Hierarchical photovoltaic string anomaly detection method and visual analysis system based on dimension reduction mode

Abstract

The application discloses a kind of hierarchical photovoltaic group string anomaly detection methods based on dimension reduction mode, compared with existing detection method, this method is based on time series electrical quantity data and weather data as sample data, combined with dimension reduction mode and hierarchical model training architecture, with weather label time series electrical quantity data dimension reduction mode chart as the input of anomaly detection model, reduce the model input cost and improve model prediction accuracy, can effectively deal with the prediction scene of real photovoltaic power plant dynamics;The application also provides a visual analysis system based on photovoltaic group string anomaly detection method, supports the group string anomaly recognition result of anomaly detection model to be shown and analyzed, through the analysis and simple annotation optimization anomaly detection model of abnormal result, help user to quickly locate abnormal group string and formulate reasonable solution through interactive operation, realize the effective investigation of photovoltaic group string fault.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power generation, and particularly relates to a hierarchical photovoltaic string anomaly detection method and visual analysis system based on dimension reduction mode. BACKGROUND

[0002] With the gradual depletion of fossil fuels and the intensification of environmental pollution problems, the use of renewable energy has become particularly important. As one of the core of renewable energy, solar energy has occupied an increasingly important position in the global energy structure due to its clean, sustainable and economic advantages. Photovoltaic power generation technology, as a key means to efficiently convert solar energy into electrical energy, is self-evident in its importance. However, with the large number of photovoltaic power stations being put into production, the core equipment photovoltaic string frequently appears abnormal. The abnormality in the photovoltaic system not only reduces the power generation efficiency, but also may bring safety hazards such as fire. At the same time, with the rapid increase of installed capacity, the traditional way of relying on manual inspection of equipment one by one has increasingly exposed the problem of labor shortage. Therefore, the development of intelligent photovoltaic string anomaly detection technology has become the current key task.

[0003] In recent years, in order to improve the performance of photovoltaic power stations, researchers have proposed many anomaly detection methods for photovoltaic strings. For example, reference 1 (Detection and analysis of hot-spot formation in solar cells) uses infrared thermal imaging technology to draw the surface temperature distribution map of solar cells under reverse bias, and proposes a fault diagnosis method based on infrared image analysis. Reference 2 (Online Fault Detection in PV Systems) studies the operating state of photovoltaic strings under different light intensity conditions, simulates by selecting the correct photovoltaic string surface temperature and irradiance, and compares the difference between the theoretical normal value and the actual measured value to confirm whether the anomaly occurs.

[0004] However, the actual photovoltaic power station system anomaly discrimination and accuracy greatly depend on the threshold set based on expert knowledge, and the detection of anomalies has a certain randomness. As disclosed in the application with the publication number CN114123971A, a photovoltaic string anomaly detection method and system based on VaDE are disclosed. Specifically, the current mean value and voltage mean value of each photovoltaic string in the photovoltaic power station within a preset time length in the detection period are obtained, and the installed capacity of each photovoltaic string is determined based on these data. Each first data index of each photovoltaic string within a preset time length in the detection period is input into a pre-established photovoltaic string anomaly detection model, and each Gaussian distribution weight corresponding to each photovoltaic string is obtained. Based on each Gaussian distribution weight, the abnormal photovoltaic string in the detection period of the photovoltaic power station is determined. Although this method improves the accuracy of anomaly detection, it relies on high-quality labeled data, which is costly to obtain in practical applications. In addition, inherent factors such as equipment defects, installation inclination, and transient disturbances such as cloud cover often cause the model to misreport, increasing the processing cost.

[0005] In addition, the combination of artificial intelligence and photovoltaic string anomaly detection improves the accuracy of the model, but due to the black box structure of the model itself, it is not conducive to user understanding of the relationship between input and output. Data visualization analysis systems are widely used in the field of deep learning to visually encode data attributes through visual charts, and to enhance accuracy and explainability by integrating human perception into data analysis through human-computer interaction, which helps users understand model behavior to some extent. Therefore, in the actual operation of photovoltaic power stations, how to combine domain knowledge for photovoltaic string anomaly detection is a problem that needs to be solved. SUMMARY

[0006] The purpose of the present application is to provide a hierarchical photovoltaic string anomaly detection method and visual analysis system based on dimension reduction mode, which combines dimension reduction mode and hierarchical model training architecture, introduces weather clustering information, and constructs an effective detection model for identifying photovoltaic string anomalies, reduces the input cost of the model and the false positive rate of the model, and combines interactive visual analysis operations. Users can intuitively understand how the anomaly detection model identifies abnormal photovoltaic strings and the possible causes of photovoltaic string anomalies, helping users quickly locate anomalies and develop corresponding processing solutions.

[0007] To achieve the above-mentioned purpose of the application, the embodiment provides a hierarchical photovoltaic string anomaly detection method and visual analysis system based on dimension reduction mode, which comprises a hierarchical photovoltaic string anomaly detection method based on dimension reduction mode and a visual analysis system based on the hierarchical photovoltaic string anomaly detection method.

[0008] In one embodiment, the hierarchical photovoltaic string anomaly detection method based on dimension reduction mode comprises the following steps:

[0009] Collecting time-series electrical quantity data and weather data of photovoltaic strings and performing data preprocessing;

[0010] Based on the preprocessed time-series electrical quantity data, a dimension reduction pattern diagram of the time-series electrical quantity data is calculated.

[0011] Based on the preprocessed weather data, clustering is performed to form weather labels, the weather labels are introduced into the dimension reduction pattern diagram of the time-series electrical quantity data, and a dimension reduction pattern diagram of the time-series electrical quantity data with weather labels is obtained.

[0012] Based on the dimension reduction pattern diagram of the time-series electrical quantity data with weather labels, an anomaly detection model is trained in a hierarchical structure, the trained anomaly detection model is used to identify abnormal photovoltaic strings, and anomaly detection results of all photovoltaic strings are obtained.

[0013] In one embodiment, the data preprocessing includes error value processing, missing value filling, and noise filtering.

[0014] The error value processing refers to eliminating error values when the time-series electrical quantity data exceeds the rated value or negative values appear.

[0015] The missing value filling fills a large number of continuous null values in the time-series electrical quantity data and weather data of the photovoltaic string by an interpolation method.

[0016] The noise filtering smoothes the filtering of the time-series electrical quantity data and weather data of the photovoltaic string by a sliding average method, thereby eliminating the interference of weather factors.

[0017] In one embodiment, the calculation of the dimension reduction pattern diagram of the time-series electrical quantity data includes performing a dimension transformation operation on the preprocessed time-series electrical quantity data and calculating the dimension reduction pattern diagram of the time-series electrical quantity data based on UMAP.

[0018] In one embodiment, the clustering of the weather data of the photovoltaic string to form weather labels includes using a KMeans clustering method to cluster and color code the preprocessed weather data to form weather labels, wherein blue represents low irradiance conditions, green represents medium irradiance conditions, and red represents high irradiance conditions.

[0019] In one embodiment, the training of the anomaly detection model in a hierarchical structure includes taking the dimension reduction pattern diagram of the time-series electrical quantity data with weather labels as the input of the anomaly detection model, grouping all the dimension reduction pattern diagrams of the photovoltaic strings according to the corresponding inverters, grouping the dimension reduction pattern diagrams of the photovoltaic strings under each group of inverters based on the weather labels, and training an anomaly detection model for each type of weather label.

[0020] In one embodiment, the anomaly detection model is a combination of three unsupervised models, including one-class support vector machine, ellipse envelope and local outlier factor, and the weighted average of the three anomaly models is taken as the output of the anomaly detection model.

[0021] In one embodiment, the identification of the abnormal photovoltaic string is based on the output results calculated by the anomaly detection model, including: an anomaly value R a and a degradation rate R d .

