Iterative estimation method for rapidly identifying model parameters of single-phase photovoltaic grid-connected inverter interfered by colored noise and having information loss

Abstract

The invention relates to an iterative estimation method for quickly identifying model parameters of a single-phase photovoltaic grid-connected inverter which is interfered by colored noise and has information loss. With the promotion of intelligent power grid construction, stable and efficient operation of a power grid is significant. The performance of single-phase power equipment serving as a power grid basic unit directly affects the electric energy quality and the power supply reliability. However, equipment is often interfered by noise in actual operation, information loss exists due to the problems of sensor faults, communication interruption and the like, and modeling and identification are difficult. According to the method, a double-rate Volterra-Wiener model of a single-phase photovoltaic grid-connected inverter based on a Volterra sequence is taken as an object, and a three-stage Levenberg-Marquardt iterative algorithm based on filtering is provided. Colored noise is eliminated through data filtering, and the system is decomposed into three subsystems by using a hierarchical principle. The Levenberg-Marquardt algorithm is insensitive to initial value selection, the convergence speed is high, and the method aims at achieving rapid and effective identification of model parameters which are interfered by colored noise and have information loss.

Description

Technical Field

[0001] The present invention relates to a method for identifying a dual-rate Volterra-Wiener nonlinear model of a single-phase photovoltaic grid-connected inverter based on a filtering-based three-stage Levenberg–Marquardt iterative algorithm. Background Art

[0002] With the full promotion of the construction of smart grids, the stability, efficiency, and accuracy of grid operation are becoming increasingly crucial. In a complex and changing power transmission network, single-phase power equipment, as a basic component of the grid, directly affects the power quality and power supply reliability. However, during actual operation, single-phase power equipment often suffers from colored noise interference and inevitably faces information loss problems due to factors such as sensor failures and communication interruptions, which pose difficulties for modeling and identification.

[0003] Existing research mostly uses Hammerstein-type nonlinear block structure models to describe the complex nonlinear dynamic characteristics of power equipment, but its parameter identification is often restricted by problems such as gradient disappearance and local convergence. Traditional gradient descent algorithms are prone to falling into suboptimal solutions when dealing with the optimization objectives of power equipment models with information loss, while methods such as extended Kalman filtering result in a decrease in identification accuracy due to colored noise and process disturbances. In addition, power equipment has significantly different dynamic characteristics during different operating periods, such as startup, steady-state operation, and load mutation stages, requiring the identification algorithm to have the ability to dynamically adjust. The present invention aims at the dual-rate Volterra-Wiener nonlinear model structure, uses data filtering methods to filter the input and output data to eliminate the interference of colored noise on parameter identification, combines the hierarchical principle, decomposes the filtered system into three subsystems, and finally realizes the fast and effective identification of the parameters of the Volterra-Wiener nonlinear model of a single-phase photovoltaic grid-connected inverter affected by colored noise and with information loss. Summary of the Invention

[0004] The present invention aims to identify the parameters of a Volterra-Wiener nonlinear model of a single-phase photovoltaic grid-connected inverter affected by colored noise and with information loss using a filtering-based three-stage Levenberg–Marquardt iterative algorithm.

[0005] The solution at the technical level is as follows: According to dual-rate sampling, data filtering methods, and the hierarchical principle, parameter estimation of a Volterra-Wiener nonlinear model of a single-phase photovoltaic grid-connected inverter affected by colored noise and with information loss is realized.

[0006] 1) Construct a dual-rate Volterra-Wiener nonlinear output error parameter identification model for a single-phase photovoltaic grid-connected inverter:

[0007] Step 1: Double-rate sampling is adopted to avoid some information loss. The double-rate system is as shown in Figure 1 the following figure:

[0008] Step 2: Build a double-rate Volterra-Wiener nonlinear identification model for a single-phase photovoltaic grid-connected inverter with a static nonlinear function based on the Volterra series as shown in the appendix Figure 2 as follows:

[0009] Step 3: Referring to the established model structure, build the expression of the Volterra-Wiener linear dynamic sub-model of the single-phase photovoltaic grid-connected inverter as follows: w(t) is regarded as the colored noise of the interference and is defined as follows: The response component x nol,k (t) of the discrete-time nonlinear Volterra system is expressed as follows: represents the coefficient of the k-th order kernel. In the double-rate Volterra nonlinear subsystem, the expression after replacing t with dt is as follows: For the K-th order Volterra series (K is the maximum nonlinear order), the expressions of x nol,k (dt) and m(dt) are as follows: where L k is the memory length of the corresponding order. After sorting, we can get: m(dt) = X T (dt)θ h . (13) For the dynamic linear subsystem in the double-rate Volterra-Wiener nonlinear system, we can get:

