Photovoltaic power station equipment anomaly detection method using nonlinear feature updating technology

Abstract

The application discloses a photovoltaic power station equipment anomaly detection method using a nonlinear feature updating technology, aims at effectively mining nonlinear relationship features between photovoltaic inverter self-monitoring data and environmental monitoring data, and realizes detection of whether the photovoltaic inverter is abnormal through real-time extraction of the most discriminative nonlinear features. Specifically, the method relates to implementing nonlinear expansion on the photovoltaic inverter monitoring data and the environmental monitoring data through a nonlinear mapping mode, and explicitly represents the nonlinear relationship features between the data. Secondly, the method realizes two-step calling of online discriminant analysis, obtains discriminative features that maximize the distinction between online sampling data and normal working condition data, and thus can most directly and most effectively detect the operation anomaly of the photovoltaic power station equipment in real time through an anomaly detection index. Compared with a traditional method, the greatest advantage of the method lies in that discriminative features are analyzed and extracted in two steps in sequence, and corresponding anomaly detection indexes are calculated.

Description

Technical Field

[0001] This invention relates to a method for detecting anomalies in photovoltaic power plant equipment, and more particularly to a method for detecting anomalies in photovoltaic power plant equipment using nonlinear feature update technology. Background Technology

[0002] A photovoltaic (PV) inverter (or solar inverter) converts the variable DC voltage generated by photovoltaic (PV) solar panels into alternating current (AC) at the grid frequency. This AC voltage can be fed back into commercial transmission systems or supplied to off-grid systems. PV inverters are a crucial component of the balance of systems (BOS) in PV array systems. They can be used in conjunction with general AC-powered equipment and are among the most closely monitored devices in the daily operation and maintenance of PV power plants. However, in actual operation, PV inverters are affected not only by their own current and voltage variations but also by the surrounding operating environment. The performance characteristics of most PV inverter systems in practical applications are nonlinear, and establishing accurate mathematical models for such nonlinear systems is typically quite difficult.

[0003] Fortunately, with the widespread application of computer technology, photovoltaic power plants are now equipped with corresponding Supervisory Control and Data Acquisition (SCADA) systems. These systems provide continuous real-time sampling data at fixed sampling intervals for operators to monitor and use. In today's world where data resources are a core competitive advantage, the sampling data provided by SCADA systems offers a solid data foundation for data-driven equipment anomaly detection. Taking a typical photovoltaic power plant operation in China as an example, the SCADA system can monitor not only the inverter's own power data but also real-time monitoring data of the environment in which the inverter operates.

[0004] In recent years, the use of real-time measurement data for anomaly monitoring of photovoltaic power plant equipment operation has gradually gained industry recognition. Furthermore, most large-scale photovoltaic power plants in my country are located in the northwest region, where the environment is relatively harsh, making inverter maintenance extremely difficult. If an inverter malfunctions but is not detected and addressed in a timely manner, it can lead to irreparable economic losses and safety hazards. With the increasing demands for reliability and safety in photovoltaic power generation systems, and considering the nonlinearity of real-time monitoring data from photovoltaic power plants, analyzing real-time monitoring data, identifying nonlinear relationships between measurement data, and constructing a real-time and efficient anomaly detection model for photovoltaic inverters is of significant practical importance. Summary of the Invention

[0005] The main technical problem this invention aims to solve is how to effectively mine the nonlinear relationship characteristics between photovoltaic inverter monitoring data and environmental monitoring data, and how to detect whether a photovoltaic inverter is abnormal by extracting the most discriminative nonlinear features in real time through a two-step process. Specifically, the method of this invention first maps the original measurement data to a high-dimensional space using a third-order polynomial nonlinear mapping, then uses a two-step online discriminative feature analysis method to update and extract the most discriminative features in real time, and finally solves the problem of detecting whether a photovoltaic inverter is abnormal by using real-time calculated anomaly detection indicators.

[0006] The technical solution adopted by the present invention to solve the above problems is as follows: a method for detecting anomalies in photovoltaic power plant equipment using nonlinear feature update technology, comprising the following steps:

[0007] Step (1): With all photovoltaic power station equipment operating normally, collect and store photovoltaic inverter monitoring data and environmental monitoring data at fixed sampling time intervals to obtain... Data vector at each sampling time Among them, the first Data vector at each sampling time Specifically, it consists of monitoring data from 7 photovoltaic inverters and 6 environmental monitoring data, numbered... , express A 1-dimensional real vector It represents the set of real numbers.

