A state monitoring method for grid-connected photovoltaic power generation system based on PCA adaptive model

Abstract

The present invention discloses a state monitoring method for a grid-connected photovoltaic power generation system based on a PCA adaptive model, which aims to comprehensively extract the distance and cross-correlation characteristics between the sampling data of the grid-connected photovoltaic power generation system, and fully consider the time-varying characteristics of the data, so as to implement more reliable state monitoring. Specifically, the method of the present invention designs a PCA model, which replaces the classical cross-correlation by a comprehensive approach of distance and cross-correlation, thereby analyzing and extracting the distance-related characteristics between the sampling data. On this basis, the monitoring indicators are adaptively calculated through an exponentially weighted moving average strategy, thereby realizing adaptive state monitoring. Compared with the classical PCA model, the distance-related characteristics between the sampling data are further considered. Secondly, the method of the present invention adaptively smoothes the monitoring indicators through a sliding window technology, and simultaneously updates the upper control limit of the monitoring indicators in real time, giving full play to the advantages of adaptive modeling.

Description

Technical Field

[0001] The present invention relates to a method for monitoring the operating status of a grid-connected photovoltaic power generation system, and in particular to a method for monitoring the status of a grid-connected photovoltaic power generation system based on a PCA adaptive model. Background Art

[0002] The development of human civilization is inseparable from the utilization of energy. As productivity increases, so does the demand for electricity. Today, in addition to natural resources, the energy sources available to humanity include solar energy, wind energy, tidal energy, geothermal energy, and nuclear energy. Among these new energy sources, solar energy is one of the least restrictive, most convenient, and most efficient and clean. Solar photovoltaic power generation primarily utilizes solar panels, which directly convert solar energy into electricity. These panels rely on relatively few external conditions, are simple to construct, and can fully utilize a variety of open spaces, such as rooftops, wilderness, and deserts. The photovoltaic panels used in grid-connected photovoltaic power generation are the foundation of power generation. Promoting the development of the photovoltaic industry will accelerate the development of photovoltaic cell materials, which will drive a new industry.

[0003] Condition monitoring of grid-connected photovoltaic power generation systems is crucial for ensuring the proper functioning of these systems. With the increasing number of grid-connected photovoltaic power plants and the ever-expanding scale of these projects, manual inspections would consume significant manpower and material resources. Furthermore, once connected to the grid, photovoltaic systems become part of the power system. Failures in the distribution network or the photovoltaic system can impact the operation of the entire system. In the field of automation research, grid-connected photovoltaic power generation systems are one of the fastest-growing power technologies. Therefore, real-time monitoring of the proper functioning of grid-connected photovoltaic systems is crucial for ensuring a continuous and stable power supply.

[0004] In recent years, the use of real-time measurement data of grid-connected photovoltaic power generation systems to monitor their operating status has gradually been recognized by the industry, and a series of photovoltaic power generation operating status monitoring methods based on data feature analysis and extraction strategies have been derived. Among them, a type of data statistical feature extraction technology represented by the principal component analysis (PCA) algorithm can perform feature transformation on data such as current and voltage of photovoltaic power generation systems, thereby implementing status monitoring using the extracted representative features. The PCA algorithm extracts potentially useful features from the DC and AC voltage and current data of the photovoltaic power generation system and generates model errors, and then uses the squared Mahalanobis distance (i.e., T 2 Monitoring indicators) and squared Euclidean distance (i.e., Q monitoring indicators) are monitored respectively.

[0005] The key to the success of this technology lies in the strong cross-correlation characteristics of the voltage and current data at the DC and AC ends, and the requirement that the data roughly obey the normal distribution. Therefore, the current methods of directly using PCA technology to monitor the operating status of grid-connected photovoltaic power generation systems have a high false alarm rate, and the reliability needs further research and improvement. Furthermore, the PCA algorithm is essentially a technology for analyzing and extracting data variance (i.e., range of variation) characteristics. Considering that the photovoltaic power generation system is directly affected by sunlight and ambient temperature, the voltage and current data at the DC and AC ends have a large range of variation, and some weak faults that affect the solar energy conversion efficiency are not easy to distinguish. In addition, due to the frequency of changes in the sampling data of the grid-connected photovoltaic power generation system, T 2 If the Q monitoring indicator relies on a fixed upper control limit, it will give a high false alarm rate. Therefore, adaptively determining the upper control limit is also one of the technical difficulties that need to be overcome for the widespread application of this technology. Summary of the Invention

[0006] The primary technical problem addressed by this invention is how to extract distance-related features between sampled data from grid-connected photovoltaic power generation systems while fully accounting for the data's time-varying characteristics, thereby enabling more reliable condition monitoring. Specifically, the method employs a PCA model that, in addition to cross-correlation, further considers distance correlation to analyze and extract distance and cross-correlation features between sampled data. Furthermore, an exponentially weighted moving average strategy is employed to adaptively calculate monitoring indicators, thereby enabling adaptive condition monitoring.

