A data-driven estimator-based control system for doubly-fed direct current generation

Abstract

The application provides a double-fed induction generator control system based on a data-driven estimator, comprising a power comparison unit, a PI controller, a rotor current comparison unit, a d-axis filter, a q-axis filter, a d-axis state estimator, a q-axis state estimator, a d-axis data-driven adaptive unit, a q-axis data-driven adaptive unit, a d-axis neural network unit, a q-axis neural network unit, a d-axis controller, a q-axis controller and a double-fed induction generator. The application can accurately estimate system uncertainty and environmental disturbance, adopts a topology structure of a stator side connected to a non-controlled rectification bridge, directly connects a direct current power grid in parallel, controls the motor system by controlling the rotor side inverter to regulate the excitation current, so as to realize direct power control of the double-fed motor direct current power generation system.

Description

Technical Field

[0001] This invention relates to the field of new energy power generation technology, and mainly studies the output power control problem of a doubly-fed induction generator connected to a DC microgrid. It also proposes a doubly-fed DC power generation control system based on a data-driven predictor. Background Technology

[0002] In recent years, doubly fed induction generators (DFIGs) have been widely used in wind power generation due to their outstanding advantages such as power decoupling control, maximum wind power capture capability and low converter cost. As a result, they have attracted widespread attention from the academic community, and their control methods are also a top priority in current research.

[0003] For doubly-fed induction generators, researchers have proposed several control methods. Vector control is a classic method commonly used in motor control, which has good steady-state characteristics, but its dynamic characteristics are not ideal. Backstepping is suitable for systems with simple models, but its effectiveness is greatly reduced for more complex systems. Direct power control is simple and has good dynamic performance, but it will generate ripple in steady state. Model predictive control can achieve good steady-state and dynamic performance at the same time, but it requires many motor parameters and a lot of calculations. Resonant control can suppress harmonics caused by uncontrolled rectifiers, improve control performance, but make the system more complex.

[0004] Doubly fed induction generators (DFIGs) are complex, high-order nonlinear, and strongly coupled electrical systems. During operation, in addition to internal disturbances caused by the inherent system model, they are also affected by environmental factors, leading to changes in motor parameters, which can worsen control performance or even cause loss of control. Therefore, addressing the model uncertainties and environmental disturbances of DFIGs is crucial for achieving good control performance. Summary of the Invention

[0005] This invention provides a doubly fed DC power generation control system based on a data-driven predictor, which can accurately estimate system uncertainties and environmental disturbances. It adopts a topology structure with an uncontrolled rectifier bridge connected to the stator side, directly connected to the DC grid, and controls the motor system by adjusting the excitation current through the control of the rotor-side inverter.

[0006] The technical means employed in this invention are as follows:

[0007] A doubly-fed DC generator control system based on a data-driven predictor includes: a power comparator unit, a PI controller, a rotor current comparator unit, a d-axis filter, a q-axis filter, a d-axis state predictor, a q-axis state predictor, a d-axis data-driven adaptive unit, a q-axis data-driven adaptive unit, a d-axis neural network unit, a q-axis neural network unit, a d-axis controller, a q-axis controller, and a doubly-fed induction generator; wherein:

[0008] The input signal of the power comparator unit is the power setpoint. With the power signal P output by the system dc The output signal is

[0009] The input signal of the PI controller is the output of the power comparator unit. The given value of the output rotor current and

[0010] The input signal of the rotor current comparison unit is the rotor current reference signal of the doubly-fed induction generator. and With the output signal of the state predictor and The output signal is

[0011] The input signal of the d-axis filter is the rotor current signal i of the doubly-fed induction generator. rd And the parameter matrix Y1, the output signal is i rd Signals g1 and Y1 generated by the filter are then used to generate signal N1.

[0012] The input signal of the q-axis filter is the rotor current signal i of the doubly-fed induction generator. rq And the parameter matrix Y2, the output signal is i rq The signals g2 and Y2 generated by the filter are then combined to generate signal N2.

[0013] The input signals of the d-axis state predictor are the output e1 of the rotor current comparator and the output signal from the data-driven adaptive unit. And the parameter matrix Y1, the output signal is the estimated value of the d-axis rotor current.

[0014] The input signals of the q-axis state predictor are the output e2 of the rotor current comparator and the output signal from the data-driven adaptive unit. And the parameter matrix Y2, the output signal is the estimated value of the q-axis rotor current.

