Fault early warning method, medium and system for photovoltaic power station inverter equipment

Abstract

The invention provides a fault early warning method for photovoltaic power station inverter equipment, a medium and a system, and belongs to the technical field of photovoltaic power stations, and the method comprises the steps: firstly collecting multi-dimensional operation data to construct a parameter matrix, and capturing a short-term operation state through time window segmentation; calculating parameter variation of adjacent windows, filtering extreme values to form a first variation matrix, calculating parameter variation rates based on three continuous windows to construct a second variation matrix, and forming a third variation matrix by applying a moving average algorithm in combination with a thermodynamic degradation mechanism; meanwhile, output current phase information is extracted, and a deviation angle and a drift rate are calculated to construct a phase change matrix. And then fusing the third variation matrix and the phase variation matrix, and constructing a comprehensive health index model by applying a self-adaptive weighting function. And finally, a health index threshold is set to trigger an early warning mechanism, and the fault time is predicted according to the index change trend, so that the technical problem of poor fault early warning timeliness of the photovoltaic power station inverter equipment is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic power stations, and more particularly, relates to a fault warning method, medium and system for an inverter device of a photovoltaic power station. Background Art

[0002] As a key device in a solar power generation system, the operation status of an inverter in a photovoltaic power station directly affects the stability and power generation efficiency of the entire power generation system. Traditional fault warning technologies for photovoltaic inverters mainly monitor a single parameter by setting a fixed threshold or use a simple statistical model based on historical data for fault judgment. Such methods are difficult to effectively capture the complex correlation characteristics among multi-dimensional parameters of the inverter device.

[0003] However, these traditional methods have obvious defects in practical applications: First, single-parameter monitoring cannot comprehensively reflect the overall operation status of the inverter; second, simple statistical models are difficult to adapt to the changes in the operation characteristics of the inverter under different environmental conditions; finally, the physical degradation mechanism of internal components of the device is ignored, resulting in the lack of theoretical support for the warning results.

[0004] Therefore, how to comprehensively consider the multi-dimensional operation parameters of the inverter, the influence of environmental factors, and the degradation mechanism of internal components to construct a fault warning model with high accuracy and strong timeliness has become the core technical problem to be solved urgently in the field of photovoltaic power station operation and maintenance management. That is to say, there is a technical problem of poor timeliness in fault warning of inverter devices in photovoltaic power stations in the prior art. Summary of the Invention

[0005] In view of this, the present invention provides a fault warning method, medium and system for an inverter device of a photovoltaic power station, which can solve the technical problem of poor timeliness in fault warning of inverter devices in photovoltaic power stations existing in the prior art.

[0006] The present invention is implemented as follows: In a first aspect of the present invention, a fault warning method for an inverter device of a photovoltaic power station is provided, including: collecting operation data of the inverter device to construct a multi-dimensional photovoltaic inverter operation parameter matrix; performing time series segmentation to construct a photovoltaic inverter parameter time window matrix; calculating the change amount of inverter parameters within adjacent time windows to filter extreme values and construct a first change amplitude matrix; calculating the parameter change rate within three consecutive time windows based on the first change amplitude matrix to construct a second change amplitude matrix; using a moving average algorithm and combining with the inverter thermodynamic degradation mechanism equation for optimization to construct a third change amplitude matrix; extracting the phase information of the inverter output current to construct a phase change matrix; fusing the third change amplitude matrix and the phase change matrix to construct a comprehensive health index model; setting a health index threshold, and triggering a warning mechanism when the index is lower than the threshold.

[0007] Among them, collecting the operating data of the inverter equipment includes input voltage, output current, power factor, harmonic content, equipment temperature, and ambient temperature. The multi-dimensional photovoltaic inverter operating parameter matrix includes a time dimension and a parameter dimension, and is a data structure formed by arranging the collected inverter operating parameters in chronological order. Each row represents a time point, and each column represents a parameter, forming a matrix.

[0008] Among them, the photovoltaic inverter parameter time window matrix refers to a subset of the inverter operating parameters intercepted within a fixed time length, used to capture the operating state of the equipment in each time period. The photovoltaic inverter parameter time window matrix reflects the short-term operating state of the inverter.

[0009] Among them, the first change amplitude matrix of the photovoltaic inverter refers to a matrix composed of the parameter differences between adjacent time windows. The first change amplitude matrix of the photovoltaic inverter reflects the short-term fluctuation characteristics of the inverter parameters and eliminates the influence of interference factors. The second change amplitude matrix of the photovoltaic inverter refers to a matrix composed of the parameter change rates within three consecutive time windows. The second change amplitude matrix of the photovoltaic inverter characterizes the accelerated change characteristics of the inverter parameters. The third change amplitude matrix of the photovoltaic inverter refers to a smoothed matrix obtained by applying a moving average algorithm to the second change amplitude matrix. The third change amplitude matrix of the photovoltaic inverter reduces the influence of random fluctuations and reflects the degree of degradation of the internal physical state of the inverter.

[0010] Among them, the phase change matrix of the photovoltaic inverter refers to a matrix composed of the phase angle offset and its change rate of the inverter output current, including two dimensions of phase offset angle and phase drift rate, used to evaluate the grid-connected synchronization control performance of the inverter. The phase change matrix of the photovoltaic inverter reflects the synchronization control performance of the inverter.

[0011] Among them, the comprehensive health index model of the photovoltaic inverter refers to an evaluation index constructed by weighted fusion of the third change amplitude matrix and the phase change matrix. The comprehensive health index model of the photovoltaic inverter quantifies the operating state of the inverter and provides a pre-judgment of the fault type.

[0012] Among them, the inverter thermodynamics degradation mechanism equation is used to describe the physical performance degradation law of the internal components of the inverter under different temperature conditions. The inputs include the internal temperature of the inverter, ambient temperature, operating duration, load rate, and heat dissipation system efficiency, and the output is the component degradation coefficient matrix. The component degradation coefficient matrix is used to construct the third change amplitude matrix of the photovoltaic inverter.

[0013] Among them, the parameter anomaly detection function is used to identify and correct extreme outliers generated during the data acquisition process, ensuring the accuracy of the change amplitude matrix. The inputs include the historical mean of the parameter, the historical standard deviation of the parameter, the environmental condition change rate, the sampling time interval, and the system working mode. The output is a parameter validity flag vector, which is used to construct the first change amplitude matrix of the photovoltaic inverter.

[0014] The second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions run on a computer, they are used to execute the above-mentioned fault warning method for a photovoltaic power station inverter device.

[0015] The third aspect of the present invention provides a fault warning system for a photovoltaic power station inverter device, including the above-mentioned computer-readable storage medium. The system can be any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is arranged inside the system, and a microprocessor for executing the program instructions stored in the computer-readable storage medium is arranged inside the system.

[0016] By constructing a multi-dimensional parameter matrix, segmenting the time window sequence, and a multi-level change amplitude matrix, the present invention realizes the comprehensive monitoring and evaluation of the operating state of the inverter device. It not only overcomes the limitations of traditional single-parameter monitoring, but also incorporates the physical state changes of the device into the warning model by integrating the thermodynamic degradation mechanism equation, significantly improving the accuracy of fault prediction. In particular, an adaptive weight fusion function is used to dynamically adjust the importance weights of different parameters, enabling the warning model to adapt to various environmental conditions and operating state changes, and enhancing the sensitivity to the degradation of the inverter synchronization control performance through the analysis of the phase change matrix.

[0017] Through the organic combination of the physical model and the data-driven method, the present invention realizes the early identification and accurate prediction of inverter faults, provides sufficient response time for the maintenance of photovoltaic power stations, and solves the technical problem of poor timeliness of fault warning for photovoltaic power station inverter devices existing in the prior art. Description of the Drawings

[0018] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention.

