ～h 4 Respectively represent i q Actual value of (2) and reference value i thereof q * I is the difference of (i) d Actual value of (2) and reference value i thereof d * Is equal to the difference between the voltage at the midpoint, the unbalance of the voltage at the midpoint, and the DC output voltage U dc Is set to the reference value of (a).
The bonus function R t Adopting a variable weight structure, and determining the weight according to the wind speed: 1) When the wind speed v w Less than the critical wind speed v wb When the loss power of PGVR is mainly caused by stator current harmonic wave, the primary task of the control strategy is to reduce the harmonic wave, and r is set at the moment 1 Maximum. 2) When the wind speed v w Greater than v wb In this case, the primary task of the control strategy is to increase the DC output voltage U dc Increase the DC output power P dc At this time, set r 2 Maximum. 3) The mid-point voltage oscillation suppression is required to be carried out in the whole variable power wind speed interval, so r is set 3 And the wind speed is changed slightly.
23 Training the agent by adopting a deep reinforcement learning TD3 algorithm:
A1. initializing the network weight parameters of the intelligent agent and setting the relevant parameters of the intelligent agent training, wherein the method comprises the following steps:
initializing the Actor main network, the weight parameter theta of the Critic main network,Target network Actor Target network corresponding to Target network Actor Target network, weight parameter theta of Critic Target network m 、/>k＝1,2；
Setting a learning rate mu of the Actor main network;
setting a discount factor gamma of deep reinforcement learning;
setting a maximum training round number N, a continuous training round number N, a maximum step number T in each training round, and updating frequencies D of a Target network, an Actor Target network and a Critic Target network;
an experience database is set up and initialized.
A2. Based on the current state S t The output of the Actor network adds action search noise N t Obtain action A t (S t |θ), there are:
A t (S t |θ)＝π(S t |θ)+N t (2)
wherein pi (S) t I θ) is the output of the Actor network.
A3. After each step of action is executed, the next environmental state S is obtained t+1 Simultaneously calculating R according to formula (1) t The obtained experience (S t ,A t ,R t ,S t+1 ) Stored in an experience database.
A4. M small batches of samples are randomly and uniformly sampled in an empirical database (S ti ,A ti ,R ti ,S ti+1 ) I.e. M empirical data sets, where i=1, 2,3, …, M.
A5. Updating weight parameters of Critic master network in a manner that minimizes errorsI.e.
Wherein M represents the size of the small batch sample; y is Y i Representing a desired objective cost function, defined as:
wherein, gamma is a discount factor; epsilon represents the target smooth noise, which is a truncated normal distribution noise, i.e
Wherein c is a truncation factor and σ is a standard deviation.
A6. If the training step number is a multiple of the update frequency D of the Critic Target network, the step A7 is entered, otherwise, the step A9 is entered.
A7. Updating θ by gradient upward for the desired return J, i.e
Wherein mu is learning rate; the expression of J is:
representing J deriving θ:
in the method, in the process of the invention,indicating pi to theta derivative.
A8. The weight parameter theta of the Actor Target network is calculated according to the formula (9) in an exponential smoothing mode m And the Critic Target network weight parameter theta mk And updating, namely:
where τ represents the soft update factor.
A9. Returning to A2 if the training step number does not reach the maximum step number T in each training round; otherwise, the current training round is finished, simultaneously the intelligent body model obtained by the training round is saved, and the average rewarding function value R of the round is calculated av And the rate of change thereof.
A10. If R is av Is continuous with the rate of change of n (n<N) less than 5% of the rounds, ending the whole training process, otherwise entering step a11.
A11. And (3) judging whether the maximum training round number N is reached, if so, ending the whole training process, and if not, returning to the step A2 to enter the next training round to continue training.
24 The trained intelligent body model is transplanted into a main control chip of the machine side converter, an actual current tracking and midpoint voltage balance control system based on the main control chip of the machine side converter is established, and current tracking and midpoint voltage balance control is achieved.