[0007] The technical solution adopted by the method of the present invention to solve the above problems is: a method for monitoring the state of a grid-connected photovoltaic power generation system based on a PCA adaptive model, comprising the following steps:

[0008] Step (1): Under the normal operating state of the grid-connected photovoltaic power generation system, data at each sampling moment is collected at fixed time intervals and stored as an 11×1 dimensional data vector; wherein the data vector at each sampling moment is composed of 11 data, which are: light intensity, ambient temperature, DC voltage, DC current, output voltage of the DC converter, three-phase voltage and three-phase current of the AC inverter.

[0009] Step (2): The data vectors x1, x2, ..., x N The data matrix X=[x1,x2,…,x N ], then according to the formula Standardize the row vectors of each row in X to obtain a new matrix Among them, x i ∈R 11×1Represents the i-th data vector, i∈{1,2,…,N}, R 11×N Represents a 11×N dimensional real matrix, R 11×1 represents a 11×1 dimensional real vector, R represents a real number set, z j ∈R 1×N represents the row vector of the jth row in the data matrix X, Represents the new matrix The row vector of the jth row in , j∈{1,2,…,11}, μ j and δ j Represents the row vector z j The mean and standard deviation of all elements in .

[0010] Step (3): Calculate the new matrix separately The covariance matrix C1∈R 11×11 And the distance correlation matrix C2∈R 11×11 The specific implementation process is shown in steps (3.1) to (3.7).

[0011] Step (3.1): According to the formula After calculating the covariance matrix C1, C2 is initialized to a 11×11 dimensional zero matrix, and j=1 and parameter k=2 are set; where the superscript T represents the transpose of a matrix or vector.

[0012] Step (3.2): According to the formula Calculate the matrix α∈R N×N The element a in the i-th row and m-th column i,m ; where i∈{1,2,…,N}, m∈{1,2,…,N}, and Respectively The i-th and m-th elements in .

[0013] Step (3.3): According to the formula Calculate the matrix β∈R N×N The element b in row i and column m i,m ;in, express The row vector of the kth row in , and Respectively The i-th and m-th elements in .

[0014] Step (3.4): According to the following formula ①, perform double centering on matrices α and β respectively to obtain matrix A∈R N×N and B∈R N×N :

[0015]

[0016] In the above formula, represents the average value vector of all column vectors in matrix α, Represents the average value vector of all column vectors in matrix β, column vector J∈R N×1 All elements in are equal to 1, α0 represents the average value of all elements in the matrix α, and β0 represents the average value of all elements in the matrix β.

[0017] Step (3.5): Calculate according to the following formula ② and The distance correlation coefficient between them is ρ(j,k):

[0018]

[0019] In the above formula, A i,m represents the element in the i-th row and m-th column of matrix A, B i,m Represents the element in the i-th row and m-th column of matrix B.

[0020] Step (3.6): Set the element in the j-th row and k-th column of C2 to ρ(j, k), and at the same time set the element in the k-th row and j-th column of C2 to ρ(j, k), then determine whether k is equal to 11; if not, set k = k + 1 and return to step (3.3); if so, execute step (3.7).

[0021] Step (3.7): Determine whether j is equal to 10; if not, first set j = j + 1, then set k = j + 1, and then return to step (3.2); if so, the distance correlation matrix C2∈R 11×11 .

[0022] Step (4): Calculate the comprehensive correlation matrix C∈R according to the formula C=C1+C2 11×11 Then, calculate the 11 eigenvalues ​​of C and their corresponding eigenvectors, and arrange the eigenvalues ​​in descending order according to their numerical values. The corresponding eigenvectors are recorded as p1, p2, ..., p 11 ; Among them, the length of the eigenvector is equal to 1, that is:

[0023] Step (5): After setting the number of characteristic components equal to d, according to the formula Calculate the monitoring index vector Q∈R N×1 ; Among them, the feature transformation matrix P=[p1,p2,…,p d ], I represents the 11×11 dimensional identity matrix, and diag{} represents the operation of converting the diagonal elements of the matrix in the braces into column vectors.