[0022] The anomaly value is obtained by selecting the anomaly detection model architecture and using the MinMax method to normalize the results, which is used to quantify the abnormality degree of each photovoltaic string, and the anomaly value R a is expressed as:

[0023]

[0024] wherein Out represents the output result of the anomaly detection model of the i-th photovoltaic string, the intermediate variable m = OCSVM|EE|LOF represents that the selected anomaly detection model is one-class support vector machine, local outlier factor and ellipse envelope, n represents the number of photovoltaic strings; Min represents the minimum value of the mean of the output results of the anomaly detection models of the n photovoltaic strings; Max represents the maximum value of the mean of the output results of the anomaly detection models of the n photovoltaic strings.

[0025] The degradation rate is approximately evaluated by quantifying the deviation of each point in the dimensionality reduction pattern diagram relative to the normal data cluster C N , including: based on the method of fitting the boundary of the distribution of the normal dimensionality reduction pattern diagram, using the same training method as the anomaly detection model to identify abnormal photovoltaic strings, selecting Gaussian mixture model as the model structure, and the Gaussian mixture model approximates the probability distribution p(x) of the normal data cluster C N by a weighted sum of multiple Gaussian distributions; when calculating the degradation rate R d , the calculation results of the Gaussian mixture model are normalized to the range of [0, 1] by using the MinMax method; and the degradation rate R d is grouped according to time t, and the minimum likelihood of all normal photovoltaic strings in each group is taken as the boundary, the proportion of the likelihood of all normal photovoltaic strings in the group relative to the minimum likelihood is calculated, and the weighted sum of the proportion of the photovoltaic string at each time point is calculated to calculate the approximate degradation rate R d , and the degradation rate R d is expressed as:

[0026]

[0027] wherein x represents a multi-dimensional data point in the dimensionality reduction pattern diagram, x ∈ C N , and CN represents a normal data cluster in the dimension reduction pattern diagram; represents a likelihood for θ, which is used to estimate the value of θ that can make x known in the dimension reduction pattern diagram, is numerically the same as the probability distribution p(x); θ represents a parameter set of the Gaussian mixture model including a mixing weight, a mean and a covariance matrix; K represents the number of Gaussian distributions, π k represents a mixing weight of the kth Gaussian distribution, represents the kth Gaussian distribution, μ k is a mean of the kth Gaussian distribution, ∑ k is a covariance matrix of the kth Gaussian distribution; MinMax represents a normalization result of the Gaussian mixture model to the range of [0, 1] by using the MinMax method when calculating the degradation rate R d , max represents a maximum value in the range of [0, 1]; R d (t) represents the degradation rate R d grouped by time t, and the time range is [1, n]; represents the likelihood grouped by time t, and the minimum likelihood of all normal photovoltaic strings in each group is represented as

[0028] In order to clearly show the hierarchical photovoltaic string anomaly detection method based on the dimension reduction pattern, a visual analysis system of the hierarchical photovoltaic string anomaly detection method is provided, the photovoltaic power generation amount prediction method is used in visual analysis of the system, and the visual analysis system comprises a global view module, a hierarchical view module, a sorting view module, a pattern view module and an analysis view module.

[0029] The global view module provides a frequency histogram and a rectangular tree diagram for showing the recognition result of the anomaly detection model on the input data, wherein the frequency histogram shows the distribution of all abnormal values of the photovoltaic strings of the photovoltaic power station calculated by the anomaly detection model, and provides an overall overview of the anomaly detection model for identifying abnormal photovoltaic strings; the rectangular tree diagram shows all abnormal values of the photovoltaic strings of the photovoltaic power station based on the hierarchical structure of the transformer-inverter-photovoltaic string of the photovoltaic power station.

[0030] The hierarchical view module is used to show the overall distribution of the dimension reduction pattern diagram of all photovoltaic strings under the inverter and the normal photovoltaic string dimension reduction pattern diagram distribution boundary obtained based on the normal string dimension reduction pattern diagram distribution boundary fitting algorithm in the form of a scatter plot and a contour plot, so as to help users understand the reason why the anomaly detection model identifies the photovoltaic string as abnormal and further optimize the anomaly detection model.

[0031] The sorting view module is used to display the original current data of each photovoltaic string under the selected inverter in the global view and the output result calculated by the anomaly detection model, and provides filtering and sorting functions to help users quickly locate the photovoltaic string of interest;

[0032] The mode view module is used to display the original time series electrical quantity data and the dimension reduction mode chart of the photovoltaic string selected in the sorting view by providing a line chart and a scatter chart, to help users mine the similarities and differences of normal and abnormal photovoltaic strings in the dimension reduction mode chart, and provide reasonable explanation for the identification result of the anomaly detection model;

[0033] The analysis view module is used to provide related abnormal characteristic analysis indicators of the specified photovoltaic string and simple data labeling function to assist users in analyzing the causes of photovoltaic string anomaly.

[0034] In one embodiment, the normal photovoltaic string dimension reduction mode chart distribution boundary is fitted based on the normal string dimension reduction mode chart distribution boundary, comprising:

[0035] The normal photovoltaic string identified by the anomaly detection model is trained to obtain a Gaussian mixture model contour chart for depicting the boundary of the normal photovoltaic string distribution, wherein the contour line boundary value k of the contour chart is calculated by the k-sigma method, the contour line is calculated based on the photovoltaic string μ-kσ, and the value range of k is [1, 5];

[0036] The further optimization of the anomaly detection model is performed by adjusting the contour line boundary value k of the contour chart and the number of sub-distributions in the Gaussian mixture model, respectively.

[0037] In one embodiment, the related abnormal characteristic analysis indicators of the specified photovoltaic string are provided, comprising: providing normal rate R n , relative power generation rate R rpg and irradiation correlation coefficient C Irr indicators and visualizing the indicators by radar chart to facilitate the estimation of the values of the indicators;

[0038] The normal rate R n represents the proportion of normal photovoltaic strings in the dimension reduction mode chart;

[0039] The relative power generation rate R rpg is the ratio of the power generation P of the photovoltaic string S under the same inverter and the same day to the optimal photovoltaic string S max , and is expressed as:

[0040]

[0041] Wherein, N represents the number of days, P s，d represents the power generation of the photovoltaic string S under the same inverter and the dth day, and Pmax，d represents the optimal photovoltaic string S max The power generation under the same inverter and on the dth day;

[0042] The irradiance correlation coefficient C Irr is the Pearson correlation coefficient between the time-series current data I and the irradiance data R of the photovoltaic string, which ranges from [0, 1], and is represented as:

[0043]

[0044] For a normal photovoltaic string S, the current data I is usually proportional to the irradiance data R, and the irradiance correlation coefficient C Irr The closer to 1 indicates that the photovoltaic string is more normal, wherein N represents time, I s，t represents the time-series current data of the normal photovoltaic string S at the moment t, represents the time-series current data mean of the normal photovoltaic string S, R t represents the irradiance data of the normal photovoltaic string S at the moment t, represents the irradiance data mean of the normal photovoltaic string S, and max represents selecting the maximum value in the range [0, 1] as the Pearson correlation coefficient;

[0045] The radar chart is used to visualize the indicators, including setting the center of the radar chart to represent the indicator value as 0, and the outermost vertex to represent the value as 1, the smaller the area of the radar chart, the more abnormal the photovoltaic string, and the radar chart uses uniform color coding to represent the specified photovoltaic string under low irradiance, medium irradiance and high irradiance conditions. The related abnormal characteristics.

[0046] Compared with the prior art, the present application has at least the following beneficial effects:

[0047] The hierarchical photovoltaic string anomaly detection method based on the dimension reduction mode provided by the present application uses time-series electrical quantity data and weather data as sample data, combines a dimension reduction mode and a hierarchical model training architecture, uses a dimension reduction mode graph of time-series electrical quantity data with weather labels as an input of an anomaly detection model, reduces the model input cost and improves the model prediction accuracy, and can effectively cope with real photovoltaic power plant dynamic prediction scenarios. The present application also provides a visual analysis system based on the photovoltaic string anomaly detection method, which supports displaying and analyzing the string anomaly recognition results of the anomaly detection model, optimizes the anomaly detection model through analysis and simple labeling of the abnormal results, helps users quickly locate abnormal strings through interactive operation and formulates reasonable solutions, and realizes effective troubleshooting of photovoltaic string faults. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced.