[0008] Specifically, the Data vector at each sampling time The 13 data points are arranged in the following order: DC input current DC input voltage AC output current AC output voltage Photovoltaic inverter operating time Photovoltaic inverter power Photovoltaic inverter temperature Photovoltaic power plant power generation Photovoltaic power plant power Ambient temperature Ambient humidity ,radiation and wind speed ;in, to For monitoring data of photovoltaic inverters, to It is environmental monitoring data.

[0009] Step (2): Process the data vectors respectively A nonlinear mapping is performed to expand the 13 data points in each data vector into 273 data points, resulting in... indivual Extended data vector of dimension The specific implementation process is shown in steps (2.1) to (2.4).

[0010] Step (2.1): Initialization Afterwards, set The numbers represent the order of the numbers. Data vector at each sampling time The 13 data points.

[0011] Step (2.2): Construct the hybrid vector ;in, Indicates calculation The square of, Indicates calculation and The product of Indicates calculation and The product of, Indicates calculation and The product of , and so on.

[0012] Step (2.3): Determine Is it less than 13? If so, then set. Then return to step (2.2) to continue constructing the next blend vector; otherwise, the resulting 13 blend vectors will be... With data vector merged into the first Extended data vector at each sampling time point ; where the superscript T denotes the transpose of a matrix or vector.

[0013] Step (2.4): Follow the steps (2.1) to (2.3) above respectively. Implement a nonlinear mapping to obtain the corresponding result. indivual Extended data vector of dimension .

[0014] Step (3): Composition of data matrix Then, according to right The row vectors of each row in the matrix are standardized to obtain a new matrix. ;in, Representing a data matrix The Middle The row vector of a row. Represent the new matrix The Middle The row vector of a row. and Representing row vectors respectively The mean and standard deviation of all elements in the dataset. , express A real matrix of dimension 1.

[0015] Step (4): Settings Represent the new matrix in sequence Column 1, Column 2 , No. The column vectors are then processed according to steps (i) to (vi) to obtain the anomaly detection index vector. Afterwards, The maximum value of the element in the middle is recorded as .

[0016] Step 1: Initialization Then, set the column vector. sum matrix .

[0017] Step 2: The Middle Delete the column vectors of the column to obtain a new matrix. .

[0018] Step 3: Solving the generalized eigenvalue problem The largest eigenvalue in the middle corresponding feature vector Then through The initial discriminant vector is calculated. ;in, express A 3D real matrix Indicates calculation The absolute value of each element in the equation.

[0019] Step 4: Initialize the discriminant vector The largest The row position of each element is recorded in a set. Then, based on the set Will The elements of the corresponding rows form a column vector. and according to the set Will The row vectors of the corresponding rows form a matrix. .

[0020] Step 5: Solving the generalized eigenvalue problem The largest eigenvalue in the middle corresponding feature vector Afterwards, through The discriminant vector is calculated. Then according to the formula Calculated anomaly detection index .

[0021] Step 6: Vectorize the anomaly detection metrics The first in Set each element to equal Then, make a judgment Is it less than If so, then set Then, set the column vector. sum matrix If not, return to step (ii); otherwise, obtain the anomaly detection index vector. .

[0022] It should be noted that the implementation process of solving the generalized eigenvalue problem in steps (iii) and (v) above is actually an implementation process of obtaining the discriminant vector through an online discriminant analysis algorithm. The goal of this online discriminant analysis algorithm is to obtain the discriminant vector through... Will and Normal data vectors are distinguished to the greatest extent possible using anomaly detection metrics, namely: ①

[0023] In the above formula, the molecule Equivalent to calculation The corresponding anomaly detection index, denominator Equivalent to calculation The sum of the anomaly detection metrics corresponding to each column vector in the table.

[0024] To maximize equation ① above, the numerator must be made smaller. Maximize while ensuring the denominator Minimize. Without loss of generality, equation ① above can be equivalently transformed into the following objective function: ②

[0025] Equation ② above can be solved using the classic Lagrange multiplier method, which requires first using the coefficients... Construct the Langron function , then calculate Compared to The partial derivatives, i.e.: ③

[0026] When equation ③ equals 0, the objective function in equation ② takes an extreme value, from which we can obtain: ④

[0027] This is the origin of solving the generalized eigenvalue problem in steps (iii) and (v). Furthermore, if both sides of equation ④ above are multiplied by the left side... And simultaneously consider the constraints. Then we can get: Therefore, eigenvalues This is equal to the objective function in equation ② above. In other words, it is the generalized eigenvalue problem. We need to find the eigenvector corresponding to the largest eigenvalue.

[0028] Step (5): After collecting the monitoring data of 7 photovoltaic inverters and 6 environmental monitoring data at the latest sampling time, perform nonlinear mapping on these 13 data according to the implementation process in steps (2.1) to (2.3) to obtain the extended vector. .

