Photovoltaic inverter LVRT control mode based on MODE algorithm and parameter identification method thereof

Abstract

The invention discloses a photovoltaic inverter LVRT control mode based on an MODE algorithm and a parameter identification method thereof, and the method comprises the steps: firstly, building a photovoltaic controller semi-physical test platform through RT-LAB, carrying out the LVRT test of different working conditions, and obtaining a data set needed by parameter identification; secondly, taking the measured values of the power and the current in the LVRT period as optimization targets, and respectively identifying control parameters of a specified power mode and a specified current mode by adopting an MODE algorithm; and finally, establishing a PSASP photovoltaic electromechanical transient model, substituting a parameter identification result, and determining an LVRT control mode of a to-be-measured working condition by calculating an identification error of a specified power mode and a specified current mode on a measured value of the to-be-measured working condition. According to the method, the control mode and the control parameters of the photovoltaic inverter can be accurately identified, and the simulation precision of the power and the current during the LVRT period is effectively improved, so that the adaptability of the identification result to different LVRT working conditions is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic inverter control, and in particular to a photovoltaic inverter LVRT control method based on a MODE algorithm and a parameter identification method thereof. Background Art

[0002] Large-scale photovoltaic grid connection increases the uncertainty of grid operation and even affects the stability and safety of power supply. In order to accurately analyze the characteristics of photovoltaic grid connection, an accurate simulation model must be established. The accuracy of the model mainly depends on the control parameters, which are often kept confidential by manufacturers and cannot be directly obtained. Parameter identification is an effective way and important means to accurately obtain the "black box" control parameters.

[0003] Existing research mainly focuses on the identification of inverter PI parameters, and its results are difficult to accurately describe the dynamic characteristics of the low voltage ride through (LVRT) process. PV inverters should have a certain low voltage ride through capability. Therefore, studying LVRT control parameters is of great significance for establishing high-precision simulation models and accurately analyzing photovoltaic grid-connected characteristics. Artificial intelligence technology plays an important role in power and integrated energy systems, and also shows significant advantages in the field of parameter identification. However, general methods only consider the overall effect in the parameter identification process, and there is a risk of overfitting in some working conditions. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a photovoltaic inverter LVRT control mode and parameter identification method based on a multi-objective differential evolution algorithm (MODE). First, the optimization range of the parameters is determined according to the LVRT control mode and the typical values ​​of the parameters; secondly, the control parameters are identified by the MODE algorithm, and the adaptive optimization operator is set by using the population diversity; finally, the control mode is judged according to the LVRT control effect of the working condition based on the identification result. This method can accurately identify the control mode and control parameters of the photovoltaic inverter, effectively improve the simulation accuracy of the power and current during the LVRT period, and thus effectively improve the adaptability of the identification results to different LVRT working conditions.

[0005] The technical solution adopted by the present invention is:

[0006] The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm include the following steps:

[0007] Step 1: First, build a semi-physical test platform for photovoltaic controllers through RT-LAB, and perform LVRT tests under different working conditions to obtain the data set required for parameter identification;

[0008] Step 2: Secondly, taking the measured values ​​of power and current during LVRT as the optimization target, the MODE algorithm is used to identify the control parameters of the specified power mode and the specified current mode respectively;

[0009] Step 3: Finally, the PSASP photovoltaic electromechanical transient model is established and the parameter identification results are substituted into it. The LVRT control mode of the working condition to be tested is determined by calculating the identification error of the actual measured value of the working condition to be tested under the specified power mode and the specified current mode.

[0010] Step 4: Verify the identification effect of the proposed multi-objective identification strategy and MODE identification method on LVRT control parameters through DE and APSO comparison.

[0011] In step 1, in order to accurately obtain the parameter identification data set, the key points required for identification are extracted according to the LVRT working condition response characteristics. Figure 1 As shown, A, B and C correspond to the three stages before the disturbance, during the disturbance and after the disturbance, respectively. Among them, the B section and the C section can be divided into the transient interval B1 and the steady-state interval B2, the transient interval C1 and the steady-state interval C2, respectively. b0, b1 and b2 are the time corresponding to the disturbance point and the start and end points of the interval B2, respectively.

[0012] The power mean and current mean of interval B2 are taken as the key points of LVRT working condition and recorded as active power key point P t , Reactive power key point Q t , Active current key point i pt and reactive current key point i qt , reactive current key point i qt ,Right now:

[0013]

[0014] Where b1 and b2 are the time corresponding to the start and end points of interval B2 respectively; p is the active current.

[0015] Extracting key points from the measured working conditions can establish an identification data set; during voltage drop, the inverter is controlled to give priority to the reactive current i q To support u t , and then take into account i within the total current range p The output of i p and i q The actual value is:

[0016] i q,cmd =min(i q,LVRT ,i q,max )(2);

[0017]

[0018] In the formula, i q,cmd Indicates reactive current output instruction; i q,LVRT Indicates reactive current low penetration instruction; i q,max Indicates the upper limit of reactive current value; i p,cmd Indicates active current output instruction; i p,LVRT Indicates active current low penetration instruction; i max Indicates the upper limit of the total current value; i p,max Indicates the upper limit of active current value.