[0020] As Figure 1 shown, it is a flowchart of a fault warning method for a photovoltaic power station inverter device provided by the first aspect of the present invention. The method includes the following steps:

[0021] S01. Collect the operation data of the inverter equipment in the photovoltaic power station, including input voltage, output current, power factor, harmonic content, equipment temperature, and ambient temperature, and construct a multi-dimensional photovoltaic inverter operation parameter matrix, where the multi-dimensional photovoltaic inverter operation parameter matrix includes a time dimension and a parameter dimension;

[0022] S02. Segment the time series of the multi-dimensional photovoltaic inverter operation parameter matrix, select the data within a fixed time window, and construct a photovoltaic inverter parameter time window matrix, where the photovoltaic inverter parameter time window matrix reflects the short-term operation state of the inverter;

[0023] S03. Calculate the change amount of the photovoltaic inverter parameters within adjacent time windows, apply a parameter anomaly detection function to filter out extreme values, and construct a first change amplitude matrix of the photovoltaic inverter, where the first change amplitude matrix of the photovoltaic inverter reflects the short-term fluctuation characteristics of the inverter parameters and eliminates the influence of interference factors;

[0024] S04. Based on the first change amplitude matrix of the photovoltaic inverter, calculate the parameter change rate within three consecutive time windows, and construct a second change amplitude matrix of the photovoltaic inverter, where the second change amplitude matrix of the photovoltaic inverter characterizes the accelerated change characteristics of the inverter parameters;

[0025] S05. Use a moving average algorithm to smooth the second change amplitude matrix of the photovoltaic inverter, and at the same time combine the thermodynamic degradation mechanism equation of the inverter to optimize the data, and construct a third change amplitude matrix of the photovoltaic inverter, where the third change amplitude matrix of the photovoltaic inverter reduces the influence of random fluctuations and reflects the degradation degree of the internal physical state of the inverter;

[0026] S06. Extract the phase information of the inverter output current, calculate the phase offset angle and the phase drift rate, and construct a phase change matrix of the photovoltaic inverter, where the phase change matrix of the photovoltaic inverter reflects the synchronous control performance of the inverter;

[0027] S07. Integrate the third change amplitude matrix of the photovoltaic inverter and the phase change matrix of the photovoltaic inverter, apply an adaptive weight integration function for multi-dimensional feature combination, and construct a comprehensive health index model of the photovoltaic inverter, where the comprehensive health index model of the photovoltaic inverter quantifies the operation state of the inverter and provides a pre-judgment of the fault type;

[0028] S08. Set a threshold for the comprehensive health index of the photovoltaic inverter. When the calculation result of the comprehensive health index model of the photovoltaic inverter is lower than the threshold of the comprehensive health index of the photovoltaic inverter, trigger a fault warning mechanism and output a warning message;

[0029] S09. Predict the occurrence time of the inverter fault and output it according to the change trend of the calculation result of the photovoltaic inverter comprehensive health index model.

[0030] Among them, the multi-dimensional photovoltaic inverter operation parameter matrix specifically refers to a data structure formed by arranging the collected inverter operation parameters in chronological order. Each row represents a time point, and each column represents a parameter, forming a matrix.

[0031] Among them, the photovoltaic inverter parameter time window matrix specifically refers to a subset of the inverter operation parameters intercepted within a fixed time length, used to capture the operation status of the device in each time period.

[0032] Among them, the first change amplitude matrix of the photovoltaic inverter specifically refers to a matrix composed of the parameter differences between adjacent time windows.

[0033] Among them, the second change amplitude matrix of the photovoltaic inverter specifically refers to a matrix composed of the parameter change rates within three consecutive time windows.

[0034] Among them, the third change amplitude matrix of the photovoltaic inverter specifically refers to a smoothed matrix obtained by applying a moving average algorithm to the second change amplitude matrix.

[0035] Among them, the phase change matrix of the photovoltaic inverter specifically refers to a matrix composed of the phase angle offset and its change rate of the inverter output current, including two dimensions of phase offset angle and phase drift rate, used to evaluate the grid connection synchronization control performance of the inverter.

[0036] Among them, the photovoltaic inverter comprehensive health index model specifically refers to an evaluation index constructed by weighted fusion of the third change amplitude matrix and the phase change matrix.

[0037] Among them, the inverter thermodynamics degradation mechanism equation is used to describe the physical performance degradation law of the internal components of the inverter under different temperature conditions. The inputs include the internal temperature of the inverter, the ambient temperature, the operation duration, the load rate, and the efficiency of the heat dissipation system. The internal temperature of the inverter is obtained from the device temperature collected in step S01, the ambient temperature is obtained from the ambient temperature collected in step S01, the operation duration is obtained from the time stamp of the operation data of the photovoltaic power station inverter device, the load rate is calculated from the output current collected in step S01, and the efficiency of the heat dissipation system is obtained from the specification parameters of the inverter device. The output is the component degradation coefficient matrix, and the component degradation coefficient matrix is used to construct the third change amplitude matrix of the photovoltaic inverter.

[0038] Among them, the parameter anomaly detection function is used to identify and correct extreme outliers generated during the data acquisition process, ensuring the accuracy of the variation amplitude matrix. The inputs include the historical parameter mean value, the historical parameter standard deviation, the environmental condition change rate, the sampling time interval, and the system working mode. The historical parameter mean value is calculated from the multi-dimensional photovoltaic inverter operation parameter matrix, the historical parameter standard deviation is calculated from the multi-dimensional photovoltaic inverter operation parameter matrix, the environmental condition change rate is calculated from the environmental temperature collected in step S01, the sampling time interval is obtained from the timestamp of the operation data of the inverter equipment in the photovoltaic power station, and the system working mode is obtained from the photovoltaic power station control system. The output is a parameter validity flag vector, and the parameter validity flag vector is used to construct the first variation amplitude matrix of the photovoltaic inverter;

[0039] Among them, the adaptive weight fusion function is used to dynamically adjust the importance weights of different parameters in the health assessment to achieve more accurate fault warning. The inputs include the historical parameter fluctuation amplitude, the current parameter change rate, the parameter correlation coefficient, the equipment operation years, and the environmental impact factor. The historical parameter fluctuation amplitude is calculated from the first variation amplitude matrix of the photovoltaic inverter, the current parameter change rate is calculated from the second variation amplitude matrix of the photovoltaic inverter, the parameter correlation coefficient is calculated from the multi-dimensional photovoltaic inverter operation parameter matrix, the equipment operation years are obtained from the installation records of the inverter equipment in the photovoltaic power station, and the environmental impact factor is calculated from the environmental temperature collected in step S01. The output is a parameter weight vector, and the parameter weight vector is used to calculate the comprehensive health index model of the photovoltaic inverter.

[0040] The specific implementation manners of the above steps are described in detail below. The specific implementation manner of step S01 is to regularly sample the inverter equipment through the data collectors arranged at the photovoltaic power station site, and the sampling time interval is set to 1 minute. The collected data items include: input voltage (DC side voltage), with a collection range of 200 - 1000V; output current (AC side current), with a collection range of 0 - 100A; power factor, with a collection range of 0.8 - 1.0; harmonic content (total harmonic distortion rate), with a collection range of 0 - 5%; equipment temperature, with a collection range of 25 - 85°C; environmental temperature, with a collection range of -20 - 50°C. The analog signals are converted into digital signals through a high-precision analog-to-digital converter and are preliminarily processed by the data preprocessing unit, including operations such as signal filtering and missing value filling. Then, the data is organized according to the time dimension and the parameter dimension to form an m×n-dimensional matrix, where m represents the number of sampling time points and n represents the number of parameters. This process uses matrix storage technology for efficient data management. The purpose of this step is to establish a complete data foundation for the inverter operation status to provide data support for subsequent analysis.

[0041] The specific implementation of step S02 is to segment the multi-dimensional photovoltaic inverter operating parameter matrix obtained in step S01 according to the time series. First, determine the time window size, which is set to 30 minutes, that is, each window contains 30 data points. The sliding window technique is adopted, and the window sliding step size is set to 5 minutes to achieve partial overlap between windows and enhance data continuity. For the data within each time window, extract the corresponding parameter subset to form a 30×n-dimensional time window matrix, where n is the number of parameters. This technique is based on the theory of segmented analysis of time series data and can effectively capture the changes in the operating state of the inverter in a short period. The role of this step is to convert long-time series data into short-term state representations, facilitating the detection of transient anomalies and development trends, and at the same time reducing the complexity of data processing.