[0015] The input signals of the d-axis data-driven adaptive unit are the output e1 of the rotor current comparator, the output g1 of the filter, and N1, and the output signal is...

[0016] The input signals of the q-axis data-driven adaptive unit are the output e2 of the rotor current comparator, the output g2 of the filter, and N2, and the output signal is...

[0017] The input signal of the d-axis neural network unit is the output of the data-driven adaptive unit. With parameter matrix Y1, the output signal is an estimate of the uncertainty of the unknown model.

[0018] The input signal of the q-axis neural network unit is the output of the data-driven adaptive unit. With parameter matrix Y2, the output signal is an estimate of the uncertainty of the unknown model.

[0019] The input signal of the d-axis controller is the output signal i of the doubly-fed induction generator. rd The given value of rotor current Output signal of d-axis state predictor Output signal of d-axis neural network and the output signal of the d-axis data-driven adaptive unit The output signal is the control signal u of the doubly-fed induction generator. rd ;

[0020] The input signal of the q-axis controller is the output signal i of the doubly-fed induction generator. rq The given value of rotor current Output signal of q-axis state predictor Output signal of q-axis neural network and the output signal of the q-axis data-driven adaptive unit The output signal is the control signal u of the doubly-fed induction generator. rq .

[0021] Furthermore, the PI controller model is as follows:

[0022]

[0023] in, This represents the given value of the q-axis rotor current. k represents the given value of the d-axis rotor current. p k represents the proportional gain coefficient. i U represents the integral gain coefficient, s represents the Laplace transform, and U represents the integral gain coefficient. dc L represents the actual value of the DC voltage. m Mutual inductance is represented by ω1, synchronous speed is represented by L. s R represents the stator inductance. s This indicates the stator resistance.

[0024] Furthermore, the d-axis filter is used to perform the following transformations:

[0025]

[0026] The q-axis filter is used to perform the following transformations:

[0027]

[0028] Where, Y1=[u rd ,σ 11 ,σ 12 ,σ 13 ,...,σ 1q ] T Y2 = [u rq ,σ 21 ,σ 22 ,σ 23 ,...,σ 2q ] T q represents the number of hidden layer nodes in the neural network, σ x Let represent the activation function of the neural network, where a is a constant greater than zero, and s represents the Laplace transform.

[0029] Furthermore, the d-axis state predictor is used to make the following estimates:

[0030]

[0031] The q-axis state predictor is used to make the following estimates:

[0032]

[0033] in, State variables representing the estimated value of the d-axis rotor current. The state variables represent the estimated value of the q-axis rotor current, and k1, ρ1, k2, and ρ2 represent the adjustment parameters.

[0034] Furthermore, the d-axis data-driven adaptive unit is used to perform the following calculations:

[0035]

[0036] The q-axis data-driven adaptive unit is used to perform the following calculations:

[0037]

[0038] in, And q represents the number of hidden layer nodes in the neural network, ψ x This represents the ideal weights of the neural network. Γ1∈R and Γ2∈R are both constants greater than zero, g(j) and N(j) are the j-th data of the record respectively, and p represents the number of data records.

[0039] Furthermore, the d-axis neural network unit is used to perform the following calculations:

[0040]

[0041] The q-axis neural network unit is used to perform the following calculations:

[0042]

[0043] Where E is a diagonal matrix of order q, E = diag{0,1,1,1,...,1}.

[0044] Furthermore, the d-axis controller is used to perform the following calculations:

[0045]

[0046] The q-axis controller is used to perform the following calculations:

[0047]

[0048] Among them, u rd (k) and u rq (k) represent the dq-axis components of the rotor voltage at time k. and Let be the state estimates of the dq-axis components of the rotor current at time k. and Let be the given values ​​of the dq-axis components of the rotor current at time k. and Let be the given values ​​of the dq-axis components of the rotor current at time k-1. and This represents an estimate of the unknown input gain. and This represents the estimated uncertainty of the model at time k.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] 1. This invention proposes a deadbeat predictive current controller based on a data-driven neural network predictor, which can realize direct power control of a doubly-fed motor DC power generation system.

[0051] 2. This invention designs two improved data-driven neural network predictors to simultaneously identify dynamic uncertainties, unknown control gains, and external disturbances in a doubly-fed DC power generation system.

[0052] 3. This invention utilizes unknown model information obtained from a neural network predictor to design a deadbeat predictive current controller. The designed controller only requires rotor current and voltage values ​​and does not require additional parameter information. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a structural diagram of the doubly fed DC power generation control system based on a data-driven predictor, as presented in this invention.