[0019] Generally speaking, for i q,cmd The operating conditions >1.1pu will be eliminated as limiting conditions. In order to maintain the LVRT capacity of the photovoltaic grid-connected system, reactive current can be used to support voltage recovery to avoid off-grid operation; i p and i q The main control methods are specified power control and specified current control, as shown in equations (4) to (5), respectively. In addition, i p It can also be achieved by crossing the front current i p0 and low voltage current limiting for control;

[0020]

[0021] In the formula, k 1,ip and k 2,ip They are low-throughput active current coefficient 1 and low-throughput active current coefficient 2 respectively; i p,set It is the low active current setting value; i p,LVRT is the active current during LVRT; k 1,iq , k 2,iq They are low-throughput reactive current coefficient 1 and low-throughput reactive current coefficient 2 respectively; i q,set It is the setting value of low reactive current; i q0 is the reactive current before crossing; i q,LVRT is the reactive current during LVRT; u lin is the low penetration threshold; u t Indicates the voltage drop value at the grid connection point.

[0022]

[0023] In the formula, k P is the low wear active power coefficient; P set is the low-crossing active power setting value; P0 is the active power before crossing; P LVRT is the active power during LVRT; k Q is the low-power reactive power coefficient; Q set is the reactive power setting value of low-crossing; Q0 is the reactive power before crossing; Q LVRT is the reactive power during LVRT.

[0024] Therefore, the control parameters required for the specified power mode are the power coefficient and the power set value, and the control parameters required for the specified current mode are the current control coefficient and the current set value. The above parameters can be identified by combining the data set and the proposed method.

[0025] In the step 2, the key points corresponding to the measured values ​​of power and current are taken as optimization targets, and the MODE algorithm is used to respectively identify the control parameters of active power and reactive power in the specified power mode and the specified current mode.

[0026] The step 2 comprises the following steps:

[0027] Step 2.1: In order to accurately identify the control mode and parameters of different LVRT working conditions of the photovoltaic inverter and ensure the adaptability of the identification results to the working conditions, the objective function is set as shown in formula (6):

[0028]

[0029] In formula (4), J1 and J2 are the total residual value and the maximum residual value respectively; g is the residual between the measured value and the identified value of the working condition g; m is the total number of working conditions; represents the sum of the absolute values ​​of residuals of m groups of working conditions; max|e g |Indicates the maximum absolute value of the operating condition residual.

[0030] Step 2.2: The MODE algorithm can simulate the multi-generation evolution process of the population to obtain individuals with the best fitness. However, unlike the natural selection mechanism of the genetic algorithm, the MODE algorithm uses the differential information between individuals for crossover mutation when searching for the best, which can expand the search range while not easily falling into the optimal state. The steps of using the MODE algorithm to identify LVRT control parameters include:

[0031] 2..2..1. Population initialization:

[0032] A random number r is set, and r∈(0,1), and an initial population T of size G is generated using typical values ​​of the parameters to be identified, as shown in equations (7) to (9).

[0033] T=[T1…T i …T G ] T’ (7);

[0034] T i =[x i1 ,…x ij ,…,x in ] (8);

[0035] x ij =minxj +r(maxx j -minx j ) (9);

[0036] Where, T i is the parent individual i of the population, i=1,2,...,G; x ij T i The value of the j-dimensional parameter, j = 1, 2, ..., n, n is T i Total dimensions; max x j is the upper bound of the j-dimensional parameter; min x j is the lower bound of the j-dimensional parameter; T represents the population set; T' represents the transposition; r(max x j -min x j ) indicates that it is used in [min x j ,max x j ] to generate x ij constraints.

[0037] 2.2.2. Individual variation:

[0038] Take any T from T a 、T b and T c , where a, b and c are not equal, calculate T b and T c The difference vector d of T a The result is recorded as T a (1) , the above process can be expressed as:

[0039] T a ( 1 )=T a +p m d=T a +p m (T b -T c ) (10);

[0040] In the formula, p m is the mutation operator; T a 、T b and T c They represent individuals with three different values, that is, three possible values ​​of the parameters to be identified.

[0041] 2.2.3. Individual crossover:

[0042] Set the crossover probability p c , take T i With T a (1)Perform crossover, i≠a,b,c, and ensure that at least one dimension is selected to obtain a new individual denoted as T a (2) ;

[0043] In order to improve the optimization speed, it is necessary to use boundary conditions to constrain it. The constraint conditions are shown in equations (11) to (12):

[0044]

[0045] In the formula, Represents individual T a (1) The result of the crossover operation; n represents the total dimension of the individual, that is, the total dimension of the parameters to be identified; Respectively represent T a (2) The values ​​in the 1st, 2nd, ..., j, ..., nth dimensions; represents the individual T constrained by boundary conditions a (2) The value of dimension j, the corresponding individual is denoted by T a (3) .

[0046] 2.2.4. Individual selection:

[0047] Compare T using the objective function J a and T a (3) , and retain individuals with small objective function values. The process is shown in equations (13) to (14):

[0048]

[0049] In the formula, J(T i ) is T i The objective function of g is the residual of working condition g; m is the total number of working conditions.

[0050]

[0051] Where, T a (4) is the individual after mutation crossover and boundary condition constraint processing; J(T a ) and J(T a (3) ) are individual T a and T a (3) The objective function of .

[0052] 2.2.5. Update T according to the selection result and return to step 2.2.2. Each traversal of T is recorded as an iteration. When the objective function corresponding to the best individual in T converges, stop the iteration.

[0053] Step 2.3: Selecting appropriate hyperparameters can effectively improve the accuracy and efficiency of the optimization results.