[0042] The specific implementation of step S03 is to calculate the change amount of the inverter parameters between adjacent time windows. For each pair of adjacent time window matrices W i and W i+1 , calculate the difference of each parameter at the corresponding time point to form a difference matrix D i . The robust statistical method is used for outlier detection, and a parameter outlier detection function is designed to filter extreme values. This function is based on the improved three-standard-deviation principle and dynamically adjusts the threshold in combination with the environmental condition change rate. The specific threshold calculation method is: Threshold = μ ± 3σ×(1 + α×ΔT), where μ is the historical parameter mean, σ is the historical parameter standard deviation, α is the environmental sensitivity coefficient with a value of 0.05, and ΔT is the environmental temperature change rate. For the data points exceeding the threshold, the historical median replacement method is used for correction, and finally, the first change amplitude matrix of the photovoltaic inverter is constructed. The purpose of this step is to obtain the short-term fluctuation characteristics of the inverter parameters and improve the data quality through outlier processing to eliminate the influence of external interference factors on the analysis.

[0043] The specific implementation of step S04 is to calculate the change rate of the parameters within three consecutive time windows based on the first change amplitude matrix obtained in step S03. For three consecutive time windows W i 、W i+1 and W i+2 , first calculate their corresponding two first change amplitude matrices D i and D i+1 , and then calculate the change rate matrix R i , and the calculation formula is: R i =(D i+1 -D i ) / D i , for D iElements with a median value of zero or close to zero are replaced with a small constant ε (set to 0.001) to avoid division-by-zero errors. The bilinear interpolation algorithm is used to process the discontinuous points in the calculation results to generate a continuously varying rate surface. Finally, a second variation amplitude matrix of the photovoltaic inverter is constructed, and this matrix captures the acceleration characteristics of parameter changes. The purpose of this step is to quantify the trend and acceleration of inverter parameter changes and provide a basis for identifying early fault symptoms.

[0044] The specific implementation of step S05 is to apply a moving average algorithm to the second variation amplitude matrix obtained in step S04 for smoothing, and at the same time optimize it in combination with the thermodynamic degradation mechanism equation of the inverter. The moving average window size is set to 5, and the weighted moving average method is used, with the weight coefficients distributed as 1:2:3:2:1 according to the distance from the center point. The thermodynamic degradation mechanism equation is based on the modified Arrhenius equation model, which describes the degradation rate of components under the action of temperature and calculates the component degradation coefficient matrix C. The smoothed data is corrected in combination with the degradation coefficient matrix, and the correction formula is: S i = A i ×(1 + β × C), where A i is the data after moving average, β is the correction coefficient, and its value is 0.15. Through this processing, a third variation amplitude matrix of the photovoltaic inverter is constructed. The purpose of this step is to reduce the influence of random fluctuations in the data and at the same time introduce the knowledge of the physical degradation mechanism of the inverter to make the data more accurately reflect the internal degradation state of the device.

[0045] The specific implementation of step S06 is to extract the phase information of the inverter output current from the original data. The discrete Fourier transform is used to perform frequency domain conversion on the time domain current signal, extract the fundamental wave phase angle, compare it with the standard phase angle (0°), and calculate the phase offset angle. The difference calculation is performed on the phase offset angles within a continuous time period to obtain the phase drift rate. The normal range of phase offset is set to ±2°, and the normal range of drift rate is set to ±0.5° / hour. A two-dimensional phase variation matrix is constructed, including two dimensions of phase offset angle and phase drift rate, to evaluate the grid-connected synchronization control ability of the inverter. The Hilbert transform technology is used to enhance the accuracy of phase extraction and reduce the influence of sampling errors. The purpose of this step is to capture the changes in the synchronization control performance of the inverter, which is a key indicator reflecting the core function of the inverter.

[0046] The specific implementation of step S07 is to fuse the third variation amplitude matrix obtained in step S05 and the phase variation matrix obtained in step S06 to construct a comprehensive health index model. An adaptive weight fusion function is used for parameter weight allocation. This function is based on the principle of grey relational analysis, calculates the correlation degree of each parameter with historical fault cases, and dynamically adjusts the weight allocation. The weight coefficient of the historical fluctuation amplitude of the parameter is 0.2, the weight coefficient of the current change rate of the parameter is 0.3, the weight coefficient of the parameter correlation coefficient is 0.15, the weight coefficient of the equipment operation years is 0.25, and the weight coefficient of the environmental impact factor is 0.1. Principal component analysis is applied for dimensionality reduction, and the first few principal components with a contribution rate of 85% are retained to construct a low-dimensional feature space. On this basis, a support vector machine regression model is used to construct a comprehensive health index calculation function, and the output value range is 0 to 1, where 1 represents complete health and 0 represents complete failure. The purpose of this step is to establish a unified health assessment index through multi-dimensional parameter fusion and quantify the overall operating state of the inverter.

[0047] The specific implementation of step S08 is to set an early warning threshold and implement fault early warning based on the health index model constructed in step S07. The early warning threshold is set to 0.75, which is determined based on the statistical analysis of a large number of historical fault cases, ensuring sufficient early warning time before a fault occurs and avoiding excessive false alarms. When the health index is lower than 0.75, the system triggers a first-level early warning; when it is lower than 0.65, a second-level early warning is triggered; when it is lower than 0.55, a third-level early warning is triggered. The early warning information includes: inverter number, current health index value, health status change trend, possible fault types and fault probabilities. The early warning information is synchronously sent to the maintenance personnel's terminal devices through various methods such as text messages, emails, and system notifications. At the same time, the system automatically generates an early warning log to record the complete early warning process. The purpose of this step is to achieve early detection and timely warning of faults and provide decision-making support for maintenance personnel.

[0048] The specific implementation of step S09 is to predict the occurrence time of inverter faults and formulate a maintenance plan based on the changing trend of the health index. A multivariate time series prediction model, including a hybrid model of autoregressive moving average model and long short-term memory neural network, is used to predict the future changing trend of the health index. Based on the prediction results, the estimated time for the health index to reach the danger threshold (0.5) is calculated as the predicted value of the fault occurrence time, and the prediction accuracy is controlled within ±48 hours; and the calculation results are output. Optionally, it also includes formulating an optimal maintenance plan according to the estimated fault time, fault type, and maintenance complexity, in combination with the current status of maintenance resources (including personnel, spare parts, tools, etc.). The maintenance plan includes: maintenance time window, maintenance item list, required resource list, expected repair time, downtime impact assessment, etc. The system automatically generates a maintenance work order and interfaces with the power station operation and maintenance management system to achieve maintenance task allocation and execution tracking. The purpose of this step is to achieve the transformation from passive response to active prevention, improve maintenance efficiency, reduce fault downtime, and lower maintenance costs.

[0049] The second aspect of the present invention provides a computer-readable storage medium, in which program instructions are stored. When the program instructions run on a computer, they are used to execute the above-mentioned fault warning method for a photovoltaic power station inverter device.

[0050] The third aspect of the present invention provides a fault warning system for a photovoltaic power station inverter device, including the above-mentioned computer-readable storage medium. The system is any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is set inside the system, and a microprocessor for executing the program instructions stored inside the computer-readable storage medium is set inside the system.

[0051] The following details the mathematical models or calculation processes involved in the present invention.

[0052] The multi-dimensional photovoltaic inverter operation parameter matrix in step S01 can be specifically expressed as follows:

[0053]

[0054] In the formula, M is the multi-dimensional photovoltaic inverter operation parameter matrix; v j，1 is the input voltage at the j-th time point; i j，1 is the output current at the j-th time point; pf j，1 is the power factor at the j-th time point; h j，1 is the harmonic content at the j-th time point; t d，j is the device temperature at the j-th time point; t e，j is the ambient temperature at the j-th time point; m is the number of sampling time points.

[0055] The parameter acquisition method is: v j，1 It is collected by a voltage sensor, with a collection range of 200 - 1000V and a sampling frequency of 1 time per minute; i j，1 It is collected by a current transformer, with a collection range of 0 - 100A and a sampling frequency of 1 time per minute; pf j，1 It is calculated by a power analyzer, and the calculation formula is where is the voltage - current phase difference at the j - th time point, with a collection range of 0.8 - 1.0; h j，1 It is measured by a harmonic analyzer, expressed as the total harmonic distortion rate, with a collection range of 0 - 5%; t d，j It is collected by a temperature sensor, and the sensor is installed on the surface of the radiator inside the inverter, with a collection range of 25 - 85°C; t e，j It is collected by an ambient temperature sensor, and the sensor is installed 1.5m around the inverter, with a collection range of - 20 - 50°C.

