Photovoltaic simulation model two-stage parameter identification method based on quantum DBO algorithm

Abstract

The invention discloses a two-stage parameter identification method for a photovoltaic simulation model based on a quantum DBO algorithm, and the method comprises the following steps: S1, constructing an electromagnetic transient simulation model of a photovoltaic grid-connected system, and determining a to-be-identified parameter and an identification range of the to-be-identified parameter; s2, determining dominant parameters of the model by adopting Sobol sensitivity analysis; s3, constructing a model response scene under a plurality of groups of to-be-identified parameter values based on a Monte-Carlo method; s4, clustering different model response scenes by adopting a DBSCAN (Density Based Spatial Clustering of Applications with Noise) clustering algorithm; s5, matching a response result obtained by actual measurement with a clustering result to find a most matched scene; and S6, performing optimization solution by adopting a QOLDBO algorithm, and identifying the key parameters. According to the method, DBSCAN and QOLDBO algorithms are adopted, the parameter identification efficiency and precision are improved, local optimum is prevented, and a better parameter identification effect is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system parameter identification, and in particular to a two-stage parameter identification method of a photovoltaic simulation model based on a quantum DBO algorithm. Background Art

[0002] As the proportion of new energy sources connected to the grid continues to increase, the proportion of photovoltaic power generation in the power system is gradually increasing. Establishing an accurate simulation model is crucial for analyzing the operating characteristics of photovoltaic power generation after it is connected to the system. In the simulation model, accurate identification of key parameters is a key step to make the model closer to actual working conditions. Photovoltaic power generation systems have a high degree of nonlinearity due to the large number of power electronic devices used, and it is difficult to directly derive the mathematical relationship between model parameters. Therefore, a more effective parameter identification method must be adopted.

[0003] At present, commonly used parameter identification methods include heuristic algorithms, least squares methods, and Bayesian analysis. Heuristic algorithms are based on intuitive or empirical construction and give feasible solutions within a certain computing time, but they cannot accurately estimate the gap between the solution and the optimal solution, and the search efficiency in high-dimensional solution space is low and it is easy to fall into local optimality. In the process of parameter identification, it usually requires multiple calls to the simulation model, resulting in low computational efficiency. As a classic regression method, the least squares method takes minimizing the sum of squares of estimation errors as the optimization goal and is widely used in linear systems. However, for parameter estimation of nonlinear systems, the least squares method is not applicable, and when the sample size is large, the solution efficiency decreases significantly. Bayesian analysis combines prior probability with observed data and uses the Bayesian formula to calculate the posterior probability, thereby inferring unknown parameters. However, Bayesian analysis faces the challenge of computational complexity in high-dimensional state space, and the Markov Chain Monte Carlo (MCMC) method may fall into low probability areas, resulting in excessively long computational time. Summary of the invention

[0004] In order to solve the problems of low computational efficiency, easy to fall into local optimum and inaccurate parameter identification results of existing parameter identification methods, the present invention proposes a two-stage parameter identification method for photovoltaic simulation model based on quantum DBO algorithm. By combining clustering screening and optimization, it breaks through the limitations of traditional parameter identification methods and achieves a dual improvement in parameter identification efficiency and accuracy to solve the above problems.

[0005] The present application discloses a two-stage parameter identification method for a photovoltaic simulation model based on a quantum DBO algorithm, comprising the following steps:

[0006] S1. Construct an electromagnetic transient simulation model for a photovoltaic grid-connected system and determine the parameters to be identified and their identification range;

[0007] S2. Use Sobol sensitivity analysis to determine the dominant parameters of the electromagnetic transient simulation model of the photovoltaic grid-connected system;

[0008] S3, randomly generate different parameters within the identification range of the dominant parameter based on the Monte-Carlo method, so as to construct the model response scenarios under different dominant parameter values;

[0009] S4, using the DBSCAN clustering algorithm to cluster different model response scenarios;

[0010] S5. Match the response results obtained by actual measurement with the clustering results to find the best matching scene;

[0011] S6. Taking the average dominant parameters of the most matching scene as the initial starting point, the QOLDBO algorithm is used to find the optimal solution and identify the key parameters.

[0012] Preferably, the photovoltaic grid-connected system electromagnetic transient simulation model includes a photovoltaic array model, an MPPT control model, an inverter model and a control link model.

[0013] Preferably, the parameters to be identified in S1 include PI control parameters, PWM control parameters, DC voltage control parameters, power factor control parameters, IV characteristic parameters of the photovoltaic module, and maximum power point tracking algorithm parameters of the inverter controller, and the identification range of the parameters to be identified is ±10% of the rated value of each parameter to be identified.