[0054] Mutation operator p m To reflect population diversity, a larger p should be set in the early stage of iteration m It can expand the search range and avoid local optimality. In the later stage of iteration, the individual diversity decreases, and the fitness of the population approaches convergence, which can increase the crossover probability p c To maintain diversity. Commonly used grid search and random search methods require manual comparison of a few parameter combinations, which reduces the possibility of finding the best parameter configuration and has low parameter adjustment efficiency. Information entropy is the most commonly used indicator to measure the purity of a sample set. The degree of data chaos is proportional to the entropy value. Therefore, information entropy can quantitatively describe the diversity of a population and its individuals.

[0055] From 2.2.1, we can see that T i Denotes the parameter to be identified, and the population size is G. Then the diversity H(T) of population T can be expressed as:

[0056]

[0057] x j =[x 1j ,...,x ij ,...,x Gj ] T (16);

[0058] In the formula, x j is the j-dimensional parameter; n represents the individual dimension; h represents x j Possible values ​​in [h0,h1]; h0 represents x j The minimum possible value; h1 represents x j The maximum possible value; p(x j =h) represents x j = the probability of h; logp(x j =h) means p(x j =h), the base is usually 2; 1j ,...,x ij ,...,x Gj Represents the values ​​of individuals 1,2,...,i,...,G in the population in dimension j.

[0059] Therefore, adaptive p m and p c Respectively expressed as:

[0060]

[0061] Where H0(T) is the diversity of the initial population; p m,min is the upper limit of mutation probability; p m,max is the lower limit of mutation probability; p c,max is the upper limit of the crossover probability; p c,min is the lower bound of the crossover probability.

[0062] Step 2.4: Set up a multi-objective identification strategy through the non-dominated sorting method as the optimization basis for the MODE algorithm.

[0063] Step 2.4.1: Traverse the initial population T and set a set of constraints on the newly generated individuals to improve the optimization efficiency, as shown in formula (18); generate offspring individuals of size G and record them as T i (4) , then the population size is expanded to 2G, as shown in formula (19):

[0064]

[0065] Where, T group is the combined population; J1(T i (4) ) represents individual T i (4) The first objective function value; J1(T i ) represents individual T i The first objective function value of J2(T i (4) ) represents individual T i (4) The second objective function value; J2(T i ) represents individual T i The second objective function value of T∪T (4) represents the combined population T and population T (3) individuals in

[0066] T1 represents the first individual in population T; T1 (4) Represents population T (4) T2 represents the first individual in population T; T2 represents the second individual in population T; Represents the population T (4) The second individual in T G represents the Gth individual in population T; Represents population T (4) The Gth individual in .

[0067] Step 2.4.2: Use quick sorting to sort T groupstratified, s is the total number of layers, then the solution set of each layer is recorded as: [T best,1 ,T best,2 ,...,T best,s ], T best,1 ,T best,2 ,...,T best,s denote the non-dominated solution sets in the 1st, 2nd, ..., sth layers respectively; the number of individuals included is [z1,z2, ...,z s ],z1,z2,...,z s represent the total number of non-dominated solutions in the 1st, 2nd, ..., sth layers respectively.

[0068] Step 2.4.3: For J1(T best,1 )、J1(T best,2 ),...,J1(T best,s ) and J2(T best,1 )、J2(T best,2 ),...,J2(T best,s ) are arranged in descending order, where J1(T best,1 )、J1(T best,2 ),...,J1(T best,s ) represent T best,1 ,T best,2 ,...,T best,s The value of the first objective function for all individuals; J2(T best,1 )、J2(T best,2 ),...,J2(T best,s ) represent T best,1 ,T best,2 ,...,T best,s The values ​​of the second objective function for all individuals.

[0069] So as to calculate individual T i The crowding distance L for J1 and J2 i,1 , L i,2 , as shown in formula (20) to formula (22):

[0070]

[0071]

[0072] L i =L i,1 +L i,2 (twenty two);

[0073] Where, L i T i The crowding distance; J1(T i+1 ) represents individual Ti+1 The first objective function value; J1(T i-1 ) represents individual T i-1 The first objective function value; maxJ1(T i ) represents individual T i The maximum value of the first objective function; minJ1(T i ) represents individual T i The first objective function minimum value; i represents the i-th individual in the non-dominated solution set; z represents the total number of non-dominated solutions;

[0074] J2(T i+1 ) represents individual T i+1 The second objective function value; J2(T i-1 ) represents individual T i-1 The second objective function value; max J2(T i ) represents individual T i The second objective function maximum value; min J2(T i ) represents individual T i The second objective function minimum.

[0075] Step 2.4.4: Crowding distance can maintain the diversity and uniformity of population distribution and avoid individual concentration. i Sort the individuals in each layer solution set and get the first G T i Constitute a new population T. After multiple updates, the individuals in T will converge to T best,1 .

[0076] As shown in Figure 2(a), if T a In T b The feasible region R(T b ), then Ta is better than T b , we say Ta dominates T b , remember T a is a non-dominated solution, T b is the dominant solution. If the non-dominated optimal solution in the population is not unique, it can be mapped to the objective space formed by J1 and J2 to form the Pareto front (PF). The non-dominated sorting method can divide the individuals in the population into multiple layers, as shown in Figure 2(b). In the multi-objective optimization process, the solution in PF is usually not unique, and the optimal solution needs to be selected based on the actual situation.