Step 3, generating a driving signal s by adopting a feedforward decoupling control strategy and an SVPWM control strategy a 、s b 、s c And driving three bidirectional switch units of the machine side converter to control the permanent magnet synchronous wind driven generator to work. The specific process is as follows:
31 2) the output A of the intelligent body inner loop current tracking and midpoint voltage balance controller obtained in the step 2 t (S t θ) there are three: output of inner loop of currentValue u d0 、u q0 And the output value z of the midpoint voltage balance ring; will u d0 Decoupling compensation term omega from d-axis feed forward e L s i q Adding to obtain a d-axis component u of the stator voltage of the permanent magnet synchronous wind driven generator d Reference value u of (2) d * The method comprises the steps of carrying out a first treatment on the surface of the Will u q0 Decoupling compensation term-omega from q-axis feedforward e L s i d +ω e ψ f Adding to obtain the q-axis component u of the stator voltage of the permanent magnet synchronous wind driven generator q Reference value u of (2) q * The method comprises the steps of carrying out a first treatment on the surface of the Wherein omega e For the rotor electric angular velocity omega of the permanent magnet synchronous wind power generator e =2pi pn (t), p is the pole pair number of the permanent magnet synchronous wind driven generator, L s Is the stator inductance, ψ f A permanent magnet flux linkage for a rotor of the permanent magnet synchronous wind driven generator;
32 The stator voltage reference value u of the permanent magnet synchronous wind driven generator obtained in the step 31) is obtained d * 、u q * Obtaining a stator voltage reference value u of the permanent magnet synchronous wind driven generator under a three-phase static coordinate system through dq/abc coordinate transformation a * 、u b * 、u c * ；
33 U) will be a * 、u b * 、u c * And the output value z of the midpoint voltage balance ring and the actual value i of the stator current of the permanent magnet synchronous wind driven generator under a three-phase static coordinate system a 、i b 、i c Sending the signal to an SVPWM module;
34 The SVPWM module is based on the wind speed v w Is modulated by the size of the (b), the modulation method is as follows:
B1. according to u a * 、u b * 、u c * And i a 、i b 、i c Partitioning the voltage quality and the current vector respectively;
B2. according to U dc For u a * 、u b * 、u c * Carrying out normalization treatment;
B3. calculating the action time of the space vector, and converting the action time into a modulation wave;
B4. when the wind speed v w Less than the critical wind speed v wb When the method is used, a seven-level switch sequence method in a continuous mode is adopted, and the value of the modulated wave is increased or decreased according to z;
B5. when the wind speed v w V is greater than or equal to v wb When in use, modulating by adopting a five-stage switching sequence method in an intermittent mode;
35 Comparing the modulated wave with a triangular carrier wave to generate a drive signal s a 、s b 、s c And driving a power switch tube of the machine side converter to control the permanent magnet synchronous wind driven generator to work.
Compared with the prior art, the invention has the beneficial effects that the current inner loop, the midpoint voltage balance loop and the rotating speed outer loop of the PID controller of the TD3 controller are used:
1. the limitation of the traditional controller in completing complex tasks is overcome. According to the method, through analyzing the operation characteristics of the PGVR under different wind speeds and designing the rewarding function matched with the modulation method according to the operation characteristics, the optimal control of multi-objective tasks is realized, the current harmonic distortion rate of the stator side of the generator can be effectively reduced, and the power loss of the converter at the machine side is reduced.
2. The influence of wind speed randomness is overcome. According to the invention, through designing a training environment of real wind conditions and actively observing wind speed change, the control system has strong self-adaptive capacity and robustness to wind speed change, and can effectively influence wind speed fluctuation and structural parameter change and uncertainty of a wind turbine generator to control the wind turbine generator, thereby realizing rapid and stable control of the low wind speed permanent magnet synchronous wind driven generator.
Drawings
Fig. 1 is a schematic diagram of a direct-drive permanent magnet wind generating set (PGVR) topology with a vienna rectifier according to the present invention.
Fig. 2 is a schematic diagram of a topology structure of a side converter of the permanent magnet direct drive wind turbine.
Fig. 3 is a schematic diagram of the control principle of the TD3 in the deep reinforcement learning according to the present invention.
Fig. 4 is a control structure block diagram of a permanent magnet synchronous wind driven generator based on a TD3 algorithm.
Fig. 5 is a block diagram of an agent according to the present invention.
Fig. 6 is a training flow chart of the agent according to the present invention.
Fig. 7 is a control structure block diagram of a permanent magnet synchronous wind power generator adopting a conventional PID-based control strategy.
FIG. 8 shows the stator side current i under the PID control strategy according to the present invention q 、i d Midpoint voltage U 0 Tracking effect versus simulation graph.
FIG. 9 shows the current distortion rate THD and the DC side output power P of the machine side converter according to the present invention and the conventional PID-based control strategy dc And a comparison of the efficiency thereof.
Wherein, the wind driven generator is 1-permanent magnet synchronous, the 2-machine side converter is 3-net side converter; 4-agent; a 5-encoder; 6-an outer ring rotation speed tracking controller; 7-an inner loop current tracking and midpoint voltage balance controller based on TD 3; 8-a first PID controller; 9-a second PID controller; 10-an inner loop current tracking controller based on a PID control strategy; 11-a midpoint voltage balance controller based on PID control; 12-a third PID controller.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, the low-wind-speed permanent magnet direct-drive wind turbine generator comprises a permanent magnet synchronous wind power generator 1, a machine side converter 2 and a grid side converter 3, wherein one end of the machine side converter 2 is connected with a stator of the permanent magnet synchronous wind power generator 1, and the other end of the machine side converter is connected with the grid side converter 3; the grid-side converter 3 is connected with an alternating current power grid through a power frequency transformer.
As shown in fig. 2, the machine side converter 2 is a vienna rectifier, which includes three filter inductors (L a 、L b 、L c ) Three bidirectional switch units (S) a 、S b 、S c ) Two output capacitors (C 1 、C 2 ) The method comprises the steps of carrying out a first treatment on the surface of the One end of each of the three filter inductors is respectively fixed with the permanent magnet synchronous wind driven generator 1The sub three-phase windings are connected; the two output capacitors are connected in series, wherein the output capacitor C is connected with the positive electrode of the direct current output end of the Vienna rectifier 1 An output capacitor C connected with the negative electrode of the direct current output end of the Vienna rectifier 2 Referred to as the lower capacitance.