[0024] Step (6): Determine the initial upper limit D of the monitoring index according to formula ③ and formula ④ as shown below: lim and the initial upper limit Q lim Then, set the weight coefficient to c, set the number of sliding window samples to L, and initialize t = 1; where 1 < c ≤ 1.2.

[0025]

[0026] In the above formula, F d,N-d,ε It represents the value of F distribution with degrees of freedom d and Nd at the confidence level ε = 99%, represents the value of the chi-square distribution with h degrees of freedom at a confidence level of ε = 99%, g = v / (2w), h = 2w 2 / v, v and w represent the mean and variance of all elements in the monitoring indicator vector Q respectively.

[0027] In step (6) above, the recommended range of values ​​for the weight coefficient c is given. The number of sliding window samples, L, needs to be determined based on different scenarios and monitoring task requirements. For example, if a grid-connected photovoltaic power generation system samples at 1-minute intervals, and the monitoring task requires providing information on whether the photovoltaic panels are abnormal within 30 minutes, then L = 30 ÷ 1 = 30.

[0028] Step (7): At the latest sampling moment, collect 11 data of the grid-connected photovoltaic power generation system, and arrange these 11 data in the order described in step (1) to form an 11×1 dimensional data vector y t ∈R 11×1 Then, according to the formula y t Each element in is normalized to obtain a column vector Among them, j∈{1,2,…,11}, y t (j) and Represents y t and The jth element in .

[0029] Step (8): According to the formula and Calculate the monitoring index D at the current sampling time respectively t and Q t Then, determine whether t is less than L; if so, set t = t + 1 and return to step (7); if not, perform adaptive smoothing on the monitoring indicators of consecutive L sampling moments according to formulas ⑤ and ⑥ as shown below, so as to obtain the smoothed monitoring indicator and

[0030]

[0031] In the above formula, n=1,2,…,L.

[0032] Step (9): Update the adaptive control upper limit of the smooth monitoring index according to formula ⑦ and formula ⑧ as shown below: and adaptive control upper limit

[0033]

[0034] Step (10): Determine whether the conditions are met: and If yes, the grid-connected photovoltaic power generation system is operating normally, and the process returns to step (7) to continue monitoring using the data at the latest sampling time; if no, the grid-connected photovoltaic system is operating abnormally, and engineering technicians are dispatched for maintenance.

[0035] Through the above-mentioned implementation steps, the advantages of the method of the present invention are introduced as follows.

[0036] First, the method of the present invention constructs a PCA model by combining a covariance matrix and a distance correlation matrix. Compared with the classic PCA model, it further considers the distance correlation characteristics between the sampled data. Secondly, the method of the present invention uses sliding window technology to adaptively smooth the monitoring indicators and simultaneously updates the upper control limit of the monitoring indicators in real time, fully leveraging the advantages of adaptive modeling. Finally, in subsequent specific implementation cases, the superiority of the method of the present invention over traditional methods will be further illustrated through examples. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the implementation process of the method of the present invention.

[0038] Figure 2 This is a comparison chart of the method of the present invention and the traditional method for monitoring the operating status of a grid-connected photovoltaic power generation system. DETAILED DESCRIPTION

[0039] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0040] The present invention discloses a method for monitoring the state of a grid-connected photovoltaic power generation system based on a PCA adaptive model. Figure 1 The schematic diagram of the implementation process shown is used to illustrate the specific implementation of the method of the present invention.

[0041] Step (1): Under the normal operating state of the grid-connected photovoltaic power generation system, the measurement data at each sampling moment is collected at a fixed time interval and stored as an 11×1 dimensional data vector; wherein the data vector at each sampling moment is composed of 11 measurement data, which are: light intensity, ambient temperature, DC voltage, DC current, output voltage of the DC converter, three-phase voltage and three-phase current of the AC inverter.

[0042] Step (2): The data vectors x1, x2, ..., x N The data matrix X=[x1,x2,…,x N ], then according to the formula Standardize the row vectors of each row in X to obtain a new matrix

[0043] Step (3): Calculate the new matrix according to steps (3.1) to (3.7) as shown below The covariance matrix C1 and distance correlation matrix C2.

[0044] Step (3.1): According to the formula After calculating the covariance matrix C1, C2 is initialized to a 11×11 dimensional zero matrix, and j=1 and parameter k=2 are set.