[0049] Figure 1 A flow chart of the photovoltaic string anomaly detection method of the present application;

[0050] Figure 2 A schematic diagram of the photovoltaic string anomaly detection method of the present application;

[0051] Figure 3 A schematic diagram of the system normal string dimension reduction mode chart distribution boundary fitting algorithm;

[0052] Figure 4 A schematic diagram of the visual analysis system of the present application;

[0053] Figure 5 A schematic diagram of the global view module;

[0054] Figure 6 A schematic diagram of the hierarchical view module;

[0055] Figure 7 A schematic diagram of the sorting view module;

[0056] Figure 8 A schematic diagram of the pattern view module;

[0057] Figure 9 A schematic diagram of the analysis view module. DETAILED DESCRIPTION

[0058] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the protection scope of the present application.

[0059] In order to effectively identify abnormal photovoltaic strings, the embodiment provides a hierarchical photovoltaic string anomaly detection method based on dimension reduction mode. The anomaly detection method combines dimension reduction mode and hierarchical model training architecture, introduces weather clustering information, constructs a detection model for effectively identifying photovoltaic string anomalies, and reduces the input cost of the model and the false alarm rate of the model. As shown in Figure 1 The photovoltaic string anomaly detection method provided by the embodiment includes the following steps:

[0060] S1, collecting time series electrical quantity data and weather data of photovoltaic strings and performing data preprocessing.

[0061] In this embodiment, the raw data consists of time-series electrical quantity data and weather data obtained from the photovoltaic power plant's photovoltaic strings. However, the raw dataset contains a large amount of abnormal data caused by equipment and communication failures. Therefore, an appropriate data preprocessing procedure needs to be designed, including:

[0062] (1) Error handling: Due to defects in the sampling equipment, the obtained time-series electrical quantity data and weather data are incorrect and do not conform to the real conditions. For example, there are current values ​​under the condition of no light at night, current values ​​exceed the rated current value or negative values ​​appear. These errors need to be eliminated.

[0063] (2) Missing value imputation: Due to equipment communication failure, the time-series electrical quantity data and weather data obtained were lost during transmission, resulting in a large number of continuous missing values ​​in the original dataset, which seriously affected the training and verification of the subsequent anomaly detection model. It is necessary to process them by linear or nonlinear interpolation methods.

[0064] (3) Noise filtering: Factors such as the stability of equipment communication and cloud cover can interfere with the raw data collected, which is reflected in the time-series electrical quantity data as noise or spikes. The moving average method is used to filter and smooth the time-series electrical quantity data, thereby eliminating interference.

[0065] Based on the above data preprocessing process, preprocessed time-series electrical quantity data and weather data are obtained.

[0066] S2. Based on the preprocessed time-series electrical quantity data, the dimensionality reduction model diagram of the time-series electrical quantity data is calculated.

[0067] In the embodiments, such as Figure 2 As shown in Figure A, the dimensionality reduction model of the preprocessed time-series electrical quantity data is obtained by performing a dimensionality transformation operation and calculating based on UMAP. Specifically:

[0068] First, the preprocessed time-series electrical quantity data is downsampled. The time granularity of the preprocessed time-series electrical quantity data is 1 minute. In order to reduce the computational cost without reducing the accuracy, the time granularity is converted to 1 hour by downsampling. The aggregation method adopts the method of calculating the mean. Assuming the total length of the entire time series is n, and the number of features used is 1 (mainly focusing on the real-time current of the photovoltaic string), the shape of the input vector after downsampling changes from (n, 1) to (n / 60, 1). Next is the dimensionality transformation. In the previous downsampling step, the number of points corresponding to one day becomes 24. Therefore, each 24 points are converted into a row, and the shape of the input vector changes from (n / 60, 1) to (n / 1440, 24). Each row thus represents the electrical quantity data of the photovoltaic string for one day. Finally, UMAP is used to convert the dimensionality-transformed data into a pattern diagram. By converting the 24-dimensional vector into a 2-dimensional vector, the shape of the input vector changes from (n / 1440, 24) to (n / 1440, 2). Projecting this onto a 2D plane yields a dimensionality-reduced pattern diagram of the time-series electrical quantity data. Each point in the dimensionality-reduced pattern diagram represents the power generation status of the photovoltaic string on a particular day.

[0069] S3. Based on the preprocessed weather data, cluster the data to form weather labels, and then incorporate the weather labels into the dimensionality reduction model of the time-series electrical quantity data to obtain the dimensionality reduction model of the time-series electrical quantity data with weather labels.

[0070] In the embodiments, such as Figure 2 As shown in B, the preprocessed weather data is clustered to form weather labels, and these weather labels are mapped back to the dimensionality-reduced pattern diagram of the time-series electrical quantity data in S2, thus completing the introduction of weather labels. Specifically:

[0071] First, the preprocessed weather data is clustered. Assuming the total length of the entire time series is n and the number of features used is 1 (mainly focusing on power plant irradiance data), the shape of the input vector is (n, 1). The irradiance data collection frequency is also 1 minute, so a downsampling method is used to convert the time granularity to the hourly level, and the vector shape becomes (n / 60, 1), reducing the computational load of the anomaly detection model and the data fluctuations caused by interference such as communication failures. Next, the irradiance data is divided by day, and the shape of a single vector becomes (24, 1) as the object of the K-means algorithm for clustering. Irradiance data sequences with similar weather conditions are grouped into the same class, resulting in three clusters representing low, medium, and high irradiance. These are encoded with blue (#7883BB), green (#86BB78), and red (#DE7676) respectively to obtain weather labels. The weather labels are mapped back to the dimensionality reduction pattern diagram of the time series electrical quantity data to obtain the dimensionality reduction pattern diagram of the time series electrical quantity data with weather labels, which is used as the input of the anomaly detection model.

[0072] S4, Dimension reduction pattern map based on time series electrical quantity data with weather label, train anomaly detection model in hierarchical structure, identify abnormal photovoltaic string, get anomaly detection result of all photovoltaic strings.

[0073] In the embodiment, as shown in C in the figure, Figure 2 When modeling the dimension reduction pattern map of all photovoltaic strings under a single inverter, since different weather conditions such as irradiance have a great influence on the distribution of string data points in the dimension reduction pattern map, there is a great change in the spatial distance between high irradiance and low irradiance points, and the outlier characteristics of abnormal photovoltaic string data points in space are easy to confuse the recognition ability of the model, therefore, the weather clustering label is introduced to group the aggregated dimension reduction pattern map of the inverter level, and an anomaly detection model is trained for each weather condition, which excludes the data distribution deviation caused by irradiance factors and further optimizes the performance of the model, specifically:

[0074] Firstly, the dimension reduction pattern map of time series electrical quantity data is taken as the input of the anomaly detection model, the hierarchical idea is introduced, and the photovoltaic strings are grouped into Inv1, Inv2,..., Invn according to the corresponding inverters. n Where n represents the total number of inverters in the input data set, Inv i represents the dimension reduction pattern map of time series electrical quantity data of all photovoltaic strings under inverter i; secondly, all dimension reduction pattern maps in each Inv i are combined together, and each point in the dimension reduction pattern map is grouped according to the corresponding weather label, and each Inv i is further divided into Inv i，low , Inv i，med , Inv i，high , which correspond to low, medium and high irradiance conditions respectively; finally, an anomaly detection model is trained for each Inv i,label (i=1,2,..., n; label=low|med|high) after two-stage grouping.

[0075] In the embodiment, a composite model is selected for anomaly detection, and the structure of the composite model can fully utilize the advantages of each model in different data distributions. For example, One Class SVM (single-class support vector machine) is a probability distribution model that does not depend on data, is suitable for data sets with unknown or complex distributions, and the introduction of kernel techniques enables it to capture nonlinear features of data and improve detection accuracy. EllipticEnvelope (elliptical envelope) uses a covariance matrix to comprehensively consider the correlation between dimensions, is simple and efficient to calculate, and is suitable for large-scale data sets that meet the multivariate normal distribution. Local Outlier Factor (local anomaly factor) evaluates the degree of anomaly of a data point by comparing the density of its local field, does not depend on the global data distribution, and can handle diverse density distribution patterns. Based on the idea of ensemble learning, One Class SVM, EllipticEnvelope, and Local Outlier Factor are combined together, and the weighted average of the three models is used as the output of the anomaly detection model.