The invention discloses a low-wind-speed permanent magnet direct-driven wind turbine control method based on a TD3 and a Vienna rectifier, which comprises the following steps of:
step 1, as shown in FIG. 3, an outer ring rotation speed tracking controller 6 of the machine side converter 2 is designed by PID control, and the output is the stator current i of the permanent magnet synchronous wind driven generator 1 q Reference value i of (2) q * 。
Step 2, as shown in fig. 3 and 4, adopts a deep reinforcement learning TD3 algorithm to design an inner loop current tracking controller and a midpoint voltage balancing controller of the machine side converter 2, and the main process is as follows:
a part capable of reflecting variable information of the low wind speed permanent magnet synchronous wind power generator 1 and the side converter 2 thereof is defined as an Environment (Environment); the part capable of executing the operation and controlling the operation of the permanent magnet synchronous wind power generator 1 and the machine side converter 2 is defined as an agent 4. As shown in fig. 3, the control principle of the deep reinforcement learning TD3 is schematically shown, and the main process is as follows: the agent 4 is based on the environmental State (State) S t Reward function (Reward) R t Output Action (Action) A t Action A t Then acts on the environment, thereby realizing the optimal environment and further effectively controlling the operation of the permanent magnet direct-driven wind turbine generator. The method comprises the following steps:
21 Constructing an intelligent agent 4 based on a TD3 algorithm according to the environmental state of the low-wind-speed permanent magnet synchronous direct-driven wind turbine generator:
the environmental conditions include wind speed v w The actual rotational speed n (t) (measured by the encoder 5) of the permanent magnet synchronous wind generator 1 and the reference value n (t) * The dq-axis component i of its stator current d Reference value i d * 、i q Reference value i q * Upper capacitance voltage U of machine side converter 2 1 And lower capacitance voltage U 2 ；U 1 And U 2 Is the middle warmerAn input to the point voltage balance controller; the environmental state is recorded as S t Then there is S t ＝[v w n(t)n(t) * i d i d * i q i q * U 1 U 2 ]Wherein, in order to realize the unit power factor operation of the permanent magnet synchronous wind driven generator 1, a d-axis current reference value i is set d * ＝0。
As shown in fig. 5, the agent 4 includes an Actor network structure and a Critic network structure, and contains a total of 6 deep neural networks, specifically:
the Actor network structure consists of a strategy network Actor main network pi (s|theta) and a Target network Actor Target network pi m (s|θ m ) Composition is prepared.
The input to the Actor main network pi (s|θ) is the ambient state S t Adding action search noise N to the output of the Actor main network pi (s|θ) t Then obtain action A t (S t θ), where θ is a weight parameter of the Actor main network, action exploration noise N t Obeying normal distribution;
actor Target network pi m (s|θ m ) The input of (2) is the environmental state S at the next moment t+1 For the Target network pi of the Actor m (s|θ m ) The output of (1) is added with the target smooth noise epsilon to obtain the motion estimation value A of the next moment t+1 (S t+1 |θ m ) Wherein θ m The epsilon obeys normal distribution as a weight parameter of an Actor Target network;
the Critic network structure consists of two evaluation networks for cost function estimation, namely Critic main networkAnd its Target network Critic Target network +.>Where s, a are general representations of motion input values to a state, deep neural network, θ m Respectively the weight parameters of the Actor network and the Target network Actor network of the Actor network; />The weight parameters of the Critic main network and the Target network Critic Target network are respectively.
Critic main network inputs S according to current Actor main network t And output action a of the network t (S t I θ), respectively output a cost functionCritic Target network inputs S according to the Actor Target network t+1 And outputting the motion estimation value A t+1 (S t+1 |θ m ) Output objective cost function respectively>
22 Design a bonus function):
reward function R t The design is as follows:
wherein r is 1 、r 2 、r 3 The weight coefficients are used for restraining stator current harmonic waves, midpoint voltage balance and direct current output voltage respectively and are used for carrying out priority division on a plurality of control tasks, and the sum of the weight coefficients is equal to 1; u (U) dc A direct current output voltage for the vienna rectifier; h is a 1 ～h 4 Respectively represent i q Actual value of (2) and reference value i thereof q * I is the difference of (i) d Actual value of (2) and reference value i thereof d * Is equal to the difference between the voltage at the midpoint, the unbalance of the voltage at the midpoint, and the DC output voltage U dc Since the dimensions of the electrical parameters differ, it is necessary to divide them on the same standard in order to assign weights.
The present invention rewards function R t Adopting a variable weight structure, and determining the weight according to the wind speed: 1) When the wind speed v w Less than the critical wind speed v wb Loss of PGVRThe power is mainly caused by stator current harmonics, the primary task of the control strategy is to reduce the harmonics, where r is set 1 Maximum. 2) When the wind speed v w Greater than v wb In this case, the primary task of the control strategy is to increase the DC output voltage U dc Increase the DC output power P dc At this time, set r 2 Maximum. 3) The mid-point voltage oscillation suppression is required to be carried out in the whole variable power wind speed interval, so r is set 3 And the wind speed is changed slightly.
23 As shown in fig. 5 and 6, training the agent 4 by using the deep reinforcement learning TD3 algorithm:
to solve the continuous state and motion problems, the agent 4 uses an Actor main network pi (s|θ) and a target network pi thereof m (s|θ m ) Fitting a deterministic strategy pi(s); using Critic home networksAnd target network thereofFitting the motion cost function Q (s, a). The training process of the Agent is a process of updating the neural network parameters in one round and is divided into action exploration and iterative optimization.