[0014] Preferably, S2 comprises the following steps:

[0015] Sobol sensitivity analysis is performed for the model Y = f(X), where the input variables X = (x1, x2, ..., x m ) is the parameter to be identified obtained by S1, m is the number of parameters to be identified obtained by S1, the value range of the input variable is the identification range of the parameter to be identified set in S1, and the output variable Y is the active power and reactive power output during the three-phase ground short circuit fault of the photovoltaic grid-connected system;

[0016] In order to more intuitively measure the influence of the model input variables on the output, the Sobol total sensitivity index is used to measure the sensitivity of the output to each input variable. The Sobol total sensitivity index is used to measure the contribution of the input to the output variance. The total sensitivity index of each parameter to be identified is obtained through Sobol sensitivity analysis, and the parameters to be identified are sorted according to the total sensitivity index. The parameters to be identified whose total sensitivity index is at least two orders of magnitude larger than the total sensitivity index of other parameters to be identified are selected as the dominant parameters.

[0017] Preferably, the calculation formula of the total sensitivity index is:

[0018]

[0019] in, Indicates that in addition to X i All input variables except X ～i When is determined, the expected value of the conditional variance of the output Y is given, A and B are sample matrices, and N0 is the number of sampling points for calculating the sensitivity index.

[0020] Preferably, S4 comprises the following steps:

[0021] S41, arranging the response results of different response scenarios, namely, the output DC voltage, active power and reactive power of the photovoltaic grid-connected system when a three-phase short circuit fault occurs under different dominant parameters into a data set;

[0022] S42. The DBSCAN algorithm is used to cluster the data set. The clustering results automatically divide the data into different clusters according to the similarity of the response scenarios. Each cluster represents a parameter combination with similar response characteristics, and the noise points represent isolated responses that do not belong to any cluster.

[0023] S43. After clustering is completed, the average dominant parameter value and the corresponding response average value are calculated for each cluster.

[0024] Preferably, clustering the data set using the DBSCAN algorithm comprises the following steps:

[0025] S421, select any unvisited point: randomly select a point p from the data set. If the point has been visited, skip it and select the next unvisited point;

[0026] S422, check ε-neighborhood: calculate the number of points in the ε-neighborhood of point p;

[0027] If the number of points in the neighborhood is greater than or equal to MinPts, then point p is marked as a core point and a new cluster is created;

[0028] If the number of points in the neighborhood is less than MinPts, the marked point p is a noise point, which may be absorbed by the subsequent clustering expansion;

[0029] S423, extended clustering: for each core point, check other points in its neighborhood;

[0030] If the points in the neighborhood are also core points, the points in its neighborhood will continue to be added to the cluster to expand the cluster;

[0031] If the point in the neighborhood is a boundary point, only the point is added to the current cluster;

[0032] Clustering is complete when all data points have been visited and there are no new points to expand the cluster;

[0033] S424. Select two parameters, domain radius ε and MinPts, through the silhouette coefficient. The silhouette coefficient calculation formula is:

[0034]

[0035] Among them, a(i) is the average distance from the data point to other points in the same cluster, b(i) is the average distance from the data point to the nearest different cluster, and the closer s(i) is to 1, the better the clustering effect of the data point is;

[0036] Obtain the combination of ε and MinPts that maximizes the average silhouette coefficient by traversing:

[0037]

[0038] By automatically selecting the best fit, there is no need to manually select the values ​​of ε and MinPts. Instead, the combination of ε and MinPts is selected based on the clustering effect.

[0039] Preferably, S5 comprises the following steps:

[0040] The improved correlation coefficient is used to calculate the similarity between the actual measurement value and each cluster scene sample. The improved correlation coefficient calculation formula is:

[0041]

[0042] Among them, r udc is the Pearson correlation coefficient of the output DC voltage, r P is the Pearson correlation coefficient of the output active power, r Q is the Pearson correlation coefficient of the output reactive power response;

[0043] The cluster with the largest correlation coefficient r(x) is selected as the best matching scenario.

[0044] Preferably, S6 comprises the following steps:

[0045] S61, initialization, using the average dominant parameter of the most matching scene as the starting point of initialization;

[0046] Normalize the average dominant parameter of the best matching scene and map it to the range of [0,1]:

[0047]

[0048] Among them, x i is one of the selected dominant parameters, x max is the maximum value of the selected dominant parameter, x min is the minimum value of the selected dominant parameter;

[0049] The normalized parameter value x' i Mapped to the quantum state angle parameter θ i , the normalized parameter value x' i will be mapped to the angle parameter θ i The range is [0,π], and the mapping formula is:

[0050] θ i =x' i π;

[0051] The quantum state angle parameter of the population initialization is obtained as:

[0052] ψ i =(cos(θ i ), sin(θ i ));

[0053] S62, quantum position update, the update formula based on the quantum state perspective is:

[0054] θ i (t+1)=θ i (t)+Δθ i ;

[0055] Among them, Δθ i is the update step size, and t is the current time.