[0077] In order to determine the final result from the Pareto frontier, the fuzzy comprehensive evaluation method can be used to quantify the objective function and select the individual with the highest target satisfaction as the optimal solution. The target satisfaction y can be expressed as:

[0078]

[0079] In the formula, max(J1) is the maximum value of J1; max(J2) is the maximum value of J2; min(J1) is the minimum value of J1; min(J2) is the minimum value of J2. In step 3, it is difficult to directly derive the control method using the working condition LVRT characteristics. From formulas (4) to (5), it can be seen that the control method can be obtained by the parameters to be identified. LVRT , Q LVRT 、i p,LVRT and i q,LVRT The simulation effect is judged, and the identification errors of active and reactive power under different control modes are calculated respectively as shown in equations (24) to (25):

[0080]

[0081] In the formula, e g,P and e g,ip are the residuals when the active power of working condition g is controlled by specified power control and specified current control respectively; e g,Q and e g,iq are the residuals when the reactive power of working condition g is controlled by specified power control and specified current control respectively; β P and β ip are the relative errors when the active power of working condition g is controlled by specified power control and specified current control; β Q and β iq are the relative errors when the reactive power of working condition g is controlled by specified power and specified current respectively; P t Indicates the key point of active power during LVRT; i pt Indicates the key point of active current during LVRT; Q t Indicates the key point of reactive power during LVRT; i qt Indicates the key points of reactive current during LVRT.

[0082] The judgment methods of active and reactive LVRT control modes are similar. Taking active as an example, the error threshold is set to 0.15. g,P |<0.15 and |e g,ip |<0.15, β P <β ip When the working condition is specified power control, β P ≥β ip The working condition is specified current control. If β P ≥β ip At the same time, I pt ≈i p0 ,i p0 Indicates the initial value of active current. If it is, it is controlled according to the current before crossing. Otherwise, there is no additional control.

[0083] In step 4, in order to verify the improvement of the adaptability of the proposed multi-objective identification strategy to the identification results, the error mean and maximum values ​​are used to analyze and evaluate the adaptability of the parameter identification results to different working conditions, and the DE algorithm and the proposed MODE algorithm are compared, where the objective function of the DE algorithm is set to J1. In order to further verify the accuracy of the MODE algorithm identification parameters, the APSO algorithm is used to compare with the MODE algorithm.

[0084] The present invention provides a photovoltaic inverter LVRT control method and parameter identification method based on MODE algorithm, and the technical effects are as follows:

[0085] Aiming at the problem that it is difficult to obtain the control mode and parameters of the electromechanical transient model of photovoltaic inverters and thus difficult to establish an accurate model to analyze the grid-connected characteristics, a control mode and parameter identification method of the electromechanical transient model of photovoltaic inverters based on the MODE algorithm is proposed. The following conclusions can be drawn from the example results:

[0086] 1) Compared with the DE algorithm, the LVRT control parameters identified by the MODE algorithm can accurately simulate power and current, effectively improving the identification accuracy.

[0087] 2) The non-dominated sorting method is used to optimize the population update process, which effectively expands the search range and reduces the risk of the algorithm falling into a local optimal solution.

[0088] 3) Compared with the APSO algorithm, the MODE algorithm can obtain stable recognition results and improve the recognition efficiency and accuracy.

[0089] 4) The present invention proposes a multi-objective identification strategy to enhance the adaptability of inverter control parameters to different working conditions. The differential evolution algorithm is simple in principle and converges quickly, and can effectively solve many different types of complex optimization problems, which is suitable for the requirements of photovoltaic inverter control parameters for identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0091] Figure 1 This is a diagram for dividing the LVRT time periods.

[0092] Figure 2(a) shows the individual T i Mapping in target space;

[0093] Figure 2(b) shows the stratification results of population T.

[0094] Figure 3 This is a flow chart of the LVRT control method of photovoltaic inverter based on the MODE algorithm and its parameter identification.

[0095] Fig. 4(a) is a population distribution diagram for identifying the parameters of the specified current control mode (before T is updated);

[0096] FIG4( b ) is a population distribution diagram for identifying parameters of a specified current control mode (after T is updated).

[0097] FIG5( a ) is a population distribution diagram for identifying parameters of a specified power control mode (before T is updated);

[0098] FIG5( b ) is a population distribution diagram for identifying parameters of a specified power control mode (after T is updated).

[0099] Figure 6(a) shows the simulation waveforms of working condition 1 under different control modes (active power control);

[0100] Figure 6(b) shows the simulation waveforms of working condition 1 under different control modes (reactive power control).

[0101] Figure 7(a) shows the simulation waveforms of working condition 2 under different control modes (active power control);

[0102] Figure 7(b) shows the simulation waveforms of working condition 2 under different control modes (reactive power control).

[0103] Figure 8 This is the identification comparison chart of DE algorithm and MODE algorithm.

[0104] Figure 9 is a comparison of the identification results of (a) PSO and MODE (operating condition 1);

[0105] Figure 9 is (b) a comparison of the identification results of PSO and MODE (operating condition 2);

[0106] Figure 9 is a comparison of the identification results of (c) PSO and MODE (operating condition 3);

[0107] Figure 9 is a comparison of the identification results of (d) PSO and MODE (operating condition 4);

[0108] Figure 9 is a comparison of the identification results of (e) PSO and MODE (operating condition 5);

[0109] Figure 9 is a comparison of the identification results of (f) PSO and MODE (operating condition 6). DETAILED DESCRIPTION

[0110] The present invention will be further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0111] The photovoltaic inverter LVRT control mode and parameter identification method based on the MODE algorithm include: first, building a photovoltaic controller semi-physical test platform through RT-LAB and performing LVRT tests on different working conditions to obtain the data set required for parameter identification; second, taking the measured values ​​of power and current during LVRT as the optimization target, the MODE algorithm is used to identify the control parameters of the specified power mode and the specified current mode respectively. Finally, the PSASP photovoltaic electromechanical transient model is established and the parameter identification results are substituted into it. The LVRT control mode of the working condition to be tested is determined by calculating the identification error of the two control modes for the measured values ​​of the working condition to be tested. The identification effect of the proposed multi-objective identification strategy and MODE identification method on the LVRT control parameters is verified by DE and APSO comparison. The example shows that the proposed method can accurately identify the control mode and control parameters of the photovoltaic inverter, and effectively improve the simulation accuracy of power and current during LVRT. The present invention provides a photovoltaic inverter LVRT control mode and parameter identification method based on the MODE algorithm, which can effectively improve the adaptability of the identification results to different LVRT working conditions.