[0055] Figure 2 This is a diagram illustrating the output power control effect in the embodiment.

[0056] Figure 3 This is a diagram showing the identification effect of the control gain β in the embodiment.

[0057] Figure 4 The diagram shows the learning effect of the neural network on f1 in the embodiment.

[0058] Figure 5 The diagram shows the learning effect of the neural network on f2 in the embodiment.

[0059] Figure 6 In the example, the predictor for i rd The recognition effect diagram.

[0060] Figure 7 In the example, the predictor for i rq The recognition effect diagram. Detailed Implementation

[0061] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0062] like Figure 1 As shown, this invention provides a doubly-fed DC power generation control system based on a data-driven predictor, comprising: a power comparator unit, a PI controller, a rotor current comparator unit, a d-axis filter, a q-axis filter, a d-axis state predictor, a q-axis state predictor, a d-axis data-driven adaptive unit, a q-axis data-driven adaptive unit, a d-axis neural network unit, a q-axis neural network unit, a d-axis controller, a q-axis controller, and a doubly-fed induction generator. Wherein:

[0063] The input signal of the power comparator unit is the power setpoint. With the power signal P output by the system dc The output signal is

[0064] The input signal to the PI controller is the output of the power comparator unit. The given value of the output rotor current and

[0065] The input signal of the rotor current comparator unit is the rotor current reference signal of the doubly-fed induction generator. and With the output signal of the state predictor and The output signal is

[0066] The input signal to the d-axis filter is the rotor current signal i of the doubly-fed induction generator. rd And the parameter matrix Y1, the output signal is i rd The signals g1 and Y1 generated by the filter are then used to generate the signal N1.

[0067] The input signal to the q-axis filter is the rotor current signal i of the doubly-fed induction generator. rq And the parameter matrix Y2, the output signal is i rq The signals g2 and Y2 generated by the filter are then used to generate the signal N2.

[0068] The input signals of the d-axis state predictor are the output e1 of the rotor current comparator and the output signal from the data-driven adaptive unit. And the parameter matrix Y1, the output signal is the estimated value of the d-axis rotor current.

[0069] The input signals to the q-axis state predictor are the output e2 of the rotor current comparator and the output signal from the data-driven adaptive unit. And the parameter matrix Y2, the output signal is the estimated value of the q-axis rotor current.

[0070] The input signals of the d-axis data-driven adaptive unit are the output e1 of the rotor current comparator, the output g1 of the filter, and N1. The output signal is...

[0071] The input signals to the q-axis data-driven adaptive unit are the output e2 of the rotor current comparator, the outputs g2 and N2 of the filter, and the output signal is...

[0072] The input signal of the d-axis neural network unit is the data-driven output of the adaptive unit. With parameter matrix Y1, the output signal is an estimate of the uncertainty of the unknown model.

[0073] The input signal of the q-axis neural network unit is the data-driven output of the adaptive unit. With parameter matrix Y2, the output signal is an estimate of the uncertainty of the unknown model.

[0074] The input signal to the d-axis controller is the output signal i of the doubly-fed induction generator. rd The given value of rotor current Output signal of d-axis state predictor Output signal of d-axis neural network and the output signal of the d-axis data-driven adaptive unit The output signal is the control signal u of the doubly-fed induction generator. rd .

[0075] The input signal to the q-axis controller is the output signal i of the doubly-fed induction generator. rq The given value of rotor current Output signal of q-axis state predictor Output signal of q-axis neural network and the output signal of the q-axis data-driven adaptive unit The output signal is the control signal u of the doubly-fed induction generator. rq .

[0076] In the study of control problems of doubly-fed induction generator (DFIG) DC power generation systems, the nonlinearity of the DFIG model and the complexity of its operating environment pose significant challenges to controller design. Accurate estimation of uncertainties and environmental disturbances is crucial. To address this, this invention proposes a deadbeat predictive current control scheme based on a data-driven predictor. The main implementation includes:

[0077] 1. A DC power generation control system for a doubly-fed motor is established based on a dual-closed-loop control method, with the outer loop using PI control to control the output power.

[0078] 2. Use a data-driven neural network predictor to identify model uncertainties and environmental disturbances in a doubly-fed DC power generation system.

[0079] 3. The inner loop adopts a deadbeat predictive current control method to ensure the system's response speed and steady-state performance.