[0056] Mapping to spatial position is:

[0057] X i (t+1)=X i (t)+R(X best (t)-X i (t))+α·rand;

[0058] Among them, X i (t) is the current position of the i-th dung beetle at time t, X best (t) is the position of the global optimal solution at the tth moment, R is the rolling control coefficient, generally between [0,1], α·rand represents a random disturbance term, α is the disturbance amplitude, and rand is a random number between [0,1], which is used to increase the diversity of solutions;

[0059] S63, scramble behavior, is used to enhance information exchange between populations and facilitate rapid convergence:

[0060] X i (t+1)=X i (t)+β(X j (t)-X i (t));

[0061] Among them, β is the weight factor that controls the proximity between the i-th dung beetle and the j-th dung beetle, X j (t) is the position of the jth dung beetle at time t;

[0062] S64, hiding behavior, expressed as:

[0063] X i (t+1)=X i (t)+H·randn;

[0064] Among them, H is the coefficient for controlling hiding behavior, and randn is a standard normal distribution random number, which is used to achieve local perturbation of the position and increase the local search capability;

[0065] S65, selecting the improved correlation coefficient value between the response result under the searched dominant parameter value and the actual measured response result as the fitness function value:

[0066] F(X) = r(x);

[0067] S66. In order to adapt to the different requirements of the early and late stages of the search process, a dynamic weight function is used to adjust the search step size and range:

[0068] X i (t+1)=X i (t)+ω(t)(X gbest (t)-X i (t))+α·rand(X tbest (t)-X i (t));

[0069]

[0070] Among them, ω(t) is a dynamic weight function, which is adjusted with the number of iterations, X gbest (t) is the global optimal solution of the current iteration, X tbest (t) is the local optimal solution of the current iteration, t max is the maximum number of iterations, ω max is the maximum weight value, ω min is the minimum weight value;

[0071] S67. Use the QOLDBO algorithm to continuously update the position of each dung beetle until one of the following conditions is met:

[0072] A. Reach the maximum number of iterations, that is, t = t max ;

[0073] B. Find the global optimal solution that satisfies the requirement, that is, the fitness function value is greater than the set threshold δ, that is, F(X)>δ.

[0074] Beneficial effects of the present invention:

[0075] (1) After using the Sobol global sensitivity analysis algorithm to determine the dominant parameters of the system model, the present invention combines the DBSCAN clustering algorithm and the QOLDBO algorithm to perform parameter identification on the dominant parameters of the model, which significantly improves the efficiency and accuracy of parameter identification of the photovoltaic grid-connected system simulation model.

[0076] (2) The present invention effectively narrows the search space for parameter identification by introducing the DBSCAN clustering algorithm, screens out a representative initial dominant parameter identification range, and avoids redundant calculations during global search.

[0077] (3) The present invention further optimizes the selected initial values ​​of dominant parameter identification through the QOLDBO algorithm to ensure that the parameter identification results are more accurate and less likely to fall into local optimality. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a flow chart of a two-stage parameter identification method of a photovoltaic simulation model based on a quantum DBO algorithm according to an embodiment of the present invention;

[0079] Figure 2 is an equivalent circuit diagram of a photovoltaic cell according to an embodiment of the present invention;

[0080] Figure 3 A topological structure diagram of a three-phase PWM inverter according to an embodiment of the present invention;

[0081] Figure 4 A schematic diagram of a dual-loop control structure according to an embodiment of the present invention; DETAILED DESCRIPTION

[0082] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and examples.

[0083] The present application embodiment discloses a two-stage parameter identification method for a photovoltaic simulation model based on a quantum DBO algorithm. The method flow is as follows: Figure 1 As shown, the following steps are included:

[0084] S1. Construct an electromagnetic transient simulation model for the photovoltaic grid-connected system and determine the parameters to be identified and their identification range.

[0085] The electromagnetic transient simulation model of the photovoltaic grid-connected system includes a photovoltaic array model, an MPPT control model, an inverter model and a control link model.

[0086] The photovoltaic array is composed of photovoltaic cells connected in series and in parallel. The equivalent circuit diagram of the photovoltaic cell is as follows: Figure 2 As shown, its mathematical model is:

[0087]

[0088] Where I is the output current of the photovoltaic cell, V is the output voltage of the photovoltaic cell, and I PV is the photocurrent of the photovoltaic cell, I0 is the reverse saturation current of the diode, q is the electron charge, R s is the equivalent series resistance, R P is the equivalent parallel resistance, A is the diode factor, k is the Boltzmann constant, T is the actual temperature, I d is the reverse saturation current of the diode at the actual temperature, is the leakage current caused by the parallel resistance of the photovoltaic cell. L For load.

[0089] Since the parallel resistance R P Usually larger than the series resistor R s If the parallel resistance R P , the simplified mathematical model of photovoltaic cells is obtained as follows:

[0090]

[0091] The relationship between the diode reverse saturation current and temperature is as follows:

[0092] I0=I on (T n / T) 3 exp[qE g (TT n ) / AkTT n ];

[0093]

[0094] Among them, I on is the reverse saturation current of the diode at room temperature, E g is the bandgap voltage constant, E g =1.12, T n is room temperature, T n =298K, V ocn It is the open circuit voltage of the battery under rated operating conditions.