[0056] This matrix construction adopts a determinant structure. Each row represents a complete set of parameters at a time point, and each column represents the change sequence of a single parameter over time. This structure facilitates subsequent time - series analysis and parameter correlation analysis, reflects the organization and management idea of multi - dimensional data, and provides basic data support for subsequent feature extraction and analysis.

[0057] The photovoltaic inverter parameter time - window matrix in step S02 can be specifically expressed as follows:

[0058]

[0059] In the formula, W k is the parameter matrix of the k - th time window; s is the sliding step, with a value of 5; w is the time - window size, with a value of 30; (k - 1)s + j represents the index of the j - th time point in the k - th window in the original time series.

[0060] The parameter acquisition method is: Extract data from the multi - dimensional photovoltaic inverter operation parameter matrix M in step S01 according to the sliding - window principle. Each time, extract data of w = 30 consecutive time points to form a time - window matrix. The sliding step s = 5 indicates that the starting points of two adjacent windows are separated by 5 time points, realizing an overlap of 25 time points between windows.

[0061] This matrix adopts the sliding - window technique, which not only ensures the continuity of data but also improves the timeliness of analysis. The window size w = 30 corresponds to 30 - minute data, which can effectively capture the change characteristics in the short term; the sliding step s = 5 causes an overlap between windows, enhancing the continuity and stability of data analysis, and conforming to the basic principle of segmented analysis of time - series data.

[0062] The first variation amplitude matrix of the photovoltaic inverter in step S03 can be specifically expressed as follows:

[0063]

[0064] In the formula, D k is the first variation amplitude matrix corresponding to the k-th time window; Δv k，j represents the difference in the input voltage at the corresponding position j between the k-th window and the (k - 1)-th window, and the calculation formula is Δv k，j = v (k-1)s+j，1 - v (k-2)s+j，1 ; the difference calculation methods of other parameters are similar.

[0065] The parameter anomaly detection function can be specifically expressed as follows:

[0066]

[0067] In the formula, F(Δp k，j ) is the anomaly detection function, which is used to judge and process the variation value Δp k，j of the parameter p at the j-th position in the k-th window; μ p is the mean value of the historical variation values of the parameter p, which is obtained by calculating , where N is the number of historical windows; Th p is the dynamic threshold, and the calculation formula is Th p = 3σ p ×(1 + α×ΔT), where σ p is the standard deviation of the historical variation values of the parameter p, which is obtained by calculating ; α is the environmental sensitivity coefficient, and its value is 0.05; ΔT is the environmental temperature change rate, which is obtained by calculating ; median(ΔP k ) is the median of all variation values of the parameter p in the k-th window.

[0068] This matrix and function design adopt a robust statistical method. By calculating the differences of parameters in adjacent windows, a variation amplitude matrix is constructed, and the three-sigma principle is combined for outlier detection. The design of the dynamic threshold takes into account the influence of environmental temperature changes on equipment parameters, improving the accuracy and adaptability of anomaly detection. The method of replacing outliers with the median ensures the continuity and smoothness of the data, effectively eliminating the influence of external interference factors.

[0069] The second variation amplitude matrix of the photovoltaic inverter in step S04 can be specifically expressed as follows:

[0070]

[0071] In the formula, R kis the second change amplitude matrix corresponding to the k-th time window; r p，k，j represents the change rate of parameter p at the j-th position in the k-th window, and the calculation formula is:

[0072]

[0073] In the formula, Δp k，j and Δp k+1，j are the change values of parameter p at the j-th position in the k-th and k+1-th windows respectively; ε is a small constant with a value of 0.001, which is used to avoid division by zero errors.

[0074] The discontinuous points in the change rate matrix are processed by the bilinear interpolation algorithm:

[0075]

[0076] In the formula, is the interpolated change rate value; (x, y) is the target point coordinate; (x1, y1), (x1, y2), (x2, y1), (x2, y2) are the coordinates of the four surrounding known points; is the corresponding known change rate value.

[0077] This matrix design adopts the principle of differential calculation. By calculating the change rate of the parameter change amount within three consecutive time windows, the acceleration characteristics of the parameter change are captured. The introduction of the small constant ε solves the mathematical problem of the denominator being zero, and the bilinear interpolation algorithm fills the discontinuous points in the data, enhancing the smoothness and continuity of the data. This design can effectively identify the trend and acceleration characteristics of parameter changes, providing data support for early fault symptom detection.

[0078] The third change amplitude matrix of the photovoltaic inverter in step S05 can be specifically expressed as follows:

[0079] S k = A k ×(1 + β×C);

[0080] In the formula, S k is the third change amplitude matrix corresponding to the k-th time window; A k is the matrix after applying the moving average algorithm; β is a correction coefficient with a value of 0.15; C is the component degradation coefficient matrix.

[0081] The calculation formula of the moving average algorithm is:

[0082]

[0083] In the formula, a p，k，j is the value of parameter p at the j-th position in the k-th window after moving average; r p，k+i，jis the change rate of parameter p at the j-th position in the (k + i)-th window; w is the weight vector, w = [1, 2, 3, 2, 1], where the weight of the center point is the largest and the weight decreases as the distance from the center point increases.

[0084] The component degradation coefficient matrix C is calculated based on the Arrhenius equation correction model:

[0085]

[0086] In the formula, c p is the component degradation coefficient corresponding to parameter p, and the calculation formula is:

[0087]

[0088] In the formula, A0 is the pre-exponential factor, with a value of 1000; E a is the activation energy, with a value of 0.7 eV; k is the Boltzmann constant, with a value of 8.617×10 -5 eV / K; T d is the device temperature, in units of K; L r is the current operating duration; L0 is the reference operating duration, with a value of 8760 hours (1 year); γ is the time aging coefficient, with a value of 0.5; δ is the temperature difference influence coefficient, with a value of 0.2; T e is the ambient temperature, in units of K; η is the heat dissipation system efficiency, with a value range of 0.7 - 0.9; λ is the heat dissipation influence index, with a value of 0.3.

[0089] This matrix design combines signal processing and physical aging models, reduces random fluctuations in data through the weighted moving average algorithm, and introduces the Arrhenius thermodynamic degradation mechanism model to correct the data in terms of physical meaning. The Arrhenius equation describes the influence of temperature on the reaction rate, and on this basis, factors such as operating duration, load rate, temperature difference, and heat dissipation efficiency are added, comprehensively considering the physical degradation process of the internal components of the inverter. This design that combines statistical methods and physical models enables the third variation amplitude matrix to more accurately reflect the true degradation state inside the device.

[0090] The phase change matrix of the photovoltaic inverter in step S06 can be specifically expressed as follows:

[0091]

[0092] In the formula, P k is the phase change matrix at the k-th time point; θ k，j is the offset angle of the j-th phase at the k-th time point; is the drift rate of the j-th phase at the k-th time point.

[0093] Phase extraction uses the discrete Fourier transform:

[0094]

[0095] Wherein, X k (j) is the spectrum of the current signal x(n) at the k-th time point; N is the number of sampling points; j is the frequency index; i is the imaginary unit.

[0096] The calculation formula for the fundamental wave phase angle is:

[0097]

[0098] Wherein, Im(X k (1)) and Re(X k (1)) are respectively the imaginary part and the real part of the complex value at the fundamental wave frequency point.

[0099] The calculation formula for the phase drift rate is:

[0100]

[0101] Wherein, t k -t k-1 is the time interval between two adjacent time points.

[0102] To improve the accuracy of phase extraction, Hilbert transform enhancement processing is adopted:

[0103]

[0104] Wherein, is the Hilbert transform of the signal x(n); represents the Hilbert transform operation.

[0105] The calculation formula for the instantaneous phase is:

[0106]

[0107] Wherein, φ(n) is the instantaneous phase of the signal.