[0112] Verification example:

[0113] In order to verify the effectiveness and adaptability of the proposed method for identifying control modes and control parameters, the specific steps are as follows: Figure 3 shown.

[0114] Step 1: Use RT-LAB to build a semi-physical test platform for the photovoltaic controller to perform LVRT testing. Under different power conditions, LVRT tests of three-phase short-circuit faults and two-phase short-circuit faults are performed respectively. The specific measured working conditions are shown in Tables 1 and 2.

[0115] RT-LAB builds a semi-physical test platform for photovoltaic controllers, including RT-ALB host computer, real-time simulator, photovoltaic controller and its host computer. RT-ALB host computer can be used to build simulation models, configure interfaces, perform simulation operations and signal recording, and communicate with the real-time simulator. The semi-physical test platform introduces some physical hardware into the simulation loop, which can overcome interference factors such as inaccurate mathematical models and truly reflect actual working conditions.

[0116] Table 1 Working condition training set

[0117]

[0118] Table 2 Working condition test set

[0119]

[0120] Step 2: After extracting the key points of the working condition, an identification data set is established, and the proposed MODE algorithm is used to identify the active and reactive LVRT control methods and parameters. Set G = 50, take reactive power as an example, the distribution of the population T objective function when updated under the specified current control method and the specified power control method is shown in Figure 4 (a), Figure 4 (b), Figure 5 (a), Figure 5 (b), and the parameter solution in the Pareto frontier is recorded as layer 1.

[0121] As shown in Figure 4(a) and Figure 5(a), the initialization population used to identify active and reactive control parameters has a large distribution range, and the number of individuals in the first five layers of non-dominated solutions is relatively small. After the MODE algorithm is used for optimization, the population distribution results of active and reactive control parameters are shown in Figure 4(b) and Figure 5(b), respectively. It can be seen that the individuals are close to the first layer of non-dominated solutions and the distribution is more concentrated.

[0122] Step 3: After obtaining the optimal solution, use the above method to identify working conditions 1 to 2 respectively. Symmetric working condition 1: P0 = 0.2pu, u t =0.5pu, asymmetric condition 2: P0 = 0.2pu, u t =0.5pu, and substitute it into the photovoltaic electromechanical model built by PSASP. The LVRT waveforms output from working conditions 1 to 2 are shown in Figures 6(a), 6(b), 7(a), and 7(b).

[0123] From Figure 6(a) and Figure 6(b), it can be seen that the active waveform of working condition 1 is closer to the measured data under the specified power control, while the reactive waveform is accurate under the specified current control mode. From Figure 6(a) and Figure 6(b), it can be seen that the control parameters of active and reactive power of working condition 2 under the specified current control mode can accurately simulate the measured waveform.

[0124] The identification results analyzed by equations (24) and (25) are shown in Tables 3 and 4. For condition 1, β P and β ip Under the specified power control, the β Q and β iq Under the specified current control, the power consumption is reduced by 32.54% and 31.76% respectively compared with the specified power control. Therefore, the active power of this working condition adopts the specified power control method, and the reactive power adopts the specified current control method. For working condition 2, β P , β ip , β Q and β iq Under the specified current control, the power consumption is reduced by 10.927%, 10.624%, 9.236% and 8.907% respectively compared with the specified power control. It can be seen that the active current and reactive current control of working condition 2 both adopt the specified current control.

[0125] Table 3 Identification error of specified power control mode during LVRT

[0126] Working conditions <![CDATA[β P / %]]> <![CDATA[β ip / %]]> <![CDATA[β Q / %]]> <![CDATA[β iq / %]]> Condition 1 1.667 2.685 32.642 31.964 Condition 2 19.480 19.269 9.385 9.141

[0127] Table 4 Identification error of specified current control mode during LVRT

[0128] Working conditions <![CDATA[β P / %]]> <![CDATA[β ip / %]]> <![CDATA[β Q / %]]> <![CDATA[β iq / %]]> Condition 1 7.340 7.246 0.102 0.204 Condition 2 8.553 8.645 0.149 0.234

[0129] Step 4: Use the error mean I1 and maximum I2 to analyze and evaluate the adaptability of the parameter identification results to different working conditions, as shown in formula (26). The DE algorithm and the proposed MODE algorithm are compared. The specific symbol definitions are shown in Table 5. Among them, the objective function of the DE algorithm is set to J1.

[0130]

[0131] Table 5 Symbolic expression

[0132]

[0133] like Figure 8 As shown in the figure, for the active power evaluation index, compared with the DE algorithm, the MODE algorithm has a higher I 1,ip increased by 0.003, while I 1,P ,I 2,P and I 1,ip The results show that the proposed method achieves better identification results in the test set working conditions, and the multi-objective strategy can improve the adaptability of control parameters to different working conditions.