[0010] Assume h 1,0 = 1. After sorting out the above equations, the expression of the complete double-rate Volterra-Wiener nonlinear output error identification model is as follows:

[0011] where I dc (dt) is the DC input current of the single-phase photovoltaic grid-connected inverter after double-rate sampling, x dt and mdt is an intermediate variable, w(t) is regarded as colored noise of interference, i ac (dt) is the AC output current of a single-phase PV grid-connected inverter after double-rate sampling, and v(t) is white noise during the identification process.

[0012] 2) Design the flow of a three-stage Levenberg–Marquardt iterative parameter identification algorithm based on filtering:

[0013] The first step: Start the parameter identification algorithm;

[0014] The second step: Let the iteration number n = 0 and set the initial values;

[0015] The third step: Obtain the DC input current I of the single-phase PV grid-connected inverter after double-rate sampling dc (dt) as input data, and the AC output current i of the single-phase PV grid-connected inverter after double-rate sampling ac (dt) as output data;

[0016] The fourth step: Construct and

[0017] The fifth step: Design a data filter Calculate i acf (dt), and

[0018] The sixth step: Calculate and

[0019] The seventh step: Define three quadratic criterion functions J fl (θ l ), J fh (θ h ) and J v (θ v );

[0020] The eighth step: Determine the damping factor λ fl (λ fl ≥0), λ fh (λ fh ≥0) and λ v (λ v ≥0);

[0021] The ninth step: Update and

[0022] The tenth step: If then n = n + 1; otherwise, obtain and End the process.

[0023] The definitions of the variables are as follows:

[0024] Define the input quantity I dc (dt), and the output quantity i ac (dt);

[0025] Define q as the data length;

[0026] Define and as the relevant information vectors;

[0027] Define θ l , θ h and θ v as the parameter vectors;

[0028] Define and as the estimated values of θ l , θ h and θ v at the nth iteration, respectively.

[0029] 3) According to the parameter estimation method of the single-phase photovoltaic grid-connected inverter dual-rate Volterra-Wiener nonlinear output error model based on the Volterra series static nonlinear block, the finally derived filtering-based three-stage Levenberg–Marquardt iterative estimation algorithm is:

[0030] Define and as follows:

[0031] Define and as follows:

[0032] Define and as follows:

[0033] Define and as follows:

[0034] Define and as follows:

[0035] Define H(θ l), H(θ h ), and H(θ v ) are as follows:

[0036] Define and as follows:

[0037] Specific steps of the above algorithm:

[0038] 1) Start the identification algorithm, let n = 1, and set the initial values: and as random vectors;

[0039] 2) Obtain the DC input current I dc (dt) of the single-phase photovoltaic grid-connected inverter after double-rate sampling as input data, and the AC output current i ac (dt) of the single-phase photovoltaic grid-connected inverter after double-rate sampling as output data;

[0040] 3) Obtain

[0041] respectively through Equation (16) and Equation (17); and

[0042] 4) Obtain fl (λ fl ≥ 0), λ fh (λ fh ≥ 0), and λ v (λ v ≥ 0);

[0043] 6) Refresh the estimated and

[0044] 7) If then n = n + 1; and repeat steps three to nine, otherwise, obtain and to end the process.

[0045] The calculation of the present invention is accurate and applicable to the parameter identification of the Volterra-Wiener nonlinear model of a single-phase photovoltaic grid-connected inverter affected by colored noise and with information loss. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The present invention will be further described below in conjunction with the accompanying drawings and example simulations.

[0047] Figure 1 It shows the schematic diagram of a dual-rate system.

[0048] Figure 2 It shows the structural diagram of a dual-rate Volterra-Wiener nonlinear identification model for a single-phase photovoltaic grid-connected inverter with a static nonlinear function based on the Volterra series.

[0049] Figure 3 It shows the random value graph of the DC input current of a single-phase photovoltaic grid-connected inverter.

[0050] Figure 4 It shows the corresponding AC output current graph of a single-phase photovoltaic grid-connected inverter used for the model experiment.

[0051] Figure 5 It is the estimation process curve graph of parameters a1, a2, b1, b2, h 1,(1) , h 2,(0,0) , h 2,(0,1) , h 2,(1,1) , c1, d1.