[0029] Step (6): Use the average value from step (3) and standard deviation right Standardization is performed to obtain column vectors. Then set and .

[0030] Step (7): Calculate the anomaly detection index according to steps (iii) to (v). Then, make a judgment Is it greater than If not, return to step (5) and continue to use the data from the latest sampling time to perform anomaly detection; if yes, proceed to step (8).

[0031] Step (8): Return to step (5) and continue to perform anomaly detection using the data from the latest sampling time. If continuous The anomaly detection index at each sampling time was greater than If the condition is normal, an abnormal alarm will be triggered; otherwise, the photovoltaic power station equipment will operate normally.

[0032] In step (8) above, the continuous The parameters for each sampling time need to be determined based on different scenarios and anomaly detection task requirements. The specific numerical value. For example, if a grid-connected photovoltaic power generation system collects data at a sampling interval of 5 minutes, and the monitoring task requires providing information on whether the grid-connected photovoltaic power generation system is abnormal within 60 minutes, then... .

[0033] The advantages of the method of the present invention, based on the above implementation steps, are as follows.

[0034] First, the method of this invention involves applying a nonlinear mapping method to nonlinearly extend photovoltaic inverter monitoring data and environmental monitoring data, explicitly representing the nonlinear relationship characteristics between the data. Second, the method of this invention uses a two-step online discriminant analysis algorithm to analyze and extract discriminant features that maximize the distinction between online sampling data and normal operating condition data, thereby enabling the most direct and effective real-time detection of operational anomalies in photovoltaic power station equipment through anomaly detection indicators. Compared with traditional methods, the greatest advantage of the method of this invention lies in the two-step sequential analysis and extraction of discriminant features and the calculation of corresponding anomaly detection indicators. Attached Figure Description

[0035] Figure 1 This is a schematic diagram illustrating the implementation process of the method of the present invention. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0037] This invention discloses a method for detecting anomalies in photovoltaic power plant equipment using nonlinear feature update technology. The following is a detailed explanation of such a method. Figure 1 The implementation flowchart shown below illustrates the specific implementation method of the present invention.

[0038] Step (1): With all photovoltaic power station equipment operating normally, collect and store photovoltaic inverter monitoring data and environmental monitoring data at fixed sampling time intervals to obtain... Data vector at each sampling time .

[0039] Step (2): Process the data vectors respectively A nonlinear mapping is performed to expand the 13 data points in each data vector into 273 data points, resulting in... indivual Extended data vector of dimension The specific implementation process is shown in steps (2.1) to (2.4).

[0040] Step (3): Composition of data matrix Then, according to right The row vectors of each row in the matrix are standardized to obtain a new matrix. ;

[0041] Step (4): Settings Represent the new matrix in sequence Column 1, Column 2 , No. The column vectors are then processed according to steps (i) to (vi) to obtain the anomaly detection index vector. Afterwards, The maximum value of the element in the middle is recorded as .

[0042] Step (5): After collecting the monitoring data of 7 photovoltaic inverters and 6 environmental monitoring data at the latest sampling time, perform nonlinear mapping on these 13 data points according to the same implementation process as in steps (2.1) to (2.3) to obtain the extended vector. .

[0043] Step (6): Use the average value from step (3) and standard deviation right Standardization is performed to obtain column vectors. Then set and .

[0044] Step (7): Obtain the anomaly detection index based on steps (iii) to (v). Then, make a judgment Is it greater than If not, return to step (5) and continue to use the data from the latest sampling time to perform anomaly detection; if yes, proceed to step (8).

[0045] Step (8): Return to step (5) and continue to perform anomaly detection using the data from the latest sampling time. If continuous The anomaly detection index at each sampling time was greater than If the condition is normal, an abnormal alarm will be triggered; otherwise, the photovoltaic power station equipment will operate normally.

[0046] When implementing sub-steps (iii) and (v) in step (4), solving the generalized eigenvalue problem requires considering... or The singular problem, specifically, the method of the present invention solves the generalized eigenvalue problem according to the steps (A) to (C) shown below.

[0047] Step (A): According to the formula Calculate the matrix Then, on the matrix Perform singular value decomposition to obtain the unitary matrix. and singular value diagonal matrix When performing singular value decomposition, only non-zero singular values ​​need to be retained.

[0048] Step (B): According to the formula Calculate the matrix Then, according to Update matrix .

[0049] Step (C): Calculation The eigenvector corresponding to the largest eigenvalue Then, according to the formula Calculate the eigenvectors .

[0050] When solving the generalized eigenvalue problem in step (v), it is only necessary to replace the above step (A) with the following: Replace with The formula in step (B) above Replace with Then, in step (C) Replace with This will solve the generalized eigenvalue problem in step (v). eigenvectors Solve for it.