[0045] Step (3.2): According to the formula Calculate the matrix α∈R N×N The element a in the i-th row and m-th column i,m .

[0046] Step (3.3): According to the formula Calculate the matrix β∈R N×N The element b in row i and column m i,m .

[0047] Step (3.4): According to the above formula ①, perform double centering on matrices α and β respectively to obtain matrix A∈R N×N and B∈R N×N .

[0048] Step (3.5): Calculate according to the above formula ② and The distance correlation coefficient ρ(j,k) between them.

[0049] Step (3.6): Set the element in the j-th row and k-th column of C2 to ρ(j, k), and at the same time set the element in the k-th row and j-th column of C2 to ρ(j, k), then determine whether k is equal to 11; if not, set k = k + 1 and return to step (3.3); if so, execute step (3.7).

[0050] Step (3.7): Determine whether j is equal to 10; if not, first set j = j + 1, then set k = j + 1, and then return to step (3.2); if so, obtain the distance correlation matrix C2.

[0051] Step (4): Calculate the comprehensive correlation matrix C∈R according to the formula C=C1+C2 11×11 Then, calculate the 11 eigenvalues ​​of C and their corresponding eigenvectors, and arrange the eigenvalues ​​in descending order according to their numerical values. The corresponding eigenvectors are recorded as p1, p2, ..., p 11 .

[0052] Step (5): After setting the number of characteristic components equal to d, according to the formula Calculate the monitoring index vector Q∈R N×1 .

[0053] Step (6): Determine the initial upper limit D of the monitoring index according to the above formula ③ and formula ④ respectively. lim and Q lim Then, set the weight coefficient to c, set the number of sliding window samples to L, and initialize t = 1.

[0054] Step (7): At the latest sampling moment, collect 11 measurement data of the grid-connected photovoltaic power generation system, and arrange these 11 measurement data in the order described in step (1) to form an 11×1 dimensional data vector y t ∈R 11×1 Then, according to the formula y t Each element in is normalized to obtain a column vector

[0055] Step (8): According to the formula and Calculate the monitoring index D at the current sampling time respectively t and Q t Then, determine whether t is less than L; if so, set t = t + 1 and return to step (7); if not, perform adaptive smoothing on the monitoring indicators of consecutive L sampling moments according to formulas ⑤ and ⑥ as shown below, so as to obtain the smoothed monitoring indicator and

[0056] Step (9): Update the adaptive control upper limit of the smooth monitoring index according to the above formula ⑦ and formula ⑧ and

[0057] Step (10): Determine whether the conditions are met: and If yes, the grid-connected photovoltaic power generation system is operating normally, and the process returns to step (7) to continue monitoring using the data at the latest sampling time; if no, the grid-connected photovoltaic system is operating abnormally, and grid-connected photovoltaic system maintenance is performed.

[0058] The method of the present invention is implemented according to the above steps (1) to (10), and finally the following is obtained: Figure 2 The operating status monitoring effect diagram is shown in the figure. Figure 2 It can be found that the method of the present invention can significantly reduce unnecessary false alarms and can timely monitor the operation abnormalities of the grid-connected photovoltaic power generation system.