[0095] The relationship between photocurrent, irradiance and temperature is as follows:

[0096]

[0097]

[0098] Among them, I pvn is the photocurrent of the battery under rated conditions, K lIt indicates the sensitivity of the photovoltaic cell output current to temperature changes. ΔT is the difference between the photovoltaic cell operating temperature and the standard test temperature (25°C in this implementation). G is the actual light intensity. n is the rated light intensity, G n =1000W / m 2 , I scn It is the short-circuit current of the battery under rated operating conditions.

[0099] The operating voltage and current of the photovoltaic array are:

[0100]

[0101] Among them, U array is the operating voltage of the photovoltaic array, I array is the working current of the photovoltaic array, U sub is the operating voltage of the photovoltaic cell, I sub is the working current of the photovoltaic cell, N s N is the number of photovoltaic cells connected in series in the photovoltaic array. p is the number of photovoltaic cells connected in parallel in the photovoltaic array.

[0102] For the MPPT control model, the photovoltaic output power will also change with the changes in light intensity, temperature and output voltage, but for a specific light intensity and temperature, there is only one corresponding maximum power point (MPP). Commonly used maximum power point tracking (MPPT) algorithms include proportional coefficient method, constant voltage tracking method, hill climbing method, fuzzy control method, perturbation observation method and conductance increment method. Among them, the perturbation observation method is the most widely used MPPT algorithm due to the small amount of measurement data and simple calculation. Therefore, the perturbation observation method is used as the MPPT algorithm in this embodiment.

[0103] The inverter model is as follows Figure 3 As shown, select the three-phase inductor current i a 、i b 、i c and DC capacitor voltage current u dc1 、u dc2 As the state variable, the state equation of the three-phase PWM inverter is obtained as follows:

[0104]

[0105] in,

[0106] x=[i a i b i c u dc1 u dc2 ];

[0107]

[0108]

[0109]

[0110]

[0111] R is resistance, L is inductance, C is capacitance, u a 、u b 、u c is the inverter output voltage, i dc1 is the DC current of the upper bridge arm, i dc2 is the DC current of the lower bridge arm.

[0112] Switching function S i (i=1,2,3) correspond to the three phases A, B, and C respectively. The relationship between T and the switching function S of the power electronic switch is:

[0113] T1=S1, T3=S2, T5=S3, T4=1-S1, T6=1-S2, T2=1-S3;

[0114] For the control link model, the photovoltaic grid-connected system of this embodiment adopts dual-loop control. Since the current inner loop is added on the basis of the voltage outer loop, the inner loop can be used to effectively suppress current disturbances. At the same time, the decoupling control of active and reactive power is achieved by adopting the feedforward decoupling control method in the dq rotating coordinate system. Figure 4 As shown, the dual-loop control controls the DC voltage and grid-connected current through the voltage outer loop and the current inner loop.

[0115] The control equation of the current inner loop is:

[0116]

[0117] Among them, k pi is the proportional constant of the current inner loop PI control link, k ii is the integral constant of the current inner loop PI control link, i dref and i qref V is given by the DC voltage outer loop and the reactive power outer loop sd is the voltage component on the d-axis, V sq is the voltage component of the q axis, s is the rotor angular position, i dref 、i qref Respectively represent the current reference values ​​on the dq axis, i d 、i q They represent the current components on the dq axis, ω is the angular velocity, V d 、V q is the voltage component on the actual dq axis.

[0118] The voltage outer loop control equation is:

[0119]

[0120] Among them, k pP k is the proportional constant of the DC voltage outer loop PI control, iP k is the integral constant of the DC voltage outer loop PI control, pQ is the proportional constant of the reactive power outer loop PI control, k iQ is the integral constant of the reactive power outer loop PI control, V dc is the DC voltage, V dc_ref is the reference DC voltage, Q is the actual reactive power, Q ref is the reference reactive power.

[0121] In the photovoltaic grid-connected system, in order to improve the accuracy of the simulation model and reduce the workload of parameter correction, the preferred parameters to be identified include the PI control parameters of the inverter controller (including the PI control parameters of the voltage outer loop and the current inner loop), PWM control parameters, DC voltage control parameters, power factor control parameters, IV characteristic parameters of the photovoltaic module (including the adjustment coefficients of the short-circuit current and the open-circuit voltage), and maximum power point tracking algorithm parameters. These parameters play a key role in the operation and dynamic response of the photovoltaic grid-connected system. In order to maintain the rationality and stability of the parameters to be identified, the identification range of these parameters to be identified is set to ±10% of the rated value of each parameter to be identified.

[0122] S2. Sobol sensitivity analysis is used to determine the dominant parameters of the electromagnetic transient simulation model of the photovoltaic grid-connected system.

[0123] In the electromagnetic transient simulation model of the photovoltaic grid-connected system, there are multiple parameters that affect the model output. If all parameters are identified, the simulation efficiency will undoubtedly be greatly reduced. Therefore, the embodiment of the present application uses a sensitivity analysis method to determine the dominant parameters of the model and identify them to reduce the workload of parameter identification. In this regard, the present embodiment adopts the Sobol global sensitivity analysis method to decompose the model and express it as a single parameter or a function of the combination of parameters, and then analyzes the influence of the parameters and the interaction between the parameters based on the variance of the single input parameter or the combination of parameters.