Claims

1. A method for detecting anomalies in photovoltaic power plant equipment using nonlinear feature update technology, characterized in that, Specifically, the steps are as follows: Step (1): With all photovoltaic power station equipment operating normally, collect and store photovoltaic inverter monitoring data and environmental monitoring data at fixed sampling time intervals to obtain... Data vector at each sampling time Among them, the first Data vector at each sampling time Specifically, it consists of monitoring data from 7 photovoltaic inverters and 6 environmental monitoring data, numbered... , express A 1-dimensional real vector Represents the set of real numbers; Step (2): Process the data vectors respectively A nonlinear mapping is performed to expand the 13 data points in each data vector into 273 data points, resulting in... indivual Extended data vector of dimensions The specific implementation process is shown in steps (2.1) to (2.4); Step (2.1): Initialization Afterwards, set The numbers represent the order of the numbers. Data vector at each sampling time The 13 data points; Step (2.2): Construct the hybrid vector ;in, Indicates calculation The square of, Indicates calculation and The product of Indicates calculation and The product of Indicates calculation and The product; Step (2.3): Determine Is it less than 13? If so, then set. Then return to step (2.2) to continue constructing the next blend vector; otherwise, the resulting 13 blend vectors will be... With data vector merged into the first Extended data vector at each sampling time point ; where the superscript T denotes the transpose of a matrix or vector; Step (2.4): Follow the steps (2.1) to (2.3) above respectively. Implement a nonlinear mapping to obtain the corresponding result. indivual Extended data vector of dimension ; Step (3): Composition of data matrix Then, according to right The row vectors of each row in the matrix are standardized to obtain a new matrix. ;in, Representing a data matrix The Middle The row vector of a row. Represent the new matrix The Middle The row vector of a row. and Representing row vectors respectively The mean and standard deviation of all elements in the dataset. , express A 3D real matrix; Step (4): Settings Represent the new matrix in sequence Column 1, Column 2 , No. The column vector of the column is then processed according to steps (i) to (vi) as shown below to obtain the anomaly detection index vector. Afterwards, The maximum value of the element in the middle is recorded as ; Step 1: Initialization Then, set the column vector. sum matrix ; Step 2: Delete The Middle The column vectors of the columns are used to obtain the new matrix. ; Step 3: Solving the generalized eigenvalue problem The largest eigenvalue in the middle Corresponding feature vector Then through The initial discriminant vector is calculated. ;in, express A 3D real matrix Indicates calculation The absolute values ​​of each element in the set; Step 4: Initialize the discriminant vector The largest The row position of each element is recorded in a set. Then, based on the set Will Elements in the same row form a column vector. and according to the set Will A matrix is ​​formed by row vectors of the same row. ; Step 5: Solving the generalized eigenvalue problem The largest eigenvalue in the middle Corresponding feature vector Afterwards, through The discriminant vector is calculated. Then according to the formula Calculated anomaly detection index ; Step 6: Vectorize the anomaly detection metrics The first in Set each element to equal Then, make a judgment. Is it less than If so, then set Then, set the column vector. sum matrix Then return to step (ii); otherwise, obtain the anomaly detection index vector. ; Step (5): After collecting the monitoring data of 7 photovoltaic inverters and 6 environmental monitoring data at the latest sampling time, perform nonlinear mapping on these 13 data points according to the same implementation process as in steps (2.1) to (2.3) to obtain the extended data vector. ; Step (6): Use the average value from step (3) and standard deviation right Standardization is performed to obtain column vectors. Then set and ; Step (7): Calculate the anomaly detection index according to steps (iii) to (v). Then, make a judgment Is it greater than If not, return to step (5) and continue to use the data from the latest sampling time to perform anomaly detection; if yes, proceed to step (8). Step (8): Return to step (5) and continue to perform anomaly detection using the data from the latest sampling time. If continuous The anomaly detection index at each sampling time was greater than If the condition is normal, an abnormal alarm will be triggered; otherwise, the photovoltaic power station equipment will operate normally.

2. The method for detecting anomalies in photovoltaic power plant equipment using nonlinear feature update technology according to claim 1, characterized in that, In step (iii) the feature vector The solution process is as follows: Step (A): According to the formula Calculate the matrix Then, on the matrix Perform singular value decomposition to obtain the unitary matrix. and singular value diagonal matrix When performing singular value decomposition, only non-zero singular values ​​need to be retained. Step (B): According to the formula Calculate the matrix Then, according to Update matrix ; Step (C): Calculation The eigenvector corresponding to the largest eigenvalue Then, according to the formula Calculate the eigenvectors .

Images