[0080] To achieve the above objectives, the present invention provides a doubly-fed DC power generation control system based on a data-driven predictor. The system consists of a main circuit and a control circuit. The control circuit consists of an outer-loop power controller and an inner-loop deadbeat predictive current controller. The inner-loop controller uses a data-driven predictor to identify model uncertainties and environmental disturbances in the doubly-fed DC power generation system.

[0081] The mathematical model of the doubly-fed induction generator is as follows, and its voltage equation can be expressed as follows:

[0082]

[0083] The flux linkage equation is expressed as

[0084]

[0085] The power equation is expressed as

[0086]

[0087] For a doubly-fed induction generator, a vector control technique with stator flux orientation is used to orient the stator flux to the d-axis. The stator flux in the dq coordinate system then has the following relationship:

[0088]

[0089] Substituting the stator flux linkage equation to reconcile the relationship between the stator and rotor currents

[0090]

[0091] Rearranging the above formulas yields a new form of the system's mathematical model.

[0092]

[0093] Where a=L m / L s ,

[0094] To facilitate controller design, the system model is rewritten as follows:

[0095]

[0096] Among them, β1=β2=β=1 / b.

[0097] Specifically, the system design includes the following:

[0098] A. Design of the power comparator unit

[0099] The input signal of the power comparator unit is the power setpoint. With the power signal P output by the systemdc The output signal is

[0100] B. Design of the PI controller

[0101] The input to the PI controller is the output of the power comparator unit. The rotor current setpoint is obtained through the following PI controller. and

[0102]

[0103] C. Design of the rotor current comparator unit

[0104] The input signal of the rotor current comparator unit is the rotor current reference signal of the doubly-fed induction generator. and With the output signal of the state predictor and Finally, the output signal of the rotor current comparator is obtained.

[0105] D. Filter Design

[0106] D1.d-axis filter design

[0107] doubly fed induction generator rotor current signal i rd The filtered output signal g1 is obtained by passing the signal through a d-axis filter. The parameter matrix Y1 is also passed through this filter to obtain the filtered output signal N1.

[0108]

[0109] D2.q-axis filter design

[0110] doubly fed induction generator rotor current signal i rq The output signal g2 is obtained by passing the signal through a q-axis filter. The parameter matrix Y2 is also passed through this filter to obtain the filtered output signal N2.

[0111]

[0112] Where, Y1=[u rd ,σ 11 ,σ 12 ,σ 13 ,...,σ 1q ] T Y2 = [u rq ,σ 21 ,σ 22 ,σ 23 ,...,σ 2q ]T And q represents the number of hidden layer nodes in the neural network, σ x Let represent the activation function of the neural network, where a is a constant greater than zero, and s represents the Laplace transform.

[0113] E. Design of State Predictors

[0114] Design of E1.d-axis state predictor

[0115] The output e1 of the rotor current comparator and the output signal from the data-driven adaptive unit Using the parameter matrix Y1 as input to the state predictor, the output signal can be obtained through the d-axis state predictor designed as follows.

[0116]

[0117] Design of E2.q-axis state predictor

[0118] The output e2 of the rotor current comparator and the output signal from the data-driven adaptive unit Using the parameter matrix Y2 as input to the state predictor, the output signal can be obtained through the q-axis state predictor designed as follows.

[0119]

[0120] Where k1, k2, ρ1, and ρ2 are adjustment parameters, and all are constants greater than zero.

[0121] F. Design of Data-Driven Adaptive Units

[0122] Design of F1.d-axis data-driven adaptive unit

[0123] By using historical accumulated data to update the neural network weights and identify unknown input gains, the output e1 of the rotor current comparator and the outputs g1 and N1 of the filter are used as inputs to the d-axis data-driven adaptive unit to obtain the output signal.

[0124]

[0125] Design of F2.q-axis data-driven adaptive unit

[0126] By using historical accumulated data to update the neural network weights and identify unknown input gains, the output e2 of the rotor current comparator and the outputs g2 and N2 of the filter are used as inputs to the q-axis data-driven adaptive unit to obtain the output signal.

[0127]

[0128] in, And q represents the number of hidden layer nodes in the neural network, ψ x This represents the ideal weights of the neural network. Γ1∈R and Γ2∈R are both constants greater than zero, g(j) and N(j) are the j-th data of the record respectively, and p represents the number of data records.

[0129] G. Design of Neural Network Units

[0130] Design of G1.d-axis neural network unit

[0131] The input to the d-axis neural network unit is the data-driven output of the adaptive unit. The output signal of the parameter matrix Y1, after passing through the following neural network unit, is an estimate of the uncertainty of the unknown model.