[0134] In order to further verify the accuracy of the MODE algorithm identification parameters, the APSO algorithm is used to compare with the proposed method. Symmetric working condition 3: P0 = 1.0pu, u t =0.35pu, Symmetrical working condition 4: P0 = 1.0pu, u t =0.5pu, asymmetric condition 5: P0 = 0.2pu, u t =0.2pu, asymmetric condition 6: P0 = 1.0pu, u t =0.2pu.

[0135] The results of calculating and analyzing the control method using equations (24) to (25) are shown in Table 6. For condition 3, β P , β ip , β Q and βiq Under the specified current control, the β is reduced by 18.299%, 20.597%, 38.065% and 37.328% respectively compared with the specified power control. For condition 4, β P , β ip , β Q and β iq Under the specified current control, the β is reduced by 3.485%, 4.592%, 36.14% and 35.543% respectively compared with the specified power control. For condition 5, β P and β ip Under the specified power control, the β Q and β iq Under the specified current control, the values ​​are reduced by 33.725% and 32.603% respectively compared with the specified power control. For condition 6, β P , β ip , β Q and β iq Under the specified current control, the power consumption is reduced by 10.798%, 8.763%, 37.083% and 35.877% respectively compared with the specified power control. It can be seen that the active power and reactive power of working condition 5 are controlled by the specified power control mode and the specified current control mode respectively, while the active power and reactive power of working conditions 3, 4 and 6 are controlled by the specified current control mode.

[0136] Table 6 Control mode identification results

[0137]

[0138]

[0139] According to the control mode and control parameter results, the simulation waveforms P, Q, i of working conditions 1 to 6 p and i q like Figure 9(a) to Figure 9(f) shown. Figure 9(a) to Figure 9(f) The specific analysis of the identification results of working conditions 1 to 6 is shown in Tables 7 and 8.

[0140] It can be seen that compared with the APSO algorithm, the proposed MODE algorithm has better β P and β ip The difference between the two algorithms is about 6% and 25% respectively, but the identification effect of working conditions 1, 2, 5 and 6 is significantly improved, making β P The decreases were 84%, 31%, 15%, and 23%, respectively. ip The β values ​​of the proposed MODE algorithm are lower than those of the APSO algorithm in conditions 1, 2, 3, 4 and 6. QThe β iq The decreases were 21%, 57%, 28%, 86% and 8% respectively. ip and β iq It can be seen that the proposed MODE algorithm has improved the identification accuracy of control parameters and the adaptability to working conditions.

[0141] Table 7 Identification error (APSO)

[0142] Working conditions <![CDATA[β P / %]]> <![CDATA[β ip / %]]> <![CDATA[β Q / %]]> <![CDATA[β iq / %]]> Condition 1 10.424 10.330 0.156 0.261 Condition 2 12.480 12.391 0.452 0.556 Condition 3 3.239 2.959 1.066 0.781 Condition 4 2.302 2.258 0.291 0.246 Condition 5 8.761 10.073 0.452 1.885 Condition 6 0.047 1.457 0.997 2.487

[0143] Table 8 Identification error (MODE)

[0144] Working conditions <![CDATA[β P / %]]> <![CDATA[β ip / %]]> <![CDATA[β Q / %]]> <![CDATA[β iq / %]]> Condition 1 1.667 2.685 0.102 0.204 Condition 2 8.553 8.645 0.149 0.234 Condition 3 3.454 3.181 0.840 0.561 Condition 4 2.882 2.842 0.075 0.032 Condition 5 7.437 6.930 0.490 1.920 Condition 6 0.036 1.546 0.780 2.279

Claims

1. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm are characterized by The following steps are involved: Step 1: Build a semi-physical test platform for photovoltaic controllers and perform LVRT tests under different working conditions to obtain the data set required for parameter identification; Step 2: Taking the measured values ​​of power and current during LVRT as the optimization target, the MODE algorithm is used to identify the control parameters of the specified power mode and the specified current mode respectively; Step 3: Establish the PSASP photovoltaic electromechanical transient model and substitute the parameter identification results. Determine the LVRT control method of the working condition to be tested by calculating the identification error of the actual measured value of the working condition to be tested under the specified power mode and the specified current mode.

2. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 1 is characterized in that: It also includes step 4: verifying the identification effect of the proposed multi-objective identification strategy and MODE identification method on LVRT control parameters through DE and APSO comparison.

3. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 1 is characterized in that: In step 1, in order to accurately obtain the parameter identification data set, the key points required for identification are extracted according to the LVRT working condition response characteristics. A, B and C correspond to the three stages before disturbance, during disturbance and after disturbance, respectively. Among them, the B segment and the C segment are respectively divided into the transient interval B1 and the steady-state interval B2, the transient interval C1 and the steady-state interval C2, and b0, b1 and b2 are the time corresponding to the disturbance point and the start and end points of the interval B2, respectively. The power mean and current mean of interval B2 are taken as the key points of LVRT working condition and recorded as active power key point P t , Reactive power key point Q t , Active current key point i pt and reactive current key point i qt , reactive current key point i qt ,Right now: Where b1 and b2 are the time corresponding to the start and end points of interval B2 respectively; p is the active current; Extract key points from the measured working conditions to establish an identification data set; during voltage drop, control the inverter to give priority to issuing reactive current i q To support u t , and thus take into account i within the total current range p The output of i p and i q The actual value is: in q,cmd =min(in q,LVRT ,in q,max )(2); In the formula, i q,cmd Indicates reactive current output instruction; i q,LVRT Indicates reactive current low penetration instruction; i q,max Indicates the upper limit of reactive current value; i p,cmd Indicates active current output instruction; i p,LVRT Indicates active current low penetration instruction; i max Indicates the upper limit of the total current value; i p,max Indicates the upper limit of active current value; For i q,cmd The working condition of >1.1pu will be eliminated as the limiting working condition; in order to maintain the LVRT capacity of the photovoltaic grid-connected system, the reactive current is used to support the voltage recovery to avoid off-grid operation; i p and i q The control methods include specified power control and specified current control, as shown in equations (4) to (5), respectively. In addition, i p It can also pass through the front current i p0 and low voltage current limiting for control; In the formula, k 1,ip and k 2,ip They are low-throughput active current coefficient 1 and low-throughput active current coefficient 2 respectively; i p,set It is the low active current setting value; i p,LVRT is the active current during LVRT; k 1,iq , k 2,iq They are low-throughput reactive current coefficient 1 and low-throughput reactive current coefficient 2 respectively; i q,set It is the setting value of low reactive current; i q0 is the reactive current before crossing; i q,LVRT is the reactive current during LVRT; u lin is the low penetration threshold; u t Indicates the voltage drop value at the grid connection point; In the formula, k P is the low wear active power coefficient; P set is the low-crossing active power setting value; P0 is the active power before crossing; P LVRT is the active power during LVRT; k Q is the low-power reactive power coefficient; Q set is the reactive power setting value of low-crossing; Q0 is the reactive power before crossing; Q LVRT is the reactive power during LVRT; Therefore, the control parameters required for the specified power mode are the power coefficient and the power setting value, and the control parameters required for the specified current mode are the current control coefficient and the current setting value, and the parameters can be identified in combination with the data set.

4. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 1 is characterized in that: The step 2 comprises the following steps: Step 2.1: In order to accurately identify the control mode and parameters of different LVRT working conditions of the photovoltaic inverter and ensure the adaptability of the identification results to the working conditions, the objective function is set as shown in formula (6): In formula (4), J1 and J2 are the total residual value and the maximum residual value respectively; g is the residual between the measured value and the identified value of the working condition g; m is the total number of working conditions; represents the sum of the absolute values ​​of residuals of m groups of working conditions; max|e g | represents the maximum absolute value of the residual error of the working condition; Step 2.2: Using the MODE algorithm to identify LVRT control parameters, including the following steps: 2.2.

1. Population initialization: Set a random number r, where r∈(0,1), and use the typical value of the parameter to be identified to generate an initial population T of size G, as shown in equations (7) to (9); T=[T1…T i …T G ] T’ (7); T i =[x i1 ,…x ij ,…,x in ] (8); x ij =minx j +r(maxx j -minx j ) (9); Where, T i is the parent individual i of the population, i=1,2,...,G; x ij T i The value of the j-dimensional parameter, j = 1, 2, ..., n, n is T i Total dimensions; max x j is the upper bound of the j-dimensional parameter; min x j is the lower bound of the j-dimensional parameter; T represents the population set; T' represents the transposition; r(max x j -min x j ) indicates that it is used in [min x j ,max x j ] to generate x ij Constraints of 2.2.

2. Individual variation: Take any T from T a , T b and T c , where a, b and c are not equal, calculate T b and T c The difference vector d of T a The result is recorded as T a (1) , the above process can be expressed as: T a (1) =T a +p m d=T a +p m (T b -T c )(10); In the formula, p m is the mutation operator; T a , T b and T c They represent three individuals with different values, i.e., three possible values ​​of the parameter to be identified; 2.2.

3. Individual crossover: Set the crossover probability p c , take T i With T a (1) Perform crossover, i≠a,b,c, and ensure that at least one dimension is selected to obtain a new individual denoted as T a (2) ; In order to improve the optimization speed, it is necessary to use boundary conditions to constrain it. The constraint conditions are shown in equations (11) to (12): Where, T a (2) Represents individual T a (1) The result of the crossover operation; n represents the total dimension of the individual, that is, the total dimension of the parameters to be identified; Respectively represent T a (2) The values ​​in the 1st, 2nd, ..., j, ..., nth dimensions; represents the individual T constrained by boundary conditions a (2) The value of dimension j, the corresponding individual is denoted by T a (3) ; 2.2.

4. Individual selection: Compare T using the objective function J a and T a (3) , and retain individuals with small objective function values. The process is shown in equations (13) to (14): In the formula, J(T i ) is T i The objective function of g is the residual of working condition g; m is the total number of working conditions; Where, T a (4) is the individual after mutation crossover and boundary condition constraint processing; J(T a ) and J(T a (3) ) are individual T a and T a (3) The objective function of 2.2.

5. Update T according to the selection result and return to step 2.2.

2. Each traversal of T is recorded as an iteration. When the objective function corresponding to the optimal individual in T converges, stop the iteration; Step 2.3: Choose appropriate hyperparameters; Step 2.4: Set up a multi-objective identification strategy through the non-dominated sorting method as the optimization basis for the MODE algorithm.

5. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 4 is characterized in that: The step 2.3 is as follows: From 2.2.1, we can see that T i Denotes the parameter to be identified, and the population size is G. Then the diversity H(T) of population T can be expressed as: x j =[x 1j ,...,x ij ,...,x Gj ] T (16); In the formula, x j is the j-dimensional parameter; n represents the individual dimension; h represents x j Possible values ​​in [h0,h1]; h0 represents x j The minimum possible value; h1 represents x j The maximum possible value; p(x j =h) represents x j = the probability of h; logp(x j =h) means p(x j =h) logarithmic function; x 1j ,...,x ij ,...,x Gj represents the value of individuals 1,2,...,i,...,G in the population in dimension j; Therefore, adaptive p m and p c Respectively expressed as: Where H0(T) is the diversity of the initial population; p m,min is the upper limit of mutation probability; p m,max is the lower limit of mutation probability; p c,max is the upper limit of the crossover probability; p c,min is the lower bound of the crossover probability.

6. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 4 is characterized in that: The step 2.4 includes the following steps Step 2.4.1: Traverse the initial population T and set a set of constraints on the newly generated individuals to improve the optimization efficiency, as shown in formula (18); generate offspring individuals of size G and record them as T i (4) , then the population size is expanded to 2G, as shown in formula (19): Where, T group is the combined population; J1(T i (4) ) represents individual T i (4) The first objective function value; J1(T i ) represents individual T i The first objective function value of J2(T i (4) ) represents individual T i (4) The second objective function value; J2(T i ) represents individual T i The second objective function value of T∪T (4) represents the combined population T and population T (3) individuals in T1 represents the first individual in population T; T1 (4) Represents the population T (4) T2 represents the first individual in population T; T2 represents the second individual in population T; T2 (4) Represents population T (4) The second individual in T G represents the Gth individual in population T; Represents population T (4) The Gth individual in Step 2.4.2: Use quick sorting to sort T group Layered, s is the total number of layers, then the solution set of each layer is recorded as: [T best,1 ,T best ,2,...,T best,s ], T best,1 ,T best ,2,...,T best,s denote the non-dominated solution sets in the 1st, 2nd, ..., sth layers respectively; the number of individuals included is [z1,z2, ...,z s ],z1,z2,...,z s denote the total number of non-dominated solutions in the 1st, 2nd, ..., sth layers respectively; Step 2.4.3: For J1(T best,1 )、J1(T best,2 ),...,J1(T best,s ) and J2(T best,1 )、J2(T best,2 ),...,J2(T best,s ) are arranged in descending order, where J1(T best,1 )、J1(T best,2 ),...,J1(T best,s ) represent T best,1 ,T best,2 ,...,T best,s The value of the first objective function for all individuals; J2(T best,1 )、J2(T best,2 ),...,J2(T best,s ) represent T best,1 ,T best,2 ,...,T best,s The values ​​of the second objective function for all individuals; So as to calculate individual T i The crowding distance L for J1 and J2 i,1 , L i,2 , as shown in formula (20) to formula (22): Where, L i T i The crowding distance; J1(T i+1 ) represents individual T i+1 The first objective function value; J1(T i-1 ) represents individual T i-1 The first objective function value of maxJ1(T i ) represents individual T i The first objective function maximum value; minJ1(T i ) represents individual T i The first objective function minimum value; i represents the i-th individual in the non-dominated solution set; z represents the total number of non-dominated solutions; J2(T i+1 ) represents individual T i+1 The second objective function value; J2(T i-1 ) represents individual T i-1 The second objective function value; max J2(T i ) represents individual T i The second objective function maximum value; min J2(T i ) represents individual T i The minimum value of the second objective function; Step 2.4.4: Crowding distance can maintain the diversity and uniformity of population distribution and avoid individual concentration. i Sort the individuals in each layer solution set and get the first G T i Constitute a new population T; after multiple updates, the individuals in T will converge to T best,1 .

7. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 6 is characterized by: In order to decide the final result from the Pareto frontier, the fuzzy comprehensive evaluation method is used to quantitatively express the objective function, and the individual with the highest target satisfaction is selected as the optimal solution; the target satisfaction y can be expressed as: In the formula, max(J 1) is the maximum value of J1; max(J2) is the maximum value of J2; min(J1) is the minimum value of J1; min(J2) is the minimum value of J2.

8. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 1 is characterized in that: In step 3, the parameters to be identified are LVRT , Q LVRT 、i p,LVRT and i q,LVRT The simulation effect is judged, and the identification errors of active and reactive power under different control modes are calculated respectively as shown in equations (24) to (25): In the formula, e g,P and e g,ip are the residuals when the active power of working condition g is controlled by specified power control and specified current control respectively; e g,Q and e g,iq are the residuals when the reactive power of working condition g is controlled by specified power control and specified current control respectively; β P and β ip are the relative errors when the active power of working condition g is controlled by specified power control and specified current control; β Q and β iq are the relative errors when the reactive power of working condition g is controlled by specified power and specified current respectively; P t Indicates the key points of active power during LVRT; i pt Indicates the key points of active current during LVRT; Q t Indicates the key points of reactive power during LVRT; i qt Indicates the key points of reactive current during LVRT.

9. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 8 is characterized in that: The judgment methods of active and reactive LVRT control modes are similar. In the active LVRT control mode, the error threshold is set to 0.

15. g,P |<0.15 and |e g,ip |<0.15, β P <β ip When the working condition is specified power control, β P ≥β ip When the working condition is specified current control; if β P ≥β ip At the same time, I pt ≈i p0 ,i p0 Indicates the initial value of active current. If it is, it is controlled according to the current before crossing. Otherwise, there is no additional control.

10. The photovoltaic inverter LVRT control method and parameter identification method based on the MODE algorithm according to claim 2 is characterized in that: In step 4, the error mean and maximum values ​​are used to analyze and evaluate the adaptability of the parameter identification results to different working conditions, and the DE algorithm and the MODE algorithm are compared, wherein the objective function of the DE algorithm is set to J1; in order to further verify the accuracy of the MODE algorithm identification parameters, the APSO algorithm is used to compare with the MODE algorithm.

Images