Claims

1. A method for monitoring the state of a grid-connected photovoltaic power generation system based on a PCA adaptive model, characterized in that: The steps are as follows: Step (1): Under the normal operating state of the grid-connected photovoltaic power generation system, the measurement data at each sampling moment is collected at a fixed time interval and stored as an 11×1 dimensional data vector; wherein the data vector at each sampling moment is composed of 11 measurement data, and the arrangement order is: light intensity, ambient temperature, DC voltage, DC current, output voltage of the DC converter, three-phase voltage and three-phase current of the AC inverter; Step (2): The data vectors x1, x2, ..., x N The data matrix X=[x1,x2,…,x N ], then according to the formula Standardize the row vectors of each row in X to obtain a new matrix Among them, x i ∈R 11×1 Represents the i-th data vector, i∈{1,2,…,N}, R 11×N Represents a 11×N dimensional real matrix, R 11×1 represents a 11×1 dimensional real vector, R represents a real number set, z j ∈R 1×N represents the row vector of the jth row in the data matrix X, Represents the new matrix The row vector of the jth row in , j∈{1,2,…,11}, μ j and δ j Represents the row vector z j The mean and standard deviation of all elements in ; Step (3): Calculate the new matrix separately The covariance matrix C1∈R 11×11 And the distance correlation matrix C2∈R 11×11 ; Step (4): Calculate the comprehensive correlation matrix C∈R according to the formula C=C1+C2 11×11 Then, calculate the 11 eigenvalues ​​of C and their corresponding eigenvectors, and arrange the eigenvalues ​​in descending order according to their numerical values. The corresponding eigenvectors are recorded as p1, p2, ..., p 11 ; Among them, the length of the eigenvector is equal to 1; Step (5): After setting the number of characteristic components equal to d, according to the formula Calculate the monitoring index vector Q∈R N×1 ; Among them, the feature transformation matrix P=[p1,p2,…,p d ], I represents the 11×11 dimensional identity matrix, diag{} represents the operation of converting the diagonal elements of the matrix in the braces into column vectors, and the superscript T represents the transpose of the matrix or vector; Step (6): Determine the initial upper limit D of the monitoring index according to formula ① and formula ② as shown below: lim and Q lim Then, set the weight coefficient equal to c, set the number of sliding window samples equal to L, and initialize t = 1; where 1 < c ≤ 1.2; In the above formula, F d,N-d,ε It represents the value of F distribution with degrees of freedom d and Nd at the confidence level ε = 99%, represents the value of the chi-square distribution with h degrees of freedom at a confidence level of ε = 99%, g = v / (2w), h = 2w 2 / v, v and w represent the mean and variance of all elements in the monitoring indicator vector Q respectively; Step (7): At the latest sampling moment, collect 11 measurement data of the grid-connected photovoltaic power generation system, and arrange these 11 measurement data in the order described in step (1) to form an 11×1 dimensional data vector y t ∈R 11×1 , and then according to the formula y t Each element in is normalized to obtain a column vector Among them, j∈{1,2,…,11}, y t (j) and Represents y t and The jth element in ; Step (8): According to the formula and Calculate the monitoring index D at the current sampling time respectively t and Q t Then, determine whether t is less than L; if so, set t = t + 1 and return to step (7); if not, perform adaptive smoothing on the monitoring indicators of consecutive L sampling moments according to formulas ③ and ④ as shown below, so as to obtain the smoothed monitoring indicator and In the above formula, n = 1, 2, ..., L; Step (9): Update the adaptive control upper limit of the smooth monitoring index according to formula ⑤ and formula ⑥ as shown below: and Step (10): Determine whether the conditions are met: and If yes, the grid-connected photovoltaic power generation system is operating normally, and the process returns to step (7) to continue monitoring using the data at the latest sampling time; if no, the grid-connected photovoltaic system is operating abnormally, and grid-connected photovoltaic system maintenance is performed.

2. A method for monitoring the state of a grid-connected photovoltaic power generation system based on a PCA adaptive model according to claim 1, characterized in that: The specific implementation process of step (3) is as follows: Step (3.1): According to the formula After calculating the covariance matrix C1, initialize C2 to an 11×11 dimensional zero matrix and set j=1 and parameter k=2; where the superscript T represents the transpose of a matrix or vector; Step (3.2): According to the formula Calculate the matrix α∈R N×N The element a in the i-th row and m-th column i,m ; where i∈{1,2,…,N}, m∈{1,2,…,N}, and Respectively The i-th and m-th elements in ; Step (3.3): According to the formula Calculate the matrix β∈R N×N The element b in row i and column m i,m ;in, express The row vector of the kth row in , and Respectively The i-th and m-th elements in ; Step (3.4): According to the following formula ⑦, the matrices α and β are double-centered to obtain the matrix A∈R N×N and B∈R N×N : In the above formula, represents the average value vector of all column vectors in matrix α, Represents the average value vector of all column vectors in matrix β, column vector J∈R N×1 All elements in the matrix are equal to 1, α0 represents the average value of all elements in the matrix α, and β0 represents the average value of all elements in the matrix β; Step (3.5): Calculate according to the following formula ⑧ and The distance correlation coefficient between them is ρ(j,k): In the above formula, A i,m represents the element in the i-th row and m-th column of matrix A, B i,m Represents the element in the i-th row and m-th column of matrix B; Step (3.6): Set the element in row j, column k of C2 to ρ(j, k), and also set the element in row k, column j of C2 to ρ(j, k). Then determine whether k is equal to 11. If not, set k = k + 1 and return to step (3.3). If so, proceed to step (3.7). Step (3.7): Determine whether j is equal to 10; if not, first set j = j + 1, then set k = j + 1, and then return to step (3.2); if so, the distance correlation matrix C2∈R 11×11 .

Images