[0132]

[0133] Design of G2.q-axis neural network unit

[0134] The input to the q-axis neural network unit is the data-driven output of the adaptive unit. The output signal of the parameter matrix Y2, after passing through the following neural network unit, is an estimate of the uncertainty of the unknown model.

[0135]

[0136] Where E is a diagonal matrix of order q, E = diag{0,1,1,1,...,1}.

[0137] H. Deadbeat Predictive Current Controller Design

[0138] Discretize the system model and multiply it by the sampling period T. s The relationship between rotor voltage and rotor current in the kth sampling period can be obtained. Based on the deadbeat principle and combined with a data-driven neural network predictor, a deadbeat predictive current controller is designed.

[0139] H1.d-axis controller design

[0140] The output signal i of the doubly fed induction generator rd The given value of rotor current Output signal of d-axis state predictor Output signal of d-axis neural network and the output signal of the d-axis data-driven adaptive unit Simultaneously, the control signal u of the doubly-fed induction generator is input to the designed d-axis controller. rd

[0141]

[0142] H2.q-axis controller design

[0143] The output signal i of the doubly fed induction generator rq The given value of rotor current Output signal of q-axis state predictor Output signal of q-axis neural network and the output signal of the q-axis data-driven adaptive unit Simultaneously, the control signal u of the doubly-fed induction generator is input to the designed q-axis controller. rq

[0144]

[0145] In the formula, u rd (k) and u rq (k) represent the dq-axis components of the rotor voltage at time k. and Let be the state estimates of the dq-axis components of the rotor current at time k. and Let be the given values ​​of the dq-axis components of the rotor current at time k. and Let be the given values ​​of the dq-axis components of the rotor current at time k-1. and This represents an estimate of the unknown input gain. and This represents the estimated uncertainty of the model at time k.

[0146] The following section uses a doubly-fed DC generator system built on the MATLAB / Simulink platform as an example to further illustrate the scheme and effects of this invention. The simulation platform includes a main circuit and a control circuit. The parameters of the doubly-fed motor model in the main circuit are selected as follows: rotor resistance R r =1.65Ω, stator resistance R s =1.37Ω, rotor inductance L r =0.1635H, stator inductance L s =0.1625H, mutual inductance L m =0.1592H, number of magnetic pole pairs is 2. The control system is built with reference to the controller designed in this invention.

[0147] The parameters for the data-driven neural network predictor and controller are selected as follows: the neural network activation function is chosen as σ(x) = (1-e^(-x) / x). -x ) / (1+e -x Adjust parameters k1 = k2 = 5000, ρ1 = ρ2 = 50000. Γ1 = Γ2 = 1500, the number of recorded data is set to 500, and the filter parameter a is set to 10.

[0148] Simulation results are as follows Figures 2-7 , Figure 2 It can be seen that the output power can follow the given value well, and it quickly returns to steady state when the load is switched, which verifies the effectiveness of the proposed method. Figure 3 It can be seen that the neural network is effective in identifying the control gain β, and can converge to the true value relatively quickly. Figures 4-5 This demonstrates the learning performance of the neural network on f1 and f2, showing that it can converge to the true value in a short time. Figures 6-7 This demonstrates the predictor's effect on rotor current i rd and i rq Good recognition effect.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A doubly-fed DC power generation control system based on a data-driven predictor, characterized in that, include: The system includes a power comparator unit, a PI controller, a rotor current comparator unit, a d-axis filter, a q-axis filter, a d-axis state predictor, a q-axis state predictor, a d-axis data-driven adaptive unit, a q-axis data-driven adaptive unit, a d-axis neural network unit, a q-axis neural network unit, a d-axis controller, a q-axis controller, and a doubly-fed induction generator; among which: The input signal of the power comparator unit is the power setpoint. With the power signal P output by the system dc The output signal is The input signal of the PI controller is the output of the power comparator unit. The given value of the output rotor current and The input signal of the rotor current comparison unit is the rotor current reference signal of the doubly-fed induction generator. and With the output signal of the state predictor and The output signal is The input signal of the d-axis filter is the rotor current signal i of the doubly-fed induction generator. rd And the parameter matrix Y1, the output signal is i rd Signals g1 and Y1 generated by the filter are then used to generate signal N1. The input signal of the q-axis filter is the rotor current signal i of the doubly-fed induction generator. rq And the parameter matrix Y2, the output signal is i rq The signals g2 and Y2 generated by the filter are then combined to generate signal N2. The input signals of the d-axis state predictor are the output e1 of the rotor current comparator and the output signal from the data-driven adaptive unit. And the parameter matrix Y1, the output signal is the estimated value of the d-axis rotor current. The input signals of the q-axis state predictor are the output e2 of the rotor current comparator and the output signal from the data-driven adaptive unit. And the parameter matrix Y2, the output signal is the estimated value of the q-axis rotor current. The input signals of the d-axis data-driven adaptive unit are the output e1 of the rotor current comparator, the output g1 of the filter, and N1, and the output signal is... The input signals of the q-axis data-driven adaptive unit are the output e2 of the rotor current comparator, the output g2 of the filter, and N2, and the output signal is... The input signal of the d-axis neural network unit is the output of the data-driven adaptive unit. With parameter matrix Y1, the output signal is an estimate of the uncertainty of the unknown model. The input signal of the q-axis neural network unit is the output of the data-driven adaptive unit. With parameter matrix Y2, the output signal is an estimate of the uncertainty of the unknown model. The input signal of the d-axis controller is the output signal i of the doubly-fed induction generator. rd The given value of rotor current Output signal of d-axis state predictor Output signal of d-axis neural network and the output signal of the d-axis data-driven adaptive unit The output signal is the control signal u of the doubly-fed induction generator. rd ; The input signal of the q-axis controller is the output signal i of the doubly-fed induction generator. rq The given value of rotor current Output signal of q-axis state predictor Output signal of q-axis neural network and the output signal of the q-axis data-driven adaptive unit The output signal is the control signal u of the doubly-fed induction generator. rq .