[0076] The anomaly detection model outputs all outliers in the dimensionality reduction pattern grouped in a hierarchical structure. In order to quantify the degree of anomaly of each photovoltaic string for anomaly detection, the anomaly value R a is defined as:

[0077]

[0078] where Out represents the output result of the anomaly detection model of the ith photovoltaic string, the intermediate variable m = OCSVM|EE|LOF represents that the selected anomaly detection model is One Class SVM, Local Outlier Factor, and EllipticEnvelope, n represents the number of photovoltaic strings; Min represents the minimum value of the mean of the output results of the anomaly detection models of n photovoltaic strings; and Max represents the maximum value of the mean of the output results of the anomaly detection models of n photovoltaic strings.

[0079] The anomaly value of each photovoltaic string is obtained by weighted summation according to the number of corresponding outliers in the dimensionality reduction pattern. The larger the anomaly value, the higher the degree of anomaly of the photovoltaic string. The anomaly threshold arr_thr of the photovoltaic string anomaly detection model is set, and the value of arr_thr is automatically calculated using the common 3σ criterion statistical method. If the anomaly value R a of the photovoltaic string is greater than the anomaly threshold arr_thr, the anomaly detection model identifies it as an abnormal string, and vice versa.

[0080] Beyond identifying whether photovoltaic (PV) strings are abnormal, the model also needs to assess the degree of efficiency loss relative to normal PV strings. This is because faults in real-world PV power plant scenarios do not necessarily lead to a decline in PV string performance; maintenance personnel primarily focus on PV strings whose power generation performance is reduced due to faults. If the degree of PV string degradation can be quantified, maintenance personnel can use this indicator to prioritize fault repairs. Therefore, the model proposes to approximate the degradation rate of PV strings by measuring the deviation of each point in the dimensionality-reduced model diagram from the normal data cluster.

[0081] In this embodiment, to calculate the deviation of data points in the dimensionality-reduced data graph of the photovoltaic string, a degradation rate is proposed for evaluation. The degradation rate is determined by quantifying the deviation of each point in the dimensionality-reduced model graph relative to the normal data cluster C. N The degree of deviation is used for approximate assessment, including: such as Figure 3 The method shown is based on fitting the distribution boundary of the normal dimensionality reduction pattern graph. It adopts the same training method as the anomaly detection model for identifying abnormal photovoltaic strings. It groups the data according to inverter and weather labels, and uses normal photovoltaic strings filtered by the threshold arr_thr as input. It selects a Gaussian mixture model as the model structure. The Gaussian mixture model (GMM) approximates the normal data cluster C by weighted sum of multiple Gaussian distributions. N Given the probability distribution p(x), the fitted GMM model can calculate the likelihood of data points in any dimensionality-reduced pattern diagram. High likelihood This indicates that the point belongs to normal data cluster C. N The probability is higher if it is, and vice versa;

[0082] In calculating the degradation rate R d The MinMax method was used to normalize the calculation results of the Gaussian mixture model to the range [0, 1]; and the degradation rate R was set as follows. d Grouping by time t, the minimum likelihood of all normal photovoltaic strings in each group is used as the boundary. The proportion of the likelihood of all normal photovoltaic strings in the group relative to the minimum likelihood is calculated. The proportions of the photovoltaic strings at each time point are weighted and summed to calculate the approximate degradation rate R. d Deterioration rate R d Represented as:

[0083]

[0084] Where x represents a multidimensional data point in the dimensionality reduction pattern diagram, x∈C N C N This represents a normal data cluster in a dimension reduction pattern diagram; This represents the likelihood with parameter θ, used to estimate the value that makes x equal to θ, given a multidimensional data point x in a reduced-dimensional model. numerically identical to the probability distribution p(x); represents the parameter set of the Gaussian mixture model including the mixing weights, means and covariance matrices; K represents the number of Gaussian distributions, and represents the mixing weight of the kth Gaussian distribution, k represents the kth Gaussian distribution, and represents the mean of the kth Gaussian distribution, k k represents the covariance matrix of the kth Gaussian distribution; MinMax represents the normalization of the calculation result of the Gaussian mixture model to the range of [0, 1] when calculating the degradation rate R d , max represents the maximum value in the range of [0, 1]; R d (t) represents the degradation rate R d grouped by time t, and the time range is [1, n]; represents the likelihood grouped by time t, and the minimum likelihood of all normal photovoltaic strings in each group is represented as

[0085] The main challenge of current photovoltaic power station operation and maintenance is that inefficient photovoltaic strings are difficult to identify. Existing research shows that machine learning and deep learning technologies have been applied to photovoltaic string anomaly detection to help identify faulty strings for maintenance personnel to repair. However, most detection methods rely on high-quality labeled data, which is costly to obtain in practical applications. In addition, inherent factors such as equipment defects, installation inclination, and transient disturbances such as cloud cover often cause the model to produce false positives, increasing processing costs. At the same time, the black box nature of the model makes the output results lack of explainability, and users are difficult to understand the influence of different input features, so as to draw effective analysis conclusions. Based on the hierarchical photovoltaic string anomaly detection method of the dimension reduction mode, the invention also provides a visual analysis system, which combines photovoltaic string anomaly detection method with visualization for effectively identifying inefficient photovoltaic strings, guiding users to quickly explore photovoltaic string patterns, improving the explainability of the model and accurately understanding the model results. Based on the visual analysis system, human perception and expert experience are combined to quickly generate valuable labeled data for developing reasonable solutions to effectively troubleshoot photovoltaic string faults. As Figure 4 shown, the visual analysis system provided by the embodiment includes a global view module, a hierarchical view module, a sorting view module, a pattern view module, and an analysis view module.

[0086] In the embodiment, as Figure 5 ​​As shown in the figure, the global view module provides a frequency histogram and a rectangular tree diagram for displaying the identification results of the photovoltaic string anomaly detection model on the input data set, and provides the actual distribution of abnormal photovoltaic strings from the perspective of abnormal values and photovoltaic power station structure. After the user selects an inverter of interest for analysis, the relevant visual data content is displayed in the hierarchical view and the ranking view.

[0087] The frequency distribution histogram shows the distribution of all abnormal values of photovoltaic strings in the photovoltaic power station calculated by the anomaly detection model. The horizontal coordinate represents the abnormal value, and the vertical coordinate represents the number of photovoltaic strings in the corresponding abnormal value range. The red vertical line indicates the photovoltaic string anomaly threshold value automatically calculated by the 3σ criterion statistical method, providing an overall overview of the anomaly identification by the model. The rectangular tree diagram hierarchically displays the abnormal values of all photovoltaic strings in the photovoltaic power station.

[0088] In the initial state, the rectangular unit in the rectangular tree diagram represents the box transformer in the photovoltaic power station. The upper left of each rectangular unit is marked with the box transformer number (such as BT001). The color coding of the rectangle represents the abnormal value, and the deeper the color, the larger the abnormal value. When a box transformer is clicked, the rectangular tree diagram displays the abnormal values of the next level of inverters, and the user can further understand the abnormal distribution of different levels.

[0089] In addition, the global view module also provides a slider function for adjusting the abnormal threshold value. The user can adjust the threshold value accurately by dragging the slider left and right or clicking the button. After adjustment, the data in the rectangular tree diagram and other views associated with it will be updated synchronously, helping the user quickly locate the abnormal string of interest; at the same time, by clicking the rectangular unit, the user can further analyze the running history of the string in other associated views. After clicking, the rectangular unit will be highlighted, and the color coding will remain consistent, improving the interaction experience and the intuitiveness of information visualization.

[0090] As shown in the figure, Figure 6 The hierarchical view module is used to display the overall distribution of the dimension reduction mode diagram of all photovoltaic strings under the inverter and the normal photovoltaic string dimension reduction mode diagram distribution boundary obtained based on the normal string dimension reduction mode diagram distribution boundary fitting algorithm in the form of scatter plot and contour plot, helping the user understand the reason why the anomaly detection model identifies the photovoltaic string as abnormal and further optimize the anomaly detection model.