[0124] The steps of Sobol sensitivity analysis include:

[0125] For the model Y = f(X), where the input variables X = (x1, x2, ..., x m) is the parameter to be identified obtained by S1, m is the number of parameters to be identified obtained by S1, the value range of the input variable is the identification range of the parameter to be identified set in S1, and the output variable Y is the active power and reactive power output during the three-phase ground short circuit fault of the photovoltaic grid-connected system.

[0126] Decompose f(X) into:

[0127]

[0128] Among them, f0 is a constant, X i is the i-th input variable, X j is the jth input variable, f i For X i The effect function when changing alone, f ij For X i and X j The effect function when changing simultaneously, f 1，2，…，m is the effect function when all variables change simultaneously.

[0129] The premise for the decomposition of function f(X) is:

[0130]

[0131] Where k = i1,…,i s , represents the subscript of the input variable to be integrated. The above formula indicates that all terms after the decomposition of the function f(X) are orthogonal.

[0132] The conditional expected value of the decomposition of function f(X) is:

[0133]

[0134] Among them, E(Y) is the overall mean of output Y, Indicates that given an input variable X i When , the conditional expected value of output Y is, Indicates that given an input variable X i and X j When the conditional expectation of output Y is j For X j The effect function when changing alone.

[0135] Assuming further that f(X) is square integrable, decompose the function f(X) into squares and integrate them in the form of variance:

[0136]

[0137]

[0138] Among them, V ar(Y) represents the total variance of the model output, that is, the variance of the output variable Y, V i The input variable X i The contribution to the total variance when acting alone, V j The input variable X j The contribution to the total variance when acting alone, V ij For X i and X j The contribution to the total variance when acting simultaneously, V 12…m represents the contribution of the higher-order interaction effects of all input variables to the output variance, Represents the input variable X i The variance of the conditional expectation of the output Y when determining, Represents the input variable X i and X j The variance of the conditional expectation of the output Y when determining .

[0139] In order to more intuitively measure the impact of the model input variables on the output, the Sobol total sensitivity index is used to measure the sensitivity of the output to each input variable. The Sobol total sensitivity index is used to measure the contribution of the input to the output variance. The expression of the Sobol total sensitivity index is:

[0140]

[0141] in, Indicates that in addition to X i All input variables except X ～i In the case of certainty, output the expected value of the conditional variance of Y.

[0142] The overall sensitivity index is calculated using the Monte Carlo estimation method:

[0143]

[0144] in, Indicates that in addition to X i All input variables except X ～i When is determined, the expected value of the conditional variance of the output Y is given, A and B are sample matrices, and N0 is the number of sampling points for calculating the sensitivity index.

[0145] After obtaining the total sensitivity index of each parameter to be identified in the electromagnetic transient simulation model of the photovoltaic grid-connected system by performing Sobol sensitivity analysis on the parameters to be identified, the parameters to be identified are sorted according to the total sensitivity index, and the parameter whose total sensitivity index is at least two orders of magnitude larger than that of other parameters is selected as the dominant parameter, thereby reducing the computational burden of parameter identification.

[0146] S3. Based on the Monte-Carlo method, different parameters are randomly generated within the identification range of the dominant parameters to construct model response scenarios under different dominant parameter values. In the above-mentioned photovoltaic grid-connected system electromagnetic transient simulation model, the output DC voltage, active power and reactive power during the photovoltaic grid-connected system three-phase ground short-circuit fault are taken as the output response of the model under different parameters. Multiple groups of different parameters are randomly generated within the identification range of the calculated dominant parameters, and the corresponding response values ​​are extracted and collected to construct model response scenarios under multiple groups of dominant parameter values.

[0147] S4. Use the DBSCAN clustering algorithm to cluster different model response scenarios.

[0148] S41. The response results of different response scenarios, namely, the output DC voltage, active power and reactive power of the photovoltaic grid-connected system when a three-phase short circuit fault occurs under different dominant parameters, are organized into a data set.

[0149] S42. The DBSCAN algorithm is used to cluster the data set. The clustering results automatically divide the data into different clusters according to the similarity of the response scenarios. Each cluster represents a parameter combination with similar response characteristics. Noise points represent isolated responses that do not belong to any cluster and are abnormal responses or less common scenarios in the model operation.

[0150] S43. After clustering is completed, the average dominant parameter value and the corresponding response average value are calculated for each cluster.

[0151] The DBSCAN algorithm (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm. Its basic principle is to identify clusters in a data set by defining the area radius of each point and the minimum number of points in the area. The DBSCAN algorithm does not require a preset number of clusters, can identify clusters of any shape, and has strong robustness.

[0152] DBSCAN has two important parameters, ε and MinPts. ε is the domain radius, which represents the other points within the ε-domain of a point, and Minpts is the minimum number of domain points, which represents the minimum number of points within the ε-domain of a point to be called a core point.