2. The doubly-fed DC power generation control system based on a data-driven predictor according to claim 1, characterized in that, The PI controller model is as follows: in, This represents the given value of the q-axis rotor current. k represents the given value of the d-axis rotor current. p k represents the proportional gain coefficient. i U represents the integral gain coefficient, s represents the Laplace transform, and U represents the integral gain coefficient. dc L represents the actual value of the DC voltage. m Mutual inductance is represented by ω1, synchronous speed is represented by L. s R represents the stator inductance. s This indicates the stator resistance.

3. The doubly-fed DC power generation control system based on a data-driven predictor according to claim 1, characterized in that, The d-axis filter is used to perform the following transformations: The q-axis filter is used to perform the following transformations: Where, Y1=[u rd ,σ 11 ,σ 12 ,σ 13 ,...,σ 1q ] T Y2 = [u rq ,σ 21 ,σ 22 ,σ 23 ,...,σ 2q ] T q represents the number of hidden layer nodes in the neural network, σ x Let represent the activation function of the neural network, where a is a constant greater than zero, and s represents the Laplace transform.

4. The doubly-fed DC power generation control system based on a data-driven predictor according to claim 3, characterized in that, The d-axis state predictor is used to make the following estimates: The q-axis state predictor is used to make the following estimates: in, State variables representing the estimated value of the d-axis rotor current. The state variables represent the estimated value of the q-axis rotor current, and k1, ρ1, k2, and ρ2 represent the adjustment parameters.

5. A doubly-fed DC power generation control system based on a data-driven predictor according to claim 4, characterized in that, The d-axis data-driven adaptive unit is used to perform the following calculations: The q-axis data-driven adaptive unit is used to perform the following calculations: in, And q represents the number of hidden layer nodes in the neural network, ψ x This represents the ideal weights of the neural network. Γ1∈R and Γ2∈R are both constants greater than zero, g(j) and N(j) are the j-th data of the record respectively, and p represents the number of data records.

6. A doubly-fed DC power generation control system based on a data-driven predictor according to claim 5, characterized in that, The d-axis neural network unit is used to perform the following calculations: The q-axis neural network unit is used to perform the following calculations: Where E is a diagonal matrix of order q, E = diag{0,1,1,1,...,1}.

7. A doubly-fed DC power generation control system based on a data-driven predictor according to claim 6, characterized in that, The d-axis controller is used to perform the following calculations: The q-axis controller is used to perform the following calculations: Among them, u rd (k) and u rq (k) represent the dq-axis components of the rotor voltage at time k. and Let be the state estimates of the dq-axis components of the rotor current at time k. and Let be the given values ​​of the dq-axis components of the rotor current at time k. and Let be the given values ​​of the dq-axis components of the rotor current at time k-1. and This represents an estimate of the unknown input gain. and This represents the estimated uncertainty of the model at time k.

Images