A1. Initializing the network weight parameters of the intelligent agent 4, and setting the relevant parameters of the training of the intelligent agent 4, including:
initializing a weight parameter theta of an Actor main network and a Critic main network,And copy the weight parameters theta to the corresponding Target network Actor Target network and Critic Target network respectively m 、/>Namely: θ m ←θ、/>k＝1,2；
Setting a learning rate mu of an Actor main network; setting a discount factor gamma of deep reinforcement learning;
setting a maximum number of rounds of training N, a maximum number of steps T in each round of training, and an updating frequency D of a Target network Actor Target network and a Critic Target network; an experience database is set up and initialized.
Action exploration:
A2. based on the current state S t To avoid pi (s|θ) falling into local optima, the output adding action of the Actor network explores noise N t Obtain action A t (S t |θ), there are:
A t (S t |θ)＝π(S t |θ)+N t (2)
A3. after each step of action is executed, the next environmental state S is obtained t+1 Simultaneously calculating R according to formula (1) t The obtained experience (S t ,A t ,R t ,S t+1 ) Stored in an experience database.
Iterative optimization: the iterative optimization is to update each weight parameter theta and theta in the intelligent agent 4 m 、The specific steps are as follows:
A4. m small batches of samples are randomly and uniformly sampled in an empirical database (S ti ,A ti ,R ti ,S ti+1 ) I.e. M empirical data sets, where i=1, 2,3, …, M.
A5. Updating weight parameters of a Critic master network by a Critic optimizer in a manner that minimizes errorsNamely:
wherein M represents the size of the small batch sample; y is Y i Representing a desired objective cost function, defined as:
wherein, gamma is a discount factor; epsilon represents the target smooth noise, which is a truncated normal distribution noise, i.e
Wherein c is a truncation factor and σ is a standard deviation.
A6. If the training step number is an integer multiple of the updating frequency D of the Critic Target network, the step A7 is entered, otherwise, the step A9 is entered.
A7. Updating θ by the Actor optimizer by gradient upward to the desired return J, i.e
Wherein mu is learning rate; the expression of J is:
representing J deriving θ:
in the method, in the process of the invention,indicating pi to theta derivative.
A8. The Actor optimizer and Critic optimizer respectively carry out weight parameter on the Actor Target network according to the formula (9) in an exponential smoothing modeθ m And Critic Target network weight parametersAnd updating, namely:
where τ represents the soft update factor.
In order to make the updating of the neural network more stable, it is prescribed that the updating frequency of pi (s|theta) is lower thanAt the same time the update frequency of the target network is also lower than that of the primary network.
A9. Returning to A2 if the training step number T does not reach the maximum step number T in each training round; otherwise, the current training round is finished, simultaneously the intelligent body model obtained by the training round is saved, and the current average rewarding function value R of the round is calculated av And its rate of change n.
A10. If the rate of change is less than 5% for n consecutive rounds, the whole training process is ended, otherwise step A11 is entered.
A11. And (3) judging whether the maximum training round number N is reached, if so, ending the whole training process, and if not, returning to the step A2 to enter the next training round to continue training.
24 The trained intelligent body model is transplanted into a main control chip of the machine side converter, an actual current tracking and midpoint voltage balance control system based on the main control chip of the machine side converter is established, and current tracking and midpoint voltage balance control is achieved.
Step 3, generating a driving signal s by adopting a feedforward decoupling control strategy and an SVPWM control strategy a 、s b 、s c Three bidirectional switching units S of the drive-side converter 2 a 、S b 、S c And controlling the permanent magnet synchronous wind power generator 1 to work. The specific process is as follows:
31 Step (2)The output A of the inner loop current tracking and midpoint voltage balance controller 7 obtained in the step 2 t (S t θ) there are three: output value u of current inner loop d0 、u q0 And the output value z of the midpoint voltage balance ring; will u d0 Decoupling compensation term omega from d-axis feed forward e L s i q Adding to obtain the d-axis component u of the stator voltage of the permanent magnet synchronous wind power generator 1 d Reference value u of (2) d * The method comprises the steps of carrying out a first treatment on the surface of the Will u q0 Decoupling compensation term-omega from q-axis feedforward e L s i d +ω e ψ f Adding to obtain the q-axis component u of the stator voltage of the permanent magnet synchronous wind power generator 1 q Reference value u of (2) q * The method comprises the steps of carrying out a first treatment on the surface of the Wherein omega e Is the rotor electric angular velocity omega of the permanent magnet synchronous wind driven generator 1 e =2pi pn (t), p is the pole pair number of the permanent magnet synchronous wind driven generator 1, L s Is the stator inductance, ψ f Is the rotor permanent magnet flux linkage of the synchronous wind driven generator 1;
32 The stator voltage reference value u of the permanent magnet synchronous wind driven generator 1 obtained in the step 31) is calculated d * 、u q * The stator voltage reference value u of the permanent magnet synchronous wind driven generator 1 under the three-phase static coordinate system is obtained through dq/abc coordinate transformation a * 、u b * 、u c * ；
33 U) will be a * 、u b * 、u c * And the output value z of the midpoint voltage balance ring and the actual value i of the stator current of the permanent magnet synchronous wind driven generator 1 under a three-phase static coordinate system a 、i b 、i c And sending the data to an SVPWM module.