[0153] Find the points in the ε-neighborhood by calculating the Euclidean distance between the data points:

[0154] N ε (p)={q∈D|Dist(p,q)≤ε};

[0155] Among them, ε is the neighborhood radius, which represents other points within the ε-neighborhood range of a point, q is another point in the data set D except point p, and D is the data set where points p and q are located.

[0156] The core conditions are:

[0157] |N ε (p)|≥MinPts;

[0158] The steps for clustering using the DBSCAN algorithm are:

[0159] S421. Select any unvisited point: randomly select a point p from the data set. If the point has been visited, skip it and select the next unvisited point.

[0160] S422, check ε-neighborhood: calculate the number of points in the ε-neighborhood of point p.

[0161] If the number of points in the neighborhood is greater than or equal to MinPts, then point p is marked as a core point and a new cluster is created;

[0162] If the number of points in the neighborhood is less than MinPts, the marked point p is a noise point, but the point p may be absorbed by the subsequent clustering expansion.

[0163] S423, extended clustering: for each core point, check other points in its neighborhood;

[0164] If the points in the neighborhood are also core points, the points in its neighborhood will continue to be added to the cluster to expand the cluster;

[0165] If a point in the neighborhood is a boundary point (i.e., a non-core point but in the neighborhood of a core point), only the point is added to the current cluster.

[0166] Clustering is complete when all data points have been visited and there are no new points that can expand the cluster.

[0167] S424. To ensure the clustering effect, two parameters, domain radius ε and MinPts, are selected through the silhouette coefficient. The silhouette coefficient calculation formula is:

[0168]

[0169] Among them, a(i) is the average distance from the data point to other points in the same cluster (intra-cluster compactness), and b(i) is the average distance from the data point to the nearest different cluster (inter-cluster separation).

[0170] The closer s(i) is to 1, the better the clustering effect of the data point is. The combination of ε and MinPts that maximizes the average silhouette coefficient is obtained by traversal:

[0171]

[0172] By automatically selecting the best fit, there is no need to manually select the values ​​of ε and MinPts. Instead, the combination of ε and MinPts is selected based on the clustering effect.

[0173] S5. Use the response results obtained by actual measurement to match the clustering results to find the most matching scene. Use the improved correlation coefficient to calculate the similarity between the actual measurement value and the response average value of each cluster sample after the above clustering. The calculation formula of the correlation coefficient is:

[0174]

[0175] Among them, x i ,y i are the values ​​of the i-th data point of the two variables, and x0 and y0 are the average values ​​of the two variables.

[0176] The corresponding responses are selected as DC voltage, active power and reactive power. On this basis, some improvements are made to the calculation of the correlation coefficient. The improved correlation coefficient calculation formula is:

[0177]

[0178] Among them, r udc is the Pearson correlation coefficient of the output DC voltage, r P is the Pearson correlation coefficient of the output active power, r Q is the Pearson correlation coefficient of the output reactive power response.

[0179] The cluster with the largest correlation coefficient r(x) is selected as the best matching scenario.

[0180] S6. Taking the average dominant parameters of the most matching scene as the initial starting point, the QOLDBO algorithm is used to find the optimal solution and identify the key parameters.

[0181] The QOLDBO algorithm (Quantum-Optimized Dung Beetle Optimization) is an algorithm based on the traditional dung beetle algorithm combined with quantum computing and adaptive strategies for optimization. Its optimization process is as follows:

[0182] S61, initialization, using the average dominant parameter of the most matching scene as the starting point for initialization.

[0183] Normalize the average dominant parameter of the best matching scene and map it to the range of [0,1]:

[0184]

[0185] Among them, x i is one of the selected dominant parameters, x max is the maximum value of the selected dominant parameter, x min is the minimum value of the selected dominant parameter.

[0186] The normalized parameter value x' i Mapped to the quantum state angle parameter θ i , the normalized parameter value x' i will be mapped to the angle parameter θ i The range is [0,π], and the mapping formula is:

[0187] θ i =x' i π;

[0188] The quantum state angle parameter of the population initialization is obtained as:

[0189] ψ i =(cos(θ i ), sin(θ i ));

[0190] S62, quantum position update, the update formula based on the quantum state perspective is:

[0191] θ i (t+1)=θ i (t)+Δθ i ;

[0192] Among them, Δθ i is the update step size, and t is the current time.

[0193] Mapping to spatial position is:

[0194] X i (t+1)=X i (t)+R(X best (t)-X i (t))+α·rand;

[0195] Among them, X i (t) is the current position of the i-th dung beetle at time t, X best (t) is the position of the global optimal solution at the tth moment, R is the rolling control coefficient, generally between [0,1], α·rand represents a random disturbance term, α is the disturbance amplitude, and rand is a random number between [0,1], which is used to increase the diversity of solutions.