[0052] Figure 6 It shows the parameter estimation error graph of the overall Volterra-Wiener nonlinear model of a single-phase photovoltaic grid-connected inverter obtained by using this algorithm.

[0053] Figure 7 It shows the comparison curve graph of the predicted AC output current of a single-phase photovoltaic grid-connected inverter and the actual AC output current of a single-phase photovoltaic grid-connected inverter obtained by using this algorithm.

Claims

1. The present invention aims to use an iterative estimation method for quickly identifying the model parameters of a single-phase photovoltaic grid-connected inverter disturbed by colored noise and with information loss, so as to effectively identify the parameters of the Volterra-Wiener nonlinear model of the single-phase photovoltaic grid-connected inverter disturbed by noise and with information loss. First, it is assumed that there is information loss during the process of collecting the DC input voltage U dc , DC input current I dc , AC output voltage u ac and AC output current i ac data. We adopt double-rate sampling to avoid some information loss. The double-rate system is shown in Figure 1. Second, a double-rate Volterra-Wiener nonlinear identification model of a single-phase photovoltaic grid-connected inverter based on a static nonlinear block of Volterra series is shown in Figure 2 as attached: Then, referring to the established model structure, the expression of the Volterra-Wiener linear dynamic sub-model of the single-phase photovoltaic grid-connected inverter is constructed as follows: w(t) is regarded as the colored noise of the interference and is defined as follows: The response component x of the discrete-time nonlinear Volterra system nol,k (t) is expressed as follows: Denote the coefficients of the k-th order kernel. In the two-rate Volterra nonlinear subsystem, the expression after replacing t with dt is as follows: For a K-th order Volterra series (where K is the maximum non-linear order), x nol,k (dt) and m(dt) are expressed as follows: Among them, L k is the memory length of the corresponding order. It can be sorted out as follows: For the dynamic linear subsystem in the dual-rate Volterra-Wiener nonlinear system, we can obtain: Assume h 1,0 = 1. After arranging the above equation, the complete expression of the dual-rate Volterra-Wiener nonlinear output error identification model is obtained as follows: Among them, I dc (dt) is the DC input current of the single-phase PV grid-connected inverter after double-rate sampling, x dt and m dt are intermediate variables, w(t) is regarded as colored noise of interference, i ac (dt) is the AC output current of the single-phase PV grid-connected inverter after double-rate sampling, and v(t) is white noise during the identification process.

2. For the iterative identification algorithm described in Claim 1, design the flow of the three-stage Levenberg–Marquardt iterative parameter identification algorithm based on filtering: The first step: Start the parameter identification algorithm; The second step: Let the iteration number n = 0 and set the initial value; Step 3: Obtain the DC input current I of the single-phase photovoltaic grid-connected inverter after double-rate sampling dc (dt) as input data, and the AC output current i of the single-phase photovoltaic grid-connected inverter after double-rate sampling ac (dt) as output data; Step 4: Construct and Step 5: Design a data filter Calculate i acf (dt), and Step 6: Calculate and Step 7: Define three quadratic criterion functions and Step 8: Determine the damping factor λ fl (λ fl ≥0), λ fh (λ fh ≥0) and λ v (λ v ≥0); Step 9: Update and Step 10: If then n = n + 1; otherwise, obtain and end the process. The definitions of the variables are as follows: Define the input quantity I dc (dt), and the output quantity i ac (dt); Define q as the data length; Definition and are relevant information vectors; Definition and be the parameter vectors; Definition and are respectively and the estimated values of the nth iteration. According to the parameter estimation method of the dual-rate Volterra-Wiener nonlinear output error model of the single-phase photovoltaic grid-connected inverter that constructs a static nonlinear block based on the Volterra sequence described in Claim 2, the finally derived three-stage Levenberg–Marquardt iterative estimation algorithm based on filtering is: The specific steps of the above algorithm: 1) Start the identification algorithm, let n = 1, and set the initial values: and are random vectors; 2) Obtain the DC input current I of the single-phase photovoltaic grid-connected inverter after double-rate sampling dc (dt) is used as input data, and the AC output current i of the single-phase photovoltaic grid-connected inverter after double-rate sampling ac (dt) is used as output data; 3) Obtain respectively through Equation (16) and Equation (17) 4) Obtained respectively through equations (18), (19) and (23)–(25) and 5) Determine the damping factor λ fl (λ fl ≥0), λ fh (λ fh ≥0) and λ v (λ v ≥0); 6) Refresh the estimates obtained through equations (26)–(34). and 7) If then n = n + 1; and repeat steps three to nine, otherwise, obtain and end the process.

Images