34 SVPWM module according to wind speed v w Is modulated by the size of the (b), the modulation method is as follows:
B1. according to u a * 、u b * 、u c * And i a 、i b 、i c Partitioning the voltage quality and the current vector respectively;
B2. according to U dc For u a * 、u b * 、u c * Carrying out normalization treatment;
B3. and calculating the action time of the space vector, and converting the action time into a modulation wave.
B4. When the wind speed v w Less than the critical wind speed v wb In this case, a seven-stage switching sequence method of a continuous system is adopted, and the modulated wave is increased or decreased in value according to z.
B5. When the wind speed v w Greater than v wb In this case, modulation is performed by a five-stage switching sequence method of the intermittent method.
35 Comparing the modulated wave with a triangular carrier wave to generate a drive signal s a 、s b 、s c Three bidirectional switching units S of the drive-side converter 2 a 、S b 、S c And controlling the permanent magnet synchronous wind power generator 1 to work.
The invention is further illustrated by the following preferred embodiment.
In order to verify the effectiveness of the control strategy based on the TD3 algorithm, the control strategy based on the TD3 algorithm is adopted for the permanent magnet synchronous wind driven generator 1 and the machine side converter 2 of the permanent magnet synchronous wind driven generator, and the comparison analysis is carried out on the experimental result with the traditional experimental result based on the PID control strategy.
As shown in fig. 7, the main idea of the control strategy of the conventional permanent magnet synchronous wind driven generator based on PID control is as follows: first, the reference rotation speed n of the permanent magnet synchronous wind power generator 1 is set r (t) and its actual rotational speed n (t) are fed into the PID-controlled outer-ring rotational speed tracking controller 6, and the output of the controller 6 is used as the current reference value i of the q-axis current tracking control of the inner-ring current tracking controller 10 q * . Then, this i is q * And the actual value i q After the difference is made, the difference is sent to a first PID controller 8, and the output of the controller 8 is decoupled with q-axis feedforward compensation term-omega e L s i d +ω e ψ f Adding to obtain u q * The method comprises the steps of carrying out a first treatment on the surface of the At the same time, d-axis current reference value i d * =0 and its actual value i d After the difference is made, the difference is sent to a second PID controller 9, and the output of the controller 9 and d-axis feedforward decoupling compensation term omega e L s i q Adding to obtain u d * Will u d * 、u q * Transforming dq/alpha beta coordinates to obtain u α * 、u β * . The input to the midpoint voltage balance controller 11 is U 2 And U 1 The difference between the two is passed through a third PID controller 12. Finally, u is α * 、u β * And the output value z of the controller 12 is modulated by the SVPWM module to generate a driving signal s a 、s b 、s c Three bidirectional switching units S of the drive-side converter 2 a 、S b 、S c Thereby controlling the permanent magnet synchronous wind power generator 1 to work.
The specific simulation parameters are shown in tables 1 and 2.
Table 1 parameters related to permanent magnet synchronous wind turbines
Table 2 relevant parameters for TD3 training
As shown in fig. 8, fig. 8a and 8b are stator side currents i of the permanent magnet synchronous wind turbine 1 according to the present invention based on TD3 control and the conventional PID control, respectively q 、i d Midpoint voltage U 0 (U 0 ＝U 2 -U 1 ) The situation is changed. As can be seen from fig. 8, the maximum tracking errors of the three are respectively 0.43A, 0.66A and 2.03V based on the TD3 control method according to the present invention, and are respectively 1.62A, 1.79A and 9.57V based on the PID control method according to the present invention, so that the control effect of the present invention is better.
As shown in fig. 9, the current distortion rate THD and the direct current side output power P of the machine side converter based on the TD3 control and the conventional PID control of the present invention dc And the efficiency case thereof, wherein efficiency η is defined as η=p dc /P w X 100%. As can be seen from FIG. 9, the present invention is employedBased on the TD3 control method, when the wind speed is 7.5m/s, the wind speed is 0.23%, 1695W and 94.1%, and based on the PID control method, the wind speed is 3.71%, 1364W and 92.3%, so that the wind speed control device is better in control effect, lower in power consumption and improved in power generation power by 24.3%.
In a word, the invention can realize the optimal control of multi-objective tasks, can inhibit the stator current harmonic distortion of the wind driven generator, reduce the power consumption, has strong self-adaptive capacity and robustness to wind speed change, can effectively cope with the wind speed fluctuation and the influence of the wind driven generator self-structural parameter change and uncertainty on the control of the generator set, and realizes the rapid and stable control of the low wind speed permanent magnet synchronous wind driven generator.