[0108] This matrix design uses signal processing theory to convert the time-domain current signal to the frequency domain through discrete Fourier transform, extract the fundamental wave phase information; and enhance the accuracy of phase extraction through Hilbert transform. The matrix includes two dimensions of phase offset angle and phase drift rate, comprehensively characterizing the grid-connected synchronization control performance of the inverter, and providing important indicators for fault warning.

[0109] The photovoltaic inverter comprehensive health index model in step S07 can be specifically expressed as follows:

[0110]

[0111] Wherein, H is the comprehensive health index, and its value range is 0 to 1; ω i is the weight of the i-th feature; f i (S, P) is the evaluation function of the i-th feature extracted from the third fluctuation amplitude matrix S and the phase fluctuation matrix P; n is the total number of features.

[0112] The calculation formula of the adaptive weight fusion function is:

[0113]

[0114] Wherein, λ i is the weight related to the historical fluctuation amplitude of the parameter, and its value is 0.2; γ i is the weight related to the current change rate of the parameter, and its value is 0.3; ρ i is the weight related to the parameter correlation coefficient, and its value is 0.15; τ i is the weight related to the operation years of the equipment, and its value is 0.25; ε i is the weight related to the environmental impact factor, and its value is 0.1.

[0115] The calculation formula of the historical fluctuation amplitude of the parameter is:

[0116]

[0117] Wherein, VAR i is the historical variance of parameter i; k λ is the adjustment coefficient, and its value is 0.5; μ λ is the mean value of the variance.

[0118] The calculation formula of the current change rate of the parameter is:

[0119]

[0120] Wherein, ΔZ i is the latest change rate of parameter i; k γ is the adjustment coefficient, and its value is 0.7; μ γ is the mean value of the change rate.

[0121] The calculation formula of the parameter correlation coefficient is:

[0122]

[0123] Wherein, Corr(i, j) is the correlation coefficient between parameter i and parameter j; m is the total number of parameters.

[0124] The calculation formula of the weight related to the operation years of the equipment is:

[0125]

[0126] where Y is the current operating years; Y max is the design life, with a value of 20 years; k τ is the adjustment coefficient, with a value of 0.3; F i is the historical failure times of parameter i; F max is the historical maximum failure times.

[0127] The calculation formula for the environmental impact factor is:

[0128]

[0129] where T d is the equipment temperature; T e is the environmental temperature; k ε is the adjustment coefficient, with a value of 0.4; μ ε is the mean temperature difference.

[0130] This model design adopts a multi-feature fusion method, and dynamically adjusts the importance of each parameter through an adaptive weight function. The weight calculation considers the historical volatility of the parameter, the current change rate, the correlation with other parameters, the equipment operating years, and the environmental impact factors, and comprehensively evaluates the contribution degree of each parameter to the health state. The model uses the Sigmoid function to normalize each factor to ensure that the output value is within the range of 0 to 1. This design realizes the quantitative evaluation of the overall health state of the inverter and provides a reliable basis for fault warning.

[0131] The setting of the health index threshold and the warning trigger condition in step S08 can be specifically expressed as follows:

[0132]

[0133] where Alert(H) is the warning function, and outputs the warning level according to the value of the health index H; Level1, Level2, and Level3 respectively represent the first-level, second-level, and third-level warnings; Normal represents the normal state.

[0134] The content of the warning information can be expressed as:

[0135] Info = {ID, H, ΔH, FT, FP};

[0136] where Info is the warning information set; ID is the inverter number; H is the current health index value; ΔH is the change trend of the health index, and the calculation formula is ΔH = H t -H t-1 ; FT is the possible fault type, determined by the fault library associated with the health index model; FP is the fault probability, and the calculation formula is FP = 1 - H.

[0137] This design is based on a threshold trigger mechanism, which divides the health index into different intervals corresponding to different warning levels. The warning information includes key information such as device identification, current status, change trend, possible fault types and probabilities, providing comprehensive decision-making basis for maintenance personnel.

[0138] The fault time prediction model in step S09 can be specifically expressed as follows:

[0139]

[0140] In the formula, T failure is the predicted fault occurrence time; T current is the current time; H is the current health index; H threshold is the fault threshold, with a value of 0.5; is the average change rate of the health index, and the calculation formula is where n is the size of the retrospective time window, with a value of 24; Δt is the sampling time interval.

[0141] The formula of the hybrid prediction model is:

[0142]

[0143] In the formula, is the predicted value of the health index at the l-th future time point; α is the hybrid weight, with a value of 0.4; ARIMA(H t , p, d, q, l) is the predicted value of the autoregressive moving average model, where p is the autoregressive order, with a value of 3, d is the differencing order, with a value of 1, q is the moving average order, with a value of 2; LSTM(H t , l) is the predicted value of the long short-term memory neural network.

[0144] The maintenance plan generation model can be expressed as:

[0145] MP = Optimize(T failure , FT, MC, MR);

[0146] In the formula, MP is the maintenance plan; Optimize is the optimization function; T failure is the predicted fault occurrence time; FT is the fault type; MC is the maintenance complexity; MR is the maintenance resource status.

[0147] This design uses a method that combines linear extrapolation and time series prediction. Based on the historical change trend of the health index, it predicts the time of fault occurrence. The hybrid prediction model combines the advantages of the ARIMA model and the LSTM neural network, improving the accuracy and robustness of the prediction. The maintenance plan generation model formulates an optimal maintenance strategy through an optimization algorithm, comprehensively considering the fault time, type, complexity, and resource status, realizing the transformation from passive response to active prevention.

[0148] Specifically, the principle of the present invention is as follows: The technical principle of the present invention is based on the deep integration of multi-dimensional data analysis and physical degradation mechanism, and realizes early fault warning by establishing a multi-level characterization model of the inverter operating state. First, multi-dimensional operating parameters of the inverter are collected to construct an operating parameter matrix to capture comprehensive information about the device operation; subsequently, through time window segmentation processing, the continuous time series is transformed into discrete state representations, facilitating the capture of the device's short-term operation characteristics.

[0149] On this basis, the present invention designs a three-level change amplitude matrix structure: The first change amplitude matrix filters extreme values by calculating the parameter change amount in adjacent time windows and applying an anomaly detection function, effectively eliminating environmental interference; the second change amplitude matrix characterizes the acceleration change characteristics of the inverter parameters by calculating the parameter change rate of three consecutive time windows, sensitively capturing fault precursors; the third change amplitude matrix combines the sliding average algorithm and the thermodynamic degradation mechanism equation, combines the data statistical characteristics with the physical mechanism, effectively reduces the influence of random fluctuations, and truly reflects the degradation degree of internal components of the device.

[0150] In particular, the present invention introduces the analysis of the inverter output current phase information, constructs a phase change matrix, and monitors the inverter synchronization control performance, which is a key fault index often overlooked by traditional methods. Finally, through an adaptive weight fusion function, multi-dimensional features are dynamically integrated to construct a comprehensive health index model, realizing fault type prediction and time prediction. The reason why this method is effective is that it not only considers the statistical characteristics of the data but also respects the physical degradation law of the device. The two complement each other, significantly improving the accuracy and timeliness of the early warning.

[0151] A specific Embodiment 1 of the present invention is provided below, and the specific implementation of each step in this Embodiment 1 is described in detail as follows.

[0152] The specific implementation of step S01 is to regularly sample data of the inverter equipment through data collectors arranged at the photovoltaic power station site. The sampling time interval is set to 1 minute. The collected data items include: input voltage (DC side voltage), with a collection range of 200 - 1000V; output current (AC side current), with a collection range of 0 - 100A; power factor, with a collection range of 0.8 - 1.0; harmonic content (total harmonic distortion rate), with a collection range of 0 - 5%; equipment temperature, with a collection range of 25 - 85°C; ambient temperature, with a collection range of -20 - 50°C. In the specific process, first, analog signals are collected through high-precision sensors, then the analog signals are converted into digital signals through high-precision analog-to-digital converters, and preliminary processing is performed by the data preprocessing unit, including operations such as signal filtering and missing value filling. Finally, the data is organized according to the time dimension and parameter dimension to construct a multi-dimensional photovoltaic inverter operation parameter matrix, which is specifically represented as follows:

[0153]

[0154] In the formula, M is the multi-dimensional photovoltaic inverter operation parameter matrix; v j，1 is the input voltage at the j-th time point; i j，1 is the output current at the j-th time point; pf j，1 is the power factor at the j-th time point; h j，1 is the harmonic content at the j-th time point; t d，j is the equipment temperature at the j-th time point; t e，j is the ambient temperature at the j-th time point; m is the number of sampling time points. This matrix adopts a determinant structure. Each row represents a complete parameter set at a time point, and each column represents the change sequence of a single parameter over time. This structure facilitates subsequent time series analysis and parameter correlation analysis. The purpose of this step is to establish a complete data basis for the inverter operation status and provide data support for subsequent analysis.