[0196] S63, scramble behavior, is used to enhance information exchange between populations and facilitate rapid convergence:

[0197] Xi (t+1)=X i (t)+β(X j (t)-X i (t));

[0198] Among them, β is the weight factor that controls the proximity between the i-th dung beetle and the j-th dung beetle, X j (t) is the position of the jth dung beetle at time t.

[0199] S64, hiding behavior, expressed as:

[0200] X i (t+1)=X i (t)+H·randn;

[0201] Among them, H is the coefficient for controlling hiding behavior, and randn is a standard normal distribution random number, which is used to achieve local perturbation of the position and increase the local search capability.

[0202] S65, selecting the improved correlation coefficient value between the response result under the searched dominant parameter value and the actual measured response result as the fitness function value:

[0203] F(X) = r(x);

[0204] S66. In order to adapt to the different requirements of the early and late stages of the search process, a dynamic weight function is used to adjust the search step size and range:

[0205] X i (t+1)=X i (t)+ω(t)(X gbest (t)-X i (t))+α·rand(X tbest (t)-X i (t));

[0206]

[0207] Among them, ω(t) is a dynamic weight function, which is adjusted with the number of iterations, X gbest (t) is the global optimal solution of the current iteration, X tbest (t) is the local optimal solution of the current iteration, t max is the maximum number of iterations, ω max is the maximum weight value, ω min is the minimum weight value.

[0208] S67. Use the QOLDBO algorithm to continuously update the position of each dung beetle until one of the following conditions is met:

[0209] A. Reach the maximum number of iterations, that is, t = tmax ;

[0210] B. Find the global optimal solution that satisfies the requirement, that is, the fitness function value is greater than the set threshold δ, that is, F(X)>δ.

[0211] The photovoltaic grid-connected system simulation model parameter identification method based on DBSCAN-QOLDBO proposed in this application, after determining the dominant parameters, uses DBSCAN to cluster the response scenarios under different dominant parameter values, performs type matching on the actual measured results to find the most matching type, and then uses the QOLDBO algorithm to use the average dominant parameter of the matching type as the initial point and the improved correlation coefficient as the fitness function to perform further parameter identification in a two-stage parameter identification method. Through cluster matching, the method of this application narrows the parameter identification range, reduces the number of subsequent optimization algorithm iterations, effectively improves the overall parameter identification efficiency, reduces the amount of calculation for subsequent parameter identification, and can prevent falling into the local optimum, achieving better parameter identification effects.

[0212] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A two-stage parameter identification method for photovoltaic simulation model based on quantum DBO algorithm, characterized in that: The following steps are involved: S1. Construct an electromagnetic transient simulation model for a photovoltaic grid-connected system and determine the parameters to be identified and their identification range; S2. Use Sobol sensitivity analysis to determine the dominant parameters of the electromagnetic transient simulation model of the photovoltaic grid-connected system; S3, randomly generate different parameters within the identification range of the dominant parameter based on the Monte-Carlo method, so as to construct the model response scenarios under different dominant parameter values; S4, using the DBSCAN clustering algorithm to cluster different model response scenarios; S5. Match the response results obtained by actual measurement with the clustering results to find the best matching scene; S6. Taking the average dominant parameters of the most matching scene as the initial starting point, the QOLDBO algorithm is used to find the optimal solution and identify the key parameters.

2. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 1 is characterized in that: The photovoltaic grid-connected system electromagnetic transient simulation model includes a photovoltaic array model, an MPPT control model, an inverter model and a control link model.

3. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 2 is characterized in that: The parameters to be identified in S1 include PI control parameters, PWM control parameters, DC voltage control parameters, power factor control parameters, IV characteristic parameters of the photovoltaic module, and maximum power point tracking algorithm parameters of the inverter controller. The identification range of the parameters to be identified is ±10% of the rated value of each parameter to be identified.

4. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 3 is characterized in that: The S2 comprises the following steps: Sobol sensitivity analysis is performed for the model Y = f(X), where the input variables X = (x1, x2, ..., x m ) is the parameter to be identified obtained by S1, m is the number of parameters to be identified obtained by S1, the value range of the input variable is the identification range of the parameter to be identified set in S1, and the output variable Y is the active power and reactive power output during the three-phase ground short circuit fault of the photovoltaic grid-connected system; The total sensitivity index of each parameter to be identified is obtained through Sobol sensitivity analysis. The parameters to be identified are sorted according to the total sensitivity index, and the parameter to be identified whose total sensitivity index is at least two orders of magnitude larger than that of other parameters to be identified is selected as the dominant parameter.

5. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 4 is characterized in that: The calculation formula of the total sensitivity index is: in, Indicates that in addition to X i All input variables except X ～i When is determined, the expected value of the conditional variance of the output Y is given, A and B are sample matrices, and N0 is the number of sampling points for calculating the sensitivity index.

6. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 5 is characterized in that: The S4 comprises the following steps: S41, arranging the response results of different response scenarios, namely, the output DC voltage, active power and reactive power of the photovoltaic grid-connected system when a three-phase short circuit fault occurs under different dominant parameters into a data set; S42. The DBSCAN algorithm is used to cluster the data set. The clustering results automatically divide the data into different clusters according to the similarity of the response scenarios. Each cluster represents a parameter combination with similar response characteristics, and the noise points represent isolated responses that do not belong to any cluster. S43. After clustering is completed, the average dominant parameter value and the corresponding response average value are calculated for each cluster.

7. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 6 is characterized in that: The clustering of the data set using the DBSCAN algorithm comprises the following steps: S421, select any unvisited point: randomly select a point p from the data set. If the point has been visited, skip it and select the next unvisited point; S422, check ε-neighborhood: calculate the number of points in the ε-neighborhood of point p; If the number of points in the neighborhood is greater than or equal to MinPts, then point p is marked as a core point and a new cluster is created; If the number of points in the neighborhood is less than MinPts, the marked point p is a noise point, which may be absorbed by the subsequent clustering expansion; S423, extended clustering: for each core point, check other points in its neighborhood; If the points in the neighborhood are also core points, the points in its neighborhood will continue to be added to the cluster to expand the cluster; If the point in the neighborhood is a boundary point, only the point is added to the current cluster; Clustering is complete when all data points have been visited and there are no new points to expand the cluster; S424. Select two parameters, domain radius ε and MinPts, through the silhouette coefficient. The silhouette coefficient calculation formula is: Among them, a(i) is the average distance from the data point to other points in the same cluster, b(i) is the average distance from the data point to the nearest different cluster, and the closer s(i) is to 1, the better the clustering effect of the data point is; Obtain the combination of ε and MinPts that maximizes the average silhouette coefficient by traversing: By automatically selecting the best fit, there is no need to manually select the values ​​of ε and MinPts. Instead, the combination of ε and MinPts is selected based on the clustering effect.

8. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 7 is characterized in that: The S5 comprises the following steps: The improved correlation coefficient is used to calculate the similarity between the actual measurement value and each cluster scene sample. The improved correlation coefficient calculation formula is: Among them, r udc is the Pearson correlation coefficient of the output DC voltage, r P is the Pearson correlation coefficient of the output active power, r Q is the Pearson correlation coefficient of the output reactive power response; The cluster with the largest correlation coefficient r(x) is selected as the best matching scenario.

9. The two-stage parameter identification method of photovoltaic simulation model based on quantum DBO algorithm according to claim 8 is characterized in that: The S6 comprises the following steps: S61, initialization, using the average dominant parameter of the most matching scene as the starting point of initialization; Normalize the average dominant parameter of the best matching scene and map it to the range of [0,1]: Among them, x i is one of the selected dominant parameters, x max is the maximum value of the selected dominant parameter, x min is the minimum value of the selected dominant parameter; The normalized parameter value x' i Mapped to the quantum state angle parameter θ i , the normalized parameter value x' i will be mapped to the angle parameter θ i The range is [0,π], and the mapping formula is: i i =x' i ·p; The quantum state angle parameter of the population initialization is obtained as: ψ i =(cos(θ) i ),sin(θ i )); S62, quantum position update, the update formula based on the quantum state perspective is: i i (t+1)=θ i (t)+Δθ i ; Among them, Δθ i is the update step size, t is the current time; Mapping to spatial position is: X i (t+1)=X i (t)+R(X best (t)-X i (t))+α·rand; Among them, X i (t) is the current position of the i-th dung beetle at time t, X best (t) is the position of the global optimal solution at the tth moment, R is the rolling control coefficient, α·rand represents a random disturbance term, α is the disturbance amplitude, and rand is a random number between [0,1]; S63, scramble behavior, is used to enhance information exchange between populations and facilitate rapid convergence: X i (t+1)=X i (t)+β(X j (t)-X i (t)); Among them, β is the weight factor that controls the proximity between the i-th dung beetle and the j-th dung beetle, X j (t) is the position of the jth dung beetle at time t; S64, hiding behavior, expressed as: X i (t+1)=X i (t)+H·randn; Among them, H is the coefficient for controlling hiding behavior, and randn is a standard normal distribution random number, which is used to achieve local perturbation of the position and increase the local search capability; S65, selecting the improved correlation coefficient value between the response result under the searched dominant parameter value and the actual measured response result as the fitness function value: F(X) = r(x); S66. Use dynamic weight function to adjust search step size and range: X i (t+1)=X i (t)+ω(t)(X gbest (t)-X i (t))+H·rand(X tbest (t)-X i (t)); Among them, ω(t) is a dynamic weight function, which is adjusted with the number of iterations, X gbest (t) is the global optimal solution of the current iteration, X tbest (t) is the local optimal solution of the current iteration, t max is the maximum number of iterations, ω max is the maximum weight value, ω min is the minimum weight value; S67. Use the QOLDBO algorithm to continuously update the position of each dung beetle until one of the following conditions is met: A. Reach the maximum number of iterations, that is, t = t max ; B. Find the global optimal solution that satisfies the requirement, that is, the fitness function value is greater than the set threshold δ, that is, F(X)>δ.

Images