Claims (1)
1. A control method of a low-wind-speed permanent magnet direct-driven wind turbine based on TD3 and Vienna rectifiers, the low-wind-speed permanent magnet direct-driven wind turbine comprises the following steps: the device comprises a permanent magnet synchronous wind driven generator, a machine side converter and a net side converter; one end of the machine side converter is connected with the stator of the permanent magnet synchronous wind driven generator, and the other end of the machine side converter is connected with the grid side converter; the machine side converter is a Vienna rectifier, and the Vienna rectifier comprises three bidirectional switch units and two output capacitors; the two output capacitors are connected in series, wherein the output capacitor connected with the positive electrode of the direct current output end of the Vienna rectifier is called an upper capacitor, and the output capacitor connected with the negative electrode of the direct current output end of the Vienna rectifier is called a lower capacitor; the method is characterized by comprising the following steps of:
step 1, designing an outer ring rotating speed tracking controller by adopting a PID control strategy, wherein the output of the outer ring rotating speed tracking controller is the q-axis component i of the stator current of the permanent magnet synchronous wind driven generator q Reference value i of (2) q * ；
Step 2, adopting a deep reinforcement learning TD3 algorithm to design an inner loop current tracking and midpoint voltage balance controller, wherein the method comprises the following steps of:
21 Constructing an intelligent agent based on a TD3 algorithm according to the environmental state of the low-wind-speed permanent magnet direct-driven wind turbine generator:
the environmental state includesWind speed v w The actual rotating speed n (t) of the permanent magnet synchronous wind driven generator and the reference value n (t) thereof * Dq-axis component i of stator current of said permanent magnet synchronous wind power generator d I d Reference value i of (2) d * 、i q I q Reference value i of (2) q * Upper capacitance voltage U of the machine side converter 1 And lower capacitance voltage U 2 The method comprises the steps of carrying out a first treatment on the surface of the The environmental state is recorded as S t Then there is S t ＝[v w n(t)n(t) * i d i d * i q i q * U 1 U 2 ]；
The intelligent agent comprises an Actor network structure and a Critic network structure, contains 6 depth neural networks in total, and specifically comprises the following components:
the Actor network structure consists of a strategy network Actor main network and a Target network Actor Target network thereof;
the input of the Actor main network is the environmental state S t Adding action search noise N to the output of the Actor main network t Then obtain action A t (S t θ), where θ is a weight parameter of the Actor main network, and motion exploration noise N t Obeying normal distribution;
the input of the Actor Target network is the environmental state S at the next moment t+1 Adding a Target smooth noise epsilon to the output of the Actor Target network to obtain an action estimated value A of the next moment t+1 (S t+1 |θ m ) Wherein θ m Epsilon obeys normal distribution for the weight parameters of the Actor Target network;
the Critic network structure consists of two sets of evaluation networks for cost function estimation, namely Critic main networkAnd its Target network Critic Target network +.>Wherein s and a are the relative states and depthsGeneral representation of motion input values for neural networks, θ m Respectively the weight parameters of the Actor network and the Target network Actor network of the Actor network; /> The weight parameters are respectively the weight parameters of the Critic main network and the Target network Critic Target network;
the Critic main network inputs S according to the current Actor main network t And output action A t (S t I θ), respectively output a cost functionThe Critic Target network inputs S according to the Actor Target network t+1 And outputting the motion estimation value A t+1 (S t+1 |θ m ) Output objective cost function respectively>
22 Design a bonus function):
the bonus function R t The design is as follows:
wherein r is 1 、r 2 、r 3 The weight coefficients are used for restraining stator current harmonic waves, midpoint voltage balance and direct current output voltage respectively and are used for carrying out priority division on a plurality of control tasks, and the sum of the weight coefficients is equal to 1; u (U) dc A direct current output voltage for the vienna rectifier; h is a 1 ～h 4 Respectively represent i q Actual value of (2) and reference value i thereof q * I is the difference of (i) d Actual value of (2) and reference value i thereof d * Is not equal to the difference value and the midpoint voltageBalance degree, DC output voltage U dc Is a reference value of (2);
the bonus function R t Adopting a variable weight structure, and determining the weight according to the wind speed: 1) When the wind speed v w Less than the critical wind speed v wb The primary task of the control strategy is to reduce harmonics when r is set 1 Maximum; 2) When the wind speed v w Greater than v wb In this case, the primary task of the control strategy is to increase the DC output voltage U dc Increasing the DC output power P of the machine side converter dc At this time, set r 2 Maximum; 3) The mid-point voltage oscillation suppression is required to be carried out in the whole variable power wind speed interval, so r is set 3 Small changes with wind speed;
23 Training the agent by adopting a deep reinforcement learning TD3 algorithm:
A1. initializing the network weight parameters of the intelligent agent and setting the relevant parameters of the intelligent agent training, wherein the method comprises the following steps:
initializing the Actor main network, the weight parameter theta of the Critic main network,Target network Actor Target network corresponding to Target network Actor Target network, weight parameter theta of Critic Target network m 、/>
Setting a learning rate mu of the Actor main network; setting a discount factor gamma of deep reinforcement learning; setting a maximum training round number N, a continuous training round number N, a maximum step number T in each training round, and updating frequencies D of a Target network, an Actor Target network and a Critic Target network; setting and initializing an experience database;
A2. based on the current state S t The output of the Actor network adds action search noise N t Obtain action A t (S t |θ), i.e.:
A t (S t |θ)＝π(S t |θ)+ N t (2)
wherein pi (S) t I theta) is the output of the Actor network;
A3. after each step of action is executed, the next environmental state S is obtained t+1 Simultaneously calculating R according to formula (1) t The obtained experience (S t ,A t ,R t ,S t+1 ) Storing the data in an experience database;
A4. the empirical database is randomly and uniformly sampled with M small batches of samples, i.e., M empirical data sets (S ti ,A ti ,R ti ,S ti+1 ) Wherein i=1, 2,3, …, M;
A5. updating weight parameters of Critic master network in a manner that minimizes errorsI.e.