[0155] The specific implementation of step S02 is to perform segmented processing on the multi-dimensional photovoltaic inverter operation parameter matrix obtained in step S01 according to the time series. First, determine the size of the time window, which is set to 30 minutes, that is, each window contains 30 data points. The sliding window technique is adopted, and the window sliding step is set to 5 minutes to achieve partial overlap between windows and enhance data continuity. For the data within each time window, extract the corresponding parameter subset to construct a photovoltaic inverter parameter time window matrix, which is specifically represented as follows:

[0156]

[0157] In the formula, W kis the parameter matrix for the k-th time window; s is the sliding step with a value of 5; w is the time window size with a value of 30; (k - 1)s + j represents the index of the j-th time point in the k-th window in the original time series. This technology is based on the theory of segmented analysis of time series data and can effectively capture the changes in the operating state of the inverter in a short period of time. The role of this step is to convert long-time series data into short-term state representations, facilitating the detection of transient anomalies and development trends, while reducing the complexity of data processing.

[0158] The specific implementation of step S03 is to calculate the change amount of the inverter parameters between adjacent time windows. For each pair of adjacent time window matrices W i and W i+1 , calculate the difference of each parameter at the corresponding time points to form a difference matrix D i , and construct the first change amplitude matrix of the photovoltaic inverter, which is specifically expressed as follows:

[0159]

[0160] In the formula, D k is the first change amplitude matrix corresponding to the k-th time window; Δv k，j represents the difference in input voltage between the k-th window and the (k - 1)-th window at the corresponding position j, and the calculation formula is Δv k，j = v (k-1)s+j，1 - v (k-2)s+j，1 ; the difference calculation methods of other parameters are similar. Next, a robust statistical method is used for outlier detection, and a parameter outlier detection function is designed for extreme value filtering:

[0161]

[0162] In the formula, F(Δp k，j ) is the outlier detection function, which is used to judge and process the change value Δp k，j of parameter p at the j-th position in the k-th window; μ p is the mean of the historical change values of parameter p, which is obtained by calculating , where N is the number of historical windows; Th p is the dynamic threshold, and the calculation formula is Th p = 3σ p ×(1 + α×ΔT), where σ p is the standard deviation of the historical change values of parameter p, which is obtained by calculating ; α is the environmental sensitivity coefficient with a value of 0.05; ΔT is the environmental temperature change rate, which is obtained by calculating ; median(ΔP k) is the median of all the varying values of parameter p in the k-th window. The purpose of this step is to obtain the short-term fluctuation characteristics of the inverter parameters, improve the data quality through outlier processing, and eliminate the influence of external interference factors on the analysis.

[0163] The specific implementation of step S04 is to calculate the change rate of the parameters within three consecutive time windows based on the first variation amplitude matrix obtained in step S03. For three consecutive time windows W i 、W i+1 and W i+2 , first calculate their corresponding two first variation amplitude matrices D i and D i+1 , then calculate the change rate matrix R i , and construct the second variation amplitude matrix of the photovoltaic inverter, which is specifically expressed as follows:

[0164]

[0165] In the formula, R k is the second variation amplitude matrix corresponding to the k-th time window; r p，k，j represents the change rate of parameter p at the j-th position in the k-th window, and the calculation formula is:

[0166]

[0167] In the formula, Δp k，j and Δp k+1，j are the varying values of parameter p at the j-th position in the k-th and k + 1-th windows respectively; ε is a small constant with a value of 0.001, which is used to avoid division by zero errors. For the discontinuous points in the change rate matrix, the bilinear interpolation algorithm is used for processing:

[0168]

[0169] In the formula, is the interpolated change rate value; (x, y) is the coordinate of the target point; (x1, y1), (x1, y2), (x2, y1), (x2, y2) are the coordinates of the four surrounding known points; are the corresponding known change rate values. The purpose of this step is to quantify the trend and acceleration of the inverter parameter changes, providing a basis for identifying early fault symptoms.

[0170] The specific implementation of step S05 is to apply the moving average algorithm to smooth the second variation amplitude matrix obtained in step S04, and at the same time optimize it in combination with the inverter thermodynamic degradation mechanism equation. The moving average window size is set to 5, and the weighted moving average method is used, with the weight coefficients distributed as 1:2:3:2:1 according to the distance from the center point. The calculation formula is:

[0171]

[0172] In the formula, a p，k，j is the value of parameter p after moving average at the j-th position in the k-th window; r p，k+i，j is the change rate of parameter p at the j-th position in the (k + i)-th window; w is the weight vector, w = [1, 2, 3, 2, 1], with the largest weight at the center point and the weights decreasing as the distance from the center point increases. Next, apply the thermodynamic degradation mechanism equation to calculate the component degradation coefficient matrix C:

[0173]

[0174] In the formula, c p is the component degradation coefficient corresponding to parameter p, calculated based on the Arrhenius equation correction model:

[0175]

[0176] In the formula, A0 is the pre-exponential factor, with a value of 1000; E a is the activation energy, with a value of 0.7 eV; k is the Boltzmann constant, with a value of 8.617×10 -5 eV / K; T d is the device temperature, in units of K; L r is the current operating duration; L0 is the reference operating duration, with a value of 8760 hours (1 year); γ is the time aging coefficient, with a value of 0.5; δ is the temperature difference influence coefficient, with a value of 0.2; T e is the ambient temperature, in units of K; η is the heat dissipation system efficiency, with a value range of 0.7 - 0.9; λ is the heat dissipation influence index, with a value of 0.3. Finally, combine the degradation coefficient matrix to correct the smoothed data and construct the third variation amplitude matrix of the photovoltaic inverter:

[0177] S k = A k ×(1 + β×C);

[0178] In the formula, S k is the third variation amplitude matrix corresponding to the k-th time window; A k is the matrix after applying the moving average algorithm; β is the correction coefficient, with a value of 0.15; C is the component degradation coefficient matrix. The purpose of this step is to reduce the influence of random fluctuations in the data and at the same time introduce the knowledge of the physical degradation mechanism of the inverter to make the data more accurately reflect the internal degradation state of the device.

[0179] The specific implementation method of step S06 is to extract the phase information of the inverter output current from the original data. Use the discrete Fourier transform to perform frequency domain conversion on the time domain current signal:

[0180]

[0181] Wherein, X k (j) is the spectrum of the current signal x(n) at the k-th time point; N is the number of sampling points; j is the frequency index; i is the imaginary unit. Extract the fundamental wave phase angle:

[0182]

[0183] Wherein, Im(X k (1)) and Re(X k (1)) are respectively the imaginary part and the real part of the complex value at the fundamental wave frequency point. Calculate the phase drift rate:

[0184]

[0185] Wherein, t k -t k-1 is the time interval between two adjacent time points. To improve the accuracy of phase extraction, Hilbert transform enhancement processing is adopted:

[0186]

[0187] Wherein, is the Hilbert transform of the signal x(n); represents the Hilbert transform operation. Calculate the instantaneous phase:

[0188]

[0189] Wherein, φ(n) is the instantaneous phase of the signal. Finally, construct the phase change matrix of the photovoltaic inverter:

[0190]

[0191] Wherein, P k is the phase change matrix at the k-th time point; θ k，j is the offset angle of the j-th phase at the k-th time point; is the drift rate of the j-th phase at the k-th time point. Set the normal range of phase offset to ±2°, and the normal range of drift rate to ±0.5° / hour. The purpose of this step is to capture the change in the inverter synchronization control performance, which is a key indicator reflecting the core function of the inverter.