[0091] After selecting the inverter number to be analyzed, the visualization analysis system will obtain the dimensionality reduction model diagrams of all photovoltaic strings under the selected inverter and merge them into a scatter plot for display. To ensure the comparability between different inverters, the x-axis and y-axis ranges of the scatter plot are uniformly set to the global maximum and minimum values ​​of the dimensionality reduction model diagrams of all photovoltaic strings in the photovoltaic power plant, plus or minus 5% of the range. In this way, when switching between different inverters, abnormal patterns can be easily detected, facilitating further analysis.

[0092] Based on the selected inverter, each point in the scatter plot is color-coded according to the weather conditions corresponding to the photovoltaic strings. The visualization analysis system divides the weather labels into three categories: blue (#7883BB) for low irradiance, green (#86BB78) for medium irradiance, and red (#DE7676) for high irradiance. This weather coding method helps users distinguish the distribution of strings under different weather conditions and understand the impact of weather on the dimensionality reduction model. Furthermore, the scatter plot supports zooming and panning, allowing users to conduct in-depth analysis of overall and local features through these interactive functions. When the mouse hovers over a point, detailed attribute information for that point is displayed, including the photovoltaic string number, time, and outliers.

[0093] To further enhance the expressiveness of the dimensionality reduction model of the photovoltaic strings, a contour map has also been added to the system. The contour map is generated using a Gaussian mixture model trained on normal photovoltaic strings identified by the anomaly detection model, and is used to depict the boundaries of the normal photovoltaic string distribution. The contour line boundary value k is calculated using the k-sigma method, with a default value of 3, indicating that the contour line is calculated based on the μ-kσ of the photovoltaic strings.

[0094] In the hierarchy view module, users can adjust the contour plot parameters using two sliders above the scatter plot. The first slider, "boundary," controls the value of k in the contour plot boundary values, ranging from [1, 5]. The second slider, "n_components," adjusts the number of sub-distributions in the GMM model, ranging from [1, m], where m is the total number of strings under this inverter. By adjusting these two sliders, users can further optimize the model's fitting effect, making it more accurately reflect the normal distribution of photovoltaic strings.

[0095] In the embodiments, such as Figure 7 As shown, the sorting view module is used to display the raw current data of each photovoltaic string under the selected inverter in the global view and the output results calculated by the anomaly detection model. It also provides filtering and sorting functions to help users quickly locate the photovoltaic strings of interest.

[0096] The sorting view module is displayed in a table format, from left to right: PV string number (Pid), current status, outlier (Ra), and degradation rate (Rd). The Pid column represents the string number under the inverter. For example, if the inverter is BT001-I001 and the Pid is PV1, then the PV string number format is BT001-I001-PV1. The second column, current status, displays the time-series current data of the PV strings, shown as a line thumbnail. The x-axis represents time, covering the period from December 9, 2022 to March 26, 2023, and the y-axis represents the current value, ranging from the minimum to the maximum current of all PV strings in the PV power plant, with a 5% range added to ensure visual contrast. R a and R d The columns display outliers and degradation rates for the photovoltaic strings, both ranging from [0, 1], presented as horizontal bar charts. The bar length represents the magnitude of the outlier and degradation rate. Users can sort the data by Pid, ​​R, etc. using the header's sorting function. a Or R d Sort the data in ascending or descending order, and drag the current timing graph of the selected photovoltaic string to the mode view for further analysis. When the amount of data exceeds the page length, users can use the pagination function at the bottom to switch between viewing the results of all photovoltaic strings.

[0097] In the embodiments, such as Figure 8 As shown, the pattern view module is used to display the original time-series electrical quantity data and dimensionality reduction pattern diagram of the photovoltaic strings selected in the sorted view by providing line charts and scatter plots. This helps users to explore the similarities and differences between normal and abnormal photovoltaic strings in the dimensionality reduction pattern diagram and provides a reasonable explanation for the identification results of the anomaly detection model.

[0098] The pattern view module consists of line charts and scatter plots. The line chart, displayed by dragging and dropping, shows the time-series current data of the photovoltaic string selected from the sorted view module. The x-axis and y-axis settings are the same as the thumbnail line chart in the sorted view module. Figure 1 The line graph in the mode view module contains two time-series current curves. The photovoltaic string coded #c77272 is the analysis string, and the photovoltaic string coded #4682b4 is the reference string. By comparing the current deviation between the two time-series current curves within the same time period, users can analyze R... a Or R d The line chart provides horizontal zoom and drag functionality, allowing users to zoom in on specific areas for in-depth analysis, based on the major reasons and domain knowledge used to explain the results.

[0099] like Figure 8As shown, the scatter plot in the model view module consists of a reduced-dimensionality plot of the analysis string on the left and a reduced-dimensionality plot of the reference string on the right. The upper left corner of both the analysis string and reference string reduced-dimensionality plots displays the photovoltaic string number, in a format such as BT001-I001-PV1, with color coding consistent with the line chart in the model view module. The reduced-dimensionality plot uses the same representation as in the hierarchical view module, displaying a scatter plot based on weather tag clustering and color coding, along with the outline of the corresponding normal photovoltaic string point distribution. However, the model view module only displays the scatter plots of the selected photovoltaic string, not all scatter plots under the entire inverter. To highlight abnormal patterns, abnormal scatter plots identified by the model are represented by star shapes, and their size is larger than that of normal photovoltaic string points. Each reduced-dimensionality plot also provides an association mapping function in the upper right corner. The first button restores the layout to the default state, and the second button is a selection function, allowing users to select a specific area in the reduced-dimensionality plot; the scatter plots on both sides corresponding to the time period will be highlighted simultaneously. Hovering the mouse over any point displays detailed information about that point, such as time and outliers, while the corresponding time period is highlighted in the line graph of the pattern view module. This design helps users analyze and uncover the potential correlation between abnormal patterns and raw current data, further interpreting the model's identification results.

[0100] In the embodiments, such as Figure 9 As shown, the analysis view module provides relevant abnormal characteristic analysis indicators and simple data annotation functions for specified photovoltaic strings to help users analyze the causes of photovoltaic string anomalies.

[0101] The left side of the analysis view module displays statistical characteristic indicators for the analysis string and reference string, helping users to interpret the causes of abnormal patterns in the photovoltaic string in detail. To reasonably explain these abnormal patterns, three important indicators for results analysis were designed and introduced: normality rate R... n Relative power generation rate R rpg Correlation coefficient C with irradiation Irr Among them, the normal rate R n This represents the proportion of normal photovoltaic strings in the dimension reduction model diagram; the relative power generation rate R rpg For photovoltaic string S under the same inverter and on the same day, relative to the optimal photovoltaic string S max The ratio of electricity generated P to the total electricity generated is expressed as:

[0102]

[0103] Where N represents the number of days, P s，d P represents the power generation of photovoltaic string S under the same inverter on day d. max，d S represents the optimal photovoltaic string. max Power generation on day d under the same inverter;

[0104] Irradiance correlation coefficient C Irr is the Pearson correlation coefficient between the time-series current data I and the irradiance data R of a photovoltaic string S, which ranges from 0 to 1, and is denoted as:

[0105]

[0106] For a normal photovoltaic string S, the current data I is usually proportional to the irradiance data R, and the irradiance correlation coefficient C Irr The closer to 1 indicates that the photovoltaic string is more normal, where N represents time, I s，t denotes the time-series current data of a normal photovoltaic string S at time t, denotes the time-series current data mean of a normal photovoltaic string S, R t denotes the irradiance data of a normal photovoltaic string S at time t, denotes the irradiance data mean of a normal photovoltaic string S, and max denotes selecting the maximum value in the range [0, 1] as the Pearson correlation coefficient.

[0107] Based on the above design of the correlation anomaly characteristic analysis indicators for the specified photovoltaic string, the value range of all feature indicators is ensured to be [0, 1], and they have the same directionality, i.e., the closer to 1, the more normal the photovoltaic string, and vice versa, which helps to enhance the expression effect of these indicators when mapped into a visual form.