Wherein Y is i Representing a desired objective cost function, defined as:
where ε represents the target smooth noise and is a truncated normal distribution noise, i.e
Wherein c is a truncation factor, and sigma is a standard deviation;
A6. if the training steps are multiples of the updating frequency D of the Critic Target network, entering a step A7, otherwise, turning to a step A9;
A7. updating θ by gradient upward for the desired return J, i.e
Wherein, the expression of J is:
representing J deriving θ:
in the method, in the process of the invention,represents pi to theta derivative;
A8. the weight parameter theta of the Actor Target network is calculated according to the formula (9) in an exponential smoothing mode m And the Critic Target network weight parameter theta mk And updating, namely:
where τ represents a soft update factor;
A9. returning to A2 if the training step number does not reach the maximum step number T in each training round; otherwise, the current training round is finished, simultaneously the intelligent body model obtained by the training round is saved, and the average rewarding function value R of the round is calculated av And the rate of change thereof;
A10. if R is av If the change rate of the continuous n rounds is less than 5%, ending the whole training process, otherwise, entering a step A11;
A11. judging whether the maximum training round number N is reached, if so, ending the whole training process, if not, returning to the step A2 to enter the next training round to continue training;
24 The trained intelligent body model is transplanted into a main control chip of the machine side converter, an actual current tracking and midpoint voltage balance control system based on the main control chip of the machine side converter is established, and current tracking and midpoint voltage balance control is realized;
step 3, a feedforward decoupling control strategy and an SVPWM control strategy are adopted to generate a driving signal to drive three bidirectional switch units of the machine side converter, and the permanent magnet synchronous wind driven generator is controlled to work, wherein the specific process is as follows:
31 2) the output A of the intelligent body inner loop current tracking and midpoint voltage balance controller obtained in the step 2 t (S t θ) there are three: output value u of current inner loop d0 、u q0 And the output value z of the midpoint voltage balance ring; will u d0 Decoupling compensation term omega from d-axis feed forward e L s i q Adding to obtain a d-axis component u of the stator voltage of the permanent magnet synchronous wind driven generator d Reference value u of (2) d * The method comprises the steps of carrying out a first treatment on the surface of the Will u q0 Decoupling compensation term-omega from q-axis feedforward e L s i d +ω e ψ f Adding to obtain the q-axis component u of the stator voltage of the permanent magnet synchronous wind driven generator q Reference value u of (2) q * Wherein ω is e For the rotor electric angular velocity omega of the permanent magnet synchronous wind power generator e ＝2πpn(t)，p、L s 、ψ f Pole pair numbers, stator inductance and rotor permanent magnet flux linkage of the permanent magnet synchronous wind driven generator are respectively set;
32 The stator voltage reference value u of the permanent magnet synchronous wind driven generator obtained in the step 31) is obtained d * 、u q * Obtaining a stator voltage reference value u of the permanent magnet synchronous wind driven generator under a three-phase static coordinate system through dq/abc coordinate transformation a * 、u b * 、u c * ；
33 U) will be a * 、u b * 、u c * And the output value z of the midpoint voltage balance ring and the actual value i of the stator current of the permanent magnet synchronous wind driven generator under a three-phase static coordinate system a 、i b 、i c Sending the signal to an SVPWM module;
34 The SVPWM module is based on the wind speed v w Is modulated by the size of the (b), the modulation method is as follows:
B1. according to u a * 、u b * 、u c * And i a 、i b 、i c Partitioning the voltage quality and the current vector respectively;
B2. according to U dc For u a * 、u b * 、u c * Carrying out normalization treatment;
B3. calculating the action time of the space vector, and converting the action time into a modulation wave;
B4. when the wind speed v w Less than the critical wind speed v wb When the method is used, a seven-level switch sequence method in a continuous mode is adopted, and the value of the modulated wave is increased or decreased according to z;
B5. when the wind speed v w V is greater than or equal to v wb When in use, modulating by adopting a five-stage switching sequence method in an intermittent mode;
35 And (3) comparing the modulated wave with the triangular carrier wave to generate a driving signal, driving three bidirectional switch units of the machine side converter, and controlling the permanent magnet synchronous wind driven generator to work.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311276120.XA CN117318553B (en) | 2023-09-29 | 2023-09-29 | Low-wind-speed permanent magnet direct-driven wind turbine control method based on TD3 and Vienna rectifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311276120.XA CN117318553B (en) | 2023-09-29 | 2023-09-29 | Low-wind-speed permanent magnet direct-driven wind turbine control method based on TD3 and Vienna rectifier |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117318553A CN117318553A (en) | 2023-12-29 |
| CN117318553B true CN117318553B (en) | 2024-03-26 |
Family
ID=89261669