[0192] The specific implementation manner of step S07 is to fuse the third change amplitude matrix obtained in step S05 and the phase change matrix obtained in step S06 to construct a comprehensive health index model:

[0193]

[0194] Wherein, H is the comprehensive health index, and the value range is 0 to 1; ω i is the weight of the i-th feature; f i (S, P) is the evaluation function of the i-th feature extracted from the third variation amplitude matrix S and the phase variation matrix P; n is the total number of features. The adaptive weight fusion function is used for parameter weight allocation:

[0195]

[0196] Wherein, λ i is the weight related to the historical fluctuation amplitude of the parameter, and the value is 0.2; γ i is the weight related to the current change rate of the parameter, and the value is 0.3; ρ i is the weight related to the parameter correlation coefficient, and the value is 0.15; τ i is the weight related to the operation years of the device, and the value is 0.25; ε i is the weight related to the environmental impact factor, and the value is 0.1. The calculation formula for the historical fluctuation amplitude of the parameter is:

[0197]

[0198] Wherein, VAR i is the historical variance of parameter i; k λ is the adjustment coefficient, and the value is 0.5; μ λ is the mean value of the variance. The calculation formula for the current change rate of the parameter is:

[0199]

[0200] Wherein, ΔZ i is the most recent change rate of parameter i; k γ is the adjustment coefficient, and the value is 0.7; μ γ is the mean value of the change rate. The calculation formula for the parameter correlation coefficient is:

[0201]

[0202] Wherein, Corr(i, j) is the correlation coefficient between parameter i and parameter j; m is the total number of parameters. The calculation formula for the weight related to the operation years of the device is:

[0203]

[0204] Wherein, Y is the current operation years; Y max is the design life, and the value is 20 years; k τ is the adjustment coefficient, and the value is 0.3; F i is the historical failure times of parameter i; F max is the historical maximum failure times. The calculation formula for the environmental impact factor is:

[0205]

[0206] Wherein, T d is the device temperature; T e is the ambient temperature; k ε is the adjustment coefficient, with a value of 0.4; μ ε is the average temperature difference. The purpose of this step is to establish a unified health assessment index through multi-dimensional parameter fusion and quantify the overall operating state of the inverter.

[0207] The specific implementation of step S08 is based on the health index model constructed in step S07, setting an early warning threshold and realizing fault early warning:

[0208]

[0209] Wherein, Alert(H) is the early warning function, which outputs the early warning level according to the value of the health index H; Level1, Level2, and Level3 respectively represent the first-level, second-level, and third-level early warnings; Normal represents the normal state. The content of the early warning information can be expressed as:

[0210] Info = {ID, H, ΔH, FT, FP};

[0211] Wherein, Info is the early warning information set; ID is the inverter number; H is the current health index value; ΔH is the change trend of the health index, and the calculation formula is ΔH = H t -H t-1 ; FT is the possible fault type, determined through the fault library associated with the health index model; FP is the fault probability, and the calculation formula is FP = 1 - H. The early warning information is synchronously sent to the maintenance personnel's terminal device through various methods such as text messages, emails, and system notifications. The purpose of this step is to achieve early detection and timely early warning of faults and provide decision-making support for maintenance personnel.

[0212] The specific implementation of step S09 is to predict the inverter fault occurrence time and formulate a maintenance plan based on the change trend of the health index. The fault time prediction model is:

[0213]

[0214] Wherein, T failure is the predicted fault occurrence time; T current is the current time; H is the current health index; H threshold is the fault threshold, with a value of 0.5; is the average change rate of the health index, and the calculation formula is Where N is the size of the backtracking time window, with a value of 24; Δt is the sampling time interval. A multivariate time series prediction model is adopted, including a hybrid model of an autoregressive moving average model and a long short-term memory neural network:

[0215]

[0216] In the formula, is the predicted value of the health index at the l-th future time point; α is the hybrid weight, with a value of 0.4; ARIMA(H t , p, d, q, l) is the predicted value of the autoregressive moving average model, where p is the autoregressive order, with a value of 3, d is the differencing order, with a value of 1, q is the moving average order, with a value of 2; LSTM(H t , l) is the predicted value of the long short-term memory neural network. The maintenance plan generation model is:

[0217] MP = Optimize(T failure , FT, MC, MR);

[0218] In the formula, MP is the maintenance plan; Optimize is the optimization function; T failure is the predicted fault occurrence time; FT is the fault type; MC is the maintenance complexity; MR is the maintenance resource status. The optimization function comprehensively considers the fault time, type, complexity, and resource status, formulates the optimal maintenance strategy, and generates a maintenance work order, including the maintenance time window, list of maintenance items, list of required resources, expected repair time, and evaluation of downtime impact, etc. The purpose of this step is to achieve the transformation from passive response to active prevention, improve maintenance efficiency, reduce fault downtime, and lower maintenance costs.

[0219] In summary, through the construction of a multi-dimensional operation parameter matrix, the embodiment of the present invention realizes segmented analysis of time series, applies the parameter change rate calculation and anomaly detection algorithm to eliminate the influence of interference factors, combines the thermodynamic degradation mechanism model to reflect the internal physical state of the equipment, extracts the current phase information to evaluate the synchronous control performance, and finally constructs a comprehensive health index model through adaptive weight fusion to achieve fault warning and maintenance plan formulation. This method fully considers the dynamic characteristics and physical degradation laws of the inverter operation, combines the ideas of data-driven and mechanism model, and can effectively improve the accuracy and timeliness of fault warning, which is of great significance to the safe and stable operation of the photovoltaic power station.

[0220] To better understand and implement the present invention, the following provides Embodiment 2 of a specific application scenario of the present invention: In a large-scale photovoltaic power station, 500 inverter devices are installed. These inverter devices often experience failures during long-term operation, resulting in reduced power generation efficiency and even shutdowns. In response to this problem, researchers designed a fault warning method based on multi-dimensional parameter fusion and conducted a six-month experimental verification on 10 inverters in this photovoltaic power station. The inverter model selected in the experiment is SG250HX, with a rated power of 250 kW and has been in operation for 3 years.

[0221] First, the researchers deployed a high-precision data acquisition system with a sampling frequency set to 1 time per minute. The parameters collected include input voltage, output current, power factor, harmonic content, device temperature, and ambient temperature, etc. Table 1 shows some of the original data collected from a certain inverter at different time points on a certain day:

[0222] Table 1 Sampling Table of Inverter Original Data

[0223]

[0224] After the acquisition was completed, the researchers set the time window to 30 minutes and the sliding step to 5 minutes, and constructed a parameter time window matrix. When calculating the parameter changes between adjacent time windows, some outliers were found. For example, at a certain moment, the output current suddenly jumped from 69.5 A to 58.2 A. It was determined as an outlier through the parameter outlier detection function and replaced with the median value of 69.3 A within this time window.

[0225] Table 2 shows some of the data of the first change amplitude matrix for a certain inverter within 5 consecutive time windows:

[0226] Table 2 Data Table of Inverter First Change Amplitude Matrix

[0227]

[0228] The researchers further calculated the parameter change rates for three consecutive time windows and constructed a second change amplitude matrix. To eliminate the influence of random fluctuations, a moving average algorithm was applied for smoothing, and at the same time, an inverter thermodynamic degradation mechanism model based on the Arrhenius equation was introduced for correction. This inverter has been in operation for 3 years (26,280 hours), and the efficiency of the heat dissipation system is 0.85. The calculated component degradation coefficient matrix is shown in Table 3:

[0229] Table 3 Component Degradation Coefficient Matrix

[0230] Parameter Degradation coefficient <![CDATA[Input voltage (c v )]]> 0.082 <![CDATA[Output current (c i )]]> 0.157 <![CDATA[Power factor (c pf )]]> 0.093 <![CDATA[Harmonic content (c h )]]> 0.128 <![CDATA[Device temperature (c td )]]> 0.065 <![CDATA[Ambient temperature (c te )]]> 0.034

[0231] Extract the phase information of the inverter output current through spectral analysis, and calculate the phase offset angle and phase drift rate. Table 4 shows the phase change matrix data of a certain inverter at different time points on a certain day:

[0232] Table 4 Inverter Phase Change Matrix Data Table

[0233] Time Phase offset angle (°) Phase drift rate (° / h) 08:00 0.52 0.08 09:00 0.59 0.07 10:00 0.67 0.08 11:00 0.74 0.07 12:00 0.83 0.09 13:00 0.91 0.08 14:00 1.02 0.11

[0234] The researchers found that the phase offset angle of the inverter continued to increase, and the drift rate tended to accelerate, but it was still within the normal range (±2°). Subsequently, the weight distribution of each parameter was carried out through the adaptive weight fusion function, and the weight distribution results learned from the historical fault data of the equipment are shown in Table 5:

[0235] Table 5 Parameter Weight Distribution Table

[0236]

[0237]

[0238] Based on the weight distribution and evaluation function of the above parameters, the comprehensive health index of the inverter is calculated to be 0.74, which is in a critical state. The system triggers a first-level warning, prompting the maintenance personnel to pay attention to the operation status of the inverter. The warning information points out that there may be problems with overheating of power devices and aging of control circuits in the inverter, and the failure probability is 26%.

[0239] Through the time series prediction model, the researchers predicted that if the inverter is not maintained, the health index will drop to 0.65 (the second-level warning threshold) after 15 days and may drop to 0.5 (the failure threshold) after 45 days. According to the prediction results, the researchers formulated a maintenance plan, suggesting replacing the power device heat sink and cleaning the heat dissipation system within 10 days, and at the same time checking the connection part of the control circuit.

[0240] During the 6-month experiment, the warning system issued a total of 87 warnings, of which 82 were confirmed to have abnormal conditions, and the warning accuracy rate reached 94.3%. Through timely maintenance, 23 possible serious faults were avoided, the downtime was reduced by about 276 hours, and the overall power generation efficiency of the power station was increased by about 3.5%.

[0241] Traditional inverter fault detection methods mainly rely on a single threshold trigger mechanism, that is, an alarm is triggered when a certain parameter exceeds a preset threshold. This method has problems such as a high false alarm rate and strong hysteresis. Another common method is regular manual inspections, which not only involve a large amount of work but also cannot detect early fault signs in a timely manner. In contrast, the fault warning method proposed by the present invention has the following advantages: First, a health index model is constructed through multi-dimensional parameter fusion, comprehensively considering the mutual influence and weight contribution of each parameter, reducing the false alarm rate caused by the fluctuation of a single parameter; Second, a thermodynamic degradation mechanism model is introduced, combining data-driven with physical models, improving the scientificity and accuracy of warning; Third, an adaptive weight algorithm is adopted to dynamically adjust the parameter weights according to the operating conditions of the equipment, making the warning more flexible and personalized; Fourth, the synchronous control performance is evaluated through phase information analysis, filling the gap in this aspect of traditional methods; Fifth, fault time prediction and automatic generation of maintenance plans are realized, transforming passive response into active prevention. Experimental results show that compared with traditional methods, the fault warning accuracy of the present invention has increased by about 18 percentage points, and the average warning lead time has been extended by 12 days, providing a strong guarantee for the safe and stable operation of photovoltaic power stations.

[0242] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 6, 7, and 8 below.

[0243] Table 6 Variable Explanation Table (First Part)

[0244]

[0245]

[0246] Table 7 Variable Explanation Table (Second Part)

[0247]

[0248] Table 8 Variable Explanation Table (Third Part)

[0249]

[0250]

[0251] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A method for fault warning of an inverter device in a photovoltaic power station, characterized in that, Including: Collecting the operation data of the inverter device to construct a multi-dimensional photovoltaic inverter operation parameter matrix; Performing time series segmentation to construct a photovoltaic inverter parameter time window matrix; Calculating the change amount of inverter parameters within adjacent time windows, filtering extreme values to construct a first change amplitude matrix; calculating the parameter change rate within three consecutive time windows based on the first change amplitude matrix to construct a second change amplitude matrix; using a moving average algorithm combined with the inverter thermodynamics degradation mechanism equation for optimization to construct a third change amplitude matrix; extracting the phase information of the inverter output current to construct a phase change matrix; fusing the third change amplitude matrix and the phase change matrix to construct a comprehensive health index model; setting a health index threshold to trigger an early warning mechanism when the index is lower than the threshold.

2. The photovoltaic power station inverter device fault warning method according to claim 1, characterized in that, Collecting the operation data of the inverter device includes input voltage, output current, power factor, harmonic content, device temperature, and ambient temperature. The multi-dimensional photovoltaic inverter operation parameter matrix includes a time dimension and a parameter dimension, which is a data structure formed by arranging the collected inverter operation parameters in chronological order. Each row represents a time point, and each column represents a parameter, forming a matrix.

3. The photovoltaic power station inverter equipment fault warning method according to claim 2, characterized in that The photovoltaic inverter parameter time window matrix refers to a subset of the inverter operation parameters intercepted within a fixed time length, which is used to capture the operation state of the device in each time period. The photovoltaic inverter parameter time window matrix reflects the short-term operation state of the inverter.

4. The method for fault warning of the photovoltaic power station inverter equipment according to claim 3, wherein The first change amplitude matrix of the photovoltaic inverter refers to a matrix composed of the parameter differences between adjacent time windows. The first change amplitude matrix of the photovoltaic inverter reflects the short-term fluctuation characteristics of the inverter parameters and eliminates the influence of interference factors; the second change amplitude matrix of the photovoltaic inverter refers to a matrix composed of the parameter change rates within three consecutive time windows. The second change amplitude matrix of the photovoltaic inverter characterizes the accelerated change characteristics of the inverter parameters; the third change amplitude matrix of the photovoltaic inverter refers to a smoothed matrix obtained by applying a moving average algorithm to the second change amplitude matrix. The third change amplitude matrix of the photovoltaic inverter reduces the influence of random fluctuations and reflects the degree of degradation of the internal physical state of the inverter.

5. The method for warning of photovoltaic power station inverter equipment faults according to claim 4, characterized in that, The phase change matrix of the photovoltaic inverter refers to a matrix composed of the phase angle offset and its change rate of the inverter output current, including two dimensions of phase offset angle and phase drift rate, which is used to evaluate the grid-connected synchronization control performance of the inverter. The phase change matrix of the photovoltaic inverter reflects the synchronization control performance of the inverter.

6. The photovoltaic power station inverter equipment fault warning method according to claim 5, wherein, The comprehensive health index model of the photovoltaic inverter refers to an evaluation index constructed by weighted fusion of the third change amplitude matrix and the phase change matrix. The comprehensive health index model of the photovoltaic inverter quantifies the operation state of the inverter and provides a pre-judgment of the fault type.

7. The method for warning of photovoltaic power station inverter equipment faults according to claim 6, characterized in that, The inverter thermodynamics degradation mechanism equation is used to describe the physical performance degradation law of the internal components of the inverter under different temperature conditions. The input includes the internal temperature of the inverter, ambient temperature, operation duration, load rate, and heat dissipation system efficiency, and the output is the component degradation coefficient matrix. The component degradation coefficient matrix is used to construct the third change amplitude matrix of the photovoltaic inverter.

8. The photovoltaic power station inverter device fault warning method according to claim 7, characterized in that, The parameter anomaly detection function is used to identify and correct extreme outliers generated during the data acquisition process, ensuring the accuracy of the variation amplitude matrix. The inputs include the historical mean of the parameter, the historical standard deviation of the parameter, the environmental condition change rate, the sampling time interval, and the system working mode. The output is a parameter validity flag vector, which is used to construct the first variation amplitude matrix of the photovoltaic inverter.

9. A computer-readable storage medium, characterized in that, Program instructions are stored in the computer-readable storage medium. When the program instructions run on a computer, they are used to execute the fault warning method of a photovoltaic power station inverter device according to any one of claims 1-8.

10. A fault warning system for an inverter device of a photovoltaic power station, characterized in that, The system includes the computer-readable storage medium according to claim 9. The system can be any one of a computer, a server, and a single-chip microcomputer. The computer-readable storage medium is set inside the system, and a microprocessor for executing the program instructions stored in the computer-readable storage medium is set inside the system.

Images