[0108] In order to more intuitively display these feature indicators, the system uses a radar chart for visualization. Since three feature indicators are involved, the background of the radar chart is a triangle, which is divided into four equal parts by alternating gray and white colors, facilitating users to quickly estimate the values of each indicator visually. The center of the radar chart represents an indicator value of 0, and the outermost vertex represents a value of 1. The smaller the area of the radar chart, the more abnormal the photovoltaic string. The radar chart is arranged in a 2x3 format, with the top and bottom rows corresponding to the analysis string and the reference string in the mode view, respectively. The color coding is consistent, and each row contains three columns, representing the photovoltaic string characteristics under low, medium, and high irradiance conditions. When the user hovers the mouse over the boundary line of the radar chart, the detailed values of each feature will be displayed.

[0109] The right part of the analysis view module also provides a data labeling interface, where users can correct the incorrect results identified by the anomaly detection model to optimize the subsequent model iteration. The labeling form contains four parts: string ID, anomaly label, inferred reason, and remarks. The string ID is used to display the selected photovoltaic string number; the anomaly label indicates whether the selected photovoltaic string has an anomaly, with the initial value being the identification result of the anomaly detection model, which can be modified by the user according to the actual analysis result; the inferred reason is a multiple-choice option, indicating the low-efficiency reason inferred through time-series current data analysis. The common low-efficiency reasons of photovoltaic strings include the following four:

[0110] 1.0 Current: Current value of the PV string for a long time (e.g. 1 day or 1 week) under sufficient weather irradiance;

[0111] 2. Dust or shading: Due to dust accumulation or shading (e.g. trees, shadows, etc.), the PV string generates power for a long time, and does not change with the weather;

[0112] 3. Internal fault: Part of the PV string components fail (e.g. short circuit, cracking, hot spot, etc.), resulting in reduced power generation, but proportional to weather changes.

[0113] 4. Single port connecting two strings: Due to engineering problems, one inverter interface connects two PV strings, resulting in a ratio of 2:1 between the PV string current and the normal PV string.

[0114] The remarks column is used to add notes on the analysis of the PV string for subsequent training of the expert model. After filling out the form, click the "Label" button to complete the labeling task of the string, and finally generate a PV string dataset containing the labeling results through the visualization analysis system export button.

[0115] In order to better illustrate the technical effects of the present application, real operation data collected from a certain centralized photovoltaic power station are used for anomaly detection. The total installed capacity of the photovoltaic power station is about 25MWp, and the overall architecture adopts the scheme of string inverter, local voltage transformation, and centralized grid connection. The power station uses more than 50,000 high-efficiency monocrystalline components of the same specification, which are built according to the hierarchical structure of component-string-inverter-transformer, and the data set is collected through the supporting photovoltaic power station NCS background monitoring system. The data set includes two parts of time series electrical quantity data and weather data: the time series electrical quantity data includes 8 transformers, 75 inverters and 1350 strings, collects the direct current voltage, three-phase current voltage, active power, reactive power, IGBT temperature, daily power generation and total power generation of the photovoltaic string running at a time granularity of 1min from December 9, 2022 to March 26, 2023, among which the photovoltaic string number format is BT[transformer ID]-I[inverter ID]-PV[string ID]; The weather data is the global irradiance, temperature and wind speed of the entire photovoltaic power station during the same period, which affect the power generation performance of the photovoltaic string, and is sampled at a frequency of 1min. Based on the developed photovoltaic string anomaly detection method and visualization analysis system, the following tasks are performed:

[0116] (1) Explore the low-efficiency strings in the power station;

[0117] After the visualization analysis system loads the collected photovoltaic power plant dataset, it first analyzes the global view module to obtain the anomaly detection model's identification results of the inefficient state of the entire photovoltaic power plant's strings. For example... Figure 4 As shown in A1, the distribution of outliers in the photovoltaic strings displayed by the frequency histogram in the global view module reveals that the outliers for the vast majority of photovoltaic strings are around 0.1. The anomaly detection model automatically calculates an anomaly threshold of 0.20 based on the 3σ criterion statistical method. Observing the position of the red threshold line in the histogram confirms that the threshold effectively distinguishes between anomalies and non-anomalies in the photovoltaic strings. Next, as... Figure 4 As shown in A2, the distribution of abnormal photovoltaic strings in physical location is explored through the rectangular tree diagram of the global view module. Referring to the mapping relationship between color and the degree of abnormality of photovoltaic strings in the reference scale, the darker the color of the rectangular unit, the more abnormal the corresponding transformer box is. It is found that the transformer box boxes BT002, BT004, and BT005 have more serious abnormalities. Clicking to enter the next level, it is found that the inverters BT002-I009 and BT002-I012 under BT002 transformer box box, the inverters BT004-I004 under BT004 transformer box box, and the inverters BT005-I017 under BT005 transformer box box have more serious abnormalities.

[0118] Furthermore, in order to understand the current situation of the photovoltaic string under inverters with more severe abnormal conditions, such as... Figure 4 As shown in A2, after selecting a specific inverter, such as the darkest one BT005-I017, the real-time current status of all photovoltaic strings from PV1 to PV18 under the inverter is displayed in the sorted view module. Figure 4 (As shown in C). According to the outlier R a Sort in descending order, and find R a Two photovoltaic strings, PV18 and PV13, had anomaly values ​​greater than the anomaly detection threshold of 0.2, with corresponding outlier values ​​of 1 and 0.26. Upon examining the real-time current time series plot, it was found that the current of PV18 was 0, while the overall current amplitude of PV13 was relatively small, clearly indicating an anomaly. This demonstrates that the anomaly detection model can identify inefficient photovoltaic strings in the dataset.

[0119] (2) Mining the features of the dimensionality reduction pattern diagram of the string;

[0120] Select inverter BT005-I017 as the analysis object in the global view module. Figure 4 A2 in the middle, such as Figure 4 As shown in B, the hierarchical view module displays a combination of dimensionality reduction diagrams of all photovoltaic strings in the inverter. Points in the combined diagram are distinguished by different color codes, classifying them as high (…). Figure 4 As shown in B1), in ( Figure 4 As shown in B2), low radiation (Figure 4 As shown in B3, the photovoltaic strings are categorized into three types (low irradiance, medium irradiance, and high irradiance) and their corresponding normal cluster distribution ranges. From the distribution boundaries of each type, the anomaly detection model fitted to the inverter effectively segments and combines the photovoltaic strings under the three weather conditions in two-dimensional space. Vertically, from bottom to top, they are arranged as low irradiance, medium irradiance, and high irradiance. There is some overlap in the distribution boundaries of each part, with a larger overlap between high and medium irradiance. Analyzing the arrangement of the dimensionality reduction points of each type of photovoltaic string reveals that the better the power generation status of the scattered points in the dimensionality reduction pattern diagram, the higher the dimensionality reduction height is directly proportional to the power generation performance of the photovoltaic string on that day. Changing the inverter object yields the same observation results.

[0121] In addition, such as Figure 4 As shown in B, the cross-over in the photovoltaic strings BT005-I017 indicates that the anomaly detection model has errors on the discrimination boundary of weather condition clustering. The overlap between high irradiance and medium irradiance is relatively higher, possibly because the fluctuation range of low irradiance is smaller than that of medium irradiance, making the difference more obvious.

[0122] like Figure 7 As shown, further exploration and analysis are conducted on each photovoltaic string under the photovoltaic string, and the outlier R of the photovoltaic string is sorted in the sorting view module. a To sort in descending order, select R. a The two lowest strings PV16(R) a =0) and PV7(R a =0) serves as the analysis string (left) and reference string (right), respectively. Observation revealed that the distribution of points in the dimensionality reduction pattern diagram of the normal string under the three weather conditions also follows the rule that the height of the dimensionality reduction space is proportional to the photovoltaic string current. The anomaly detection model identifies anomalous star-shaped data points as deviating significantly from their respective data clusters in spatial distance, indirectly confirming the reliability of the anomaly detection model. The analysis string is then modified to R under the inverter. a The highest PV18 has a string current of almost 0, R a =1 indicates that the anomaly detection model considers the photovoltaic string's current performance to be abnormal every day, and in the dimensionality reduction model, all scatter points are almost clustered in the same local area, which is consistent with the actual situation. However, if the analysis string is replaced with another photovoltaic string PV13(R) under the inverter that is judged to be abnormal... a =0.26), such as Figure 4 As shown in D, compare the timing current diagrams of PV13 and PV7 in the comparison mode view module. Figure 4 In D1), it was found that the current data of PV13 was lower than that of the normal photovoltaic string PV7 at some time points, indicating that the overall power generation performance was low. Figure 4 D2 andFigure 4 The reduced dimension pattern figure corresponding to the observation PV13 in D3 is found. The red part (high irradiance) and the green part (medium irradiance) of the photovoltaic string are found to be overall downshifted compared to the right reference string. For example, the high irradiance scatter points appear in the medium irradiance range, and the medium irradiance scatter points appear in the low irradiance range. All the points with severe downshift are identified as abnormal points by the anomaly detection model. According to the rule that the spatial height is proportional to the photovoltaic string current, it is concluded that the photovoltaic string performance under high irradiance and medium irradiance is lower than the normal state, while it remains normal under low irradiance. Mapping back to the time series current figure for comparison and verification, the actual result is consistent with the above inference.

[0123] (3) Comparative analysis of string anomalies

[0124] In the sorting view module, the abnormal value R a is arranged in descending order, and the photovoltaic string with high abnormal value is selected in combination with the time series electrical quantity data, and the radar chart is used for detailed analysis and simple labeling (E in the figure). Figure 4 The following is the specific situation of each type of anomaly in the analysis process:

[0125] (3.1) Long-time zero current. For example, photovoltaic string BT005-I017-PV18. This type of anomaly is characterized by a certain period of time (several days or more) in which the current is 0 or close to 0 under sufficient environmental factors such as irradiance. In the time series current, it is characterized by a certain period of time (several days or more) in which the current is 0 or close to 0 under sufficient environmental factors such as irradiance. In the reduced dimension pattern figure in the pattern view module, the data points are clustered in a small range on both sides of the reduced dimension space, and the radar chart in the analysis view corresponding to it shows a point under the three types of weather conditions, i.e. each index value is 0. When labeling this data, the "long-time zero current" option is checked.

[0126] (3.2) Dust or shading. For example, BT002-I008-PV2. This kind of anomaly is analyzed by analyzing the radar chart on the left side of the view module, and it is found that the irradiance correlation coefficient of this low-efficiency photovoltaic string is reduced, such as the irradiance correlation coefficient of BT002-I008-PV2 under low, medium and high weather conditions is 0.94, 0.75 and 0.73 respectively, indicating that the real-time current value of the photovoltaic string is reduced by the influence of weather irradiance, which may be due to the accumulation of grass, buildings, dust and excrement, etc. reduce the weather irradiance that the photovoltaic string contacts, and the stronger the weather irradiance, the greater the impact, but there is still an approximately proportional relationship, which is manifested in the dimension reduction mode chart as a large number of downshift of the scatter points under medium and high irradiance and is identified as an anomaly by the anomaly detection model. In addition, the relative power generation rate of the photovoltaic string is 0.47, 0.33 and 0.2 respectively, indicating that the current value of the photovoltaic string is lower than that of the normal photovoltaic string, and the stronger the weather irradiance, the higher the current value of the photovoltaic string. The results of the time series current chart in the mode view module indeed show that the overall amplitude of the current is reduced and the deviation of the high irradiance part is larger. When labeling this kind of data, check the "dust or shading" option.

[0127] (3.3) Internal defects. For example, BT001-I008-PV1. Compared with the second type of anomaly, the causes of this type of anomaly include diode breakdown, surface cracking or hot spot, etc., which can cause part of the components of the photovoltaic string to fail to work normally, reducing the current amplitude of the photovoltaic string and seriously affecting the correlation between current and irradiance. The irradiance correlation coefficients of the string BT001-I008-PV1 are 0.95, 0.73 and 0.41 respectively, which are also reduced and the stronger the irradiance, the more the reduction, which is due to internal defects causing part of the components of the photovoltaic string to fail to work normally, even if they are exposed to light, they cannot generate current, thereby reducing the impact of irradiance on the current value of the photovoltaic string, and it is more prominent under high irradiance conditions. Similarly, the relative power generation rate is reduced to 0.58, 0.45 and 0.33, and there are a large number of downshifts in the medium and high irradiation points of the dimension reduction mode chart and are identified as anomalies. Unlike the second type of time series current value, which tends to a lower value regardless of the irradiance, the current of the internal defect string is proportionally reduced, because the internal defect causes part of the components of the photovoltaic string to fail to work, which is equivalent to a photovoltaic string with reduced component number generating power. When labeling this kind of data, check the "internal defect" option.

[0128] (3.4) Single port connected to two strings. For example, BT004-I003-PV18. This kind of anomaly is more special, which is found in the field research. The reason is that the photovoltaic string is limited by the actual environment during installation, resulting in that a single interface in the inverter is connected to two strings. This kind of photovoltaic string is 2:1 in the timing current compared with the normal photovoltaic string, and all the scatter points in the whole dimension reduction mode diagram appear to move up. In addition, the relative power generation rate of this kind of photovoltaic string in the mode view module is close to 1, while that of the normal photovoltaic string is close to 0.5. When labeling this kind of data, the reason is checked in the "single port connected to two strings" option.

[0129] The above specific embodiments have described the technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the most preferred embodiment of the present application, and is not intended to limit the present application. Any modifications, supplements, and equivalent replacements made within the principle range of the present application shall be included in the protection scope of the present application.

Claims

1. A hierarchical photovoltaic string anomaly detection method based on dimensionality reduction mode, characterized in that, Includes the following steps: Collect time-series electrical quantity data and weather data of photovoltaic strings and perform data preprocessing; Based on the preprocessed time-series electrical quantity data, a dimension reduction model diagram of the time-series electrical quantity data is calculated. Weather labels are formed by clustering the preprocessed weather data. The weather labels are then incorporated into the dimensionality reduction pattern diagram of the time-series electrical quantity data to obtain the dimensionality reduction pattern diagram of the time-series electrical quantity data with weather labels. Based on the dimensionality reduction pattern diagram of time-series electrical quantity data with weather labels, where each point in the dimensionality reduction pattern diagram represents the power generation status of a photovoltaic string on a certain day, an anomaly detection model is trained using a hierarchical structure. This training includes: using the dimensionality reduction pattern diagram of time-series electrical quantity data with weather labels as input to the anomaly detection model; grouping the dimensionality reduction pattern diagrams of all photovoltaic strings according to their respective inverters; further grouping the dimensionality reduction pattern diagrams of photovoltaic strings under each group of inverters based on weather labels; training a separate anomaly detection model for each type of weather label dimensionality reduction pattern diagram; and using the trained anomaly detection model to identify abnormal photovoltaic strings, thus obtaining the anomaly detection results for all photovoltaic strings.

2. The hierarchical photovoltaic string anomaly detection method according to claim 1, characterized in that, The data preprocessing includes: error value handling, missing value imputation, and noise filtering; The error value handling refers to the process of identifying and eliminating errors when the sequential electrical quantity data exceeds the rated value or becomes negative. The missing value imputation uses interpolation methods to fill in a large number of consecutive missing values ​​in the time-series electrical quantity data and weather data of the photovoltaic string; The noise filtering uses a moving average method to smooth the time-series electrical quantity data and weather data of the photovoltaic string, thereby eliminating the interference of weather factors.

3. The hierarchical photovoltaic string anomaly detection method according to claim 1, characterized in that, The calculation of the dimensionality reduction model diagram of the time-series electrical quantity data includes: performing a dimensionality transformation operation on the preprocessed time-series electrical quantity data and calculating the dimensionality reduction model diagram of the time-series electrical quantity data based on UMAP.

4. The hierarchical photovoltaic string anomaly detection method according to claim 1, characterized in that, The method of clustering weather data based on photovoltaic strings to form weather tags includes: using the KMeans clustering method to cluster and color-encode the preprocessed weather data to form weather tags, where blue represents low irradiance conditions, green represents medium irradiance conditions, and red represents high irradiance conditions.

5. The hierarchical photovoltaic string anomaly detection method according to claim 1, characterized in that, The anomaly detection model combines three unsupervised models: single-class support vector machine, elliptical envelope, and local anomaly factor. The weighted average of the three anomaly models is used as the output of the anomaly detection model.

6. The hierarchical photovoltaic string anomaly detection method according to claim 1, characterized in that, The identification of abnormal photovoltaic strings is based on the output of an anomaly detection model, and the output includes: anomaly value R. a and degradation rate R d ; The outlier selection anomaly detection model architecture is obtained by normalizing the results using the MinMax method, which is used to quantify the anomaly degree of each photovoltaic string. The outlier R is... a Represented as: Where Out represents the output of the anomaly detection model for the i-th photovoltaic string, the intermediate variable m = OCSVM|EE|LOF indicates that the selected anomaly detection model is a single-class support vector machine, local anomaly factor, and elliptical envelope, and n represents the number of photovoltaic strings; Min represents the minimum value of the average output of the anomaly detection models for the n photovoltaic strings; Max represents the maximum value of the average output of the anomaly detection models for the n photovoltaic strings. The degradation rate is quantified by comparing each point in the dimensionality reduction model with the normal data cluster C. N The degree of deviation is used to approximate the assessment, including: a method based on fitting the distribution boundary of the normal dimensionality reduction pattern graph, using the same training method as the anomaly detection model for identifying abnormal photovoltaic strings, and selecting a Gaussian mixture model as the model structure. The Gaussian mixture model approximates the normal data cluster C by weighted sum of multiple Gaussian distributions. N The probability distribution p(x) is used to calculate the degradation rate R. d The MinMax method was used to normalize the calculation results of the Gaussian mixture model to the range [0, 1]; and the degradation rate R was set as follows. d Grouping by time t, the minimum likelihood of all normal photovoltaic strings in each group is used as the boundary. The proportion of the likelihood of all normal photovoltaic strings in the group relative to the minimum likelihood is calculated. The proportions of the photovoltaic strings at each time point are weighted and summed to calculate the approximate degradation rate R. d Deterioration rate R d Represented as: Where x represents a multidimensional data point in the dimensionality reduction pattern diagram, x∈C N C N This represents a normal data cluster in a dimension reduction pattern diagram; This represents the likelihood with parameter θ, used to estimate the value that makes x equal to θ, given a multidimensional data point x in a reduced-dimensional model. Numerically, it is the same as the probability distribution p(x); θ represents the parameter set of the Gaussian mixture model, including the mixture weights, mean, and covariance matrix; K represents the number of Gaussian distributions, π k This represents the mixture weights of the k-th Gaussian distribution. Let μ represent the k-th Gaussian distribution. k Let ∑ be the mean of the k-th Gaussian distribution. k Let be the covariance matrix of the k-th Gaussian distribution; MinMax represents the condition in calculating the degradation rate R. d The MinMax method is used to normalize the calculation results of the Gaussian mixture model to the range [0, 1], where max represents the maximum value within the range [0, 1]. R d (t) represents the degradation rate R. d Grouped by time t, with a time range of [1, n]; This represents the likelihood grouped by time t, and the minimum likelihood of all normal photovoltaic strings in each group is expressed as:

7. A visualization analysis system for a hierarchical photovoltaic string anomaly detection method based on dimensionality reduction mode, characterized in that, The system uses the hierarchical photovoltaic string anomaly detection method based on dimensionality reduction mode as described in any one of claims 1 to 6 in the visualization analysis. The visualization analysis system includes: a global view module, a hierarchical view module, a sorted view module, a pattern view module, and an analysis view module. The global view module provides frequency histograms and tree diagrams to display the anomaly detection model's identification results on the input data. The frequency histogram shows the distribution of outlier values ​​of all photovoltaic strings in the photovoltaic power station calculated by the anomaly detection model, providing an overall overview of the anomaly detection model's identification of outlier photovoltaic strings. The tree diagram displays the outlier values ​​of all photovoltaic strings in the photovoltaic power station based on the hierarchical structure of the photovoltaic power station's transformer-inverter-photovoltaic string. The hierarchical view module is used to display the overall distribution of the dimensionality reduction pattern diagram of all photovoltaic strings under the inverter in the form of scatter plots and contour plots, and to obtain the distribution boundary of the dimensionality reduction pattern diagram of normal photovoltaic strings based on the normal string dimensionality reduction pattern diagram distribution boundary fitting algorithm. This helps users understand the reasons why the anomaly detection model identifies photovoltaic strings as abnormal and further optimize the anomaly detection model. The sorting view module is used to display the original current data and the output results of the anomaly detection model calculation for each photovoltaic string under the selected inverter in the global view, and provides filtering and sorting functions to help users quickly locate the photovoltaic string of interest; The pattern view module is used to display the original time-series electrical quantity data and dimensionality reduction pattern diagram of the photovoltaic strings selected in the sorted view by providing line charts and scatter plots, which helps users to explore the similarities and differences between normal and abnormal photovoltaic strings in the dimensionality reduction pattern diagram, and provides a reasonable explanation for the identification results of the anomaly detection model. The analysis view module provides relevant abnormal characteristic analysis indicators and simple data annotation functions for specified photovoltaic strings, assisting users in analyzing the causes of photovoltaic string anomalies.

8. The visualization analysis system according to claim 7, characterized in that, The method of obtaining the distribution boundary of the normal photovoltaic string dimensionality reduction pattern diagram based on fitting the distribution boundary of the normal string dimensionality reduction pattern diagram includes: The normal photovoltaic strings identified by the anomaly detection model are trained to obtain the Gaussian mixture model to generate a contour map, which is used to depict the boundary of the distribution of normal photovoltaic strings. The contour line boundary value k of the contour map is calculated by the k-sigma method, which means that the contour line is calculated based on the μ-kσ of the photovoltaic string. The value range of k is [1, 5]. The further optimization of the anomaly detection model is achieved by adjusting the contour line boundary value k of the contour map and the number of sub-distributions in the Gaussian mixture model.

9. The visualization analysis system according to claim 7, characterized in that, The aforementioned provision of relevant abnormal characteristic analysis indicators for specified photovoltaic strings includes: providing the normality rate R. n Relative power generation rate R rpg Correlation coefficient C with irradiation Irr The indicators are visualized using radar charts, making it easier to estimate the values ​​of each indicator. The normality rate R n This represents the proportion of normal photovoltaic strings in the dimension reduction diagram; The relative power generation rate R rpg For the photovoltaic string S under the same inverter and on a certain day, relative to the optimal photovoltaic string S max The ratio of electricity generated P to the total electricity generated is expressed as: Where N represents the number of days, P s，d P represents the power generation of photovoltaic string S under the same inverter on day d. max，d S represents the optimal photovoltaic string. max Power generation on day d using the same inverter; The irradiation correlation coefficient C Irr Let I be the Pearson correlation coefficient between the time-series current data I and the irradiance data R of the photovoltaic string, with a range of [0, 1], and be expressed as: For a normal photovoltaic string S, the current data I is usually proportional to the irradiance data R, and the irradiance correlation coefficient C Irr The closer the value is to 1, the more normal the photovoltaic string is, where N represents time and I represents time. s，t This represents the time-series current data of a normal photovoltaic string S at time t. R represents the average time-series current data of a normal photovoltaic string S. t This represents the irradiance data of the normal photovoltaic string S at time t. denoted as the mean irradiance data of a normal photovoltaic string S, and max represents the maximum value selected in the range [0, 1] as the Pearson correlation coefficient; The visualization of indicators via radar charts includes: setting the center of the radar chart to represent an indicator value of 0, the outermost vertex to represent a value of 1, the smaller the area of ​​the radar chart, the more abnormal the photovoltaic string is, and the radar chart uses a uniform color code to represent the relevant abnormal characteristics of a specified photovoltaic string under low, medium and high irradiance conditions.

Images