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311276120.XA Active CN117318553B (en) | 2023-09-29 | 2023-09-29 | Low-wind-speed permanent magnet direct-driven wind turbine control method based on TD3 and Vienna rectifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117318553B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011226326A (en) * | 2010-04-16 | 2011-11-10 | Hitachi Ltd | Control device and method for wind power generator group |
| CN108988395A (en) * | 2018-10-13 | 2018-12-11 | 曲阜师范大学 | A kind of topological structure and control method of the directly driven wind-powered current transformer of vertical axis permanent magnet |
| CN109104107A (en) * | 2018-10-13 | 2018-12-28 | 曲阜师范大学 | The topological structure and control method of permanent magnet direct-drive type wind-driven generator generator-side converter wear |
| CN115360741A (en) * | 2022-10-20 | 2022-11-18 | 中国华能集团清洁能源技术研究院有限公司 | Wind storage flexible control method and device based on deep reinforcement learning in spot-cargo scene |
| CN115903457A (en) * | 2022-11-02 | 2023-04-04 | 曲阜师范大学 | Low-wind-speed permanent magnet synchronous wind driven generator control method based on deep reinforcement learning |
| CN116796644A (en) * | 2023-07-05 | 2023-09-22 | 杭州电子科技大学 | Wind farm parameter identification method based on multi-agent SAC deep reinforcement learning |
-
2023
- 2023-09-29 CN CN202311276120.XA patent/CN117318553B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011226326A (en) * | 2010-04-16 | 2011-11-10 | Hitachi Ltd | Control device and method for wind power generator group |
| CN108988395A (en) * | 2018-10-13 | 2018-12-11 | 曲阜师范大学 | A kind of topological structure and control method of the directly driven wind-powered current transformer of vertical axis permanent magnet |
| CN109104107A (en) * | 2018-10-13 | 2018-12-28 | 曲阜师范大学 | The topological structure and control method of permanent magnet direct-drive type wind-driven generator generator-side converter wear |
| CN115360741A (en) * | 2022-10-20 | 2022-11-18 | 中国华能集团清洁能源技术研究院有限公司 | Wind storage flexible control method and device based on deep reinforcement learning in spot-cargo scene |
| CN115903457A (en) * | 2022-11-02 | 2023-04-04 | 曲阜师范大学 | Low-wind-speed permanent magnet synchronous wind driven generator control method based on deep reinforcement learning |
| CN116796644A (en) * | 2023-07-05 | 2023-09-22 | 杭州电子科技大学 | Wind farm parameter identification method based on multi-agent SAC deep reinforcement learning |
Non-Patent Citations (1)
| Title |
|---|
| 基于功率反馈的直驱式永磁发电机组最大功率跟踪控制;黄河;付立军;侍乔明;王刚;;电气传动;20140120;44(第01期);53-57 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117318553A (en) | 2023-12-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Feng et al. | Direct torque control with variable flux for an SRM based on hybrid optimization algorithm | |
| CN108418256B (en) | Virtual synchronous machine self-adaptive control method based on output differential feedback | |
| CN112448399B (en) | Doubly-fed wind power plant subsynchronous oscillation suppression method based on analog inductance | |
| CN114512986B (en) | Parameter optimization design method for passive LCL filter of grid-connected inverter | |
| CN111654052B (en) | Flexible direct current converter modeling device and method based on dynamic phasor method | |
| WO2018145498A1 (en) | Reinforcement learning algorithm-based self-correction control method for double-fed induction wind generator | |
| CN109599889B (en) | Fuzzy active disturbance rejection based ride-through control method and system under unbalanced voltage | |
| CN113469332A (en) | Virtual synchronous generator inertia damping self-adaptive control method based on fuzzy nerves | |
| Aissaoui et al. | Application of fuzzy logic in control of electrical machines | |
| Buduma et al. | Robust control and Kalman MPPT for grid-assimilated wind energy conversion system | |
| Kamalapur et al. | A novel neuro fuzzy method DTC of induction motor drive for efficient torque response | |
| CN113098058B (en) | Self-adaptive optimization control method, device, equipment and medium for moment of inertia | |
| CN115903457B (en) | Control method of low-wind-speed permanent magnet synchronous wind driven generator based on deep reinforcement learning | |
| Rashid et al. | Adaptive predictive controller design for nonlinear grid-connected PMSG based wind energy conversion system using dynamic matrices | |
| CN117318553B (en) | Low-wind-speed permanent magnet direct-driven wind turbine control method based on TD3 and Vienna rectifier | |
| CN110518625B (en) | Grid-connected inverter direct-current component suppression method with variable learning rate BP-PID control | |
| Sun et al. | Design and optimization of control parameters based on direct-drive permanent magnet synchronous generator for wind power system | |
| CN108400599B (en) | Coordinated control method for UPFC in wind power generation grid-connected system | |
| Parida et al. | A new self-adjusting PI controller for power control in a wind turbine generator | |
| Shen et al. | HOSMD and neural network based adaptive super-twisting sliding mode control for permanent magnet synchronous generators | |
| Alzain et al. | Optimization of sliding mode control based on BAT-algorithm for the DFIG-WT | |
| CN111045329B (en) | Double-fed fan digital physical hybrid simulation method based on self-adaptive mode switching | |
| Chakraborty et al. | A novel application of adaptive filtering algorithm for LVRT capability enhancement of grid-connected DFIG-based wind energy conversion systems (WECS) | |
| Moghadasian et al. | Intelligent sensorless speed control of six-phase induction machine | |
| Tamaarat | Active and reactive power control for DFIG using PI, fuzzy logic and self-tuning PI fuzzy controllers |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |